I Modeling barriers to cost change in solar and nuclear energy technologies by Philip Eash-Gates B.S. Engineering Science, Trinity University (2008) Submitted to the Institute for Data, Systems, and Society in partial fulfillment of the requirements for the degree of Master of Science in Technology and Policy at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY June 2019 C Massachusetts Institute of Technology 2019. All rights reserved. Signature redacted Author Technology and Policy Program May 10, 2019 Signature redacted Certified by Jessika E. Trancik Associate Professor, Institute for Data, Systems and Society Thesis Supervisor Signature redacted Accepted by Noelle Eckley Selin Director, Technology and Policy Program MASSACHUSETTS INsTITUTEl Associate Professor, Institute for Data, Systems and Society and OF TECHNOLOGY Department of Earth, Atmospheric and Planetary Sciences JUN 0 4 2019 LIBRARIES ARCHIVES Modeling barriers to cost change in solar and nuclear energy technologies by Philip Eash-Gates Submitted to the Institute for Data, Systems, and Society on May 10, 2019, in partial fulfillment of the requirements for the degree of Master of Science in Technology and Policy Abstract The cost of photovoltaic systems has declined more rapidly than other electricity production technolo- gies, while nuclear plant costs have risen. Changing costs have contributed to global energy transitions in the past, and our capacity to decarbonize the electricity sector will depend on the cost of low- carbon electricity production technologies like photovoltaic and nuclear energy. Understanding the mechanisms behind historical cost evolution and potential future improvement can inform the design of energy technologies and the policies that advance them. This thesis investigates historical barriers and future opportunities for cost reduction in solar and nuclear power. By developing innovative mathematical and conceptual models, we address the following questions: (1) How can "plug-and-play" design improve costs in photovoltaic systems? (2) What were the sources of cost escalation and overruns in nuclear power plant construction? We address these questions in chapters 2 and 3. Chapter 2 assesses the potential for plug-and-play designs to reduce non-module costs in photovoltaic systems. This work advances use of the design structure matrix for studying cost change in energy technologies by evaluating design factors across multiple systems. We identify the cost components with significant latent potential for improvement-profit, installation labor, overhead, electrical balance of system, and customer acquisition-and show that plug-and-play designs have advantageous effects on their constituent parts. A conventional small-scale photovoltaic project contains nearly 6oo interactions across 30 or more system elements; we show that plug-and-play designs can reduce the number of interactions by two-thirds and elements by half. Several mechanisms are important to the cost change potential of plug-and-play technology: eliminating various project tasks or shifting their responsibility to the consumer removes the associated overhead and profit of installation firms; pre-assembly of system components and standardization of project tasks eliminates installation labor costs; reduction and simplification of BOS electrical components lowers equipment costs; and standardization of system design precludes time-intensive tasks involved in customer acquisition. We compare the advantages of prevailing plug-and-play designs and consider future opportunities for technological innovation and policy advancement. Chapter 3 examines the engineering assumptions underlying many nuclear cost models using histor- ical cost data from the U.S. nuclear industry. We show that expectations for technological improvement may have underestimated factors external to hardware design. By mapping separate cost trajectories for standard plant designs, we find that nth-of-a-kind (NOAK) plants have been more expensive than first-of-a-kind (FOAK) plants, counter to traditional expectations. Indirect costs external to technologi- cal design were responsible for most of the cost rise observed between 1976 and 1987. Decomposition of cost changes in the reactor containment building shows that while safety was a significant factor 3 driving cost increases, non-safety factors were comparably influential. Comparing productivity data from recent U.S. plant construction to industry expectations, we find that material deployment rates are up to thirteen times slower than cost estimating guidelines suggest. We discuss which technologies could potentially lower the impact of external, previously cost-increasing factors, with the support of regulatory changes and R&D. Thesis Supervisor: Jessika E. Trancik Title: Associate Professor, Institute for Data, Systems and Society 4 Acknowledgments I am deeply grateful to my community of mentors, colleagues, friends, and family for their support and inspiration throughout the last two years. I would like to thank my advisor, Jessika, for her mentorship and encouragement of my academic endeavors. Thank you to my research colleagues for your wisdom and guidance, especially Goksin Kavlak, James McNerney, Micah Ziegler, and Ajinkya Kamat. With gratitude for the vibrant community within the Technology and Policy Program and the Institute for Data, Systems, and Society I offer my appreciation to Barb DeLaBarre, Frank Field, Ed Ballo, Noelle Selin, Jeremy Rossen, Fran Marrone, Kim Strampel, Laura Dorson, and Beth Milnes. To my dear friends and classmates, thank you for adding joy to my time in TPP. My family has always been a well of strength and love; thanks Charlotte, Matt, Merrill, Mom, and Dad. Alex, without fail you accompany me through thick and thin-may it be true that "the Road goes ever on and on." Finally, I dedicate this thesis to my wife and dearest companion. Hannah, each day with you is a gift. 5 THIS PAGE INTENTIONALLY LEFT BLANK 6 Contents List of Figures 9 List of Tables 11 1 Introduction 13 1.1 Research m otivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.2 Background ...... ... .. ........................................ 14 1.3 Research contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 1.4 Thesis overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 2 Effects of plug-and-play photovoltaic design on balance of system costs 21 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.2 D ata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 2.3 Effects of plug-and-play innovations on photovoltaic system cost . . . . . . . . . . . . 27 2.3.1 PV system designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2.3.2 Analysis of photovoltaic projects using design structure matrices . . . . . . . . 29 2.3.3 Cost change potential in photovoltaic systems . . . . . . . . . . . . . . . . . . 35 2.3.4 Prospective cost change in the balance of system . . . . . . . . . . . . . . . . . 38 2.4 D iscussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 S2-1 Expanded design structure matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 3 Sources of cost overruns in nuclear power plant construction 55 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 .58 3.2 D ata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Sources of Cost Change in Nuclear Construction . . . . . . . . . . . . . . . . . . . . . 59 3.3.1 Cost Evolution of Nuclear Plants of Standard Design . . . . . . . . . . . . . . . 59 3.3.2 Sources of Cost Change in Nuclear Plant Construction . . . . . . . . . . . . . . 62 3.3.3 Sources of Cost Change in Containment Buildings . . . . . . . . . . . . . . . . 65 7 3.3.4 Evaluation of Nuclear Construction Productivity . . . . . . . . . . . . . . 69 3.3.5 High-level Mechanisms of Containment Building Cost Change . . . . . . . 71 3.3.6 Opportunities for Future Cost Reduction in Nuclear Construction . . . . . 75 3.4 D iscussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 S3-1 Full list of total plant cost accounts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 S3-2 Indirect cost attribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 S3-3 Containment cost model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 S3-4 Containment building cost data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 S3-5 Indirect containment building cost change . . . . . . . . . . . . . . . . . . . . . . . . 90 S3-6 Containment building sensitivity analysis . . . . . . . . . . . . . . . . . . . . . . . . . 91 S3-7 High-level mechanisms of containment building cost change . . . . . . . . . . . . . . 93 S3-7.1 Variable classification scheme to assign low-level to high-level mechanisms . . . 93 S3-7.2 Retrospective analysis of high-level mechanisms . . . . . . . . . . . . . . . . . 94 S3-8 Prospective analysis of containment cost change . . . . . . . . . . . . . . . . . . . . . 100 S3-8.1 Assumptions for prospective analysis . . . . . . . . . . . . . . . . . . . . . . . 100 S3-8.2 Prospective analysis with indirect cost change . . . . . . . . . . . . . . . . . . 100 S3-8.3 Prospective analysis of high-level mechanisms . . . . . . . . . . . . . . . . . . 100 8 List of Figures 2-1 Design structure matrix for a conventional small-scale PV system . . . . . . . . . . . . 32 2-2 Design structure matrix for a large plug-and-play PV system . . . . ..... . . . . . . 33 2-3 Design structure matrix for a small plug-and-play PV system . . . . . . . . . . . . . . . 34 2-4 Elements by cost component in PV systems . . . . . . . . . . . . . . . . . . . . . . . . 41 2-5 Interactions by cost component in PV systems . . . . . . . . . . . . . . . . . . . . . . 43 S2-1 Expanded design structure matrix for a conventional small-scale PV system . . . . . . . 51 S2-2 Expanded design structure matrix for a large plug-and-play PV system . . . . . . . . . 52 S2-3 Expanded design structure matrix for a small plug-and-play PV system . . . . . . . . . 53 3-1 U.S. nuclear construction costs and learning rates . . . . . . . . . . . . . . . . . . . . 62 3-2 Nuclear plant cost change, 1976 to 1987 . . . . . . . . . . . . . . . . . . . . . . . . . . 64 3-3 Nuclear plant indirect costs, 1987, and cost change, 1976-1987 . . . . . . . . . . . . . . 66 3-4 Historical construction productivity change in nuclear and at large . . . . . . . . . . . 70 3-5 Contributions of low-level mechanisms to direct containment cost increase . . . . . . . 71 3-6 Contributions of low-level mechanisms by type to direct containment cost increase . . . 72 3-7 Contributions of high-level mechanisms to direct containment cost increase . . . . . . 73 3-8 Contributions of low-level mechanisms to future direct containment cost improvement 77 3-9 Contributions of high-level mechanisms to future direct containment cost improvement 78 S3-1 Contributions of EEDB cost accounts to cost change . . . . . . . . . . . . . . . . . . . 82 S3-2 Contributions of low-level mechanisms to total containment cost increase, high indirect 91 S3-3 Contributions of low-level mechanisms to total containment cost increase, low indirect 92 S3-4 Contributions of high-level mechanisms to total containment cost increase, low indirect 93 S3-5 Contributions of high-level mechanisms to total containment cost increase, high indirect 94 S3-6 Sensitivity of containment low-level mechanisms to uncertainties in ironworker wages . 95 S3-7 Sensitivity of containment low-level mechanisms to uncertainties in concrete worker wages 95 S3-8 Sensitivity of containment low-level mechanisms to uncertainties in steel prices . . . . 96 S3-9 Sensitivity of containment low-level mechanisms to uncertainties in structural steel prices 96 9 S3-loSensitivity of containment low-level mechanisms to uncertainties in concrete prices . . 97 S3-11 Sensitivity of containment low-level mechanisms to uncertainties in steel deployment .. .. ............. . . . .. . .. .. ... .. .97 rates . . . . . . . . . . . S3-12 Sensitivity of containment low-level mechanisms to uncertainties in structural steel de- ploym ent rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 S3-13Sensitivity of containment low-level mechanisms to uncertainties in concrete deployment rates ...... ... .. .. . . . . ........ .. ....... . . . . . . . . . . ... ... ..98 S3-14Contributions of low-level mechanisms to prospective containment cost change . . . . 101 S3-15Vogtle construction site satellite image . . . . . . . . . . . . . . . . . . . . . . . . . . 102 10 List of Tables 2.1 Project elements in PV systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 2.2 Project interactions in PV systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 2.3 Solar PV cost components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 2.4 Project elements per cost component in PV systems . . . . . . . . . . . . . . . . . . . 40 2.5 Project interactions per cost component in PV systems . . . . . . . . . . . . . . . . . . 42 3.1 Nuclear learning rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 S 3 -1 Containment building cost data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 S3-2 Classifiers used to identify applicable high-level mechanisms . . . . . . . . . . . . . . 99 S3-3 High-level mechanisms and related measures of innovative activity . . . . . . . . . . . 99 S3-4 Prior assignments of low-level mechanisms to high-level mechanisms . . . . . . . . . . 103 S3-5 Patents documenting R&D activity related to passive cooling containment design change 104 S3-6 Publications documenting R&D activity related to passive cooling containment design change ........ ... ........................................... 104 S3 -7 Assignments of low-level mechanisms to high-level mechanisms . . . . . . . . . . . . . 105 S3-8 Data sources for assignments of low-level mechanisms to high-level mechanisms . . . . 1o6 S3-9 Containment building cost data for prospective cost change analysis . . . . . . . . . . 107 S3-1oAssignments of low-level mechanisms to high-level mechanisms for prospective cost change analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 11 THIS PAGE INTENTIONALLY LEFT BLANK 12 Chapter 1 Introduction i.i Research motivation Decarbonization of electricity production is a critical step toward addressing the risks of climate change [1]. The power sector is responsible for the largest share of greenhouse gas emissions and also presents the greatest mitigation opportunity [2]. Improvements to global electricity access and electrification of transportation and heating are expected to spur growth in electricity demand [31, which heightens the importance of reducing emissions from our electricity systems if we are to mitigate the worst effects of climate change. Several low-carbon electricity production technologies have reached levels of maturity needed to contribute toward climate solutions [4]. Solar photovoltaic (PV) systems and nuclear power plants have attractive attributes for climate change mitigation. The two technologies have similar lifecycle impacts on global warming, roughly one tenth that of coal (e.g. [5]). Nuclear energy has other attractive attributes including the potential to provide a base-load electricity supply without supplemental storage technology, significant uranium resources, low fuel price volatility, and operating costs below those of fossil-fired power plants [6, 7]. Solar energy is the earth's most abundant energy resource, has equitable geographic distribution, and is readily scalable to levels necessary to meet global growth in electricity demand [4, 8]. Although the characteristics of both technologies are well-suited to alleviating climate change, global nuclear power capacity has not grown appreciably in the last two decades, while PV has increased by over 450GW [9, 10, 11, 12], with cost improvement being an influential factor in the divergent deployment rates. Our ability to effectively decarbonize the electricity sector will be dependent upon the cost of low- carbon electricity production technologies like PV and nuclear. Limitations on the adoption of these two technologies could drive up the cost of climate mitigation considerably [2]. The cost of these technologies relative to market alternatives will be an important determinant of rate of deployment [13]. Future changes in affordability will affect market penetration by influencing the actions of utilities, 13 firms, and individuals [14, 15]. Energy system costs are not static, and the causes of cost change warrant investigation. Changing costs have been responsible for major energy transitions and industrial revolutions in the past [16, 17]. Previous work has related energy technology costs to factors like the production of the technology or level of research investment (e.g. [18, 19]). Despite substantial work to study trends in technological cost, further advances in modeling techniques are needed to better understand the causes and mecha- nisms of cost reduction [20]. In this work we seek to advance approaches to cost change modeling to understand historical cost change and evaluate the potential for future reduction. 1.2 Background The cost of PV systems has declined more rapidly than other electricity production technologies, while nuclear plant costs have risen [21]. Previous work has begun to examine what the causes of historical cost change and opportunities for future reduction might be in these technologies. Cost change in photovoltaic systems Compared with other electricity production technologies, solar PV has declined most rapidly [21].1 This quick pace of cost decline and improvements in other aspects of performance have spurred industry growth [22]. The cost decline in PV systems can be attributed to changes in many of the inputs to PV projects and how they interact with each other [23]. One component, however, has been responsible for most of the cost improvement over the last four decades: PV modules [24]. Recent analysis indicates that improved module efficiency was the leading cause of the precipitous decline in module cost and that a combination of public and private R&D played the most significant role in driving overall technological change [20]. While the role of modules in historical cost improvement for PV systems has been substantial, modules now constitute a relatively small share of total cost - 13% of a residential PV project in the U.S. in 2017 [25]. 'Balance of system' (BOS) costs comprise the remainder and have been slower to change, with improvement occurring at roughly half the rate of modules [26]. While module prices are global, BOS costs vary by location, causing significant discrepancies in system price across and even within countries [23, 26, 24]. Recognizing the rising relative importance of BOS costs, recent work has focused on strategies for improving BOS design and processes (e.g. [27]). Strategies for reducing BOS costs will be necessary to achieve widespread affordability and meet market penetration targets in climate change mitigation scenarios. PV systems are a field-built technology, assembled of factory-produced hardware by a team of laborers at a project site. The mass-manufactured elements of PV systems require inputs of raw materials and 'Comparison to coal, natural gas, nuclear fission, wind, and solar thermal. 14 labor, using factory equipment for fabrication and assembly. These inputs and processes incur significant cost. However, the resulting hardware costs have been declining rapidly [20, 24] and now represent about a third of total system cost [23]. While off-the-shelf PV systems that reduce non-hardware costs exist, they have been most commonly deployed in off-grid applications which constitute less than 1% of the global PV capacity [28]. The design, construction, testing, and inspection of grid-connected systems are coordinated and performed by human actors in the field (or based on site visits) through manual processes, subject to site-specific conditions [29, 30]. The aspects of PV projects which require custom and manual activities may contribute to the slower cost decline in the BOS [24]. In an effort to reduce BOS costs and installation challenges with these manual processes, federally-funded R&D initiatives have sought to advance the use of automation and standardization in PV projects, setting a goal of developing low-cost "plug-and-play" systems that can be purchased, installed, and operational in a single day [31]. Plug-and-play PV technologies have potential to reduce BOS costs through novel system designs. Plug-and-play systems utilize pre-assembled, standardized hardware which can be installed without professional assistance, thereby simplifying interactions among system hardware, actors, and the project site. Examples of innovations used in these systems include AC modules 2, touch-safe electrical con- nection ports, hardware-integrated electrical grounding, pre-manufactured racking systems, and smart devices for grid interface that automate regulatory compliance tasks [33, 15, 34, 35, 36]. Through the use of pre-assembled hardware, plug-and-play technology shifts tasks of interconnecting materials and equipment from the field to the factory, where productivity is typically higher [37], and enables partial or full automation of tasks. Automation and hardware pre-assembly have played an important role in cost reduction for elements of PV systems (e.g. [38]) and in certain field-build technologies (e.g. [39]). Similarly, standardization is important to enabling cost reducing mechanisms, such as scale economies and knowledge transfer [40]. Will plug-and-play PV systems be small, portable, and interconnected by a standard power cord like many home appliances (e.g. 33, 15]), or will they be designed as versions of existing systems, employing smart interconnection devices [34, 35]? Previous studies on plug-and-play PV are divided by their emphasis on either the former or the latter of these two predominant technological designs. However, taken together, the two system types may access a broader market than either would separately (e.g. [15, 34]) and tradeoffs in system size, cost, and complexity of installation will allow deployment across differing applications. Regulatory control of the two designs will also affect where and how these two types of plug-and-play systems are installed. What will be the cost of plug-and-play technology, and how will it reduce PV system costs in the future? Cost estimates for the U.S. range from $1.5o/W in the near-term [34] down to $o.25/W in the mid-term [15]. A limited number of plug-and-play products are commercially available and provide an 2AC modules are PV modules with an AC microinverter directly mounted to the module and capable of producing AC power with no external DC power [32]. 15 indication of current market pricing. In the U.S., typical prices are near $2.So/W (e.g. [41]) which is comparable to professionally installed system costs. Some markets, however, have already seen significant reduction in relative price, such as the U.K. where systems may be purchased for less than $1.oo/W (e.g. [42]). Existing literature on plug-and-play systems tend to emphasize installation labor and regulatory costs as primary opportunities for cost improvement but have less to say about other major sources of system cost. An appraisal of the effects of plug-and-play design on all cost categories is needed to better understand the technology's cost reduction potential. One tool that is particularly useful for analyzing detailed design and processes information is the design structure matrix or DSM [43]. The DSM highlights the architecture of a system by representing the N elements that comprise that system in an N x N matrix, mapping the interactions among all elements. As a tool for studying products and processes, DSM has been used for a variety of analyses related to system configuration: highlighting patterns in system design [43], feasibility assessment of various product features [44], calculation of product or process attributes (e.g. weight, power con- sumption, cost, or sales price) [43, 44], process schedule and budget optimization [43], and more [45]. DSM applications and extensions have grown significantly since the tool's inception, with innovative analysis across a wide range of industries [45]. There is precedent for applying DSM in energy systems modeling, including power plant delivery [46] and smart grid design [47]. In this work, we model PV projects using DSM to evaluate opportunities for plug-and-play systems to improve costs. Cost change in nuclear power plants Globally, nuclear construction costs have increased over the span of the last fifty years, with more rapid escalation in the U.S. than in other countries [48]. In the U.S., rising costs and project delays have proven problematic to expanding nuclear power capacity since the 1970s [49, 50, 51]. A survey of the plants which began construction after 1970 shows an average overnight cost overrun (realized costs relative to initial budget) of 241% [51]. Since the 1990s, two nuclear projects have begun construction, both two-reactor expansions of existing generating stations. The VC Summer project in South Carolina was abandoned in 2017 with sunk costs of $9B, and the Vogtle project in Georgia is severely delayed. Current estimates place the total price of the Vogtle expansion at $25B ($11,ooo/kW), almost twice as high as the initial estimate of $14B, and costs are anticipated to rise further [52, 53]. Challenges in nuclear construction are not unique to the U.S.; cost escalation, cost overrun, and schedule delays are common globally and studies have suggested that costs have risen more in nuclear projects than in other types of infrastructure [54.]3 Although costs have risen historically, nuclear industry, government, and research agencies continue to expect cost reductions in nuclear construction (e.g. [55, 56, 57, 58, 59, 60]). Such entities invest in the development and commercialization of next-generation reactor designs with the expectation 3 Comparison to solar farms, transmission projects, wind farms, other thermal power plants, mining projects, roads, bridges and tunnels, railway networks, and hydroelectric dams. 16 that successive plants of uniform design will cost less than first-of-a-kind plants [61, 62, 63, 57, 64]. This notion is applied generally to all commercial reactors, though the sources have anticipated cost reductions that are greatest for small modular reactors (SMR), due to expected learning effects in factory settings [65, 66, 59]. The first SMR has yet to be built. Retirement of existing nuclear plants has motivated goals to add new capacity. In the U.S., nuclear power constitutes 20% of the electricity supply, down from a peak of 23% in 1995, and 6o% of low- carbon electricity [67]. Low-cost domestic natural gas and declining costs of renewable electricity supply have put several plants at risk of premature retirement, while equipment replacements to extend plant lifetimes have proven challenging [68]. Four U.S. plants have shut down despite possible license extensions, and closure of 15-20 more plants is expected by 2030 [69]. Other countries with aging nuclear infrastructure (e.g., Spain, the UK) are facing similar challenges [il]. This outlook stands in contrast to nuclear power's expected role in many decarbonization scenarios (e.g., [70, 71]). Various hypotheses on the causes of nuclear construction cost increases are presented in previous literature. These studies fall into two groups: studies of nuclear technology cost trends and associated learning over time; and engineering cost models of nuclear plants for a given design, at a given point in time. By studying time series of overnight capital costs, studies in the first group have shown that nuclear costs in the U.S. increased before and after Three Mile Island [72], that cost trends differ across countries [73, 74], and that construction costs have increased even in countries with comparatively short construction times [75]. Previous work has shown a cost reducing effect when the same firm built multiple plants of the same model in France [76], and costs remained stable in Japan between 1980 and 2011, probably owing among other factors to supportive national policies [74]. However, the majority of studies document increases in cost of construction and conclude that the nuclear experience has been one of limited or even negative learning [75, 77, 78, 72, 79]. Cost increases and plant performance decreases have been associated with reactor upscaling, with a lack of technology standardization, fragmented industry structure and plant ownership, and with increasing plant complexity [75, 80, 49, 81]. Studies in the second group develop engineering cost models of nuclear reactors and plants to provide construction cost benchmarks in the U.S. [82, 83, 84, 85, 86, 87, 88, 89] and other countries (e.g. [90]). Other, forward-looking studies have outlined design and construction strategies for cost reduction, such as modularization, off-site manufacturing, passive cooling, and advanced construction materials [91, 63, 6o, 62]. However, the focus of these studies has been on aggregated measures for plant cost change, which are important for comparing technologies but can mask the contribution of individual developments, such as changes in design or labor productivity, to cost trends. Bottom-up engineering and top down cost models are also used to develop standards for estimating individual nuclear plant costs [92, 64] or forecasting costs of specific reactor technologies [57, 58, 59, 63]. In response to cost uncertainty, such guidelines have been developed to minimize financial risk and provide consistent comparison among available technologies. Similarly, cost estimating guidelines 17 are used in models for projecting industry-wide growth and cost change at a national or global scale across nuclear and non-nuclear energy technologies [19, 18], and in global planning for climate change mitigation [71, 93]. Although empirical studies of nuclear construction indicate that costs have escalated as industry experience has grown, cost estimating guidelines used in these studies generally assume costs will decline with experience. Studies that directly test the validity of modeling assumptions with empirical evidence are currently largely missing. 1.3 Research contributions This work seeks to determine the drivers of technological change that influence the cost of deploying low-carbon electricity systems. In particular we focus on PV systems and nuclear power plants and aim to better understand the past trends in the cost and barriers to cost reduction. We also evaluate what opportunities exist for future cost improvement in these two technologies. While previous work has suggested that plug-and-play PV systems will be significantly less expensive than conventional small-scale systems, even establishing cost targets, such efforts have not studied and classified the potential sources of cost change. Such study is needed, however, to understand why plug-and-play systems stand to help reduce cost. Here we model two types of plug-and-play projects using design structure matrices and map project inputs and their interactions. We evaluate which components are costly and have not substantially improved over time (see also [24]), then determine the effect of plug-and-play design on those components. Previous research has studied two dominant plug-and-play designs separately, but has not performed side-by-side comparisons. We study both system types and contrast them with conventional PV systems. Our work advances the use of DSM for studying cost in energy systems or other technologies by using a new combination of system classifications. We add to current knowledge of plug-and-play technology by studying system design across multiple project dimensions (hardware, site, actors, and tools and equipment), field vs. factory assembly of hardware, various interaction types (material, spatial, energy, and information), levels of standardization and automation, and effects on cost components. This DSM mapping lays the foundation for future analysis of plug-and-play systems, such as optimization of product attributes (e.g. cost, weight, system efficiency, durability, etc.) and installation processes (e.g. scheduling, project management, permitting, inspection, and interconnection). We also study nuclear cost change using U.S. construction cost data from five decades, modeling the cost evolution of entire plants and of one major plant component, the reactor containment building. We present a collection of insights that motivate us to revisit common assumptions about the role of hardware design in influencing plant costs, in comparison to external, often site-specific or regional factors. Contrary to the commonly expected cost declines for plants of the same design class, we find that costs have instead risen in the U.S., both for entire plants and for containment buildings. Increased stringency of regulations was only one of a number of drivers, which also include declining productivity 18 and increasing commodity usage. We discuss the implications of our findings for the widespread use of nuclear cost estimating guidelines and examine opportunities and challenges for future cost reduction. 1.4 Thesis overview The subsequent two chapters of this thesis evaluate the drivers of technological change that influence the cost of deploying PV systems and nuclear power plants. The chapters are based on two papers that are in preparation [94, 95]. Chapter 2 In the second chapter we evaluate the potential for plug-and-play PV systems to improve BOS cost through design innovations. We find significant technological potential to improve five components which are costly and which have not substantially improved over time-profit, installation labor, over- head, electrical BOS, and customer acquisition. We map 575 interactions among 31 project elements necessary for a conventional small-scale PV system installation and show how two plug-and-play de- signs reduce the number of elements by 19% or 45% and interactions by 40% or 69%. Specifically, we find several mechanisms are important to the cost change potential of plug-and-play technology: elim- inating various project tasks or shifting their responsibility to the consumer removes costs associated with overhead and profit of installation firms; pre-assembly of system components and standardiza- tion of project tasks eliminates installation labor costs; reduction and simplification of BOS electrical components lowers equipment costs; and standardization of system design precludes custom and man- ual tasks involved in customer acquisition. Finally, we discuss the benefits and challenges of the two plug-and-play designs as well as possible implications of the technology for various project stakeholders. Chapter 3 In the third chapter, we investigate the cost change drivers in nuclear power plant construction. We examine historical evidence for the engineering assumptions in many cost models and study how expectations regarding technological improvement may have contributed to an underestimation of factors external to hardware design. Separating the cost trajectories of different plant designs used in the U.S., we evaluate the expectation that repeated construction of plants of standard design will achieve cost reduction, observing that nth-of-a-kind (NOAK) plants have been more, not less expensive than first-of-a-kind (FOAK) plants. Analysis of historical cost account data shows that indirect costs caused over 70% of the rapid cost rise in the period 1976-1987. Decomposition of overnight cost changes in the reactor containment building shows that inflation-adjusted costs more than doubled during the period 1976-2017, from a combination of new safety requirements, non-safety-related construction productivity decreases, and design changes that improved safety but raised the material needs of 19 structures; safety and non-safety factors were comparably influential. We find that actual construction productivity observed in recent U.S. plants is up to thirteen times lower than industry expectations. A prospective analysis of technologies that enhance productivity and reduce commodity use indicates substantial opportunity to reduce the impact of external, previously cost-increasing factors, though R&D and regulatory change would be needed to support their adoption. 20 Chapter 2 Effects of plug-and-play photovoltaic design on balance of system costs Balance of system (BOS) costs constitute the majority of photovoltaic system costs, have not declined as rapidly as module costs, and cause locational differences in costs. Strategiesf or reducing BOS costs will be necessary to achieve widespread affordability and meet market penetration targets in climate change mitigation scenarios. Plug-and-playP V systems seek to reduce BOS costs through pre-assembled hardware which can be installed without professionala ssistance, thereby simplifying interactionsa mong the elements involved in a project: system hardware, the project site, and human actors. Yet how do currently proposed plug-and-play products alter PV system designs and costs, and how significant are the cost reductions that may resultf rom these alterations?P revious work suggests that plug-and-plays ystems will be dramatically less expensive than conventional small-scale systems. Some studies have established cost targets that these systems are expected to reach. But these studies have emphasized the importanceo f installationl abor and PII without detailed review of other key components. Moreover past studies haven't focused on understanding the mechanisms by which plug-and-play systems can reduce cost. To fill these gaps and evaluate cost change opportunities, we model two types of plug-and-playp rojects using design structure matrices, advancing the use of this tool for studying technology cost by mapping a new combination of system relationships. We evaluate which components are costly and have not substantially improved over time, then determine the effect of plug-and-play design on those components. We map 575 interactionsa mong 31 project elements necessary for a conventional small-scale PV system installation and show how two plug-and-play designs reduce the number of elements by 19% or 45% and interactions by 40% or 69%. Further, these plug- and-play systems increase interaction standardizationb y one third or one half and increase interaction automationt hreefold or sevenfold. We show that plug-and-playd esigns have significantp otential to improve the five cost components with high latent opportunityf or cost change-profit, installation labor; overhead, electricalB OS, and customer acquisition. Specifically, we find that several mechanisms are important to the 21 cost change potential of plug-and-play technology. First, eliminatingv arious project tasks or shifting their responsibilityt o the consumer removes the associated overhead and profit of installationf irms. Second, pre- assembly of system hardwarea nd standardizationo f project tasks eliminates installationl abor costs. Third, simplification of BOS electrical hardware has potential to lower equipment costs. Fourth, standardization of system design precludes labor-intensive tasks involved in customer acquisition. We compare the benefits and challenges of the plug-and-play designs and discuss possible implications of the technology for various project stakeholders. 1. 2.1 Introduction Solar PV costs and the role of the balance of system Solar photovoltaic (PV) systems have declined in cost more quickly than other energy technologies [21]. Such rapid cost declines, coupled with performance improvements, have spurred market expansion [22]. Sustained growth of renewable electricity supply is a key strategy for reducing greenhouse gas emissions to alleviate the impacts of climate change [2]. Thus, our ability to reduce emissions of greenhouse gases while simultaneously meeting growing global energy demand will be dependent upon further technological improvement and cost reduction of electricity production technologies like solar PV. The cost decline in PV systems can be attributed to changes in many of the inputs to PV projects and how they interact with each other [23]. We call these inputs elements, which are the physical items and people involved in the project-PV hardware (e.g. modules, racking, inverters), infrastructure at the project site (e.g. roof, grounding, electrical distribution equipment), human actors (e.g. electricians, system designers, permitting personnel), and their tools and equipment. When two elements interface with each other in the process of creating a working PV system-whether through a material connection, spatial adjacency, an electricity flow, or an information transfer-we call this an interaction. Elements are typically items of cost, while interactions can affect the cost intensity of elements. Groupings of elements and interactions together form cost components, the major categories of costs for installations. 2 Although costs have declined across all components over time [23], a single cost component has been responsible for three quarters of the cost improvement over the last 40 years: PV modules [24]. Despite historical prominence in cost improvement for PV systems, modules now constitute a rela- tively small share of total cost, 13% of a residential PV project in the U.S. according to a recent national benchmark [25]. "Balance of system" (BOS) costs comprise the remainder and have been slower to change, with improvement occurring at roughly half the rate of modules [26]. While module prices are global, BOS costs vary by location, causing significant discrepancies in system price across and even 'A version of this chapter is in preparation for journal submission with co-authors Ajinkya Shrish Kamat, Goksin Kavlak, Magdalena M. Klemun, and Jessika E. Trancik 2 [94]There are eleven primary cost components for small-scale PV installations: modules, inverter system, structural balance of system, electrical balance of system, supply chain, sales tax, installation labor, PIH (permitting, inspection, and interconnection), customer acquisition, overhead, and profit. 22 within countries [23, 26, 24]. Strategies for reducing BOS costs will be necessary to achieve widespread affordability and meet market penetration targets in climate change mitigation scenarios. PV systems are a field-built technology, assembled of factory-produced hardware by a team of skilled and unskilled laborers at a project site. The mass-manufactured elements of PV systems require inputs of raw materials and labor, using factory equipment for fabrication and assembly. These inputs and processes incur significant cost, however, the resulting hardware costs have been declining rapidly [20, 24] and now represent about a third of total system cost [23]. While off-the-shelf PV systems that reduce non-hardware costs exist, they have been most commonly deployed in off-grid applications which constitute less than 1% of the global PV capacity [28]. The design, construction, testing, and inspection of grid-connected systems are coordinated and performed by human actors in the field (or based on site visits) through manual processes, subject to site-specific conditions [29, 30]. The aspects of PV projects which require custom and manual activities may contribute to the slower cost decline in the BOS [24]. In an effort to reduce BOS costs and installation challenges with these manual processes, federally-funded R&D initiatives have sought to advance the use of automation and standardization in PV projects, setting a goal of developing low-cost "plug-and-play" systems that can be purchased, installed, and operational in a single day [31]. Plug-and-play photovoltaic systems Plug-and-play PV systems have potential to reduce BOS costs through design innovation. Such systems utilize pre-assembled, standardized hardware which can be installed without professional assistance, thereby simplifying interactions among system hardware, actors, and the project site. The use of pre-assembled hardware shifts tasks of interconnecting materials and equipment from the field to the factory, where productivity is typically higher [37], allowing for partial or full automation of tasks. Plug-and-play systems employ an array of innovations intended to simplify and standardize PV projects, enabling individuals without specialized training to deploy them. One technology commonly used in plug-and-play system are AC modules-PV modules with an AC microinverter directly mounted to the module and capable of producing AC power with no external DC power [32]. The use of AC modules simplifies wiring and reduces the hardware needed to demonstrate compliance with electrical codes [33]. Touch-safe electrical connection ports, integrated grounding wires, and module frame grounding are designed to simplify, standardize, and improve the safety of electrical wiring. Modules can be quickly deployed using pre-manufactured racking systems, integrated hook and clamp con- nections, or module-integrated railless systems. Pre-attached sealing putty reduces or eliminates the need for flashing or other waterproofing measures. System PHI is standardized and automated with the help of a PV utility interface device in some designs, while others use a simple power cord for interconnection at an electrical outlet. Two key strategies for plug-and-play PV designs are standardization and automation, which have been credited for improved cost performance in product manufacturing [96]. Standardization is an 23 important factor in enabling cost reducing mechanisms, including scale economies and knowledge transfer [40]. Many elements of the BOS are challenging to encapsulate into solutions that are effective across projects [24], but efforts are underway in the U.S. and Germany to standardize PV system design in residential systems to reduce locational differences in system costs [26]. Automation and pre-assembly of hardware have played an important role in cost reduction for elements of PV systems (e.g. [38]) and in certain field-build technologies (e.g. [39]). However, automation is not prevalent in the design, assembly, testing, and inspection processes used in current PV projects. For example, the regulatory process of permitting, inspecting, and interconnecting PV systems is notorious for creating workflow challenges, schedule uncertainty, and delayed grid deployment of PV systems, especially as interconnection applications have increased over time [971. Recognizing the opportunities afforded by automating human processes in PV projects, some entities have begun to automate routine tasks. One of the largest U.S. utilities, Pacific Gas and Electric, cut permitting, inspection, and interconnection (PII) cost by 78% and duration by 70% through automation [97]. Recent work has suggested that the cost-reducing innovations that enable plug-and-play systems also can expand the market for solar PV. Mundada et al. suggest a market opportunity of nearly 60GW in the U.S., showing how plug-and-play systems can be attractive to the average American household as well households with more limited capital, as they can be expanded module-by-module over time, improve affordability [15]. They also note that portability of plug-in units allows households which rent their homes to adopt solar PV. Faunhofer USA discuss how simplified plug-and-play designs can expand the market of installers to include other trades like roofing contractors, HVAC installers, and electricians, thereby encouraging competition and streamlining cost structures [34], another significant opportunity for system cost reduction [98]. Literature on plug-and-play PV is divided by technological design: Will systems be small, portable, and interconnected by a standard power cord like many home appliances (e.g. [33, 15]), or will they be more thoughtfully designed versions of existing PV systems, employing smart interconnection devices [34, 35]? There may be opportunity for both systems, with heterogeneity in system size3 providing access to multiple market segments (e.g. [15, 34]). Tradeoffs in system size, cost, and complexity of installation will allow deployment of these two classes of design in differing applications. Regulatory control of the two designs will also affect where and how plug-and-play systems are installed. As previous research has studied large or small plug-and-play systems separately, an opportunity exists to evaluate these designs side-by-side. In this work we consider the effects of plug-and-play design on cost throughout the various dimensions of a PV project by reviewing both designs and comparing them to conventional PV systems. 3A plug-and-play system that is interconnected to an electrical outlet using a power cord will be limited by the existing overcurrent protection device, which must be rated greater than or equal to 125% of the maximum current in the branch circuit [99]. An existing 15 Amp circuit breaker would therefore limit system size to 1440W (15A x 120V + 125%) [33]. This is the case for many countries North and South America where noV or 120V electrical systems are common, while much of the rest of the world operates on 220V to 24ov and, conditional on electrical codes, could interconnect approximately twice the system capacity 24 A number of countries, jurisdictions, and utilities have regulatory standards that inhibit the use of plug-and-play systems as currently envisioned, although solutions are being developed. Mundada et al. [33] and Fraunhofer USA [34] discuss the PII requirements for small and large systems and review technical and policy innovations needed to overcome regulatory barriers. Researchers at Fraunhofer USA have prepared standards that can be adopted by jurisdictions or utilities to ensure safety while creating compliance pathways for plug-and-play systems [36]. PV utility interface devices (also called solar connection devices or meter collars) are one technical solution to enable regulatory compliance. These units provide direct interconnection between a PV system and an existing utility meter in the form of a listed device with standard mating plugs. The PV utility interface integrates power flow, communication, and overcurrent protection into a single device. Existing standards [36] and commercially available devices [ioo] use a connector inlet with four power pins (leg 1, leg 2, neutral, ground). By interconnecting directly at the meter, this allows for power flow to the residence or back to the grid, while avoiding potential challenges with existing electrical service panels (overloads, lack of breaker space). Similarly, a set of four communication pins allow for two-way communication with the PV system, which can be used by the owner and the utility for a variety of monitoring purposes. The PV utility interface provides integrated overcurrent protection that is readily accessible for use in P11 processes. Smart devices like the PV utility interface or advanced inverter systems are expected to enable self-testing and self-commissioning of plug-and-play equipment. With inherent protocols for testing the safety and functionality of the system, and two-way communication capabilities, such devices could automate regulatory approval with the authority having jurisdiction (AHJ) and utility, decreasing PlI duration and costs. Cost analysis for plug-and-play systems What costs are expected for plug-and-play systems? Mundada et al. estimate a nearly 60GW market potential for plug-and-play systems in the U.S. [15] using near- and medium-term costs ranging from $o.25/W to $1.25/W, but offer no justification for the selected costs. The work of Fraunhofer USA targets system costs of $1.5o/W for plug-and-play systems by 2020 based on a bottom up estimate with assumed future costs [101, 34]. A limited number of plug-and-play products are commercially available and provide an indication of current market pricing. In the U.S., typical prices are near $2.5O/W (e.g. [41]) which is comparable to professionally installed system costs. Some markets, however, have already seen significant reduction in relative price, such at the U.K. where systems may be purchased for less than $1.oo/W (e.g. [42]). Higher pricing in the U.S. may be a result of regulatory challenges in PII, whereas policies are more favorable in other countries such as the U.K., Netherlands, Czech Republic, and Switzerland [33]. While previous work has suggested that plug-and-play systems will be significantly less expensive than conventional small-scale systems, even establishing cost targets, such efforts have not studied the sources of potential cost reduction in detail. Existing literature on plug-and-play systems tend to 25 emphasize installation labor and PII as primary opportunities for cost improvement but have less to say about other major sources of cost. An appraisal of the effects of plug-and-play design on all cost components is needed to better understand the technology's cost reduction potential. Modeling a technology as a network presents an opportunity to study design architectures with multiple interacting parts, and one such tool-the design structure matrix (DSM)-is useful for an- alyzing detailed information related to system design and processes [43]. The DSM highlights the architecture of a system by representing the N elements that comprise that system in an N x N matrix, mapping the interactions among all elements. As a practical product and process planning tool, DSM has been used for a variety of analyses related to system configuration: pattern indication in system design [43], feasibility assessment of various product features [44], calculation of product or process attributes (e.g. weight, power consumption, cost, or sales price) [43, 44], process schedule and budget optimization [43], and more [45]. DSM applications and extensions have grown significantly since the tool's inception, with analysis across a wide range of industries [45]. There is precedent for applying DSM in energy systems modeling, including power plant delivery [46] and smart grid design [47]. We evaluate opportunities for cost improvement by modeling plug-and-play projects using a DSM, evaluating the cost change potential of various BOS cost components, and determining the effect of plug-and-play design changes on those components. Rather than estimating the cost change potential of plug-and-play designs, we focus on identifying the mechanisms through which these designs could achieve cost reduction, and their ability to capture latent potential. We show that for costly components which have been slow to improve, plug-and-play designs reduce, standardize, or automate element interactions or eliminate elements entirely. Section 2.2 reviews the data we use to study the effects of plug-and-play design, section 2.3 summarizes the methods and results of our analysis, and section 2.4 discusses our findings and their limitations while considering the future of plug-and-play technologies. 2.2 Data Our analysis of the sources of cost reduction potential in plug-and-play systems uses two datasets: (i) equipment and project-level data on PV system design and installation, including technical and process information and (2) historical cost data for residential PV systems in the U.S. To model project design and installation, we draw from an array of recent literature and project- related documents including technology handbooks (e.g. [30, 29]), national codes (e.g. [99, 102]), technical reports (e.g. [34, 27]), local permitting guidelines (e.g. [103, 104]), utility interconnection policies (e.g. [105, io6]), design standards (e.g. [36, 107]), system diagrams (e.g. [1o8, 109]), research publications (e.g. [35, 33]), national solar benchmark reports (e.g. [110, 23] installer surveys (e.g. [ii]), product specifications (e.g. [100, 112]), installation manuals (e.g. [113, 114]), and visual documentation of system installation processes (e.g. [n, 116]). We capture common practices and specifications in conventional residential PV system design and installation, while acknowledging that 26 current industry practice and system design are not only diverse, but continuously evolving. Where heterogeneity exists and there is adequate data on frequency of use, we use system information that is most representative of the industry. Where rates of occurrence are not readily available, we make selections in accordance with best practices in engineering and project management. Our research identifies two dominant plug-and-play system designs, and we collect information for both, while recognizing that additional configurations exist. For our analysis of cost reduction potential, we use historical data from previously published studies, disaggregated into cost components according to the most important categories of expense. We draw upon previously published studies performed by national laboratories [11, 110, 23]. We use gross domestic product (GDP) price indices from the U.S. Bureau of Economic Analysis to adjust for the effects of inflation [117]. 2.3 Effects of plug-and-play innovations on photovoltaic system cost We evaluate opportunities for cost improvement by modeling plug-and-play projects using design structure matrices, connecting elements of PV projects and their interactions to cost, and evaluating the impact of design changes on key cost components. 2.3.1 PV system designs Will plug-and-play systems be thoughtfully redesigned versions of existing PV systems that employ smart interconnection devices (e.g. [34, 35]) or will they be small, portable, and interconnected by a standard power cord like many home appliances (e.g. [33, 15])? To study the effect of plug-and-play systems on cost, we prepare DSM for three different PV systems: 1. A conventional small-scale PV system representative of the current U.S. residential market. This is the benchmark design for comparison. 2. A plug-and-play system of similar scale, requiring owner assembly, and interconnected using a smart PV utility interface. 3. A comparatively smaller plug-and-play system, substantially pre-assembled, and interconnected through a power cord and standard electrical outlet. We call these systems conventional, plug-and-play large, and plug-and-play small. The conventional system design is based upon the prevailing system architecture currently used in the U.S. [23]. The two plug-and-play system designs are based upon the dominant architectures discussed in literature and expected in the U.S. market [33, 15, lol]. The three systems are described in detail below. 27 The benchmark DSM is based upon a conventional PV system with DC power optimizers and a string inverter [23]. We model a single external AC disconnect as well as disconnect hardware internal to the inverter (e.g. [112]) and electric service panel. We assume all electrical hardware requires skilled electrical labor (e.g. [104]). We allow for custom layout of the module racking system and roof penetrations with flashing for waterproofing [113]. PII is accomplished through separate applications to the AHJ and electric utility, which both require onsite inspection (e.g. [103, io6]). These modeling choices are reflective of common practice in the U.S., although requirements may vary by jurisdiction or project site and would affect the system design and installation. The plug-and-play large DSM is based upon a PV system installed by the owner using AC modules, an AC combiner, and a PV utility interface capable of automating the PII process. Our model assumes an external disconnect is integrated into the AC combiner (e.g. [114]). Touch-safe interconnection ports are used for electrical and communication connections among modules and BOS electrical devices, replacing custom installation of wires, junction boxes, and electrical connectors (e.g. [118]). The PV utility interface is the only hardware which requires skilled electrical labor. It connects to the existing electric meter and serves as a port of electrical and communication interface, as well as providing overcurrent protection (e.g. [loo]). We assume an automated electronic commissioning and PII process that tests and checks the system for conformity to code requirements, although other methods of verification may also be used. 4 Thus, no onsite inspection is required and PII is accomplished remotely via two-way communication through the PV utility interface supported with visual documentation from the owner. The modeled system is of comparable size to conventional residential systems (e.g. 3 to 10 kW [23]). We allow for custom layout of the module racking system and roof penetrations with sealing putty tape for waterproofing. The plug-and-play small system we model has all or nearly all hardware pre-assembled and mounted to module units: AC modules, an AC combiner, touch-safe connection ports, and a standard plug that is compatible with conventional wall outlets. Consequentially, the installation process is simplified to fastening the module units to the roof, plugging the units into each other, and plugging the system into a dedicated outdoor outlet. The plug-and-play small system has no external disconnect, only automated internal disconnect used to meet code requirements for anti-islanding [33].5 Power from 4Four methods of code verification are described in recent standards for plug-and-play systems [36]: A. Verification by design: documentation is provided showing system characteristics conform to code requirements and are enforced at time of installation through a clearly identified means. B. Verification by electrical self-test (remotely communicated): a commissioning process checks the as-built system or external data sources to confirm that the system conforms to code requirements. The self-test will be summarized in a report that includes documentation that pre-installation regulatory approval has been granted, confirmation that the as-built system meets code requirements, and the results and calculations which support the test conclusions. C. Verification by remote visual documentation: photo documentation is provided showing system characteristics conform to code requirements. (generally used to verify workmanship, e.g. array layout, cable management, required plaques) D. Verification by visual inspection: traditional inspection should be performed for compliance with code that cannot be completed using the other three methods (conductor connections to terminals blocks, service panel wiring) Different verification strategies may be used for different code requirements on the same system. sRegulations require that in the event of a grid outage any interconnected independent inverters quickly shut off [99]. 28 the unit flows through the existing branch circuit to the existing electrical service panel, relying on an existing breaker for overcurrent protection. We assume that the owner does not need to take any steps for P11-that the system is listed by a Nationally Recognized Testing Laboratory for conformity to code [331, that the electric utility allows installation of small systems without an inspection or an interconnection agreement (e.g. [1051), and that the roof mounting is non-permanent and thus does not require structural permitting [331. The system we model is less than 1,440W in capacity, based upon maximum load allowed in a branch circuit with a 15 amp breaker [99, 331. 2.3.2 Analysis of photovoltaic projects using design structure matrices How can we evaluate the ability of plug-and-play systems to achieve cost reductions? PV installations are complex engineered systems, whose costs are influenced by numerous interrelated elements: PV hardware (e.g. modules, racking, inverters), the project site (e.g. roof, grounding, electrical distribution equipment), human actors (e.g. electricians, system designers, permitting personnel), and their tools and equipment. Any design change will affect multiple elements of this system. Modeling these systems as a network presents an opportunity to study design architectures and map the interactions among system hardware, the project site, and actors. One such tool, the design structure matrix (DSM), is useful for managing large data sets of transdisciplinary information related to system design and processes [431. The DSM highlights the architecture of a system by representing the N elements that comprise that system in an N x N matrix, mapping the interactions among all elements. Relating hardware, site infrastructure, and actors to overall system cost components, we can use DSM to study how plug-and-play systems change the design of conventional small-scale PV systems. Specifically, we can use it to address key cost components and cost reduction strategies identified in literature: overhead and profit, P11, installer labor, customer acquisition, BOS economies of scale, standardized hardware connections, vertical consolidation of hard and soft components [119, 120, 25, 26]. DSM can strengthen our understanding of how hardware and installation of the system can be standardized and automated, enabling cost reducing mechanisms like scale economies and knowledge transfer. Consistent system modeling enables direct comparison of conventional and plug-and-play designs. Applying the design structure matrix to PV systems We develop a DSM layout, used in figures 2-1 to 2-3, to concisely convey multiple dimensions of cost- relevant information for the three designs, beginning with system elements. Elements included in the analysis constitute four domains: PV hardware, site-specific infrastructure, human actors involved in the project, and tools and equipment used. Each of the N elements appear as a row and a column in the N x N matrix. Integrated sub-elements which are pre-assembled and connected to other hardware are distinguished from primary project elements by the shading of the on-diagonal cells. The elements 29 within each domain are ordered roughly according to project sequencing. Only elements that are directly involved in the PV system are included. For example, the plug-and-play small DSM includes an electrical outlet, which is used for interconnection, but not any other electrical distribution infrastructure that interacts with the outlet, such as ground, overcurrent protection, or an electrical service panel. Thus, the DSM for conventional, plug-and-play large, and plug-and-play small systems have differing elements. See figures S2-1 to S2-3 in the supplementary materials for versions of the DSM with all elements shown ("absent" elements denoted as such). We include hardware which are significant to the cost, operation, and safety of the PV systems, leaving out items of less significance based on our review of available technical, cost, regulatory, and engineering data (see discussion in section 2.2. The decision not to specify hardware of less consequence (e.g. inverter temperature management system) affects the number of interactions and elements shown in the DSM, which may influence our results. We select elements based upon common practice in the U.S., however, the set of PV hardware used in an installation can vary significantly by region and according to local requirements; an example of this is the number of AC and DC disconnects required by the utility and AHJ [121, 331. Interactions between system elements are important drivers of cost, so we provide multi-dimensional data on each interaction in the DSM. Direct interactions between elements are signified by data in off- diagonal cells, while indirect and downstream interactions are omitted. Because of the significance of standardization and automation in enabling cost-improving mechanisms [96, 40, 38, 39], we classify each interaction according to its degree of standardization (custom or standard) and automation (manual or automated) using numerals 1 through 4. Previous work has shown the value of DSM for studying automation and standardization in system design [45]. Custom interactions are unique or project-specific in design or implementation, as when an electrician cuts and installs site-specific lengths and configurations of wire and conduit. Standard interactions are fixed or pre-defined and do not vary across project sites, such an overcurrent protection device (breaker) "snapping into" an electrical service panel via a standard connection. Manual interactions require human action, as when a permitting official gathers project information by visiting the site and examining the system. Automated interactions proceed in part or in whole without human involvement, such as when a smart meter or communication device transmits information to a utility. Elements may interact with each other in more than one manner, so to quantify the total number of interactions with full granularity, we designate four types of interactions which are prevalent in DSM product development studies [122]: material connections, spatial adjacency, energy flows, and information transfers. These four types of interactions are coded by color and their relative position within a 2 x 2 grid that represents the interactions of any two elements in the DSM. While material connections and spatial adjacency are two-way in nature, energy and information may either flow uni- or bi-directionally. The directionality of an interaction is signified by its position relative to the elements involved: an interaction is an output from the element which is figured in the same column as the interaction and is an input to the element 30 which is figured in the same row as the interaction. 6 Thus, bi-directional interactions will appear in two locations in the DSM. The system boundary for this analysis is the local project level, capturing downstream effects of plug-and-play design changes where most costs are incurred. Excluded are supply chain, manufactur- ing firm, and certain aspects of the installation firm; however, the costs of these excluded elements and interactions are proportionate to or reflected within the costs of other items in the DSM (e.g. manufacturing costs are reflected within the cost of the PV hardware, supply chain costs are roughly proportional to hardware costs [23], so our analysis includes the primary drivers of cost within these excluded domains, albeit with less granularity than incorporating the domains outright. A portion of the plug-and-play pre-assembly activities are captured as automated interactions between primary project elements and integrated sub-elements. Tasks of factory pre-assembly are likely to include partial or full automation, thus, we categorize interfaces among pre-assembled hardware as automated. Effects of plug-and-play on the total system Figures 2-1 to 2-3 show that plug-and-play designs change the BOS architecture and project flow sig- nificantly. To evaluate the effect, we quantify the elements and interactions according to the relevant typologies (element pre-assembly, interface type, degree of standardization, degree of automation, and domain) and compare the play-and-play installations to the conventional. Table 2.1 summarizes the number of elements in conventional, plug-and-play large, and plug-and- play small projects, identifying substantial reductions in project complexity for plug-and-play designs. We identify thirty-one elements in a conventional PV project, of which only two are integrated (or "pre-assembled") sub-elements. The plug-and-play large design eliminates 19% of the system elements and the plug-and-play small design eliminates 45%, with both systems utilizing significantly more integrated sub-elements than the conventional system. Both plug-and-play designs reduce the number of actors involved in the project: 50% and 80% reductions for large and small systems, respectively. The plug-and-play small design decreases the number of PV hardware elements by 29%, while the large design has the same quantity of hardware elements as a conventional project. In Table 2.2 we quantify the interactions needed for conventional, plug-and-play large, and plug- and-play small installations, finding considerable improvements in project workflow for plug-and-play designs. The conventional PV project we study has 575 interactions among PV hardware, the project site, actors, and their tools and equipment. Approximately half (54%) of these interactions are standardized while almost none are automated (6%). The plug-and-play designs eliminate many of the interactions: 40% and 69% fewer for large and small system projects, respectively. Both plug-and-play systems use less than half as many custom interactions and manual interactions as a conventional system. The 6We have chosen to use the "inputs in rows" convention, although "inputs in columns" DSM are also prevalent [43]. 31 Sit Actors PV hardware Actors W E 0C E.Q. U 0 B Conventional .2 0 cr ie W C S 0. _0PV system p C r E ME 0 E E 6 .._mrir 0 S f2 . . ~-,. . - . . I I I Roof penetration sealing U I I I 1111111110 Roof fasteners Module racking Module fasteners Modules Power optimizer String inverter Ground fault protection .! Arc fault protection External disconnect Overcurrent protection 5 3 3 Electrical service subpanel Conduit Junction boxes Wiring Grounding conductor I I I U I Electrical connectors I r N I WEI W -tffi Ro of t Main electrical service panel -- J Ground Owner Sales and marketing team Project manager System designer Laborer Electrician Permit application reviewers Permitting inspector Utility application reviewers Utility inspector Tools and equipment FI F ~ T iT LJE I = T [IT I Elements Interfaces Standardization and automation UProject Material connection 1 Standard and automatedelement 2 Custom and automated EEn ergy flow 3 Standard and manualEintegrated Information transfer 4 Custom and manualsub-element S4pSatial adjacency Figure 2-1: Design structure matrix for a conventional small-scale PV system. 32 PV hardware Site Actors E E 9: * E Plug-and-play .2 t :3 PV system 0 C a) a 0 M- (la rge) C o Ea) C ; o ) - t 2) 0 MI 7S mE C , a 0 0 2 Roof penetration sealing 3 11 3 1i I Roof fasteners 3 1 3 3~ ~ Module racking Module fasteners Modules Microinverters 4L AC combiner ilL Eli SGround fault protection Arc fault protection i r > External disconnect r PV utility interface Overcurrent protection Conduit Wiring Grounding conductor Electrical connectors Communication cabling M R oof I I. I I V) Electrical meter in I F- * - - ~ ~.u- L~ L~ ru Owner I Ins I b I I I I I I Sales and marketing team 'I a I 11 System retailer Electrician Permit application reviewers Utility application reviewers Tools and equipment Elements Interfaces Standardization and automation EProject Material connection I Standard and automatedelement 2 Custom and automated -> Energy flow 3 Standard and manual f -) Information transfer 4 Custom and manualIntegrated sub-element Spatial adjacency Figure 2-2: Design structure matrix for a large plug-and-play PV system. 33 PV hardware Site Actors C E U) 40 E 4- 40 Plug-and-play Cu 00 C LQ CuO o) Cu U0 in C PV system C C C CLD C C Cu -0Q) (U0 :2d 0)Cu U 0 ECu C0 42 M (small) Cu a) Cu Cu Cu .tou 0 4C Cu -ot9c/ > :82 Cu C: . 3 0 Cu E 0 U ~ 0 0- 0 ~ 0 o tn Cu 0 0 2 ~ 2 2 . < 0L Roof penetration sealing EI! U 17Roof fastenersModule racking 1 --- ~I1-*-I i--p-r--i-i---i E-- lI Module fasteners (U Modules II Ii ~r4L -- ~5-- -3 Microinverters 1 ~1u AC combiner I 1 1 1 Ground fault protection Arc fault protection Wiring - -- - Grounding conductor Electrical connectors rim Roof (A Electrical outlet Owner 0 Sales and marketing team System retailer Tools and equipment Elements Intertaces Standardization and automatior Material connection 1 Standard and automated Project element ' E-> 2 Custom and automatedEnergy flow 3 Standard and manual Ellntegrated -> Information transfer 4 Custom and manualsub-element 11, Spatial adjacency Figure 2-3: Design structure matrix for a small plug-and-play PIV system. 34 -I Classification Conventional PnP large PnP small A PnP large A PnP small count count/ count count/ count count/ countpnP - Acount/ countpnP - Acount/ total total total countconv count0c,, countcov countconv Elements by type Primary 29 94% 16 64% 8 47% -13 -45% -21 -72% Integrated sub-elements 2 6% 9 36% 9 53% 7 350% 7 350% Elements per domain PV system 17 55% 17 68% 12 71% 0 0% -5 -29% Site 3 10% 2 8% 2 12% -1 -33% -1 -33% Actors 10 32% 5 20% 2 12% -5 -50% -8 -80o% Tools & equipment 1 3% 1 4% 1 6% 0 0% 0 0% Total elements 31 100% 25 100% 17 100% -6 -19% -14 -45% Abbreviations: PnP = play-and-play; conv = conventional Table 2.1: Project elements in conventional, plug-and-play large, and plug-and-play small designs for PV systems. interface types which are most greatly reduced by plug-and-play innovations are spatial adjacency and information transfer, with the plug-and-play small design causing a more significant impact than the large. The greatest reduction in number of interactions (absolute) across all domains occurs between the PV hardware and the actors: 1oS fewer interactions for the large plug-and-play design and 159 fewer for the small. The domains experiencing the greatest share of reductions (relative) differs for the large and small plug-and-play projects, with former occurring between the project site and the actors and the latter occurring within the domain of actors (actors +-> actors). Taken together, figures 2-1 to 2-3 with tables 2.1 and 2.2 show that the innovations used in plug-and- play systems will have significant effects on the design, installation, and regulatory approval process for PV systems. However, to determine if these innovations are suitable strategies for cost reduction requires relating these changes to cost components. 2.3.3 Cost change potential in photovoltaic systems To evaluate if plug-and-play designs are suitable for cost reduction we first analyze recent cost data for residential PV systems in the U.S. from the period 2010 to 2018, disaggregated by expense category, to estimate the relative importance of various cost components to future cost reduction. Cost components represent different groupings of production inputs or categories of cost, such as installation labor, equipment costs, and sales tax. Following the classification systems used in national benchmarks of photovoltaic cost (e.g. [25, 23]), we identify eleven cost components that comprise the total installed cost of a residential PV system, to which we assign indices, i. Some components have significantly declined in cost over the period we study, while and others have not: for example, modules and sales tax are 81% and 80% less costly in 2018 than in 2010, while electrical BOS and profit are 12% and 1% less costly. Under a "business as usual" framework, short-term cost change is assumed to continue at a 35 Classification Conventional PnP large PnP small A PnP large A PnP small count count/ count count/ count count/ countpnP - Acount/ countp,,p - Acount/ total total total countconv countconv countconv countconv Interactions by degree of standardization Standard 308 54% 254 74% 147 82% -54 -18% -161 -52% Custom 267 46% 91 26% 32 18% -176 -66% -235 -88% Interactions by degree of automation Automated 35 6% 87 25% 84 47% 52 149% 49 140% Manual 540 94% 258 75% 95 53% -282 -52% -445 -82% Interactions by interface type Material connection 170 30% 144 42% 79 44% -26 -15% -91 -54% Spatial adjacency 282 49% 146 42% 81 45% -136 -48% -201 -71% Energy flow 25 4% 21 6% 14 8% -4 -16% -11 -44% Information transfer 98 17% 35 10% 6 3% -63 -64% -92 -94% Interactions per domain PV system - PV system 163 28% 154 45% 91 51% -9 -6% -72 -44% PV system -* Site 18 3% 15 4% 11 6% -3 -17/0 -7 -39% PV system -> Actors 71 12% 37 11% 16 9% -34 -48% -55 -77% PV system -+ T&E 31 5% 20 6% 8 4% -11 -35% -23 -74% Site -* PV system 16 3% 14 4% 10 6% -2 -13% -6 -38% Site 4 Site 0 0% 0 0% 0 0% 0 0 Site -* Actors 18 3% 4 1% 4 2% -14 -78% -14 -78% Site -* T&E 6 1% 4 1% 2 1% -2 -33% -4 -67% Actors -+ PV system 120 21% 49 14% 16 9% -71 -59% -104 -87% Actors -* Site 29 5% 5 1% 4 2% -24 -83% -25 -86% Actors +-+ Actors 40 7% 11 3% 3 2% -29 -73% -37 -93% Actors -- T&E 12 2% 4 1% 2 1% -8 -67% -10 -83% T&E -+ PV system 30 5% 20 6% 8 4% -10 -33% -22 -73% T&E -> Site 6 1% 4 1% 2 1% -2 -33% -4 -67% T&E - Actors 15 3% 4 1% 2 1% -11 -73% -13 -87% T&E +T&E o 0% 0 0% 0 0% 0 0 Total interactions 575 100% 345 100% 179 100% -230 -40% -396 -69% Abbreviations: PnP = play-and-play; conv = conventional; T&E = tools and equipment; PII = permitting, inspection, and interconnection Table 2.2: Project interactions in conventional, plug-and-play large, and plug-and-play small designs for PV systems. 36 similar rate for each component, relatively quick or relatively slow, until there are significant changes to the mechanisms of cost reduction. As strategies already exist for the components with rapidly declining cost, we focus on the ability of plug-and-play innovations to activate or accelerate the decline in slowly changing components. Thus, two cost ratios are important to such future improvement: (1) the component cost in the current period, Ci,2 , to the total system cost in the current period, Z'1 Ci,2 , which represents the importance of the component as a share of the total system cost and (2) the current component cost, Ci,2 , to the initial component cost, Ci,, which represents the residual share of cost relative to its 2010 value. We define a combined metric for latent cost reduction potential, Ki, which we compute as a scaled product of these two cost ratios. To the second ratio we apply a power factor, bi, which takes a value of zero if the technical minimum of cost component i is approximately zero and takes a value of one if technical minimum is significantly greater than zero. This product is scaled by a constant a such that the values of Ki to sum to 1. Equation 2.1 depicts the computation of the latent cost reduction potential: Ci, 2 Ci, (2.1) Ki I( -(.1 2 il1 Ci,2 Ci,1 This metric assumes there is always cost reduction potential until a cost component hits its minimum. Based on a literature survey of technical minima, we select values of bi equal to zero for the following cost components: 1. Installation labor costs, which can be entirely eliminated if system hardware is substantially pre- assembled, designed for touch-safe electrical features, and interconnected using standard plugs 33, 15] 2. P11, which is precluded if the item 1 above is met, the system is tested and listed Nationally Recognized Testing Laboratory for conformity to code (e.g. [33], and the AHJ and utility policies allow (e.g. [105]) 3. Structural BOS, which can be reduced to an adhesive layer pre-affixed to PV modules which would eliminate all module racking, racking fasteners, module fasteners, roof penetrations, and the associated waterproofing [34] We use a value of one for bi for the remaining components. The values of Ki are influenced by the selection of the time period (2010 to 2018), and we expect that cost components which have experienced variable cost change over time would yield somewhat different results if an alternate period were studied. Table 2.3 presents each components' latent cost reduction potential, inflation-adjusted cost data for years 2010 and 2018, share of 2018 system cost, and residual share of 2018 cost relative to 2010 value. The table provides definitions of each cost component and listings of elements and interactions that 37 are included in each component, which is used for the analysis in section 2.3.4 below. We find that installation labor (18%) and firm overhead and profit (14% and 21%) account for the largest shares of latent cost reduction potential, followed by electrical BOS (11%) and customer acquisition (8%). Three cost components have very little latent potential: sales tax (1%), P11 (4%), and the inverter system (4%). The remaining three components-structural BOS, supply chain, and modules-have modest latent potential (6% to 7%). Thus, profit, installation labor, overhead, electrical BOS, and customer acquisition present the largest opportunity for future cost reduction strategies focusing on costly components that have been slow to improve through other innovations. 2.3.4 Prospective cost change in the balance of system Here we compare the designs of conventional projects to plug-and-play projects and study the changes in elements and interactions that comprise each cost component, evaluating which components are likely to improve. We first decompose each cost component into its constituent parts (elements and interactions) using the groupings identified in table 2.3. Next, we compare the results of this decom- position across the three modeled systems, using the latent cost change potential of each component (computed in section 2.3.3) to evaluate if plug-and-play designs can significantly improve key cost components. As the inputs to cost components, elements and interactions carry cost significance. Elements are physical items and people involved in the project; the former may have to be purchased, incurring costs to a component through the purchase and other means (e.g. supply chain and profit), and the latter may earn wages and accrue other costs (e.g. payroll) in association with cost components. Although not all elements will have equal cost, eliminating an element will generally reduce cost, ceteris paribus. The cost intensity of the elements in a cost component is affected by interactions, and, all else equal, eliminating interactions will typically decrease cost. For example, reducing interactions can result in: 1. Decreased level of material usage if fewer lengths of wire or conduit are needed to connect various electrical hardware elements 2. Reduced labor costs if a worker has fewer time-intensive tasks to complete 3. Less overhead if wear and tear on tools and equipment is reduced The attributes of interactions and elements may also affect their cost significance. Standardizing an interactions may enable cost reducing mechanisms such as scale economies and knowledge transfer [40]. Similarly, automation and pre-assembly of hardware in manufacturing has played an important role in past cost reduction for elements of PV systems (e.g. [38]) and in other field-build technologies (e.g. [39]). Standardizing and automating PII, a target of plug-and-play designs, has been shown to cost by 78% and duration by 70% in the case of one utility [97]. Elements and interactions may be appear in one cost component, more than one, or none. For example, the element "wiring" incurs cost in BOS electrical, sales tax, supply chain, and profit costs components, whereas the element 38 Cost 2010 2018 Share Residual Latent Definition Included elements Included interactions Component cost cost of 2018 cost 1 cost system share reduction cost potential Modules $2.47 $0.47 18.0% 19.0% 5.8% Price of DC module to first buyer. Modules. N/A Inverter $0.48 $o.18 6.9% 37.3% 4.4% Price of inverter and its integrated hardware String inverter (conventional), Interaction cells with automated material system to first buyer. Historical cost change analysis microinverters and AC combiner (PnP) connections with the string inverter (conventional) and conventional DSM uses string inverter, with integrated disconnect (PnP large), or microinverter and AC combiner PnP DSMs use microinverters with AC integrated ground and arc fault (PnP)-exduding wiring and grounding. combiner. protection (all). Structural $0.36 $o.1o 3.8% 27.5% 6.5% Price of module racking, module fasteners, Module racking, module fasteners, roof Interaction cells with automated material BOS roof fasteners, and roof penetration sealing to fasteners, and roof penetration sealing. connections with the module racking, fasteners, or first buyer. roof sealing. Electrical $0.22 $0.19 7.3% 88.0% 10.8% Price of all non-inverter and non-module Power optimizers (conventional), Interaction cells with automated material BOS electrical hardware to first buyer. external disconnect (conventional), connections with the electrical BOS elements. wiring, conduit, electrical connectors, overcurrent protection, service panels, grounding conductor, communication wiring (PnP large), PV utility interface (PnP large). Supply $1.11 $0.35 13.4% 31.5% 7.2% Added cost for shipping and handling of All PV hardware. None. Supply chain interactions are beyond the chain equipment (e.g. 16% in 2018). Additional system boundary supply chain costs for modules and inverters (e.g. 35% and 20%, respectively in 2018). Sales tax $0.33 $0.06 2.5% 19.8% 0.8% Sales tax on equipment weighted by state All PV hardware. N/A installed capacity (9% in 2010, 7% in 2018 when weighted by state). Installation $0.75 $0.28 10.7% 37.5% 18.1% Direct hourly cost of electrician and laborer Laborer (conventional) and electrician Interactions with laborer and electrician. labor wages plus indirect labor burden (e.g. FICA, (conventional and PnP large). Owner's '0 workers compensation) labor (PnP) is excluded as a cost component. PII $0.20 $o.06 2.3% 30.5% 3.9% Applications and labor for permitting (AHJ) Permit and utility application reviewers, Interactions with permit and utility application and interconnection (utility) as well as permit and utility inspectors, project reviewers, permit and utility inspectors. associated staff time for paperwork and manager. inspection. Typical permit fee is $200 in 2018 and $490 in 2010. Customer $0.77 $0.30 11.5% 39.0% 7.6% Cost of marketing, advertising, sales calls, Conventional: sales and marketing, Conventional: interactions with sales and acquisition sales-related site visits, preparing bids, and project manager, system designer. PnP: marketing, system designer, and project manager contract negotiation. sales and marketing. (with owner and site only). PnP: interactions with sales and marketing. Overhead $0.39 $0.29 11.1% 74.3% 14.0% Cost of administration and buisness Conventional: sales and marketing, Conventional: interactions with sales and overhead. Includes payroll, facilities, project manager, system designer, marketing, project manager, system designer, information technology, office expenses and laborer, electrician, tools and laborer, electrician, tools and equipment (except staff for administration, legal, finance, and equipment. PnP: sales and marketing, permitting and utility). PnP: interactions with business management. system retailer, electrician (large), tools sales and marketing, system retailer, electrician and equipment. Many elements are (PnP large), tools and equipment. Many beyond the system boundary. interactions are beyond the system boundary Profit $0.33 $0.33 12.6% 98.6% 21.0% Fixed profit margin applied to all direct All PV hardware. Conventional: sales Interaction cells with automated material expenses incurred by firms: 17% (system and marketing, project manager, connections. Conventional: Interactions with sales retailer and electrician for PnP, PV installer laborer, electrician, permitting and and marketing, system designer, project manager, for conventional). inspection elements. PnP: electrician laborer, electrician, permitting and inspection (PnP large), sales and marketing. elements. PnP: Interactions with electrician (PnP large), sales and marketing. Total $7.41 $2.61 100% 35.3% 100% Cost data are reported in 2018$/WDc and are derived from [111, 11o, 23] for U.S. residential PV installations. Note 1: relative to 2010 component cost. Table 2-3: Solar PV cost components. Cost component Conventional PnP large PnP small A PnP large A PnP small count count/ count count/ count count/ countenP - Acount/ countpnp - Acount/ total total total countconv countconv countconv countconv Module 1 3% 1 4% 1 6% 0 0% 0 0% Inverter system 3 10% 5 20% 4 24% 2 67% 1 33% Structural BOS 4 13% 4 16% 4 24% 0 0% 0 0% Electrical BOS 9 29% 7 28% 3 18% -2 -22% -6 -67% Supply chain 17 55% 17 68% 12 71% 0 0% -5 -29% Sales tax 17 55% 17 68% 12 71% 0 0% -5 -29% Installation labor 2 6% 1 4% 0 0% -1 -50% -2 -100% PHI 5 16% 2 8% 0 0% -3 -60% -5 -100% Customer acquisition 3 10% 1 4% 1 6% -2 -67% -2 -67% Overhead 6 19% 4 16% 3 18% -2 -33% -3 -50% Profit 26 84% 19 76% 13 76% -7 -27% -13 -50% Abbreviations: PnP = play-and-play; conv = conventional; PII = permitting, inspection, and interconnection Table 2.4: Project elements per cost component in conventional, plug-and-play large, and plug-and-play small designs for PV systems. Elements may contribute cost to more than one cost component. "Count/total" refers to the share of elements per cost component relative to the total number of elements in the PV system (31 for conventional, 25 for plug-and-play large, and 17 for plug-and-play small). "roof" (a pre-existing structure) adds no expense. Similarly, an electrician's labor (e.g. interactions with electrical hardware) incurs expense in installation labor, overhead (payroll), and profit costs components, whereas the owner's labor (e.g. interactions with the plug-and-play PV hardware) adds no cost. Effects of plug-and-play on cost component elements Figure 2-4 and table 2.4 identify the quantity and type of elements in each cost component, showing that for all BOS (non-module) cost components, plug-and-play designs reduce the number of elements, increase the share of pre-assembled elements, or accomplish both. Installation labor, PII, and customer acquisition have the greatest percent declines in elements for both plug-and-play designs (50% to ioo%). The plug-and-play small system entirely eliminates the elements of installation labor and PII, and it converts the inverter system into a fully integrated sub-element that is pre-assembled to the module unit. The inverter system is the only cost component with an increase in quantity of elements, caused by the addition of an AC combiner in both plug-and-play systems and an integrated external disconnect in the plug-and-play large system. Effects of plug-and-play on cost component interactions The cost component most affected by interaction changes for plug-and-play designs are installation labor, PII, customer acquisition, overhead, and profit. In figure 2-5 and table 2.5 we show that eight cost components experience pronounced changes through expanded use of automation, improved standardization, a reduction in overall interactions, or a combination of these three changes. Additional 40 >1 I Modules Inverter system Structural BOS 1- 5- 4- 4- 3- 3. 2- 2 1 - 0- I I I 0 Conventional PnPig PnPsm Conventional PnPg PnPsm Conventional PnPig PnPsm Electrical BOS Supply chain Sales tax 8 15- 15- 6- 10- 10- 4- 4 - Eo 5- ' 5- Er- ,, 0- , 0- 1M Conventional PnPig PnPsm Conventio nal PnPg PnPsm Conven tional PnPig PnPsm Installation labor PH1 Customer acquisition M2- 5- 3 E z 4- 2 3- 2- 0C 0 0 I N. Conventional PnPig PnPsm Conventional PnPig PnPsm Convent ional PnPi, PnPsm Overhead Profit 6- 1*25- Element class 20 - 4- I.0 15 - 7 Integrated sub-element2- 10 - Primary element5- 0 ,1 0- Convention al PnPig PnPsm Conventional PnPig PnPm System design Figure 2-4: Elements by cost component in PV systems. Plug-and-play designs reduce the number of elements in many PV system cost components and pre-assemble much of the hardware. Four trends are evident for cost components in plug-and-play large and small systems relative to conventional: the number of elements in BOS components is reduced (-16% or -43% on average), the share of pre-assembly is increased (eightfold and twelvefold on average), or both; the small system has a more pronounced effect and eliminates installation labor and PII entirely; time-intensive components such as installation labor, PII, and customer acquisition see the greatest decrease in relative share (6o% and 90%); the majority of BOS hardware is pre-assembled (53% and 75%, a fourfold and fivefold increase). Abbreviations: PnP = play-and-play; conv = conventional; lg = large; sm = small. 'Integrated sub-elements' are hardware elements which are pre-assembled and connected to other hardware. 'Primary elements' are the main inputs to the project. 41 Cost component Conventional PnP large PnP small A PnP large A PnP small count count/ count count/ count count/ countPnP - Acount/ countPnP - Acount/ total total total countconv countconv countconv countconv Module 0 0% 0 0% 0 0% 0 0 Inverter 14 2% 27 8% 27 15% 13 93% 13 93% Structural BOS o 0% 4 1% 16 9% 4 16 Electrical BOS 0 0% 42 12% 40 22% 42 40 Supply chain 0 0% 0 0% 0 0% 0 0 Sales tax o 0% 0 0% 0 0% 0 0 Installation labor 93 16% 24 7% 0 0% -69 -74% -93 -100% PII 113 20% 8 2% 0 0% -105 -93% -113 -100% Customer acquisition 49 9% 1 0% 1 1% -48 -98% -48 -98% Overhead 271 47% 81 23% 27 15% -190 -70% -244 -90% Profit 319 55% 98 28% 84 47% -221 -69% -235 -74% Abbreviations: PnP = play-and-play PV; conv = conventional PV; PII = permitting, inspection, and interconnection Table 2.5: Project interactions per cost component in conventional, plug-and-play large, and plug-and-play small designs for PV systems. Interactions may contribute cost to more than one cost component. "Count/total" refers to the share of interactions per cost component relative to the total number of interactions in the PV system (575 for conventional, 345 for plug-and-play large, and 179 for plug-and-play small). effects within this group of cost components are also notable. There is a prominent reduction in labor-intensive interactions which incur cost, which we observe in installation labor, PII, customer acquisition, overhead, and profit. For example, the plug-and-play large and small designs eliminate 74% and ioo% of the installation labor interactions that incur cost, and these are replaced by a fewer number of pre-assembly interactions within the inverter system, structural BOS, and electrical BOS cost components. The interaction shift toward pre-assembly enables automation and standardization of these tasks upstream in manufacturing, as well as increased standardization downstream at the project site. The plug-and-play designs also minimize or eliminate cost-incurring site visits in installation labor, PII, customer acquisition, and overhead (and thus eliminate the profit associated with these direct costs). We observe a sizable reduction in cost-incurring information exchanges within three components: PII, caused by eliminating the need for inspectors; customer acquisition, as there is no need to collect and distribute information for project bids; and overhead, prompted by eliminating the roles of project manager and system designer which have information-intensive responsibilities. Three components are not reflected in our analysis of interactions: module, sales tax, and supply chain costs. The module cost component is based exclusively upon the sale price of the DC module and thus has no post-manufacturing interactions. Sales tax has no associated interactions, as it is an indirect markup on PV system materials and equipment. The supply chain necessary to deliver PV hardware is beyond the system boundary of our model, however, as supply chain costs are approximately proportional to hardware costs [231, our analysis includes its primary cost drivers. Finally, some elements of project overhead are also beyond the system boundary (refer to table 2.3 for a list of elements that comprise overhead and information on which are included in the DSM). 42 Degree of automation Degree of standardization Interface type E 20- 10 - 15 -- 10 C _ 0105 t__20-A Interaction class 0- 404 Automated 3 CO' -0 100 C 0 CO Manual Standard 25 - 0 0 - 75 -- Custom S5490 00 001j- - Material connection 60 - E 30- Spatial adjacency z 0 - Energy flow E :e 0 300 V .- Information transfer 50C c3 10- S0- 300- 200- a-100- 0- ElPConv. PPgPnP~,, Conv. PnPig PnPm Conv. PnPig Pnl .m System design Figure 2-5: Interactions by cost component in PV systems. Plug-and-play designs standardize, automate, and re- duce the number of interactions in many PV system cost components. Five trends are evident for cost components in plug-and-play large and small systems relative to conventional: the number of interactions in BOS components is reduced (-67% or -77% on average), although interaction increase in BOS hardware (fourfold and fivefold on average); the share of automation is increased (fourfold and fivefold on average); the share of standardization is increased (52% and 43% on average); the small system has a more pronounced effect and eliminates installation labor and PII entirely; the number of information transfers decreases the most (-89% and -95%), followed by spatial adjacency and material connections, while number of energy flows increases (275% and 225%). Abbrevi- ations: PnP = play-and-play; conv = conventional; lg = large; sm = small. Automated' interactions proceed in part or in whole without human involvement. 'Manual' interactions require human action. 'Standard' interactions are fixed or pre-defined and do not vary across project sites. 'Custom' interactions are unique or project-specific in design or implementation. 43 Total effects of plug-and-play on components influential to cost change Tables 2.3 to 2.5 and figures 2-4 and 2-5 show that plug-and-play designs have significant potential to improve the five cost components with the most latent opportunity for cost change: profit, installation labor, overhead, electrical BOS, and customer acquisition. Several mechanisms are responsible: 1. Eliminating various project tasks or shifting their responsibility to the consumer removes the associated overhead and profit of installation firms 2. Pre-assembly of system hardware and standardization of project tasks eliminates installation labor costs 3. Reduction and simplification of BOS electrical hardware lowers equipment costs 4. Standardization of system design precludes many custom and manual tasks which comprise customer acquisition Profit. Plug-and-play systems have a substantial cost-reducing potential for firm profit on system direct costs, as such systems reduce or eliminate the services required of a PV installation firm. This is significant, as profit has the largest value of Ki (latent cost reduction potential), because it has not substantially declined over the period we study and represents a large fraction of total system cost. The combined impacts of plug-and-play large and small systems, respectively, on the project elements and interactions that incur profit are: - Elements: 27% and 5o% reductions in project elements, with increases of 44% and 65% in the share of pre-assembled system hardware, and - Interactions: 69% and 74% reductions in interactions, with 39% and 49% increases in the share of standardization and 70% and 94% increases in the share of automation. Profit has received little attention in plug-and-play literature, although it has been shown that aligning profit ratios in the PV industry with more established industries would substantially reduce cost (e.g. [34]). Feldman et al. (2013) suggest that higher profit margins for PV firms are driven by local market factors such as high electricity prices and low competition [123]. Hoepfner (2016) notes that plug-and- play technology has the potential to expand the market of installers to include trades with more efficient cost structures (e.g. roofing contractors, HVAC installers, electricians), perhaps to a margin of 7% on direct costs [34] compared to the 17% margin earned on average by residential installers [23]. However, apart from decreasing profit margin, this analysis shows that plug-and-play systems may also reduce the cost of equipment and services provided-the direct costs in proportion to which a profit margin must be earned-likely creating a second profit-reducing effect. By eliminating various project tasks or shifting their responsibility to the consumer, plug-and-play designs can decrease the associated profit markup. A third profit-decreasing effect is possible: remaining direct expenses (e.g. hardware costs, marketing) will be shifted to a centralized firm responsible for system pre-assembly and sale, which 44 affords opportunities for scale economies and knowledge transfer due increased standardization and automation. It is unknown to what extent plug-and-play systems will displace the work of installation firms vs. serving new market segments (e.g. low-income households and renter-occupied housing [15]); to the extent that the former occurs, installation firms will likely lose operating profits, which may affect their ability repay investors or make strategic investments [123]). Overhead. Firm overhead, like profit, is an indirect cost component responsible for a large share of system costs which benefits from the breadth of plug-and-play innovations. Large and small plug-and- play systems reduce elements in the DSM responsible for overhead costs by one third or one half, and interactions by 70% or 90%. Important aspects of firm overhead like local facilities, administrative staff, and infrastructure are beyond the system boundary of our analysis, yet would be eliminated entirely by the transition away from a local installation firm. The transition to a centralized plug-and-play assembly firm as the primary system provider yields greater market potential and economies of scale over a local PV installer firm, which may further drive down overhead costs. Installation labor. The cost component most deliberately targeted by plug-and-play design for improvement is installation labor, and we find that plug-and-play innovations are well-suited to this intent. Standardization and safety improvements (e.g. touch-safe electrical connections) allow in- stallation tasks to be performed by individuals without specialized training. The plug-and-play small design requires no professional labor, achieving 1oo% labor cost savings relative to conventional sys- tems, estimated at $o.28/WDC [23]. A plug-and-play large system eliminates the work of a general laborer and requires an electrician only for installing the PV utility interface, which requires 74% fewer system interactions than installation of a conventional PV system. The cost of installation for this device is expected to be roughly $200 [34] or about $o.03/WDC, representing a savings Of $0.25/WDC over current labor costs [23]. Further, the remaining interactions are simplified through standardized material connections and constraining the work to a single, predefined location at the project site. By connecting directly to a utility meter (e.g. [ioo]), a PV utility interface may also avoid unforeseen labor costs for wiring or other upgrades in existing buildings with infrastructure that lacks sufficient load capacity or is not code compliant. Whether a reduction in PV installer jobs will occur will depend on the extent to which plug-and-play systems displace the work of installation firms vs. creating and serving a new market. Factory pre-assembly of plug-and-play equipment is likely to create new employment opportunities. Customer acquisition. Plug-and-play designs can eliminate the majority of the costs of customer acquisition: sales calls, sales-related site visits, preparing bids, and negotiating contracts. For both plug-and-play systems this results in a two-thirds decrease in elements which incur costs and a 98% reduction in interactions, eliminating requirements for site visits. Only marketing and advertising costs 45 remain, which we expect to be standardized, presuming a transition to online retailers and brick and mortar stores [15]. Electrical BOS. We expect the changes in the electrical BOS to be cost favorable, particularly for the plug-and-play small design, due to decreases in the amount of electrical hardware and consolidation of elements during pre-assembly. The labor and capital infrastructure needed to pre-assemble the hardware will add cost to the electrical BOS, but we expect this effect to be secondary to the reduction of major elements such as an electrical service panel and external disconnect. The tasks of integrating hardware are likely to be standardized and partially automated, enabling new opportunities for cost reduction over time through economies of scale and learning by doing. Total effects of plug-and-play on other cost components Our analysis shows that plug-and-play designs also affect four of the six remaining cost components, which have relatively lower untapped potential for cost change. PII costs will be entirely eliminated for plug-and-play small systems and nearly eliminated for plug-and-play large systems, as the admin- istrative burden will be shifted from the installation firm to the system owner (no-cost labor) and no site visits are needed by the utility or the AHJ. Inverter costs are expected to rise by $o.21/WDC due to the shift from DC power optimizers and a string inverter to microinverters and an AC combiner.7 The structural BOS is pre-assembled to the modules for plug-and-play small systems, likely adding an incremental assembly cost, while the racking of the plug-and-play large system is not significantly different than a conventional system. Sales tax costs are quite small though may see a minor increase due to the added cost of microinverters and an AC combiner, partially offset by reductions in BOS electrical costs. Our analysis shows no cause for change in module prices. Supply chain costs are outside the DSM system boundary, but may improve due to standardization of hardware, which could enable scale economies for delivery of uniform equipment and reduction in overstock burden because of hardware standardization; one source expects an increase due to direct distribution to end customer [34] although another anticipates purchase through large online and brick and mortar retailers with streamlined distribution [is] which could improve supply chain costs. 2.4 Discussion In this paper we show that plug-and-play designs have significant potential to affect the five cost components with the most latent opportunity for improvement-profit, installation labor, overhead, electrical BOS, and customer acquisition. We map 575 interactions among 31 project elements necessary 7A recent national benchmark suggests $o.18/Woc for the microinverter systems and $o.39/WDC for string inverter systems with DC power optimizers [23]. In addition to project installation benefits, microinverters also enhance the output power efficiency of the PV system by eliminating DC line losses [124, 125]. 46 for a conventional small-scale PV system installation and show how two plug-and-play designs reduce the number of elements by 19% or 45% and interactions by 40% or 69%. Further, these plug-and- play systems increase interaction standardization by one third or one half and increase interaction automation threefold or sevenfold. Several mechanisms are important to the cost change potential of plug-and-play technology: eliminating various project tasks or shifting their responsibility to the consumer removes the associated overhead and profit of installation firms; pre-assembly of system hardware and standardization of project tasks eliminates installation labor costs; simplification of BOS electrical hardware has potential to lower equipment costs; and standardization of system design precludes time-intensive tasks involved in customer acquisition. Our findings provide a new perspective that is not fully captured in the existing literature on plug-and-play systems, which tend to emphasize installation labor and PII as primary opportunities for cost improvement but have relatively little to say about customer acquisition, overhead, profit, and BOS electrical costs. While plug-and-play is extremely effective for these two "targeted" cost components, PHI is of little consequence to cost change and installation labor is only one of five components with significant latent cost improvement potential. Moreover, our findings provide new insight on the types of cost-reducing mechanisms that would allow plug-and-play systems to reduce cost across all system cost components. The two plug-and-play systems we study have differing impacts on cost and project performance. The plug-and-play small design-a portable, substantially pre-assembled system that is interconnected through a power cord and standard electrical outlet-eliminates more elements of cost and reduces, standardizes, and automates more system interactions. Compared to the plug-and-play large design- an innovatively redesigned but conventionally-sized system that is interconnected using a smart PV utility interface-the small system is simpler to install, requires no skilled labor, and eliminates the PII process. Despite these benefits, the plug-and-play small systems are significantly constrained in capacity by the existing electrical infrastructure at the site (e.g. maximum power of ioooW to 1440W in countries with 11oV or 12oV electrical systems [33]). If installed incorrectly, plug-and-play small systems could represent a safety hazard by overloading electrical circuitry, particularly due to the lack of PII oversight, though technical solutions to overcome this risk exist and more are in development [33]. By comparison, the plug-and-play large system is unconstrained in capacity by existing branch circuitry and overcurrent protection, carries lower associated safety risk, and is likely to benefit from scale economies in the supply chain due to greater quantities of identical hardware items (e.g. modules, microinverters, racking). Our review suggests that both designs have significant cost reduction potential and their differing attributes will make plug-and-play technology attractive across a greater range of applications. Our work advances the use of DSM for studying cost in energy systems or other technologies by using a new combination of system classifications. We provide new understanding of the impact of plug-and-play technology by studying system design across multiple project dimensions (hardware, site, actors, and tools and equipment), field vs. factory assembly of hardware, various interaction types 47 (material, spatial, energy, and information), levels of standardization and automation, and effects on cost components. This DSM mapping lays the foundation for future analysis of plug-and-play systems, such as optimization of product attributes (e.g. cost, weight, system efficiency, durability, etc.) and installation processes (e.g. scheduling, project management, and P11). There are limitations to our approach. DSM modeling requires selecting specific technological designs, omitting others. We choose two plug-and-play systems that are representative of the seemingly best designs available through current technology, although alternate configurations exist. 8 Similarly, "conventional" practice includes not one, but many system designs, and our benchmark representing just one common approach. We assume regulatory standards that allow installation of plug-and- play systems as modeled, 9 but where regulations differ, projects may require additional elements and interactions not included in the DSM. Next, our choice of system boundary excludes aspects of some cost components: administrative staff and firm infrastructure that impact overhead costs, the equipment supply chain (an entire cost component), and the labor and capital used in manufacturing. Although we discuss possible effects of plug-and-play designs on these items, future work could provide additional insight by expanding the system boundary to add these domains. Our method for determining the most important components for future cost change is adequate for developing a general ranking, however, these estimates should not be construed as a detailed evaluation of the economic or technical cost reduction potentials of each component. Finally, as individual elements and interactions in our analysis have varying effects on cost, changes in their quantity and attributes do not translate proportionally to cost change. What are the implications of plug-and-play technology for the various project stakeholders? Reduc- tions in PV system costs will save money for consumers who purchase them and will increase total installed solar capacity, all else equal. Secondary effects from increasing installed capacity and R&D to create plug-and-play systems are expected to further decrease costs [126], imparting additional savings to the end-user. Given the low lifecycle impact of PV systems on global warming [127], deployment of plug-and-play technology can lower the emissions intensity of our electricity systems, helping to mitigate the effects of climate change on society [2]. But will such benefits occur at the expense of solar installers? Whether a reduction in PVjobs will occur will depend on the extent to which plug-and-play systems displace the work of installation firms vs. creating and serving a new market. One study suggests plug-and-play systems will grow the solar market and increase jobs in manufacturing and distribution [15]. Plug-and-play large systems have the potential to expand the supply of installers to include additional trades (e.g. roofing contractors, HVAC installers, electricians) [34]. As these trades typically have more efficient cost structures than solar firms [23, 34], this competition may challenge existing business models in the PV industry. At this time, it is unknown whether plug-and-play systems will displace the work of installation firms or serve new market segments (e.g. low-income households "e.g. DC panels with a string inverter instead of an inverter system with microinverters and an AC combiner. 91.e. non-skilled labor for installation of electrical equipment that is appropriately tested and listed, automated PIH for large systems, and no PH1 or interconnection agreement for tested and listed small systems. 48 and renter-occupied housing [151); to the extent that the former occurs, installation firms will likely lose operating profits, which may affect their ability repay investors or make strategic investments [123]). We model plug-and-play projects that can be installed using existing technologies at sites with traditional infrastructure, but new developments in PV technology and building codes may provide ad- ditional cost improvement opportunities for this technology. For example, recently developed adhesive mounting schemes for PV modules could eliminate all module racking, racking fasteners, module fas- teners, roof penetrations, and the associated waterproofing [341, simplifying plug-and-play installation further and nearly eliminating BOS structural costs. Policy developments in building codes and elec- tric utility standards could further improve the installation process. At this time, plug-and-play large systems would require the services of a skilled electrician to install the PV utility interface; however, the electrical meters of tomorrow may include interconnect sockets for distributed energy resources (similar to a PV utility interface), which could eliminate the need for a skilled electrician during PV system installation. Current international building codes include voluntary standards for solar-ready buildings, which incorporate infrastructure for interconnection of PV systems at the time of construction [128, 102], and some jurisdictions already require new buildings to be built to such standards (e.g. [129, 130]). Other technologies under development may further facilitate the adoption of plug-and-play technology, including smart AC combiners, cable management devices, and software platforms for PII automation [34, 101]. These findings have implications on regulatory policy and system design. Plug-and-play technology presents significant opportunity for cost reduction in PV systems, yet will require novel approaches to PII to ensure safe and efficient market adoption. Ongoing development of the design strategies discussed here and future innovations are needed to advance plug-and-play PV to further technical maturity 49 Supplemental Materials Effects of plug-and-play photovoltaic design on balance of system costs Section S2-1 provides DSM figures with all elements shown, allowing for identification of "absent" elements in each system design. S2-1 Expanded design structure matrices Figures S2-1 to S2-3 are DSM which show the "absent" elements in each system to facilitate comparison of the designs and identification of elements that are eliminated and added as a result of plug-and-play innovations. These figures supplement the condensed versions provided in section 2.3.2. 50 SSittt e AAccttoorrsPV hardware s i E Conventional E a PV system .2e E E -o .c2 -E5 Roof penetration sealing 0I I I 0 I I Rooffasteners Module racking I L - mi Module fasteners Modules Microinverters AC combiner Power opt imiz er String inverter n Groundfaut protection Arcfauftprotection > External disconnect I- I ~- - - - - PV utility interface Overcurrent I _protection Electrical service subpanel Conduit Junction boxes Wiring Grounding conductor Electrical connectors Communication cablingL iJ J Roo f Elect rical outlet . aneec rical service panel Electrical meter Ground- Owner Sales and marketing team System retailer Project manager System designer Laborer Electrician 6 Permit application reviewers Permitting inspector Utility application reviewers Utility InspectorI I I I Elements Interfaces Standardization and automation E Material connection 1StandardandautomatedProject element lw 2 Custom and automated' 4 Energy 3 Standard and manual + Information transfer 4 Custom and manual NIntegratedlsub-element Spatial adjacency Albsent element Figure S2-1: Expanded design structure matrix for a conventional small-scale PV system. 51 . 11- PV hardware Site Actors E E t2 i Plug-and-play 5 .02 S PV system .2 (large) E E5 S - E E o E Roof penetration sealing U I I I I Roof fasteners Module racking Module fasteners Modules Microinverters AC combiner Power optima er String inverter . . . . . . . . . . 5 Groundfault protection Arc fault protection > External disconnect PV utility interface Overcurrent protection Electricalservicesubpanel Conduit Junction boxes Wiring Grounding conductor Electricalconnectors Communication cabling am 1111 I I I U I Roof Electricaloutlet Main elect rical service panel Electrical meter Ground I Owner Sales and marketing team System retailer Project manager System designer Laborer Electrician Permit application reviewers Permitting inspector Utility application reviewers Utility inspector r r Tools and equipment I1 1 1 1 1 1 0 1 N I 1 0 1 I T 7 7 1 Elements Interfaces Standardlzation and automation Material connection I Standard and automated Project element 2 Custom and aut ornated SEnergy floiown 3 Standard and manual+Informat transfer 4 Custom and manual N Integratedsub-element Spatial adjacency lAbsent element Figure S2-2: Expanded design structure matrix for a large plug-and-play PV system. 52 PV hardware Site Actors E It IT E Plug-and-play Se * Z PV system E2 o B (small) E .2 E S ECE Er: E E 19 Roofpenetrationsealing 0 I I I I Rooffastenees Module racking Module fasteners Modules Microinveters AC combiner Power optimizer String inverter Ground fault protection Arc fault protection > External disconnect PVutilityinterface Overcurrent protection Electricalservicesubpanel Conduit Junction boxes Wiring Grounding conductor Electrical connectors Communication cabling .. ... .. ... .. . . Rodf Elect rkaloutlet t!Main electrical service panel Electricalmeter Ground t EI OwnerII Sales and marketing team System retailer Project manager System designer toLaborer Electrician Permit application reviewers Permitting inspector Utility application reviewers Utility Inspector Tools and equipmentI l III ! IIII I [T11 IMI1I]9777 7 i1i1I1i1i1IILiE Elements Interfaces Standardization and automation 'Mraterialconnection I standardand automatedProject element +Energy 2 Custom and automatedfow 3Standardand manual SInformation transfer 4 Custom and manualIntegrated sub-elemeat Spatial adjacency Absent element Figure S2-3: Expanded design structure matrix for a small plug-and-play PV system. 53 THIS PAGE INTENTIONALLY LEFT BLANK 54 Chapter 3 Sources of cost overruns in nuclear power plant construction Projections of nuclear plant costs in the U.S. have often failed to predict cost overruns. The drivers of these overruns have often been assumed to be externally-imposed safety requirements, but this has not been demonstrated empirically. Here we examine these assumptions and revisit nuclear plant cost modeling using historical data from five decades. First, by separatingt he cost trajectories of different U.S. reactor designs, we observe that nth-of-a-kind (NOAK) plants have been more, not less expensive thanfirst-of-a-kind (FOAK) plants. Breakdowns of cost indicate that factors external to the choice of reactor, such as rising expenses for engineerings ervices, labor supervision, and construction support infrastructure, contributed well over half of the rapid cost rise in the period 1976-1987. Decomposing overnight cost changes in the reactor containment building, a major safety-grade structure, we find that real costs more than doubled from 1976 to 2017. While some of this cost rise came from external safety requirements, a significantp ortion came from safety-improvements undertaken independently by plant designers, as well as non-safety-related decreases in construction productivity. Notably, actual productivity in recent U.S. plants is up to thirteen times lower than industry expectations. Safety was thus an important but by far not the only factor driving cost increases. Technologies to enhance productivity and reduce commodity usage, some of which are already used in the construction industry, could reduce the impact of external, previously cost-increasing factors, but R&D and regulatory efforts may be needed to support their development and adoption in the nuclear industry1. 1A version of this chapter is in preparation for journal submission. Authors include Philip Eash-Gates*, Magdalena M. Klemun*, Goksin Kavlak, James McNerney, Jacopo Buongiorno, David A. Petti, and Jessika E. Trancik [95]. *Authors contributed equally 55 3.1 Introduction Nuclear power is viewed as a potential solution for minimizing greenhouse gas emissions from electricity generation and industrial process heating, as required to meet mid-century climate policy goals (e.g. [62, 131, 132]). In addition to low life-cycle emissions, other attractive attributes include base-load electricity supply, significant uranium resources, low fuel price volatility, and operating costs below those of fossil-fired power plants [6, 7]. In the U.S., however, rising construction costs and project delays have presented a pervasive problem in expanding nuclear capacity since the 1970S [49, 50, 51]. A survey of plants begun after 1970 shows an average overnight cost overrun of 241% [51]. Despite historical precedence for rising costs, nuclear industry, government, and research agencies continue to forecast cost reductions in nuclear construction (e.g. [55, 56, 57, 58, 59, 60]). These entities make significant investment in the development and commercialization of next-generation reactor designs based on the expectation that successive plants of standard design will cost less than first-of-a-kind plants [61, 62, 63, 57, 64]. This notion is applied generally to all commercial reactors, though the anticipated cost reductions are greatest for small modular reactors (SMR) due to expected learning effects in factory settings [65, 66, 59]. The first SMR has yet to be built. The goal to expand nuclear capacity is also motivated by retirements of existing plants. In the U.S., nuclear power plants provide 20% of the electricity supply, down from a peak of 23% in 1995, and 6o% of low-carbon electricity [67]. Low-cost domestic natural gas supply and declining costs of renewable power have put several plants at risk of premature retirement, and equipment replacements to extend plant lifetimes have proven challenging [68]. Four U.S. plants have shut down despite possible license extensions, and closure of 15-20 more plants is expected by 2030 [69]. Other countries with aging nuclear infrastructure (e.g., Spain, the UK) are facing similar challenges [11]. This outlook stands in contrast to the projected role of nuclear power in many decarbonization scenarios (e.g., [70, 71]). The early history of nuclear technology in the U.S. contrasts with recent experience in a number of ways. The U.S. pioneered the technology in the 1950s for naval submarine use and to this day generates more electricity in nuclear plants than the three next leading countries: France, China, and Russia [133]. Prominent global organizations such as the International Atomic Energy Agency and the Generation IV International Forum grew from U.S. initiatives and retain U.S. leadership within their organizations. U.S. federal investment in nuclear research and development is second highest among International Energy Agency member countries [134], and international cost estimating guidelines are based heavily upon U.S. reactor design and construction practices (e.g. the work of the Economic Modeling Working Group of the Generation IV International Forum [571. Rapid capacity growth in the 196os was accompanied by significant unit upscaling, followed by operational improvements and rising capacity factors [72]. However, in addition to rising project durations and costs, studies on thermal pollution and low-level radiation became a source of public controversy in the 1970S [135]. Following the 1979 Three Mile Island accident, a long hiatus of nuclear construction began. Despite political 56 attempts to revive the U.S. nuclear industry (e.g. the U.S. Energy Policy Act of 2005), new projects have struggled. Previous literature has presented various hypotheses on the causes of nuclear construction cost increases. These studies fall into two groups: 1) studies of nuclear technology cost trends and associated learning over time; 2) engineering cost models of nuclear power plants for a given design, at a given point in time. By studying time series of overnight capital costs, studies in the first group have shown that nuclear costs in the U.S. have increased before and after Three Mile Island [72], that cost trends differ across countries [73, 741, and that construction costs have increased even in countries with comparatively short construction times [75]. Previous work has shown cost reduction occurred in France when the same firm built multiple plants of the same model [76], and costs remained stable in Japan between 1980 and 2011, owing among other factors to supportive national policies [741. However, the majority of studies document increases in cost of construction and conclude that the nuclear experience has been one of limited or even negative learning [75, 77, 78, 72, 79]. Cost increases and plant performance decreases have been associated with reactor upscaling, with a lack of technology standardization, with fragmented industry structure and plant ownership, and with increasing plant complexity 2 [75, 80, 49, 81]. By developing engineering cost models of nuclear reactors and plants, studies in the second group have provided cost benchmarks for reactor construction in the U.S. [82, 83, 84, 85, 86, 87, 88, 89] and other countries (e.g. [90]). Other, forward-looking studies have outlined design and construction strategies for cost reduction, such as modularization, off-site manufacturing, passive cooling, and advanced construction materials [91, 63, 60, 62]. However, the focus of these studies has been on aggregated measures for plant cost change, which are important for comparing technologies but can mask the contribution of individual developments, such as changes in design or labor productivity, to cost trends. Both bottom-up engineering and top down models are also used to develop standards for estimating individual nuclear plant costs [92, 64] or forecasting costs of specific reactor technologies [57, 58, 59, 63]. In response to cost uncertainty, such guidelines have been developed to minimize financial risk and provide consistent comparison among available technologies. Similarly, cost estimating guidelines are used in models for projecting industry-wide growth and cost change at a national or global scale across nuclear and non-nuclear energy technologies [19, 18], and in global planning for climate change mitigation [71, 93]. Although empirical studies of nuclear construction indicate that costs have escalated as industry experience has grown, cost estimating guidelines used in these studies generally assume costs will decline with experience. Studies that test the validity of modeling assumptions against empirical evidence are currently largely missing. In this paper we begin to address these gaps by using U.S. construction cost data from five decades 2Studies often consider increases in number of plant components, new control systems, redundancy in equipment, and added safety features to be indicative of increasing complexity 57 to model the cost evolution of entire plants and of one major plant component, the reactor containment building. We present a collection of insights that motivate us to revisit common assumptions about the role of hardware design in influencing plant costs, in comparison to external, often site-specific or regional factors. Contrary to the commonly expected cost declines for plants of the same design class, we find that costs have instead risen in the U.S. Next, we examine what types of costs contributed most to cost increases, using cost accounting data on individual plant components and the services needed to install these components. Motivated by evidence that labor intensive and safety-related plant components accounted for the largest increases, we then study in detail the containment building, one of the most expensive components and a component with significant safety requirements. Here we find that increased stringency of regulations was only one of a number of drivers, which also include declining productivity and increasing commodity usage. We discuss implications of our findings for the widespread use of nuclear cost estimating guidelines and examine opportunities and challenges for future cost reduction. 3.2 Data Collection of empirical cost data from nuclear projects is challenging. Primary data sources are scarce compared to other technologies, as relatively few nuclear plants have been completed by only a handful of companies, and the average plant is over forty years old. In addition, the use of best-case data or data from other construction projects, which is not reflective of the nuclear industry, is common in bottom- up cost modeling [refs](e.g. [57, 64]). Changes to project design, schedule, and cost mid-stream are frequent and create another obstacle to finding data that is representative of an entire project. To address the above issues we collect data from a broad array of sources and check empirical data against hypothetical and best-case assumptions. For our analysis of nuclear learning rates and NOAK cost trends, we use databases of construction data, including International Atomic Energy Agency reactor information, historical government reports, and published data from academic and industry literature (e.g. PRIS [136], U.S. EIA [137, 51], Koomey and Hultman [72]). To study construction productivity changes in the U.S., we derive material deployment and labor data from reports by the Bureau of Labor Statistics [138]. Our evaluation of cost estimating guidelines is based upon series of reports prepared under the U.S. Department of Energy and by industry consortia (e.g. [64, 57]). For containment cost decomposition, we turn to EEDB data on commodity costs, labor costs, labor productivity, and structure dimensions of light-water containment buildings constructed during the 1970s and 198os, and fill in the gaps with U.S. Geological Survey commodity price data. In later years, our sample size shrinks commensurate with the contraction of the U.S. industry. Structure dimensions for 2017 containment buildings are extracted from nuclear engineering specifications and architectural drawings of the Westinghouse APiooo LWR, the only plant design currently under construction in the U.S. We also draw on construction and engineering reports from the VC Summer project in South 58 Carolina and the Vogtle project in Georgia, and on interviews with a Vogtle project representative [1391. The resulting data set is, to the extent possible, representative of the nuclear industry in all three years studied here. In 1976 and 1987, several plants were under construction, and our data represents an industry average [82]. In 2017, only one plant was under construction, and our productivity data are representative of this project. Commodity cost data, which are less project-specific than productivity, represent average price differentials between standard and nuclear commodities, and projects using the same containment design are under way in other countries. These factors allow us to extract more general insight from a U.S. centric analysis. 3.3 Sources of Cost Change in Nuclear Construction We first examine trends in overnight construction costs. We then decompose cost changes of individual plants over time into specific technical and economic causes. 3-3.1 Cost Evolution of Nuclear Plants of Standard Design Various studies have shown empirically that industry-wide average overnight construction costs for nuclear plants have increased over time, yet an assumption of cost-decline persists in many forward looking models. In one such study, Rubin et al. [79] apply an experience curve model to a combined set of U.S. and French nuclear cost data from a study by Grubler [75], and find that with every doubling of total installed nuclear capacity, costs rose by 38% on average. Experience curve models express technology unit costs, C, such as the costs of an electric generation system in units of $/kWe, as a function of cumulative installed capacity (GWe), x: C = Coxb (3-1) where CO is the unit cost of a reference system and b is the rate of cost change. Eq. 3.1 is commonly used to compute a technological learning rate, LR, the fractional reduction of cost expected with each doubling of cumulative capacity: 3 LR = 1 - 2b (3.2) 3The term 'experience curve' describes industry-level learning with exogenous effects, but is commonly conflated with the term 'learning rate.' Due to its wide recognition and use in literature, we use 'learning rate' to summarize industry-wide rates of cost change, though arguably it should be reserved for firm-level production. 59 In the case of cost increase, the value of the learning rate will be negative. Cooper [78] identifies negative learning effects for France and the U.S., with a sharp worsening in both countries after 35GWe of cumulative capacity Cooper also identifies negative learning effects for U.S. utilities and builders, finding that costs rose with experience for every entity Lovering et al. [73] calculate cost change by country-specific eras, identifying cost stabilization or decline for some subsets of reactors. Lang [48] uses data from Lovering et al. to compute learning rates for seven countries and the global industry, finding evidence of early learning (cost decrease) followed by a reversal and subsequent cost increase. The findings of Lovering et al. and Lang for early cost reduction include data from "turnkey" projects subsidized by manufacturers to gain market presence, which do not reflect actual costs [73] and are excluded from others' analysis of nuclear costs (e.g. [51, 140, 141]). Despite historical cost escalation, various prospective models and cost estimating guidelines still use an assumption of positive learning (that costs will decline with experience). Cost estimating guidelines typically assume regulatory stability as a precondition for positive learning effects, whereas projections of future industry growth expect cost reductions from learning even amid changing regulatory environ- ments (e.g. the various IPCC emission scenarios [71]). In its cost guidelines for advanced reactors, Oak Ridge National Laboratory [92, 64] recommends learning rates of 6% (equipment) and 2% (labor) for NOAK reactors of standard design after 4.5GWe cumulative installed capacity Similarly, the Economics Modeling Working Group (EMWG) of the Generation IV International Forum [57] established a global guideline of 6% (total) cost reduction with every doubling of cumulative capacity after 8GWe. The academic and scientific communities use similar learning rates to assess the role of nuclear power in future energy strategies and greenhouse gas mitigation scenarios. Published estimates range from 1% to 1o%, with SMRs at the upper end (e.g. the expert elicitation of Abdulla et al. [59]). The emissions scenarios behind IPCC climate modeling rely on nuclear learning rates between 1% and 7%, while the global scenarios of the International Atomic Energy Agency (IAEA) expect higher learning effects [71, 93, 58]. McDonald and Schrattenholzer [18] suggest learning rates for a multitude of energy-related technologies, including nuclear at 6%. However, it has been demonstrated that the reported values are not empirical, but assumed [142, 143, 144, 75]. The prevalence and breadth of prospective studies using positive learning rates may contribute to the common assumption that nuclear plant costs decline with experience. We begin by examining the trends in nuclear construction costs with time, focusing on whether n-th of a kind (NOAK) power plants of the same type achieve a cost reduction relative to (FOAK) first of a kind plants. We focus in on this question because we posit that the expectation of cost reduction achieved by NOAK plants may help explain the divergence between projected and observed cost changes over time. We use Eqs. 3.1 and 3.2 to compute empirical learning rates for the U.S. nuclear experience as a whole and for individual reactor designs. We study all standard designs constructed in sufficient capacity to achieve hypothesized nth-of-a-kind (NOAK) cost reductions. A commonly assumed threshold for NOAK cost reduction is 8GWe cumulative capacity, which was established by the EMWG [57]. The assumption 6o for NOAK cost reduction is based upon construction of nearly identical plants. In practice, however, there have been some design variations within the standard design classes we study (e.g. in cooling systems, thermal output, and net electric capacity [136]). In Fig. 3-1a, we prepare an experience curve for 107 U.S. nuclear plants of various designs. Similar curves are shown in work by Grubler [75], Koomey and Hultman [72], Cooper [78], and Lang [48]. Here, we plot the construction cost of nuclear plants against their cumulative built capacity, aggregating plants of all types by year. We see that costs rose very rapidly, echoing previous findings [75, 77, 78, 72, 791. Next we examine the assumption of model-specific cost reduction, plotting experience curves sep- arately for each prominent technology class in Fig. 3-lb. Although the rapid rise in costs across all nuclear plants is well known, the assumption that cost reductions may still have occurred for particular classes of plants persists. The practice of including cost data from all reactor types may contribute to this, as historical reports of construction costs (e.g. [137, 51]) and previous publications (e.g. [72, 18, 19, 73, 78, 751) lack information on reactor type. Berthelemy and Rangel [76] is the only study we find which quantifies the cost effect of model standardization at an industry-wide level. Their results show that standardization decreases cost relative to overall industry trends, while design innovation increases construction costs. Fig. 3-1 shows the overnight cost of construction and estimated learning rates (a) across the U.S. nuclear experience and (b) for the four standard reactor designs installed in excess of 8GWe in the U.S. In all cases, costs rose with experience. We estimate a learning rate of -115% for the entire industry, indicating that plant costs increased more than twofold for each doubling of cumulative U.S. capacity Over the period studied, 86% of the cost change is associated with rising industry experience. Studying plants with standard reactor design, we find that NOAK costs exceed FOAK costs for all models, and in most cases the FOAK plant cost the least. Learning rates vary by nuclear technology from -94% to -31%, suggesting that design standardization slowed the rate of cost escalation relative to the industry at large (learning rate of -115%), but was unable to reverse it. While it frequently stated as a "well known" fact that NOAK plants are less expensive than FOAK plants (e.g. [63, 62]), these findings indicate that NOAK cost reductions are not a certain consequence of design standardization, and plants of standard design are subject to the same endogenous factors affecting other plants. However, our results should be interpreted within the context that not all plants within each model are perfectly identical and the design differences may have contributed to the unexpected cost increases. Table 3.1 compares our results to previous empirical studies, established cost estimating guidelines, and projections of future nuclear technology growth for the U.S. and the world. Empirical learning rates are calculated through statistical analysis of observed cost data from completed projects, the result of evaluating Eqs. 3.1 and 3.2 to compute LR, whereas the other approaches used in the literature are not necessarily based in nuclear cost experience. Cost estimating guidelines and future growth models use learning rates based upon assumed technological learning to show how nuclear plant costs 61 . . I ii lil i i i i i i 10,000 00 10,000 Industry guideline for 00 achieving NOAK costs ka 001 5,000 0 5,000(N m 00 0 0 C-) Browns r,- 0 Ferry-1 2,000 2,000|I r, H m Indian C 0 1 ,000 C 0 Point-2C-) C 0U 1,000 6- Palisades -A-Combustion Engr 2LP, -49% LR -0-Gen Electric BWR-4, -50% LR 500 = + 1.11 log(Capacity) 500log(Cost 3.47 x Robinson-2 -0-Westinghouse 3LP, -94% LR Learning Rate: -115% R' = 0.859 -O-Westinghouse 4LP, -31% LR 15 30 45 60 75 90 105 0.5 1 2 5 10 20 30 Cumulative Capacity Installed (GW) Cumulative Capacity Installed (GW) (a) (b) Figure 3-1: U.S. nuclear construction costs and learning rates. Data are ordered by construction start date. (a) Average overnight plant cost, aggregated by construction year. Minimum and maximum values represent the highest and lowest cost reactors, respectively, in each year, with no min/max markers for years with only one reactor. Fitted line represents ordinary least squares regression of log-transformed data. Experience across all designs results in an estimated learning rate of -115%. (b) Individual overnight cost for all standard plant designs achieving the capacity threshold at which NOAK cost reduction is expected (8GWe [57]). Learning rates vary by technology from -94% to -31%, and NOAK costs exceed FOAK costs. Large markers and labels indicate FOAK plant for each technology. Cost volatility is correlated with variations in construction duration. are expected to decline as experience increases; typically the former estimates costs of individual plants and technologies, while the latter projects cumulative impact of the technology within long-term energy strategies and markets. These assumed nuclear learning rates may be derived from comparable technologies (e.g. coal power plants), expert elicitation, or the authors' judgment. 3.3.2 Sources of Cost Change in Nuclear Plant Construction To shed light on the causes of cost escalation, we decompose overnight construction into its cost components, beginning with the period 1976-1987, for which we have reliable cost data on all plant components [82, 83, 86, 85, 89]. We examine the contributions of 61 different cost accounts from the Department of Energy's Energy Economic Data Base (EEDB) to cost increase. These accounts, shown as ci in Eq. 3.3 below, represent individual plant components and services needed to install these 62 Approach Study Reactor Modelsa Market Time Learning Data Sourced Periodb Ratec Empirical This work Various U.S. 1971-1996 -115% [72, 137, 51, 136] Empirical This work Standard U.S. 1971-1996 -56%e [72, 137, 51, 136] Empirical This work Standard, CE 2LP U.S. 1971-1985 -49% [72, 137, 51, 136] Empirical This work Standard, GE BWR-4 U.S. 1972-1990 -50% [72, 137, 51, 136] Empirical This work Standard, WH 3LP U.S. 1971-1987 -94% [72, 137, 51, 136] Empirical This work Standard, WH 4LP U.S. 1973-1996 -31% [72, 137, 51, 136] Empirical Cooper [78] Various U.S. 1971-1996 Negative Empirical Rubin [79] Various U.S., France 1971-1996 -38% [75] Empirical Lang [48] Various U.S. 19 7 0-1 9 9 6 f -102% [73] Empirical Lang [48] Various World 19 7 0-2015if -23% [73] Cost guideline ORNL [92, 64] Standard, advanced U.S. 1989- 6%/2%- Cost guideline EMWG [57] Standard, advanced U.S.->4Worldh 2007- 6% Future growth' McDonald [18] Various World 1975-1993 6% [19]1 Future growth Abdulla [59] Standard, SMR U.S.---)Worldh 2012- 1 0 %/W Expert elicit. Future growth IPCC [93] Various, advanced World 2000-2100 4-7%k Future growth IPCC [71] Various, conventional World 1990-2100 1-5%1 Future growth IAEA [58] Various, advanced World 2000-2100 7-10%1 'Abbreviations: CE = Combustion Engineering; GE = General Electric; WH = Westinghouse; BWR = boling water reactor; LP = loop; SMR = small modular reactor. bBased upon first year of commercial operation. cDue to its wide recognition and use in literature, the term 'learning rate' is used here to summarize industry-wide rates of cost change. Arguably this term should be reserved for firm-level production. Positive rates indicate positive learning (cost decrease); negative rates indicate negative learning (cost increase). dNo reference is provided for studies which do not use or do not disclose a data source. eMean value of four U.S. standard designs. Median value is -50%. fWe omit the result for an earlier period, which includes turnkey project prices that do not reflect actual costs. 9Equipment and labor, respectively. hThe U.S. industry is the basis for this study, but findings are applied globally. Often cited as empirical, this study instead reports assumed learning rates [142, 143, 144]. 'Expert responses ranged from 0-1% to 15-20%, with 10% as the median. kBased upon climate stabilization scenarios of the IPCC Climate Change Synthesis Report [93] as studied by [145]. 'Based upon scenario AlT of the IPCC Special Report on Emission Scenarios [71] as studied by [58]. Table 3.1: Nuclear learning rates components. QE is the electrical output of the plant. (3.3) i1 Fig. 3-2 depicts the effects of the most important accounts, and SI section S3-1 provides a full listing of the 61 cost accounts. The overall trend is cost increase, with few accounts experiencing minor cost decline, suggesting that any positive learning effects are outweighed by other factors. Further, a diversity of accounts contribute to the total cost escalation, indicating that the cause cannot be easily attributed to any one source. However, grouping accounts into direct and indirect categories, we identify that changes in indirect expenses were the greatest. 63 Direct and indirect cost accounts, i Home Office Services Field Job Supervision Temp Construction Fac Payroll Insurance & Taxes Nuclear Steam Supply Sy Construction Tools & Equip Air Water + Steam Serv Sy Reactor Containment Bldg Other Reactor Plant Equip Elect Struc + Wiring Contnr Turbine Generator Field QA/QC Turbine Room + Htr Bay U I ndirect Cost Yardwork E Direct Cost Mechanical Equipment 0% 5% 10% 15 % 20% 25% Contribution to cost change Figure 3-2: Nuclear plant cost change, 1976 to 1987. Indirect cost accounts comprise 72% of the total cost change. The four largest contributors to cost increase are indirect accounts: home office engineering services, field job supervision, temporary construction facilities, and payroll insurance and taxes. Only accounts with a cost change contribution exceeding 1% are included (see Fig. S3-1 in the Supplementary Material for a full list of accounts). Indirect costs caused most (72%) of the cost increase during period 1 (1976-1987). But why did indirect costs rise so dramatically, while the modeled reactor design (Westinghouse 4-loop) remained the same? The literature presents many hypotheses, but little quantitative evidence. The account from EEDB [89] in 1988, the last year the database was updated, suggests a multitude of causes: prolif- eration of regulations, codes and standards; owner/designer overreaction to the rapid appearance of these regulations, codes and standards; rework caused by field interferences, constantly changing de- signs in response to new requirements and inadequate engineering-to-construction lead times; extreme precision required in analyses, coupled with inflexible design and construction quality assurance re- quirements; management preoccupation with regulatory inspection, enforcement personnel site visits and prudency reviews; and low worker morale, caused by all of the above. To quantify which aspects of the technology were most responsible for the rise in indirect expenses, we delve further into the EEDB model and attribute indirect costs to plant components. Indirect costs are comprised of construction support activities such as engineering, administration, and construction supervision. Direct costs are the costs of materials and labor needed for physical components like reactor equipment, structures, control and monitoring systems, and assemblies. We estimate the amount of indirect costs incurred by each direct cost componentby aggregating the indirect expenses 64 into cost "bases", B(, according to the construction inputs that incur them: site labor, materials, factory equipment, and safety-related components. We assume each direct account is responsible for a share of the indirect costs base that is the proportion of its construction inputs to the total construction inputs for each input category, 6>. For instance, indirect costs incurred by the fuel storage building are proportional to the ratio of fuel storage labor, material, equipment, and safety-related costs to total plant costs in these four categories. The total indirect cost incurred by an account, Zi, then, is the sum of the products of each account's share and the indirect cost base for each cost category: Zi = C" BI (3.4) We assume that the ratio of indirect to direct costs is constant within each of the four input categories across all accounts, mimicking the methods used in EEDB. A complete description of the method and our assumptions is provided in SI section S3-2. Fig. 3-3 shows the results of redistributing indirect costs to individual plant components. We find that labor-intensive components tend to incur more indirect costs. Several safety-related components, most importantly the containment building, also incur high indirect costs. These results reflect the observation that components with more and longer installation steps also require more engineering and construction supervision to ensure that these steps are completed according to standards. The three plant components that were most influential in causing indirect cost change-the nuclear steam supply system (NSSS), the turbine generator, and the containment building-also contributed most to direct cost increase. In section 3.3.2 we focus on direct containment building costs in further decomposing cost changes into underlying engineering choices and productivity trends because we can model these costs using historical and recent design drawings. The use of design drawings enables us to extend our analysis from the historical period 1 (1976-1987) to the year 2017. We also discuss why the main conclusions we draw hold for total containment costs, not just indirect costs, using the indirect cost data currently available, while acknowledging uncertainties. 3.3.3 Sources of Cost Change in Containment Buildings While our analysis of total plant costs in section 3.3.2 has identified rising indirect costs and expensive plant components as drivers of cost change, another important question is which technical and economic developments have caused these drivers. To address this question we develop a detailed engineering cost model of the containment building, the second most expensive component of a nuclear power plant, and study the drivers of its cost change. Although this case study can only explain a fraction of overall plant cost change, we can begin to understand the underlying mechanisms. We focus on 65 Direct cost accounts, i E D 1.6 Reactor containment Nuclear Steam Supply Sy building ($ Reactor Containment Bldg *0 1.5 Other Reactor Plant Equip 245/k ~ - o Q Turbine GeneratorFuel storage U- 1.4 0 Electrical - Air Water + Steam Serv Syt building ($29/k We) 0 distribution Elect Struc + Wiring Contnr U- 1.3 structures - Mechanical Equipment U- 0 Reactor instruments ($113/kWe) Turbine Room + Htr Bay 1.2 & control ($47/kWe) Yardwork t50 ) T Service systems Other Turbine Plant Equip Control Rm/D-G Building -C 1.1 ($111/kWe) - Prim Aux Bldg + Tunnels *0 Station service U Reactor 0 equip, Condensing Systems1.0 - equip ($31/kWe) ' . misc ($138/kWe) Radwaste Processing Power & Control Wiring 0.9 Heat rejection, Waste Process Building 0 mechanical equip Rx Instrumentation + Cntrl 5) 0.8 QV ($87/kWe) Feed Heating System o General Safeguards System " Direct Cost .E 0.7 Turbine generator components Main Heat Xfer Xport Sys ($175/kWe) Mn Steam + Fw Pipe Enc. 0.6 * Safety-related Fuel Storage Bldg * Indirect Cost, NSSS Components Station Service Equipment Attributed ($207/kWe) Instrumentation + Control - 0% 20% 40% 60% 80% 100% 0% 2% 4% 6% 8% 10% 12% Labor intensity (labor cost/total direct cost) Contribution to cost change (a) (b) Figure 3-3: Nuclear plant indirect costs, 1987, and cost change, 1976-1987. (a) Attribution of indirect expenses to the direct cost accounts that incur them reveals that labor-intensive components and safety-related components represent a disproportionately large share of indirect expenses relative to their cost. The containment building incurs more indirect expenses than any other component. Results are shown for year 1987, though are similar in 1976; (b) Indirect cost accounts comprise 72% of the total cost change between 1976 and 1987. The containment building is responsible for the largest share of cost change due to indirect expenses. Only accounts with a cost change contribution exceeding 1% are included. light-water reactors because this is the dominant technology in the U.S. Containment buildings are airtight steel and concrete structures that form the outermost layer of a nuclear reactor. They are designed to shield the environment from the escape of radioactive gases or materials during an accident, to protect the reactor against missile and aircraft impacts, and to provide structural support for the nuclear steam supply system. We focus on the containment for two reasons: 1) as the largest safety-grade structure of a nuclear power plant, comprised mostly of steel and concrete, the containment constitutes a useful lens to study field construction challenges and changing safety paradigms that also affect other plant components; 2) as a symmetric structure with comparatively simple geometry, design parameters can be more easily extracted from publicly available design drawings than for other components. Our cost model separately accounts for material and labor costs to construct the foundation, shell, 66 and dome of the containment. We write total containment construction costs as Ccontainment = Cfoundation + Cshell + Cdome (3.5) We focus on steel, rebar and concrete and omit the costs of other, less costly materials (e.g. materials used for formwork). We use thin shell approximations to estimate the volumes of individual structures (see SI section S3-3). To study the effects of labor productivity trends on costs, we develop a model with deployment rates of construction materials as variables. For structures made of materials i, this deployment rate is the ratio of material volumes V to the quantity of labor (in person-hours) needed to deploy these volumes, ri: Vi = __ This modeling choice results in a cost equation for direct construction costs of the form shown in Eq. 3.6 below, where costs are modeled as a sum of products of structure volumes V, material prices pi, volumetric material fractions fi, and per-volume labor costs ('). Vi Ccontainment = V fipi + -- (3.6) We =1QE ( Vi In SI section S3-5 we use a simple expansion of this model to estimate indirect containment costs and to draw conclusions about overall plant construction costs. We select our periods of study based upon availability of data and to align with major shifts in U.S. nuclear construction: From 1976 to 1987 (period 1), the era of transition characterized by changing public opinion, rising nuclear regulations, and the events surrounding the Three Mile Island accident. From 1987 to 2017 (period 2), the recent era characterized by protracted construction periods, devel- opment of new generations of reactor design, the long hiatus in nuclear project development, and an attempt to revive the nuclear construction industry. Populating our cost model with values from these periods, we can ask how much each variable contributed to the cost increase of the containment building. Even with a cost model in hand, attributing cost increases to particular variables is non-trivial. We draw on a recently developed method for decomposing the sources of cost changes in a technology [20]. We model the cost of a technology as a sum of a set of cost components, C(r) = Zi Ci(r), if these cost components can be written as products of functions (giy) of a vector of explanatory variables (r) that affect costs. The elements of this vector represent material prices, wages, and engineering design parameters. The central challenge addressed by this method is to estimate the individual contributions of multiple changing variables to total cost change when data are only available at discrete times. This situation precludes integration of the total differential, and requires choosing a suitable approximation of the contribution of the y-th variable to cost change: 67 /gjy(rfy)\ (ACy)(t1 , t2 ) iC I n , (37) giy(r j where di is a representative value of the i-th cost component during the time period [20]. We refer to changes in the variables in vector r as low-level mechanisms of cost change. As shown in Figs 3-5 and 3-6, some low-level mechanisms were significantly more influential for cost increase than others. These mechanisms include changes in material deployment rates, structure thicknesses, and steel prices. However, the importance of these mechanisms changed over time. During period 1 (1976-1987), the design of the containment structure stayed the same, and cost increase was caused primarily by declining deployment rates. Although concrete and steel worker productivity declined by comparable amounts (-40% for steel, -50% for concrete during the 1976-1987 period), steel worker productivity made a larger contribution to cost increase due to the higher wage paid to steel workers. We study nuclear productivity decline in more detail in section 3.3.4 During period 2 (1987-2017), a new containment design was adopted by Westinghouse (the AP- 1ooo), and the resulting changes to dimensions, material usage and labor needs drove cost increase. The switch from active to passive cooling, a design that reduces the need for operator intervention during emergencies by taking advantage of natural forces, required the separation of the steel liner from the concrete shield building to form two free standing structures. This change enabled natural air convection between the two layers, but also required thicker structures. Layers previously acting together to resist external and internal forces now needed to hold up independently [146, 147, 148]. The thickness of the steel shell, which was five times greater in 2017 as it was in 1987, made the single largest contribution to cost increase (60%). Period 2 caused the majority of cost increase (80%) during the 1976-2017 period. Our results illustrate trade-offs that can result from innovations that affect many variables simulta- neously. Switching to a free-standing containment steel vessel in period 2 allowed the use of cheaper steel, as well as more rapid steel shell deployment, but the cost-reducing effect of these changes was offset by increasing structure thicknesses and the resulting higher overall material and labor costs. Avoiding a cost increase over the 1987-2017 period despite increased commodity use would have re- quired massive improvements in labor productivity (a ten fold increase in steel and rebar deployment rates, over the 1987-2017 period, in addition to a 20% increase in concrete deployment rates). While our analysis of direct cost change in containment buildings covers only 3-4% of total plant costs in 1976 and 1987 , the costs of civil works in total account for 30-50% of total nuclear power plant costs [62]. The conclusions drawn from this case study-e.g., on the effects of increased commodity usage-can therefore add insight on drivers of cost change in other field-constructed plant components such as spent fuel handling buildings, turbine generator buildings, and cooling towers. 68 Contrary to the years 1976 and 1987 we do not have information on total indirect costs in 2017 because the Vogtle plant is still under construction. We exclude indirect containment costs in Fig. 3-5 but examine the effects of currently available indirect cost data on total containment cost change in Supplementary section S3-5. Given the range of possible ratios between direct and indirect costs, total containment costs depend sensitively on assumptions regarding indirect costs. However, mechanisms that are influential for direct containment cost change (deployment rates, steel prices, structure thick- nesses) also tend to be influential for total cost change. Deployment rates are slightly more important because they affect a larger fraction of total costs. Commodity prices become less influential. We explore the effects of uncertainties in material deployment rates, commodity prices, and wages using a sensitivity analysis (see SI section S3-6). Cost change results are most sensitive to uncertainties in variables related to the use of steel (wages, prices and deployment rates), but our major conclusions are unaffected by these uncertainties. 3.3.4 Evaluation of Nuclear Construction Productivity In section 3.3.2 we find that declining construction productivity was the main source of containment cost increase in period 1 (1976-1987). What caused this decline? Previous work points to a general decrease in U.S. construction productivity during the period [62, 511, but has not looked at the nuclear industry specifically. To study this, we look at the evolution of material deployment rates in nuclear construction. Using data from EEDB reports [82, 83, 84, 85, 86, 87, 88, 89] and recent APiooo engineering construction reports [149, 150], we compute the ratio of the volume of materials (steel or concrete) installed to the total hours of labor needed for installation. We compare these deployment rates to two benchmarks: An index of material deployment rates in the construction industry as a whole, and deployment rates assumed in nuclear industry cost estimation guidelines. Material deployment rates in the construction industry decreased over the period of study, falling about 14%, as shown in Fig. 3-4. Nevertheless, deployment rates in nuclear construction declined more dramatically, with a precipitous drop between 1979 and 1980 following the Three Mile Island accident. Compared with the construction industry at large, nuclear deployment rates declined five to six times more quickly, and was a primary cause of nuclear cost increase. Labor interviews provide insight into some of the causes of declining productivity [151], pointing to problems experienced in the field. Craft laborers, for example, were unproductive during 75% of scheduled working hours, primarily due to construction management and workflow issues, including lack of material and tool availability, overcrowded work areas, and scheduling conflicts between crews of different trades. Material deployment rates in the U.S. nuclear industry have been considerably lower than those assumed by the industry for cost estimation purposes (e.g. EMWG [57]). Industry average rates in the post-Three Mile Island period were two to three times slower for steel and concrete. More recently, rates at the Vogtle and VC Summer project sites have been three to four times slower for steel, and eight 69 1.6 - - - - - - - - - - - - - - - - - - - - - -- Nuclear industry cost estimation guidelines (2007) 1.4 4-1 0 Nuclear industry cost 0 estimation guidelines (2007) 1.0 SNuclea - - - ndustry -0.8 N median ~0 - 0. . Summer & Vogtle 0.2 . -0- Nuclear steel - Nuclear concrete -0- U.S. construction industry 0.0 1976 1986 1996 2006 2016 Figure 3-4: Historical construction productivity change in the nuclear industry and at large. Material deployment rates are normalized to their distinct 1976 values (and nuclear and non-nuclear productivity are not equal in 1976). Nuclear productivity declined sharply in association with the Three Mile Island accident and has decreased further in the decades following. Deployment rates used in cost estimating [57i] are disjoint from the last five decades of experience. Data are derived from EEDB [82, 83, 84, 85, 86, 87, 88, 89], recent engineering reports from the VC Summer and Vogtle projects [149, 15o], and the U.S. Bureau of Labor Statistics [138]. to thirteen times slower for concrete. This disparity between projections of productivity and actual experience has contributed significantly to cost overruns. These trends are observed despite recent efforts to improve productivity through modular design. Instead of using standard reinforced concrete, which is constructed on-site using elaborate formwork, the shield building in the AP-iooo is comprised of prefabricated steel-plate composite (SC) modules. SC modules have two steel layers and tie bars that act as concrete reinforcement, reducing the time needed for formwork and rebar placement [152]. Smaller modules are assembled into larger modules on-site and then lifted into place. However, placement is only one of many steps needed to install a module, which also involves welding, piping, cabling, and other tasks. Although SC modules were used in two major structures on the AP-iooo nuclear island (the contain- ment and auxiliary buildings), the effect of modular construction on the average steel deployment rate across the nuclear island was not enough to raise productivity over previous years. Poor workmanship, 70 1976-1987 (2 A Steel shell wall thickness T_ 4% oWf tota-l) 1987-2017 (76% of total) Overall: 1976-2017 (100%) A Concrete depl. speed A Steel dome, thickness A Foundation height A Concrete fraction I A Struct. steel depl. speed A Base shape corr. factor A Concrete worker wage A Shield building wall height A Ironworker wage I A Steel module price markup A Rebar price I A Conversion efficiency A Concrete price A Shield building, inner radius A Heat output A Steel shell found. thickness Steel shell dome, outer height A Steel shell, inner radius A Steel shell wall height A Steel shell, outer radius A Base shape corr. factor A Shield building, outer radius A Top shape corr. factor A Shield building thickness Cost increase: Cost increase: Cost increase: - A Steel shell depl. speed 25% from 1976 61% from 1987 102% from 1976 - A Steel price -60 -40 -20 0 20 40 60 -60 -40 -20 0 20 40 60 -60 -40 -20 0 20 40 60 % contribution to containment building cost increase Figure 3-5: Percentage contributions of containment building low-level mechanisms to the increase in direct containment costs during period 1 (1976-1987, left), period 2 (1987-2017, middle), and during both time periods (1976-2017, right), listed in the order of their contributions to total cost increase during the 1976-2017 period. Period 1 caused 24% of total containment cost increase, while period 2 caused 76%. Shape correction factors ('corr. factors') account for the change in containment design and geometry in time period 2. The full set of variable names is given in SI Table S3-1. Shape correction factors are explained in SI section S3-3. and the extra steps needed for quality control of the modules, are among the possible causes of low productivity [139]. 3.3.5 High-level Mechanisms of Containment Building Cost Change What were the drivers behind low-level mechanisms of cost change discussed in section 3.3.2? A com- mon view is that safety regulations have increased the costs of nuclear power plant construction [140, 77, 153, 132]. Here we examine this view more closely, using our case study of the containment building to attribute changes in the variables (low-level mechanisms) to higher-order processes that likely caused these changes (high-level mechanisms). The goal of this analysis is to provide a quantitative estimate of the role of engineering design in changing containment costs, as compared to external factors such as regulatory oversight. 71 1976-1987 (24% of total) 1987-2017 (76% of total) Overall: 1976-2017 (100%) Structure Thicknesses Deployment Rates Material Composition Laborer Wages Thermoelectric Design Other Plant Geometry Material Prices 0 50 100 0 50 100 0 50 100 % contribution to containment building cost increase Figure 3-6: Aggregations of the percentage contributions shown in Fig. 3-5 according to variable types. 'Structure thicknesses' account for the contributions of changing thicknesses of steel and concrete layers, 'deployment rates' account for changing steel and concrete deployment rates, 'material composition' represents the changing concrete fraction in reinforced concrete, 'wages' include changing labor rates for steel and concrete workers. 'Thermoelectric design' accounts for changes in the thermal power output and the efficiency of the plant, 'other plant geometry' accounts for the changing geometry of the containment building (see shape correction factors in Fig. 3-5), and 'material prices' account for changing concrete and steel prices. Variable types are listed in the order of their contributions to total cost change over the 1976-2017 period. We assign changes that require significant modifications of the containment building design and construction process to the mechanism 'Research and development (R&D)'. While construction projects are inherently site-specific and on-site adjustments are common, the mechanism R&D accounts for more fundamental changes that require longer term, off-site activities. For example, changing the design by separating the steel liner and concrete shield building required years of R&D by Westinghouse, as documented in containment-related patents and journal papers published during the 1970-2017 period (see Supplementary Table S3-5). To account for different ways in which safety requirements can induce cost change we define three types of R&D: 1) 'R&D, technical performance' accounts for changes made to enhance plant performance as a whole, for instance by improving operational simplicity and constructibility; 2) 'R&D, flexible safety' (R&D FS) accounts for changes made in response to safety regulations specifying a performance metric (e.g. the design pressure of the containment steel vessel), but not a prescribed set of engineering parameters (e.g., the thickness of a wall) to achieve this metric. While these changes were often motivated by safety requirements, companies still had flexibility regarding the design choices to make to ensure compliance. 3) 'R&D, prescribed safety' (R&D PS), in contrast, refers to changes in response to more narrowly defined, prescriptive safety requirements. An example is provided by Westinghouse's increase of the steel fraction in the concrete shield building in response to an NRC aircraft impact assessment. Although Westinghouse had considered their design suitable to withstand a commercial aircraft impact, the company was asked by NRC to revisit the design, leaving few options but to use more steel [154]. 72 1976-1987 (24% of total) 1987-2017 (76% of total) Overall: 1976-2017 (100%) R&D, technical performance Non-safety R&D, flexblessafet R&D, prescribed safety Safety R&D, flexible safety Unlearning despite doing Process interference, safety Other -20 0 20 40 60 -20 0 20 40 60 -20 0 20 40 60 % contribution to containment building cost increase Figure 3-7: Percentage contributions of containment building high-level mechanisms to cost increase during period 1 (1976-1987, left), period 2 (1987-2017, middle), and during the overall time period (1976-2017, right). Safety-related mechanisms (light bars) caused 88% of direct cost increase in the first period, 45% in the second time period, and a little over half (54%) of total direct cost increase over the 1976-2017 period. The significant contribution of structure thicknesses to cost increase in period 2 (see Fig.3-6), although rooted in a flexible safety requirement, does not result in a large contribution of the mechanism 'R&D, flexible safety' to cost increase. This is because this mechanism also caused cost-decreasing effects (decreasing steel price, increasing steel deployment rates) during the same time period. We define three additional high-level mechanisms to account for changes driven by non-R&D related processes. The first mechanism, 'Process interference, safety' (PIS)', represents the effects of on-site NRC and other safety-related personnel on the construction process. Construction supervision, quality assurance and control by NRC regulators can interfere with regular construction workflows, thereby slowing productivity (see Supplementary Table S3-8). The second additional high-level mechanism represents a negative form of learning-by-doing, where the performance of workers decreases during routine construction activities. We refer to this mechanism as 'Unlearning-despite-doing' instead of the previously used 'negative learning' [75] to draw a distinction between UDD and learning-by-doing as a concept in economic theory. Learning-by-doing is tied to the notion that change accumulates as a result of an activity, such as working [155]. UDD, in contrast, attributes performance decreases to parasitic processes (e.g. decreasing morale) that did not originate in construction activities, but were also not counteracted by them. These processes may have diminished productivity gains from problem-solving activities during routine, sequential work steps, which is often seen as the source of learning-by-doing [15s, 156]. Finally, we use the mechanism 'Other' to refer to changes that were affected by broader economic trends rather than by the nuclear industry itself, such as wages and commodity prices. We describe the assignment process in SI section S3-7, and a complete list of assignments for the low-level mechanisms shown in Fig.3-5 is given in SI Table S3-7. In time period 1 (1976-1987), we assign material deployment rates to 'R&D, FS', UDD, and PIS to account for several parallel developments that affected labor productivity Following the Three Mile Island accident, NRC regulations required 73 increased documentation of safety-compliant construction practices, prompting companies to develop quality assurance programs to manage the correct use and testing of safety-related equipment and nuclear construction materials. In contrast to the more informal practices used in the 196os, new NRC quality assurance standards became increasingly specific in the 1970s, regulating construction steps such as concrete placement and rebar testing and leaving limited implementation flexibility [140]. We therefore categorize the slowdowns in productivity caused by these programs as R&D PS. Time period i also saw the inception of the NRC's resident inspector program, putting NRC directly on site to monitor plant and construction activities [157]. Since these interferences directly affected construction practices [157, 158], we categorize material deployment rates as PIS. Finally, we account for productivity changes that cannot be fully explained by safety by assigning deployment rates to the mechanism'UDD'. A worker survey from six nuclear power plants constructed during time period 1, for instance, attributes 27% of unproductive time to a lack of material and tool availability, indicating supply chain management issues that are largely independent from safety requirements (e.g. inadequate supplies kept in stock [151]). While the dimensions of the containment building were constant in time period 1 (1976-1987), they changed in pursuit of general performance goals (design simplicity, constructibility) and safety goals in time period 2 [159]. Based on the information contained in patents and journal papers (Tables S3-5 and S3-6), we assign thicknesses, radii and heights equally to R&D TP and R&D FS. Since the use of cheaper steel is a consequence of the design change, the steel price decline in period 2 is also assigned to R&D TP and R&D FS. We also change the assignment of the steel shell deployment rate to R&D TP and R&D FS. in time period 2. This choice reflects the design change to the free standing steel vessel that allowed faster steel deployment, which was driven by both general and safety-related performance goals. Concrete and rebar prices are assigned to 'Other'. In time period 2, the only mechanism assigned to R&D PS is the concrete fraction in the shield building, reflecting NRC's requirement for a more robust shield building to withstand the impact of a commercial airplane crash [154]. A list of other assignments and underlying data sources is given in SI Tables S3-7 and S3-8. The contributions of high-level mechanisms to containment cost increase are shown in Fig. 3-7. Altogether, different safety-related mechanisms contributed roughly half to direct cost increase over the 1976-2017 period. Contrary to the common assumption that safety requirements were the most important cost-increasing factor, other developments that were not prescribed by the regulator were also influential, including decreasing productivity, and design changes to improve cooling performance. Prescriptive safety requirements were more influential during time period 1 than they were during time period 2, reflecting the trend away from prescriptive safety requirements in the 198os. 'Unlearning- despite-doing' and 'Process interference, safety' were important in both time periods, together causing almost 70% of total cost change during the 1976-2017 period. The large influence of these procedural and often site-specific mechanisms points to the importance of innovation efforts in these areas. 74 3.3.6 Opportunities for Future Cost Reduction in Nuclear Construction We conclude our analysis by examining scenarios for future reductions in containment construction costs. The goal of this analysis is to understand whether some of the cost-increasing factors largely external to initial plant design (e.g. on-site productivity) can be addressed through targeted innovation efforts, or whether these factors are likely to continue causing cost uncertainty. Each scenario corre- sponds to a set of changes to the variables in the containment cost model relative to their values in 2017. These 'prospective low-level mechanisms' represent the estimated effect of different innovation efforts to reduce costs. These efforts affect multiple variables, either directly or indirectly through interde- pendencies between variables. The cost change model serves as a tool to compare the contributions of individual variable changes to cost reductions. We also assign the prospective low-level mechanisms to different types of higher order innovation processes (high-level mechanisms) to estimate their potential contribution to future cost change. We consider three scenarios (see SI Table S3- 9 for assumptions made in each scenario). In scenario 1, we assume that cost improvement is pursued along multiple dimensions, reducing material usage, commodity cost, wages, and increasing productivity. This scenario is represented as 20% change of all variables in a cost-reducing direction (e.g., structure thicknesses and commodity prices decrease by 20%, while deployment rates increase by 20%.) Scenario i is conceptual in nature, since no real-world design change will induce equal percent-changes across all variables, and is meant to probe the relative impact of different variables on total containment costs. We change the conversion efficiency and the concrete fraction in reinforced concrete by less than 20% (by 2% and 5%, respectively) to reflect current engineering constraints, for instance in increasing conversion efficiencies without major design changes (see SI section S3-8.1). While scenario 1 is designed to show the effect of changing multiple variables without specifying design changes, scenarios 2 and 3 represent more targeted efforts to increase productivity. In scenario 2, we assume that on-site deployment rates improve due to adoption of advanced manufacturing and construction management techniques. Research interest in employing advanced manufacturing in large-scale construction projects has increased in recent years, particularly for concrete structures [16o]. We draw on a recent publication demonstrating 3D printing of a dome-like structure to estimate improved concrete deployment rates and capital costs of automation equipment [161] (see SI section S3-8.1 and SI Table S3-9 for details). Progress in developing automated rebar placement systems has been slower, and we therefore turn to innovations in process management (e.g. optimized rebar delivery and placement planning as demonstrated in [162]) rather than to hardware innovations. Scenario 3 is focused on advanced construction materials. Due to their higher strength, these materials have been shown to reduce commodity use and on-site rebar congestion in high-rise buildings and bridges (e.g., [163, 164]). We model a combination of high-strength reinforcement steel (HSRS), and ultra high-performance concrete (UHPC) as these materials have been shown to achieve greatest 75 strength at lowest cost when used together [165]. HSRS provides up to 40% more yield strength ((i.e., the stress at which a predetermined amount of permanent deformation occurs) than conventional rebar [166], which is equivalent to a proportionate reduction in rebar amounts per unit of concrete volume. We model the use of HSRS ('Grade loo rebar') as an increase in fcon, the volumetric concrete fraction in reinforced concrete. UHPC, a steel fibre-reinforced cement composite material, has compressive and tensile strengths 2-3 times that of standard concrete [167, 168]. These properties allow a reduction in concrete volumes for the same structural stability. Both materials, however, are more expensive than standard concrete. We account for these price differences accordingly Lastly, we model the increase in productivity as a linear function of the rebar volume reduction on the construction site, which leads to a 5% increase in material deployment rates. We assume that all deployment rates are equally affected by the reduction in on-site rebar congestion. The model is described in SI section S3-8.3. The prospective scenarios achieve cost reductions of 30-50% relative to 2017 containment costs, though no scenario leads to a reduction relative to 1976 costs. In scenario 1, reductions in rebar use in reinforced concrete (represented by f.con) and in steel worker wages are more influential than all other changes, together causing over 40% of total direct cost reductions (see Fig.3-8). Even in a hypothetical scenario where steel for the containment vessel was free, changing the rebar fraction in the shield building and foundation would remain the dominant cost changing effect due to labor costs. These results demonstrate the large-scale dimensions and labor intensity of nuclear structures which inherently limit materials-related cost-reduction opportunities. Scenario 2 results in a reduction of containment construction costs to approximately two thirds of estimated costs in 2017 (-35%). This effect is driven primarily by faster concrete deployment. Capital costs for automation equipment are relatively insignificant. Scenario 2 nevertheless represents a 30% cost increase relative to 1976 costs. Even if we assume steel productivity reaches 1976 levels (peak levels in our record) without specifying a technology to achieve this, costs would still increase by almost 20% over 1976 costs. Despite the large reduction in commodity usage, scenario 3 only reduces costs by a little over one third (-37% from 2017 levels). This result is rooted in the high cost of UHPC. Due to the expensive fabrication of UHPC steel fibers, the price of UHPC is currently lo times that of standard nuclear concrete [169, 164, 163] (see SI section S3-8.1). There is also an approximately so% price premium on high-strength rebar relative to standard rebar. These prices partially offset the effects of reduced material intensity achieved through higher strength materials. Cost reductions in this scenario could reach 5o% if the costs of advanced materials reached current prices of nuclear commodities. Costs have declined in European countries that scaled UHPC production earlier [164], but similar cost declines have yet to be achieved in North America. The prospective analysis highlights the challenges in reducing the costs of a field-constructed, site- specific technology that also requires high safety standards. Scenario 1 shows that the rebar fraction 76 Scenario 1: Broad improvement Sce nario 2: Increased productivity Scenario 3: Advanced materials A Concrete fraction A Ironworker wage A Foundation height A Concrete worker wage A Shield building thickness A Concrete depl. speed A Struct. steel depl. speed A Steel shell wall thickness A Steel price - A Conversion efficiency A Steel dome, thickness - A Rebar price - A Steel shell depl. speed - A Steel module price markup - A Shield building, inner radius - A Concrete price - A Automation hardware - A Steel shell found. thickness - A Base shape corr. factor - A Top shape corr. factor - Steel shell dome, outer height - A Base shape corr. factor - A Steel shell, inner radius - A Steel shell wall height - A Steel shell, outer radius - A Shield building height - Cost reduction: Cost reduction: Cost reduction: A Shield building, outer radius - 54% from 2017 35% from 2017 37% from 2017 - Ii I A Heat output - - - 0 20 0 20 40 60 80 -60 -20 20 60 %c ontribution to containment building cost decrease Figure 3-8: Percentage contributions of containment building low-level mechanisms to cost reductions under three improvement scenarios representing innovation efforts focused on broad improvement across all variables, on productivity increases through automation and better construction management, and on advanced construction materials. In scenario 1 (left), several variables (structure thicknesses, deployment rates, and prices) improve by 20% (i.e. they change by 20% in a cost-reducing direction). In scenario 2 (middle), material deployment rates are assumed to improve due to automation and construction process management. In scenario 3 (right), standard nuclear concrete and rebar is replaced with advanced construction materials, which allows a reduction in commodity usage. The improvement scenarios are defined relative to the 2017 containment cost data used in Fig. 3-5. in the concrete shield building is twice as influential as other variables, yet changing this variable is difficult given current safety standards. Scenario 2 makes clear that advanced manufacturing methods would need to encompass both concrete and steel to deliver more significant cost reductions. Studying the the properties of printed metals (e.g., their microstructure, corrosion cracking, and irradiation effects) under nuclear operating conditions is an area of active research, but no commercially available product currently exists [170]. We use the assignment scheme presented in section S3-7.1 to relate low-level to high-level mech- anisms, but make two adjustments to account for the forward-looking focus of the analysis. First, we remove the high-level mechanism research & development, prescribed safety ('R&D PS') because we focus on improvements to individual variables that are unlikely to be directly prescribed by the 77 Scenario 1 Scenario 2 Scenario 3 R&D, technical performance R&D, flexible safety Learning-by-doing Knowledge spillovers Otler -20 0 20 40 60 -20 0 20 40 60 -50 0 50 100 %c ontribution to containment building cost increase Figure 3-9: Percentage contributions of high-level mechanisms to containment building cost decrease for different innovation strategies. The assignments of low-level to high-level mechanisms are given in SI Table S3-10. regulator. Second, we include an additional mechanism to account for the transfer of knowledge and hardware innovations from other technology industries to the nuclear industry. We refer to this mech- anism as 'Knowledge spillover' (KS). Knowledge spillover is similar to learning-by-copying in the sense that capabilities developed in one domain are adopted by another [171], but the process of doing so may go beyond copying. Nuclear companies adopting advanced materials will likely need to adapt their use to nuclear construction and inspection processes, rather than just copying what other industries do. We account for the latter by assigning changes due to the switch to advanced materials not only to to 'Knowledge spillover', but also to 'R&D, flexible safety'. (Note that knowledge transfer is also implicitly included as a high-level mechanism in our historical cost change analysis. However, due to lack of evidence for this mechanism having caused any of the cost-increasing changes in the variables we estimate its effect as zero and therefore do not include it in figures or discussion thereof.) We assign increases in material prices to the same high-level mechanisms that enable the switch to advanced materials (R&D FS and KS). In scenario i, where we do not specify a technology innovation to achieve the assumed variable changes, we use historical associations between low- and high-level mechanisms as our 'best guess' for the future. We find that all roads to cost reduction lead through different forms of R&D, while the contribution of other high-level mechanisms varies depending on the scenario. LBD is slightly more important when improvements to the construction process are adopted (scenario 2). Knowledge spillovers are less important for cost reductions in scenario 3 (advanced materials) than in scenario 2 because the cost decrease enabled by knowledge spillovers and the use of advanced materials is simultaneously diminished by the higher prices of these materials. 78 3.4 Discussion In this paper we examine assumptions about nuclear plant cost modeling using historical data from five decades to study the causes of rising nuclear construction costs in the U.S. We find a strong correlation between the increased application of nuclear technology and cost escalation, even among plants of standard design. Decomposing individual plant costs, we identify declining labor productivity and increasing commodity usage in safety-grade structures as major drivers of cost increase over time, which we show through a case study of the reactor containment building. The findings of this research lead us to revisit how expectations regarding technological improvement may have contributed to an underestimation of cost factors external to hardware design. While it is acknowledged that construction costs increased for nuclear plants generally, substantial reductions within a given design class ('nth-of-a-kind' plants) are still commonly expected in engineering cost models. We review nuclear cost estimating practices and industry growth projections, identifying a common expectation that learning effects drive down cost as experience grows [92, 64, 18, 19, 59, 71, 93, S8, 63, 62, 74]. The notion that improved plant designs can solve cost issues once new designs can be standardized and production scaled has driven substantial public and private R&D investment, but it is unclear what the net effect of such investment has been. While previous empirical evidence shows that costs rise with experience [75, 78, 791, our work demonstrates that this effect was also true for nth-of-a-kind (NOAK) plants of standard reactor technology in the U.S., indicating that cost reductions from standardization should not be expected as an inherent consequence of standardization or industry experience. However, our results should be interpreted within the context that not all plants within each standard design are perfectly identical and the design differences may have contributed to the unexpected cost increases. Rising costs are often assumed to be associated with increasing stringency of safety regulations (e.g. [140, 771. Here we estimate that prescriptive safety requirements caused roughly half of the direct containment cost increase between 1976 and 2017. Design choices and productivity declines, which are quoted less often as reasons for cost escalation, played a similarly significant role. We show that nuclear productivity has declined faster than that in the construction industry, and that actual productivity at nuclear construction projects is significantly below industry expectations. The widespread use of expectations that do not match actual experience may be a contributing factor in cost overruns, and suggests the importance of a comprehensive update using empirical, country-specific productivity data where available. We decompose plant cost change to identify causes of escalation in the 1970s and 198os. The large contribution of the costs of services and supervision tasks ('indirect costs') points to the potential of intensifying complications among safety regulations, design complexity (e.g. numbers of interacting parts), equipment supply chains, and construction management. However, further research would be needed to identify the causes. Previous papers have hypothesized that increased application of the 79 technology and cost escalation is a consequence of the complexity of nuclear technology, in which learning effects are negated by increases in system complexity associated with industry growth [172, 78, 80]. Our findings suggest that engineering and economic models used to project future construction costs and industry growth should be reexamined in light of the limitations of assumed learning rates and engineering design solutions to exogenous challenges. As an example, we find the expectation of cost declines from learning has likely contributed to historical overestimation of nuclear capacity growth for climate mitigation: in the year 2000, the IPCC projected the contribution of nuclear power to climate mitigation efforts would increase by 650GWe by year 2020, using an economic model with learning effects to estimate adoption of various generation technologies [71]. In 2018, the global nuclear operating capacity is 36oGWe, scarcely higher than the year 2000 capacity of 35OGWe [9, 10, ii] indicating a 64oGWe shortfall in expected capacity growth and associated emissions mitigation. By comparison, solar and wind power have met learning rate expectations outlined by the IPCC, and grew in combined capacity by 885 GWe from 2000 to 2017 [12]. We also examine avenues for future construction cost reduction, focusing in particular on variables that have contributed to rising costs in the past. Our scenarios suggest that the use of high-strength com- posite materials, as well as advanced manufacturing techniques, could significantly reduce construction costs below estimated 2017 levels. Knowledge transfer from other industries, for instance in the form of advanced manufacturing techniques, could be particularly impactful if it enables automated control of process parameters, thereby reducing the costs of human-led construction supervision. However, our scenarios require further examination to answer whether they might practically be achieved, and whether standards can be satisfied. while our analysis identifies the rebar density in reinforced concrete as the most influential variable for cost decrease, this is also a variable constrained by safety regulations, most notably the requirement for containment structures to withstand commercial aircraft impacts. 80 Supplemental Materials Sources of cost overruns in nuclear power plant construction Section S3-1 provides additional information on our analysis of total plant cost change during period 1 (1976-1987). Section S3-2 lays out the approach used for the attribution of total indirect costs to individual plant components. Section S3-3 describes the cost change model we use to quantify the factors ('low-level mechanisms') driving cost increases in U.S. light-water reactor containment buildings over the 1976-2017 period. Section S3-4 gives the data sources used to populate the containment cost model. Finally, section S3-7 describes how we assign the contributions of low-level mechanisms to cost change to higher-order improvement processes ('high-level mechanisms'), and provides data to support the assignments. S3-1 Full list of total plant cost accounts Fig S3-1 below gives the contributions of all 61 cost accounts modeled in the Energy Economic Database to 1976-1987 cost change [82, 83, 84, 85, 86, 87, 88, 89], supplementing the shorter list shown in Figs 3-2 and 3-3b. S3-2 Indirect cost attribution A series of Energy Economic Data Base (EEDB) reports commissioned by the U.S. Department of En- ergy estimates the cost of nuclear plants over time by modeling the cost of major direct and indirect construction components [82, 83, 84, 85, 86, 87, 88, 89]. Here we develop a method to aggregate the indirect expenses according the construction inputs that incur them, which allows us to assign fractions of total indirect costs to the plant components and activities incurring these costs. We then calculate the total cost change of each plant component as the sum of the contributions of its direct costs and share of indirect costs. The cost of each construction component is reported in a series of 61 EEDB accounts, i, with costs and quantities of the construction inputs: labor, materials, and factory equipment. Using explanations of the account modeling contained in the EEDB reports, we identify the indirect expense accounts that are incurred by craft labor hours, L, and each category of construction input costs, w. We assign craft labor hours of the accounts to variable ri. The total cost of an account, Ci, is the sum of its three cost cate- gories: labor, material, and factory equipment, expressed as Ci ,Iabo, + Cidma,, ,ial + Cij)factory equipment* We define a new cost category to designate whether or not accounts expenses are safety-related, Cj,,,f,,Y and we set its value equal to zero dollars for non-safety-related accounts, and equal to the 81 Direct and indirect cost accounts, i Direct cost i E- D Home Office Services Nuclear Steam Supply Sy Field Job Supervision Te p onstruc on Fac Reactor Containment Bldg Payro Insurance Taxes Other Reactor Plant Equip Nuclear Steam Supply Sy Turbine Generator Construction Tools & qui_ Air Water + Steam Serv Sy Air Water + Steam Serv y Reactor Containment Bldg Elect Struc + Wiring Contnr?t her Reactor Plant Equip Mechanical Equipment E ect Struc + Wiring Contnr Turbine Room + Htr Bay Turbine Generator Yardwork Field QA/QC Turbine Room + Htr Ba Other Turbine Plant Equip Yardwor Control Rm/D-G Building Mechanicl Eauipment Prim Aux Bldg + Tunnels Field ce Expenses Condensing Systems Control Rm/D-G Building Radwaste Processing OthOethre Tu Itrea ste PPlrnoc eEsqssiunigp - Power & Control Wiring Plant Startup & est Waste Process Building Prim Aux Bldg + Tunn Is Rx Instrumentation + Cntrl Home Office BWA Heating System Conde nsirlt Systems Feed Rx Instrumentati +sCntr- Safeguards System Permits Ins & Local Taxes Main Heat Xfer Xport Sys Waste Process Building Mn Steam + Fw Pipe Enc. Power & Control Wiring Bldg Mn Steam + Fw Pipe Enc. Fuel Storage Instrumentation + control Station Service Equipment Feed Heating System Instrumentation + Control Fuel Storage Bldg Reactor Equipment Reactor Equipmerit Equip Com munications Equip Communications Home Office Constr Mg mt Turbine Plant Misc Items Safeguards System Admin + Service Big Station Service Equipment Switchgear Main Heat Xfer Xport Sys Structures Turbine Plant Misc Items Wastewater Treat Equip Reactor Plant Misc Items Switchgear Protective Equipment Protective Equi ment Ultimate Heat Sink Struct Transportation& Lig Eqpt Treat Equip Structures Wastewater Security Building Transportation & Lift Eqpt Admin + Service BIg Security Building Furnishings + Fixtures Emerg Feed Pump Bldg Switchboards Furnishings + Fixtures Technical Support Center Waste Water Treat Switchboards Ultimate Heat Sink St ruct Technical Support Center Manway Tunnels Rca) Waste Water Treat Non-Essen Swgr Bldg Manway Tunnels (Rca) Fire PuP House Fuel Handling + Storage Contain Hatch Shld U Indirect Cost Non-Essen Swgr Bldg * Direct Cost Transportation Fire Pump House Pipe Tunnels Intk Contain Hatch ShIdContr Rm Emg Air - Direct Cost " Indirect Cost, - Emerg Feed Pump Bldg Pipe Tunnels Reactor Plant Misc Items Elec. Tunnels Attributed Fuel Handling + Storage Contr Rm Emg Air Intk 0% 5% 10% 15% 20% 25% 0% 2% 4% 6% 8% 10% 12% Contribution to cost change Contribution to cost change (a) (b) Figure S3-1: Contributions of EEDB cost accounts to cost change during period 1 (1976-1987) for (a) separate direct and indirect cost accounts and (b) attribution of indirect expenses to direct cost accounts that incur them. total account cost, C,, for safety-related accounts. First, we compute the "indirect cost bases," the total indirect cost incurred by each cost category and 82 by craft labor hours. In Eq. S3-1 we calculate the value of B 1,, the indirect cost bases for the four cost categories, as the sum of total indirect costs incurred by the respective input categories, as described in the EEDB reports. The craft labor hours cost base, BL, is comprised only of field job supervision, as shown in Eq. S3-2. In Eq. S3-3 we compute the amount of indirect costs incurred by and assignable to each direct cost component. For each cost category, an account is responsible for a share of the indirect cost base that is the proportion of its cost to the total direct cost of that category, or . Similarly, an account is responsible for a share of the craft labor indirect cost base that is the proportion of its hours to the total direct craft labor hours, or i. The total indirect cost incurred by an account, then, is the sum of the products of each account's share multiplied by the indirect cost base for each cost category and for craft labor. This computation assumes that indirect expenses are incurred uniformly by dollar and by labor hour. Variable Unit Cost of account i in cost category e Cio $ Total cost of account i Ci $ Craft labor in account i Ti PH Indirect costs incurred by category Be, $ Indirect costs incurred by craft labor BL $ Indirect costs incurred by account i Zi $ PH = person-hour Equations: B&)a Ci (S3-1) BL Cifield.job.supervision (S3-2) Zi= EB + i BL V i E 9J (S3-3) Indices: i nuclear construction accounts w cost categories: {site labor, material, factory equipment, safety-related} L site labor 83 Sets: =-s et of accounts i that are direct costs (e.g. containment building, nuclear steam supply system, turbine generator) - = set of accounts i that are indirect costs (e.g. home office engineering services, field job supervision, payroll taxes & insurance) J, = set of indirect cost accounts i that are incurred by direct costs in category co (e.g. for osite.labor: payroll insurance & taxes; for tosafety: field office quality assurance, home office quality assurance) S3-3 Containment cost model We model containment building costs as the sum of labor and commodity costs incurred by the foundation, the shell and the dome. We focus on structures where the geometry can be modeled based on publicly available design drawings. Ccontainment = Cfoundation + Cshell + Cdome (S3-4) We model the foundation as a circular area delineated by the outer radius reon0ut of the containment shell (i.e. the radius of the concrete layer on the outside of the shell). We focus on concrete, rebar and steel, and omit wood and metal used for formwork due to data limitations. This decision simplifies our model without significantly altering the results of the cost change analysis as formwork costs contribute less than 5% to individual containment building component costs. To model specific costs (in S/We) we divide the cost equation by the electrical power output of the plant, QE. We write Cfoundation = 2 -[(rconout ) 7(hfoundfcopcon + hfound(l - fcon)PreSte el) QE +2lrrsteeloutpsteeldfoundsteelg3+ TsteelWsteel93 + TreSteelWreSteel + TconWcon] (S3-5) where the first line represents material costs and the second line represents labor costs. r 0 out is written in units of m, material prices are written in USD/m3 , task durations ri are written in hours, and wages wi are written in $/hour. A detailed list of all variables is given in Supplementary Table S3-1. The shape correction factor g3 accounts for the switch from a flat steel plate at the bottom of the containment, which is used in 1976 and 1987, to a hemispheroid in 2017. We address this design change by setting g3 = 0 in 2017, and account for the steel bottom layer in the cost equation for the dome instead (see equation S3-3). This allows for a simple formulation of the model despite the geometry 84 change as we can "switch off" a cost component instead of revising the entire equation. Since we want the duration of individual construction steps to be dependent on variables determining the geometry of the foundation, we can break down task durations ri further and write V= - d(S3-6) Vsteel Vsteel s (S3-7) Ts tee / Substituting for ri, we get Cfoundation = - (rcon.out ) 2 7(hfoundffconPcon + hfound(1 - fcon)PreSteel) QE VsubCon VsubConSteel + Wcon + WreSteel] (S3-8) vcon vresteel Variables vi represent the speed of deploying concrete, steel, and rebar in m3/person-hour. Variables Vi represent the total amount of concrete, steel and rebar used for the substructure in m3 . Since VsubCon is equal to r 20 o~ rdconfcon we can simplify the above equation and write Cfoundation - ( [fconPcon + Presteel - fconPresteel + WreSteel qQR VreSteel fconWreSteel + fconWcon (S3-9) VreSteel Vcon where q, the overall efficiency of a nuclear power plant is defined as the ratio of the electrical energy output and the rate of thermal energy output from the reactor, QR QE (S 3 -10)QR Next we look at the superstructure. It consists of the shell and the dome, each containing concrete and steel elements. 85 Containment shell. We model individual layers of the shell, where di denotes the thickness of each layer. We write 1 Cshell = [2 7rconoutdshellConhshellConfconPcon qQR +2TrconoutdsheIIConhsheIICon0- fcon)PreSteelmmod.reSteel + 2 rrsteelOutdshellSteelhshellSteelPsteeI+ TconWcon + TreSteelWreSteel + TsteelWsteel] (S3-11) where reonOut is the radius of the concrete shell, fc,, is the volumetric concrete fraction, PreSteel is the price of the steel rods (rebar), and rsteelOut is the radius of the steel shell (which is equal to the inner radius of the concrete shell). We treat the assembly of the rebar structure, the placement of the concrete and the assembly and welding of the steel vessel components are separate work steps. The factor mmod.reSteel accounts for the higher per unit-costs of steel used in prefabricated modules, which are used in the 2017 containment design (mmod.reSteel is simply the ratio of module to non-module steel). Substituting the material deployment speeds from above we can rewrite this as 1 Cshell 2= [ frconoutdshellConhshellSteeIfconPcon r1QR 2 +2- rconOutdshellConhshellSteel(1- fcon)Pre.steelMmod.reSteel +2 7rrsteeloutdsteelhshellSteelpsteel+ Vsh Wcon + WreSteel + shellSteelVsteelWsteel Vcon VreSteel (S3-12) Using the breakdown of the volumes already included in the first terms of the equation we can write ((2,rrconOutdshellConhshell)(fconPcon+ PresteeCl -shfceolnlP resteel + f [on Wcon QRq Vcon +WreS+te Wel e~_tfeclc oonWnrree~~teeeeil)) + (2(-d7) rcon. indsteel hshellSteel)(Ps tee I + WWsste e el ] ((SS3-13) VreSteel VreSteel Vsteel Since de0o , the thickness of the concrete layer of the shell, is equal to the difference between reonOut and reonin, and dsteel is the difference between rconIn and rsteelln, we can simplify this to read 86 [(2rrconout(rconOut - rCcosnhIne)hlslh ell)(fconPco[n + +PreSteel- fconPreSteel + f onWcon QR1 Vcon WreSteel _fconWre~teei + ) Wst eel+ con(con.in - rsteelIn)hsheulsteel)(Psteel+ )A VreSteel VreSteel Vsteel (S3-14) Containment dome. For the reinforced concrete layer of the dome we use a hemispherical shape with radius reonout as a model. We subtract a smaller (inner) hemisphere from a larger (outer) hemisphere to compute the volume of the concrete dome wall. This choice is based on the drawing provided in the 1977 EEDB report [82]. Since the steel layer is significantly thinner than the concrete layer, we use a thin shell approximation of the volume of a spherical shell to compute steel costs. This simplifies the cost equation by reducing the number of terms. To account for the containment design change during period 2 (1987-2017) we include two shape correction factors in the dome cost equation (gi,g2). We keep the same functional form in all three years, but use g, and g2 to correct numerically for the effects of the changing building geometry on costs. The first shape correction factor, gi, accounts for the fact that the steel shell is flat at the bottom in 1976 and 1987, resulting in only one hemisphere, while the 2017 design features two semi-spheroids at the top and at the bottom. We therefore write gi as the ratio of a surface area of a sphere to the area of the dome shape we want to model in each year: 47r 2 -1 (S3-15) AdomeSteel(t) (St15 In 1976 and 1987, the dome volume we want to model is a hemisphere, so Adomesteel(t) = 2,r 2 for t=1976 and t=1987, and g, = 2. Fore t=2017, the steel layer changes to a hemispheroid both at the top and at the bottom. For spheroid where the height is smaller than the radius, the surface area is h2 1+ e AdomeSteel(t) 2,r 2 + sT-- In , (S3-16) e 1-e where e =1 - h (S3-17) r2 In this case g, decreases relative to 1976 and 1987 (gi = 1.36). (It is not equal to one as there are now two hemispheroids, but each with a surface area smaller than a hemispherical surface.) Consistent with the approach taken for the steel layer of the dome, we keep the same functional 87 form for the concrete layer of the dome (a hemisphere) in all three years, but correct for the design change during period 2 using 92. This correction factor is the ratio of the volume we want to model in each year, to that of a hemisphere: VdomeConcrete (, (3-18) 3 meaning that g2 = 1 is 1 in 1976 and 1987 and g2 = 0.34 in 2017 due to the smaller volume of the APiooo's truncated cone. We compute this volume as 2 7 2 VdomeConcrete( 01 ) = '3[ rocnouthtconeout - r coneOut(htconeOut - htconeIn)], (S3-19) where htconeout=5.59, rtconeout = 5.34, and htconein = 11.43 [173]. Based on this formulation, the hemisphere volume cancels out in 2017 and the cost equation repre- sents the volume of a truncated cone. Using gi and g2, the cost equation for the dome becomes: 3 Cdome=1[27[(3 r33 QE 3 conOut - ronnIn)fc pcong2 + (rcon t - ro0n0n )(1 - fcon)PreStee1920 0 0 + steelOut ddomeSteelPsteell + TconWcon + TreSteelWreSteel + TsteelWsteel] (S3-20) 91 Substituting for QE and ri we can rewrite this as Cdome = 27r [(r3e - rtQn, )(f c p c ,g + presteelg2 - fconPre.s3eelg20 1 0 0 4 2 -I cgc~o+ WreSteel fcon)r-eostejec9n s~teree st.es teel stee9 4rr2 Wte reSteel reSteel 2) + steelOut ddomeSteel(Psteel + VSteel (S3-21) vcon VreSteel VreSteel 91 VSteel 88 S3-4 Containment building cost data The data used to calculate the containment building costs is given in Table S3-1 below. Low-level mechanisms Unit 1976 1987 2017 Thermoelectric efficiency (rq) unitless 33.4% 33.5% 32.9% Thermal output (QR) MW 3425 3417 3400 Steel shell deployment rate (v.steel) m3/PH 0.000345 0.000219 0.00132 Concrete deployment rate (v.con) m3/PH 0.212 0.113 0.019 Rebar deployment rate (v.reSteel) m3/PH 0.00243 0.00155 0.00132 Foundation height (h.found) m 3.05 3.05 5.55 Concrete fraction in reinforced concrete (f.con) unitless o.96 0.958 0.941 Shape correction factor, dome, steel layer (gi) unitless 2 2 1.36 Shape correction factor, dome, concrete layer (g2) unitless 1 1 0.347 Shape correction factor, foundation, steel (g3) unitless 1 1 0 Wage, concrete installation (wage.con) 2017$/PH 31.96 32.92 44.76 Wage, steel installation (wage.steel) 2017$/PH 45-43 46.51 49.16 Shield building, wall height (h.shellCon) 2017$/PH 45.4 45.4 52.5 Shield building, inner radius (r.conIn) m 21.3 21.3 21.2 Shield building, outer radius (r.conOut) m 22.7 22.7 22.1 Shield building, thickness (d.shellCon) m 1.37 1.37 0.91 Steel shell, foundation thickness (d.foundSteel) m o.oo635 o.oo635 0.0413 Steel shell, inner radius (r.steelln) m 21.34 21.34 19.81 Steel shell, outer radius (r.steelOut) m 21.35 21.35 19.86 Steel shell, wall height (h.shellSteel) m 45.4 45.4 42.7 Steel shell, wall thickness (d.shellSteel) m 0.00953 0.00953 0.0445 Steel dome, outer height (h.domeSteelOut) m 21.3 21.3 11.5 Steel dome thickness (d.shellSteel) m 0.0127 0.0127 0.0413 Modularity price factor (m.mod.reSteel) unitless 1 1 1.46 Steel shell price (p.steel) 2017$/m 3 212000 18oooo 47000 Rebar price (p.reSteel) 2017$/m 3 11800 10100 13600 Concrete price (p.con) 2017$/m 3 157 128 247 Containment building cost, direct 2017$/We 53717 67060 109645 Containment building cost, total 2017$/We 76725 166878 198210-305871 PH = person-hour. References: 1976: [82], 1987: [89, 174], 2017: [136, 173, 60, 175, 57, 174, 150, 176] Table S3-1: Containment building cost data. The range for total cost data in the last row is for an indirect cost ratio of o.98 (low end) to 2.3 (high end). 89 S3-5 Indirect containment building cost change Several types of indirect costs are proportional to direct labor costs (e.g., construction supervision costs, QA/QC costs) because they are directly caused by activities accompanying construction. We therefore use the same cost model presented in section 3.3.2 of the main article, but multiply craft labor costs by the ratio of indirect labor hours to direct labor hours, Mid to compute a rough estimate of indirect costs. Ccontainment = Vi fipii(1 + Mind)] (S3-22) We i= QE [ i Figs S3-2 and S3-3 show low-level mechanisms of cost change for two different choices of mid in 2017, spanning the range from engineering estimates from the unfinished Vogtle project (see Fig.S3-2), and empirical values from the late 198os (see Fig. S3-3). For the low end, a ratio of approximately 1 (0.99) was computed using information from a construction planning report for the three-year phase leading up to hot functional testing at the Vogtle site in Georgia [150]. The value 0.99 is the ratio of total estimated indirect labor hours to total craft labor hours during this construction phase (note that we only use the ratio in our model and continue using our bottom-up estimate for containment building-related craft labor hours). Indirect labor hours include field indirect labor costs (provision of construction services, construction of temporary construction facilities), field non-manual services (field engineering, construction supervision, project controls, field procurement and subcontracts, con- struction management, quality control, environmental safety, quality assurance, construction document control, support services), and home office engineering. Using this value for the variable mid in 2017, costs decline by 90% during period 2. For the high end we assume a constant ratio between 1987 and 2017 (mind=2.29). The 1987 value is computed from the EEDB report, and the attribution of total indirect costs to individual plant components (see S3-2). Under a constant indirect cost ratio, costs in crease by 83% during period 2 (compared to a 19% increase for a decreasing indirect cost ratio, as shown S3-2). These results point to significant improvements in the indirect cost ratio are required to avoid significant overall containment cost escalation under a design change that increases material intensity (period 2). The high-level causes of total (direct and indirect) containment cost increase are shown below, using the same range of indirect cost ratios as for the analysis of low-level causes. We assign the variable mind equally to 'R&D, prescribed safety', 'R&D, flexible safety', 'Unlearning despite doing', and 'Process interference, safety' (see Table S3-7). For the low end of the indirect cost range, the decrease in mid decreases the contribution of safety-related mechanisms from 64 to -3%. Unlearning-despite-doing and direct safety interference remain influential due to the contribution of material deployment rates to cost increase. Material deployment rates remain important because they affect both direct and indirect cost 90 1976-1987 (83% of total 1987-2017 (17% of total) Overall: 1976-2017 (100%) A Concrete depl. speed A Steel shell wall thickness A Struct. steel depl. speed A Foundation height A Steel dome, thickness A Concrete fraction A Concrete worker wage Indirect cost factor A Base shape corr. factor A Shield building height A Ironworker wage A Steel module price markup A Conversion efficiency A Rebar price A Concrete price A Shield building, inner radius A Heat output A Steel shell found. thickness Steel shell dome, outer height A Steel shell, inner radius A Steel shell wall height A Base shape corr. factor A Steel shell, outer radius A Shield building, outer radius A Top shape corr. factor A Shield building thickness Cost increase:- Cost increase. Cost increase: A Steel price 118% from 1976- 11 ro 1987 141% from 176 A Steel shell depl. speed -60 -40 -20 0 20 40 0 -200 -100 0 100 200 -1 -0 -40 -20 0 20 40 60 % contribution to containment building cost increase Figure S3-2: Percentage contributions of containment building low-level mechanisms to cost increase during period 1 (1976-1987, left), period 2 (1987-2017, middle), and during both time periods (1976-2017, right), for an indirect cost ratio mind of o.84 in 1976, 2.29 in 1987, and o.99 in 2017. We assume that indirect labor is compensated at the same rate as the direct labor type supervised, documented or controlled by indirect labor. Low-level mechanisms are listed in the order of their contributions to overall (1976-2017) cost increase. Variables names starting with d, h, and r represent structure thicknesses, heights, and radii. Variables labeled as v refer to material deployment rates. Variables glg2 and g3 represent shape correction factors to account for the change in containment design and geometry in time period 2. components. The contribution of direct safety-related mechanisms ('R&D, prescribed safety', 'Process interference, safety') is less dependent on the inclusion or exclusion of indirect costs (35-40% of total costs with, 35% without indirect costs. S3-6 Containment building sensitivity analysis Here we examine the effect of uncertainties in our data on the results of the containment cost change analysis. We focus on uncertainties in deployment rates, wages and commodity prices. We consider structure dimensions to be less uncertain as these are derived from design drawings. The sensitivity analysis shows that the results are most sensitive to uncertainties in all variables 91 1976-1987 (44% of total) 1987-2017 (56% of total) Overall: 1976-2017 (100 A Concrete depl. speed - ---F A Steel shell wall thickness - Indirect cost factor A Foundation height A Concrete fraction A Struct. steel depl. speed A Steel dome, thickness A Concrete worker wage A Base shape corr. factor A Shield building height A Ironworker wage A Conversion efficiency A Shield building, inner radius A Steel module price markup A Rebar price A Heat output A Concrete price A Steel shell found. thickness Steel shell dome, outer height A Steel shell, inner radius A Steel shell wall height A Steel shell, outer radius A Base shape corr. factor A Shield building, outer radius A Top shape corr. factor U A Steel price Cost increase: Cost increase: - U Cost increase: - A Shield building thickness 118% from 1976- 70% from 1987- 270% from 1976- A Steel shell depl. speed , M , , - 30 -40 -20 0 20 40 60 -60 -40 -20 0 20 40 -60 -40 -20 0 20 40 60 % contribution to containment building cost increase Figure S3-3: Percentage contributions of containment building low-level mechanisms to cost increase during period 1 (1976-1987, left), period 2 (1987-2017, middle), and during both time periods (1976-2017, right), for an indirect cost factor mind of 0.84 in 1976, and a constant indirect cost ratio of 2.29 in 1987 and 2017. We assume that indirect labor is compensated at the same rate as the direct labor type supervised, documented or controlled by indirect labor. Low-level mechanisms are listed in the order of their contributions to overall (1976-2017) cost increase. Shape correction factors ('corr. factors') account for the change in containment design and geometry in time period 2. 92 1976-1987 (22% of total) 1987-2017 (78% of total) Overall: 1976-2017 (100%) R&D, general performance R&D, direct safety R&D, indirect safety Unlearning despite doing Direct safety interference Other -20 0 20 40 60 -50 0 50 -50 0 50 % contribution to containment building cost increase Figure S3-4: Percentage contributions of containment building high-level mechanisms to cost increase when including indirect costs, for indirect cost factor of 0.99 in 2017 (low end of the range, see S3-5 for high end results. Direct safety-related mechanisms ('R&D, prescribed safety', 'Process interference, safety') contribute 41% to cost increase. All safety-related mechanisms ('R&D, prescribed safety', 'R&D, flexible safety', 'Process interference, safety') contribute -3%. related to steel, the most expensive commodity (see Figs S3-6-S3-13). Among these variables, the wage paid to ironworkers is the most influential variable because it affects both components containing steel and components containing structural steel. In time period 1 (1976-1987), ironworker wages become a similarly important cost increasing mechanism as declining productivities. In time period 2, ironworker wages become almost as important for cost change as the lower-ranking structure thicknesses (e.g. the foundation height). Overall, however, these uncertainties do not affect the conclusions drawn in the main article. S3-7 High-level mechanisms of containment building cost change In this section we explain how we connect low-level mechanisms to high-level mechanisms. S3-7.1 Variable classification scheme to assign low-level to high-level mecha- nisms We assign low-level mechanisms to high-level mechanisms using a two-step method. First, we classify each variable in the cost equation according to fundamental properties of the process or technology component that is described by the variable. This step identifies the improvement mechanisms that a variable is amenable to because of the technology property it describes. For instance, variables such as conversion efficiencies that represent fundamental, physical properties determining the performance of a technology will often require R&D to change. The requirement for R&D is a more general characteristic that can be assigned to a variable, independent on the time period and location considered. 93 1976-1987 (22% of total) 1987-2017 (78% of total) Overall: 1976-2017 (100%) R&D, general performance R&D, direct safety R&D, indirect safety Unlearning despite doing Direct safety interference Other -20 0 20 40 60 -50 0 50 -50 0 50 % contribution to containment building cost increase Figure S3-5: Percentage contributions of containment building high-level mechanisms to cost increase when including indirect costs, for indirect cost ratio of 2.29 in 1987 and in 2017 (high end of the range, see S3-4 for low end. Direct safety-related mechanisms ('R&D, prescribed safety', 'Process interference, safety') contribute 35% to cost increase. All safety-related mechanisms ('R&D, prescribed safety', R&D flexible safety, 'Process interference, safety') contribute 65%. We therefore refer to the first step as 'prior assignment'. The variable classifiers used to generate the prior assignment are described in Table S3-2 below. In the second step, we use data from each time period analyzed to either accept or reject the prior assignments. The goal of this assignment step is to support each assignment with evidence for different types of innovative activity in a specific time period. If evidence for a certain type of innovative activity that corresponds to a high-level mechanism cannot be provided, the assignment is removed. For instance, if a variable is amenable to R&D driven changes according to the prior assignment, but neither patents nor journal papers indicate research activity during a specific time period, we do not assign the variable to R&D and instead assign changes to a different high-level mechanism for whose influence during a specific time period we have evidence. We refer to the result of this part of the procedure as 'posterior assignment'. Overall, we use this procedure to compute rough estimates of the contributions of different high-level mechanisms, in order to address related questions about the impact of safety-related activities. A list of the different types of innovative activity we consider for each high-level mechanism, and the measures used to quantify different activities, is given in Table S3-3 below. Prior assignment. The prior assignment is given in Table S3-4 below. S3-7.2 Retrospective analysis of high-level mechanisms Evidence for R&D activities. Table S3-5 below shows the patents we use to document different types of R&D activity related to the containment building. In period 2 (1976-1987), we assign structure thicknesses and dimensions to R&D TP due to the changes documented in the below patents. Design 94 1976-1987 1987-2017 Overall_(t976-2017 A Steel shell wall thickness A Concrete dep . speed A Steel dome, thickness A Foundation height A Concrete fraction A Base shape corr. factor A Concrete worker wage A Structural steel depl. speed A Shield building height A Rebar price A Ironworker wage A Steel module price markup A Concrete price A Conversion efficiency A Shield building, inner radius A Heat output A Steel shell foundation thickness A Steel shell dome outer height A Steel shell, inner radius A Steel shell wall height A Steel shell, outer radius A Base shape corr. factor A Shield building, outer radius A Top shape corr. factor A Shield building thickness A Steel price A Steel shell depl. speed 4 -50 0q -100 -50 0 50 100 -100 -50 0 50 100 -100 50 100 % contribution to containment building cost Increase Figure S3-6: Sensitivity of containment building low-level mechanisms to uncertainties in ironworker wages. We vary ironworker wages by +/-io% relative to the base case results shown in 3-5 in the main article. 1976-1987 1987-2017 Overall (1976-2017) A Steel shell wall thickness A Concrete depl. speed U A Steel dome, thickness A Foundation height A Concrete fraction I A Base shape corr. factor A Concrete worker wage A Structural steel dep. speed A Shield building height A Rebar price I A Ironworker wage I 1-1 A Steel module price markup A Concrete price A Conversion efficiency A Shield buildin inner radius Heat output A Steel shell foundation thickness A Steel shell dome outer heightA Steel shell, inner radius A Steel shell wall height A Steel shell, outer radius A Base shape corr. factor A Shield building, outer radius A Top shape corr. factor A Shield building thickness A-Steel price A Steel shell depl. speed -100 -50 0 50 100 -100 -50 0 50 100 -100 -50 0 50 100 % contribution to containment building cost increase Figure S3-7: Sensitivity of containment building low-level mechanisms to uncertainties in concrete worker wages. We vary concrete worker wages by +/-io% relative to the base case results shown in 3-5 in the main article. 95 1976-1987 1987-2017 Overall (1976-2017) A Steel shell wall thickness A Concrete dept. speed A Steel dome, thickness A Foundation height A Concrete fraction -. A Base shape corr. factor A Concrete worker wage A Structural steel dept. speed A Shield building height A Rebar price 4 A Ironworker wage A Steel module price markup A Concrete price A Conversion efficiency A Shield building, inner radius Heat output A Steel shell foundation thickness A Steel shell dome outer height A Steel shell, inner radius A Steel shell wall height A Steel shell, outer radius A Base shape corr. factor A Shield building, outer radius A Top shape corr. factor A Shield building thickness A teel price A Steel shell dept. speed -1 00 -50 0 50 100 -100 -50 0 50 100 -100 -50 0 50 100 % contribution to containment building cost increase Figure S3-8: Sensitivity of containment building low-level mechanisms to uncertainties in steel prices. We vary steel prices by +/-io% relative to the base case results shown in 3-5 in the main article. 1976-1987 1987-2017 Overall (1976-2017) A Steel shell wall thickness A Concrete dept. speed A Steel dome, thickness A Foundation height A Concrete fraction A Base shape corr. factor A Concrete worker wage A Structural steel dep. speed p- A Shield building height A Rebar price A Ironworker wage A Steel module price markup A Concrete price A Conversion A efficiencyShield buildin inner radius A Heat output A Steel shell foundation thickness - A Steel shell dome outer height A -Steel shell, inner radius A Steel shell wall height A Steel shell, outer radius A Base shape corr. factor A Shield building, outer radius A Top shape corr. factor A Shield buildingthickness A-Steel price A Steel shell depl. speed -100 -50 0 50 100 -100 -50 0 50 100 -100 -50 0 50 100 % contribution to containment building cost increase Figure S3-9: Sensitivity of containment building low-level mechanisms to uncertainties in structural steel prices. We vary structural steel prices by +/-io% relative to the base case results shown in 3-5 in the main article. 96 1976-19871 1987-20j7 Overall (1976-2017 A Steel shell wall thickness A Concrete depi. speed 0 A Steel dome, thickness A Foundation height A Concrete fraction A Base shape corr. factor * A Concrete worker wage U A Structural steel dep. speed U U A Shield building height I A Rebar price I I A Ironworker wage A Steel module price markup A Concrete price A Conversion efficiency A Shield buildin inner radius - A Heat output A Steel shell foundation thickness A Steel shell dome outer height A Steel shell, inner radius A Steel shell wall height A Steel shell, outer radius A Base shape corr. factor A Shield building, outer radius A Top shape corr. factor A Shield building thickness A Steel price A Steel shell depl. speed -5 -100 -50 0 50 100 -100 -50 0 50 100 -100 -50 0 50 100 %c ontribution to containment building cost increase Figure S3-10: Sensitivity of containment building low-level mechanisms to uncertainties in concrete prices. We vary concrete prices by +/-io% relative to the base case results shown in 3-5 in the main article. 1976-1987 1987-2017 0 verall (1976-2017) A Steel shell wall thickness A Concrete depl. speed A Steel dome, thickness A Foundation height A Concrete fraction A Base shape corr. factor U A Concrete worker wage U A Structural steel depl. speed U A Shield building height U A Rebar price A Ironworker wage A Steel module price markup A Concrete price A Conversion efficiency A Shield building, inner radius Heat output A Steel shell foundation thickness A Steel shell dome outer height A Steel shell, inner radius A Steel shell wall height A Steel shell, outer radius A Base shape corr. factor A Shield building, outer radius A Top shape corr. factor A Shield building thickness A Steel price A Steel shell depl. speed -100 -50 0 50 100 -100 -50 0 50 100 -100 -50 0 50 100 % contribution to containment building cost increase Figure S3-11: Sensitivity of containment building low-level mechanisms to uncertainties in steel deployment rates. We vary steel deployment rates by +/-io% relative to the base case results shown in 3-5 in the main article. 97 1976-1987 1987-2017 Overallf(1976-2017) A Steel shell wall thickness A Concrete depl. speed El A Steel dome, thickness A Foundation height A Concrete fraction I A Base shape corr. factor A Concrete worker wage U A Structural steel dep. speed B A Shield building height A Rebar price BaE A Ironworker wage IIF A Steel module price markup I A Concrete price I A Conversion efficiency A Shield building, inner radius - Heat output A Steel shell foundation thickness A Steel shell dome A outer hei htSteel shell inner radius A Steel shell wall height A Steel shell, outer radius A Base shape corr. factor A Shield building, outer radius A Top shape corr. factor A Shield building thickness A Steel price A Steel shell depl. speed -100 -50 0 50 100 -100 -50 0 50 100 -100 -50 0 50 100 % contribution to containment building cost increase Figure S3-12: Sensitivity of containment building low-level mechanisms to uncertainties in structural steel de- ployment rates. We vary structural steel deployment rates by +/-io% relative to the base case results shown in 3-5 in the main article. 1976-1987 1987-2017 Overall (1976-2017) A Steel shell wall thickness A Concrete dept. speed A Steel dome, thickness A Foundation height fl A Concrete fraction A Base shape corr. factor A Concrete worker wage A Structural steel depl. speed A Shield building height A Rebar price I A Ironworker wage I A Steel module price markup A Concrete price A Conversion efficiency A Shield buildin inner radius Heat output A Steel shell foundation thickness A Steel shell A dome outer heightSteel shell, inner radius' A Steel shell wall height A Steel shell, A outer radiusBase shape corr. factor A Shield building, outer radius A Top shape corr. factor A Shield building thickness A Steel price A Steel shell depl. speed -100 -50 0 50 100 -100 -50 0 50 100 -100 -50 0 50 100 % contribution to containment building cost increase Figure S3-13: Sensitivity of containment building low-level mechanisms to uncertainties in concrete deployment rates. We vary concrete deployment rates by +/-1o% relative to the base case results shown in 3-5 in the main article. 98 Table S3-2: Classifiers used to identify applicable high-level mechanisms Variable classifiers Description High-level mechanisms Internal Influenced by technology industry R&D, EOS, LBD, KS External Influenced by factors outside of technology industry Other, EOS for bulk purchases Hard Affects costs through physical technology attribute Embodied in hardware R&D, EOS, LBD KS Soft Affects costs through non-physical technology attribute Not embodied in hardware LBD, KS Technology Describes cost determinant of final product Process Describes cost determinant of manufacturing LBD, KS or installation proess LBD, KS Feature Descriptive attribute of technology or process R&D Cost component Non-descriptive summand in cost equation, could be broken down further All mechanisms Scale Measure for size of process or product EOS Usage Measure for per-unit input consumption EOS, LBD Installation One installation can be completed at a time LBD, KS Preparation Multiple tasks can be completed simultaneously EOS, LBD, KS Administrative Can be directly affected by policy regulation (e.g. fees) Sourcing Business processes Table S3-3: High-level mechanisms and related measures of innovative activity Mechanism Activity Measure R&D Patenting Patent counts and citations Journal papers Paper count and citations EOS Upscaling of factories Factory output LBD Embodied in hardware Direct regulation Changing regulations Policy documents drawings in these patents show a separation of the steel liner and the concrete shield building to allow for natural air circulation between the liner and the shield building, which is the main driver of increasing structure thicknesses and dimensions during period 2. Assignment result. The posterior assignment of low-level mechanisms to high-level mechanisms for time periods 1 and 2 is given in Table S3-7 below. 99 S3-8 Prospective analysis of containment cost change S3-8.1 Assumptions for prospective analysis Assumptions for prospective cost reduction scenarios are given in Table S3-9. In scenario 1, all vari- ables are changed by 20% in a cost reducing direction, with the exception of variables facing known engineering constraints. Conversion efficiency is an example. The efficiency improvements of 2-3% assumed here can be achieved by the use of hydrophobic condenser coatings [181]. Larger efficiency changes (as estimated for advanced Brayton cycles [182]), in contrast, would require design changes beyond the scope of our model. We also constrain the reduction in rebar density to that estimated for high strength rebar. Using Grade loo instead of Grade 60 rebar is estimated to cut rebar volumes by - 40% for the same strength, which corresponds to a 2.5% change in the concrete fraction. In scenario 2, we use the demonstrated volumetric deposition rate from [161], scaled down in proportion to the higher density of concrete (p = 2700kg/m3 ) as compared to polyurethane (p = 28kg/M 3) used in the original reference [161] (which means that we assume the same mass deposition rate can be achieved regardless of the material deposited). To compute the improved rebar placement rate, we apply the 30% improvement reported in [162] to the 2017 rebar deployment rate. We hold the steel deployment rate constant for lack of automation options. In scenario 3,we use the reported price of Ductal, the standard UHPC supplier in the U.S. [163], which is roughly io times that of standard nuclear concrete. We use a rebar price premium of 5o%. S3-8.2 Prospective analysis with indirect cost change Figure S3-14 gives the contributions of individual low-level mechanisms to future cost reductions under the three scenarios described in the main text, including indirect cost change. In scenario 1, the variable mind replaces the foundation thickness as the third most important cost-reducing factor. Other changes are described in the main article. S3-8.3 Prospective analysis of high-level mechanisms The data used to populate the containment cost model in scenarios 1-3 are given in Table S3-9 below, and the assignments of low-level mechanisms to high-level mechanisms for the three scenarios considered are given in Table S3-10. Modeling the effect of reduced rebar congestion on productivity. The use of high-strength rebar (HSR) leads to a reduction of rebar volumes needed for the same yield strength. Rebar congestion is a common problem on nuclear construction sites, and its reduction is expected to increase labor productivity by freeing up space and shortening distances over which commodities and equipment need to be transported. Due to the lack of estimates for this increase in the literature we develop a 100 simple model where deployment rates increase as a function of the volumetric work space difference between a construction site where Grade 60 rebar is used, and one where HSR is used (Grade ioo). We set the ratio of the increased (VHSR) to the standard deployment rate (vR) as equal to the ratio of the construction space after and before the switch to HSR (VHSR and VR, respectively): VHSR _ Vtotal - VHSR (S3-23) yR Vtotal - VR where 60 VHSR = (S3-24) The total area occupied by the Vogtle 3&4 construction site is roughly 65 acres, and based on satellite pictures we assume that about 1/8 of that area is used for the containment building (see S3-15 below). Scenario 1: Broad imorovement Scenario 2: Increased oroductivity Scenario 3: Advanced materials A Concrete fraction A Ironworker wage A Foundation height A Indirect cost factor A Concrete worker wage A Shield building thickness A Concrete depl. speed -7 A Struct. steel depl. speed A Steel shell wall thickness A Conversion efficiency A Steel shell depl. speed A Steel price A Steel dome, thickness A Rebar price A Shield building, inner radius A Steel module price markup A Concrete price A Automation hardware A Steel shell found. thickness A Base shape corr. factor A Top shape corr. factor Steel shell dome, outer height A Base shape corr. factor A Steel shell, inner radius A Steel shell wall height A Steel shell, outer radius A Shield building height -Cost reduction: -Cost reduction. Cost reduction. A Shield building, outer radius -59% from 2017 55% from 2017 42% from 2017 A Heat output -60 -40 -20 0 20 40 -60 -40 -20 0 20 40 60 -60 -40 -20 0 20 40 60 % contribution to containment building cost decrease Figure S3-14: Percentage contributions of containment building low-level mechanisms to cost decrease for scenar- ios 1 (20% change of several variables in a cost reducing direction), scenario 2 (factor 3 increase in productivity for all materials) and scenario 3 (use of advanced materials). Low-level mechanisms are listed in the order of their contributions to overall cost reduction. 101 Figure S3-15: Alwin W. Vogtle Units 3&4 construction site, Google Maps Satellite, October 2018. We assume a "construction space" that is 2.5 meters high. These assumptions lead to an increase in material deployment rates by 5%. These assumptions come with uncertainties (e.g., we don't have exact information on the height of the space in which Vogtle construction workers operate), but they are sufficient to compute a first order estimate of the effect of HSR on deployment rates. For a range of heights (1.5m-3.5m) and a range of construction area fractions used for the containment building (o.1-o.5), we compute a 13% increase (lower end of the range) and a 1% increase (upper end of the range) in the deployment rate. These changes would not alter our conclusions regarding the relatively minor effect of reduced congestion. For a 1-13% increase across all material deployment rates, cost reductions in scenario 3 would reach 28-32% relative to 2017, instead of 29% for the middle value used in the main paper. 102 Low-level mechanisms High-level mechanisms A Thermoelectric efficiency (q) R&D GP A Thermal capacity (QR) R&D TP A Steel shell deployment rate (v.steel) R&D TP, R&D PS, UDD, PIS A Concrete deployment rate (v.con) R&D TP, R&D PS, UDD, PIS A Rebar deployment rate (v.reSteel) R&D GP, R&D PS, UDD, DRI A Foundation height (h.found) R&D TP, R&D FS A Concrete fraction in reinforced concrete (f.con) R&D TP, R&D FS , R&D DS A Shape correction factors (gl, g2, g3) R&D TP, R&D FS A Wages (wage.con, wage.steel) Other A Shield building, wall height (h.shellCon) R&D TP, R&D FS A Shield building, inner radius (r.conIn) R&D TP, R&D FS A Shield building, outer radius (r.conOut) R&D TP, R&D FS A Shield building, thickness (d.shellCon) R&D TP, R&D PS, R&D FS A Steel shell, foundation thickness (d.foundSteel) R&D TP, R&D PS, R&D FS A Steel shell, inner radius (r.steelIn) R&D TP, R&D FS A Steel shell, outer radius (r.steelOut) R&D TP, R&D FS A Steel shell, wall height (h.shellSteel) R&D TP, R&D FS A Steel shell, wall thickness (d.shellSteel) R&D TP, R&D DS, R&D FS A Steel dome, outer height (h.domeSteelOut) R&D TP, R&D FS A Steel dome thickness (d.shellSteel) R&D TP, R&D PS, R&D FS A Modularity price factor (m.mod.reSteel) LBD, R&D GP A Steel shell price (p.steel) EOS, R&D TP, R&D FS, Other A Rebar price (p.reSteel) EOS, R&D TP, R&D FS, Other A Concrete price (p.con) EOS, R&D TP, R&D FS, Other A Indirect cost ratio (m.ind) EOS, R&D TP, R&D PS, R&D FS UDD, DRS, Other Table S3-4: Prior assignments of low-level mechanisms to high-level mechanisms. In cases where multiple high- level mechanisms are assigned, ioo% is divided equally among all mechanisms. The meaning of the abbreviations for high-level mechanisms is as follows: 'R&D, TP' refers to research and development, technical performance; 'R&D, FS' refers to research and development, flexible safety; 'R&D, PS' refers to research and development, prescribed safety; 'UDD' refers to unlearning despite doing; 'PIS' refers to 'Process interference, safety'. The high- level mechanism 'Other' is not abbreviated. The contributions of the high-level mechanisms for this assignment are shown in Fig. 3-7 in the main text. Variables that remain constant during period i are not assigned to high-level mechanisms because they do not contribute to cost change during this period. 103 Table S3-5: Patents documenting R&D activity related to passive cooling containment design change in period 2 1987-2017, in chronological order. Westinghouse patents related specifically to the passive cooling of nuclear fuel, of the reactor vessel, and of small modular reactor designs are excluded from the list due to the paper's focus on mechanisms of containment cost change. The column entitled 'Year' refers to the year of patent publication. Year Patent Number Title Assignee USPTO 1988 4753771 Passive safety system for a pressurized nuclear reactor Westinghouse 1991 5049353 Passive containment cooling system Westinghouse 1994 5345482 Passive containment cooling water distribution device Westinghouse 1997 5612982 Nuclear power plant with containment cooling Westinghouse 2015 9177675 Passive containment air cooling for nuclear power plants Westinghouse Table S3-6: Publications documenting R&D activity related to passive cooling containment design change in period 2 1987-2017, in chronological order. Authors are listed in the references and are affiliated with Westinghouse. Publications by non-Westinghouse authors are excluded because it is likely they did not result directly from R&D driving the containment design change during period 2 (1987-2017), which is the focus here. Year Journal Title 2001 Proceedings of ICONE 9 APiooo status overview [177] 2002 Proceedings of ICONE lo Westinghouse APiooo Advanced Plant Simplification Results, Measures, and Benefits [159] 2002 Advanced Nuclear Safety features and research needs of Westinghouse advanced reactors [178] 2006 Nuclear Engineering & Design Westinghouse APlooo advanced passive plant [60] 2008 Nuclear Engineering & Design APlooo will meet the challenges of near-term deployment [179] 2011 Energy Procedia The APioooTM Reactor: Passive Safety and Modular Design [180] 104 Table S3-7: Assignments of low-level mechanisms to high-level mechanisms during period 1 and period 2. In cases where multiple high-level mechanisms are assigned, 1oo% is divided equally among all mechanisms. The meaning of the abbreviations for high-level mechanisms is as follows: 'R&D, TP' refers to research and development, technical performance; 'R&D, FS' refers to research and development, flexible safety; 'R&D, PS' refers to research and development, prescribed safety; 'UDD' refers to unlearning despite doing; 'PIS' refers to 'Process interference, safety'. The high-level mechanism 'Other' is not abbreviated. The contributions of the high-level mechanisms for this assignment are shown in Fig. 3-7 in the main text. Variables that remain constant during period i are not assigned to high-level mechanisms because they do not contribute to cost change during this period. Low-level mechanisms Period 1 (1976-1987) Period 2 (1987-2017) A Thermoelectric efficiency (q) R&D TP R&D TP A Thermal capacity (QR) R&D TP R&D TP A Steel shell deployment rate (v.steel) R&D PS, UDD, PIS R&D TP, R&D FS A Concrete deployment rate (v.con) R&D PS, UDD, PIS UDD, PIS A Rebar deployment rate (v.reSteel) R&D PS, UDD, PIS UDD, PIS A Foundation height (h.found) constant R&D TP, R&D FS A Concrete fraction in reinforced concrete (f.con) R&D FS, R&D PS R&D TP, R&D FS, R&D PS A Shape correction factors (gi, g2, g3) constant R&D TP, R&D FS A Wages (wage.con, wage.steel) Other Other A Shield building, wall height (h.shellCon) constant R&D TP, R&D FS A Shield building, inner radius (r.conIn) constant R&D FP, R&D FS A Shield building, outer radius (r.conOut) constant R&D TP, R&D FS A Shield building, thickness (d.shellCon) constant R&D TP, R&D FS A Steel shell, foundation thickness (d.foundSteel) constant R&D TP, R&D FS A Steel shell, inner radius (r.steelln) constant R&D TP, R&D FS A Steel shell, outer radius (r.steelOut) constant R&D TP, R&D FS A Steel shell, wall height (h.shellSteel) constant R&D TP, R&D FS A Steel shell, wall thickness (d.shellSteel) constant R&D TP, R&D FS A Steel dome, outer height (h.domeSteelOut) constant R&D TP, R&D FS A Steel dome thickness (d.shellSteel) constant R&D TP, R&D FS A Modularity price factor (m.mod.reSteel) constant R&D TP A Steel shell price (p.steel) Other R&D TP, R&D FS, Other A Rebar price (p.reSteel) Other Other A Concrete price (p.con) Other Other A Indirect cost ratio (m.ind) R&D PS, R&D FS R&D PS, R&D FS UDD, PIS UDD, PIS 105 High-level mechanisms Period 1 (1976-1987) R&D TP R&D FS Concrete fraction in reinforced concrete (increased seismic and tornado loads [140] R&D PS Deployment rates ("mandatory implementation of improved safety features" [140]) UDD Deployment rates (inadequate construction management [151]) PIS Deployment rates (start of NRC resident inspector program in late 1970s [157] Other Material prices [refs] High-level mechanisms Period 2 (1987-2017) R&D TP Structure thicknesses (goals for design change described in [148, 147, 146, 159]) Steel price, steel deployment rate R&D FS Structure thicknesses (see above) Steel price, steel deployment rate R&D PS Concrete fraction [154] UDD Concrete/rebar deployment rate [15o] PIS Concrete/rebar deployment rate ([1501, NRC inspection throughout construction process [158]) Other Material prices Table S3-8: Data sources for assignments of low-level mechanisms to high-level mechanisms during period 1 and 2. In cases where multiple high-level mechanisms are assigned, ioo% is divided equally among all mechanisms. The meaning of the abbreviations for high-level mechanisms is as follows: 'R&D, TP' refers to research and development, technical performance; 'R&D, FS' refers to research and development, flexible safety; 'R&D, PS' refers to research and development, prescribed safety; 'UDD' refers to unlearning despite doing; 'PIS' refers to 'Process interference, safety'. The high-level mechanism 'Other' is not abbreviated. 1o6 Low-level mechanisms Unit 2017 Scenario 1 Scenario 2 Scenario 3 Thermoelectric efficiency (q7) unitless 32.9 33.8% 32.9% 32.9 % Thermal output (QR) MW 3400 3400 3400 3400 Steel shell deployment rate (v.steel) m3/PH 0.00132 0.001584 0.00132 0.0149 Concrete deployment rate (v.con) m3/PH 0.019 0.023 5.3827 0.022 Rebar deployment rate (v.reSteel) m3/PH 0.00132 0.001584 0.001716 0.0149 Foundation height (h.found) m 5.55 4.44 5.55 2.15 Concrete fraction in reinforced concrete (f.con) unitless 0.941 0.965 0.941 0.965 Shape corr. factor, dome, steel layer (gi) unitless 1.36 1.36 1.36 1.36 Shape corr. factor, dome, concrete layer (g2) unitless 0.347 0.347 0-347 0.347 Shape corr. factor, foundation, steel (g3) unitless o o 0 0 Wage, concrete installation (wage.con) 2017$/PH 44.76 35.81 44.76 44.76 Wage, steel installation (wage.steel) 2017$/PH 49.16 39.33 49.16 49.16 Shield building, wall height (h.shellCon) 2017$/PH 52.5 52.52 52.5 52.5 Shield building, inner radius (r.conln) m 21.2 21.37 21.2 21.74 Shield building, outer radius (r.conOut) m 22.1 22.7 22.1 22.1 Shield building, thickness (d.shellCon) m 0.91 0.73 0.91 0.35 Steel shell, foundation thickness (d.foundSteel) m 0.0413 0.0330 0.0413 0.0413 Steel shell, inner radius (r.steelln) m 19.81 19.81 19.81 19.81 Steel shell, outer radius (r.steelOut) m 19.86 21.35 19.81 19.86 Steel shell, wall height (h.shellSteel) m 42.7 42.7 42.7 42.7 Steel shell, wall thickness (d.shellSteel) m 0.0445 0.03556 0.0445 0.0445 Steel dome, outer height (h.domeSteelOut) m 11.5 21.3 11.5 11.5 Steel dome thickness (d.shellSteel) m 0.0413 0.0330 0.0413 0.0413 Modularity price factor (m.mod.reSteel) unitless 1.46 1.17 1.46 1.46 Steel shell price (p.steel) 2017$/m 3 47332 37866 47332 47000 Rebar price (preSteel) 2017$/m 3 13609 10887 13609 20413 Concrete price (p.con) 2017$/m 3 247 197.56 247 2600 Indirect cost ratio (m.ind) 2017$/m 3 1 0.8 1 1 Containment building cost, direct 2017$/We 109645 50235 71335 69481 Containment building cost, total 2017$/We 116338 81362 88707 116392 PH = person-hour. References: Scenario 1: [181, 183] Scenario 2: [161, 162, 82]; Scenario 3: [168, 169, 165, 167, 166, 164] Table S3-9: Containment building cost data for the prospective cost change analysis. The column entitled '2017' serves as the base case and is the same as in the last column of Table S3-1. In scenario 1 (left), multiple variables (structure thicknesses, deployment rates, and prices) improve by 20% (i.e. they change by 20% in a cost-reducing direction). In scenario 2 (middle), material deployment rates are assumed to reach peak historical productivity levels observed in 1976. In scenario 3 (right), standard nuclear concrete and rebar is replaced with advanced construction materials, which allows a reduction in commodity usage. Volumetric UHPC prices are assumed to be 20 times that of standard nuclear concrete [164, 176]. High-strength rebar is assumed to cost So% more than standard rebar [168]. The variable r.conIn increases in scenarios i and 3 to reduce the shield building wall thickness (r.conOut is therefore constant in these scenarios). 107 Table S3-10: Assignments of low-level mechanisms to high-level mechanisms for the prospective cost change analysis. We only show low-level mechanisms that change. In cases where multiple high-level mechanisms are assigned, 100% is divided equally among all mechanisms. The meaning of the abbreviations for high-level mechanisms is as follows: 'R&D, TP' refers to research and development, technical performance; 'R&D, FS' refers to research and development, flexible safety; 'KS' refers to knowledge spillovers.The high-level mechanism 'Other' is not abbreviated. The contributions of the high-level mechanisms for this assignment are shown in Fig. 3-9 in the main text. Low-level mechanisms Scenario i Scenario 2 Scenario 3 A Thermoelectric efficiency (q) R&D TP constant constant A Steel shell deployment rate (v.steel) LBD constant constant A Concrete deployment rate (v.con) LBD R&D TP/FS, KS, LBD constant A Rebar deployment rate (v.reSteel) LBD KS, LBD constant A Foundation height (h.found) R&D TP/FS constant R&D FS , KS A Reinforced con., con. fraction (f.con) R&D TP/FS constant R&D FS, KS A Wages (wage.con, wage.steel) Other constant constant A Shield building, inner radius (r.conIn) R&D TP/FS constant R&D FS, KS A Shield building, thickness (d.shellCon) R&D TP/FS constant R&D FS, KS A Steel shell, wall thickness (d.shellSteel) R&D TP, FS constant R&D FS, KS A Steel dome thickness (d.shellSteel) R&D TP, FS constant R&D FS, KS A Steel shell price (p.steel) Other constant constant A Rebar price (p.reSteel) Other constant KS, Other A Concrete price (p.con) Other constant KS, Other A Automation equipment (p.automare) Constant constant KS 108 Bibliography [1] Joeri Rogelj, Gunnar Luderer, Robert C Pietzcker, Elmar Kriegler, Michiel Schaeffer, Volker Krey, and Keywan Riahi. "Energy system transformations for limiting end-of-century warming to below 1.5 C". Nature Climate Change 5.6 (2015), p. 519. [2] Rajendra K Pachauri, Myles R Allen, Vicente R Barros, John Broome, Wolfgang Cramer, Renate Christ, John A Church, Leon Clarke, Qin Dahe, Purnamita Dasgupta, et al. Climate change 2014: synthesis report. Contribution of Working Groups I, II and III to the fifth assessment report of the Intergovernmental Panel on Climate Change. IPCC, 2014. [3] International Energy Agency. World Energy Outlook 2018. 2018, p. 661. DOI: https: //doi. org/ https://doi.org/10.1787/weo-2018-en. URL: https://www.oecd-ilibrary.org/content/ publication/weo-2018-en. [41 Mark Z Jacobson. "Review of solutions to global warming, air pollution, and energy security". Energy & Environmental Science 2.2 (2009), pp. 148-173. [5] National Renewable Energy Lab. Life Cycle Greenhouse Gas Emissionsf rom Electricity Generation. https: //www. nrel. gov/docs/fy3osti/57187. pdf. [Online; accessed 27-April-2019]. 2013. [6] Daniel Weisser. "A guide to life-cycle greenhouse gas (GHG) emissions from electric supply technologies". Energy 32.9 (2007), pp. 1543-1559. [7] T. Bruckner, I.A. Bashmakov, Y. Mulugetta, H. Chum, A. de la Vega, J. Edmonds Navarro, A. Faaij, B. Fungtammasan, A. Garg, E. Hertwich, D. Honnery, M. In eld, S. Kainuma, Khennas, S. Kim, H. B. Nimir, K. Riahi, R. Strachan, R. Wiser, and X. Zhang. "Energy Systems. In: Climate Change 2014: Mitigation of Climate Change. Contribution of Working Group III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change." Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. (2014). [8] Joel Jean, Patrick R Brown, Robert L Jaffe, Tonio Buonassisi, and Vladimir Bulovik. "Pathways for solar photovoltaics". Energy & Environmental Science 8.4 (2015), pp. 1200-1219. 109 [9] World Nuclear Association. Nuclear Power in Japan. 2018. URL: http: //www .world-nuclear. org/information-library/country-profiles/countries-g-n/japan-nuclear-power. aspx (visited on 2018). [lo] International Atomic Energy Agency. Power Reactor Information System: Reactor overview and Nuclear share. Tech. rep. 2018. URL: https: //pris. iaea. org. [il] Mycle Schneider, Antony Froggatt, Julie Hazemann, Tadahiro Matsuta, M.V. Ramana, Juan Rodriguez, and Andreas Rudinger. The world nuclear industry status report 2017. Tech. rep. 2017. URL: https://www.worldnuclearreport.org/IMG/pdf/20170912wnisr2017-en-r.pdf. [12] International Renewable Energy Agency. Trends in Renewable Energy (Installed Capacity). 2018. URL: https: //www. irena .org/ourwork/Knowledge-Data-Statistics/Data-Statistics (visited on 2018). [13] Gert Jan Kramer and Martin Haigh. "No quick switch to low-carbon energy". Nature 462.7273 (2009), p. 568. [14] M Granger Morgan, Ahmed Abdulla, Michael J Ford, and Michael Rath. "US nuclear power: The vanishing low-carbon wedge". Proceedingso f the NationalA cademy of Sciences 115.28 (2018), pp. 7184-7189. [is] Aishwarya S Mundada, Emily W Prehoda, and Joshua M Pearce. "US market for solar photo- voltaic plug-and-play systems". Renewable energy 103 (2017), pp. 255-264. [16] Vaclav Smil. Energy transitions:h istory, requirements, prospects. ABC-CLIO, 2010. [17] A Grubler, X Bai, T Buettner, S Dhakal, DJ Fisk, T Ichinose, JE Keirstead, G Sammmer, D Satterthwaite, NB Schulz, et al. "Global Energy Assessment-Toward a Sustainable Future". International Institutef or Applied Systems Analysis and Cambridge University (2012), pp. 1307- 1400. [18] Alan McDonald and Leo Schrattenholzer. "Learning rates for energy technologies". Energy policy 29.4 (2001), pp. 255-261. [19] Nikolaos Kouvaritakis, Antonio Soria, and Stephane Isoard. "Modelling energy technology dy- namics: methodology for adaptive expectations models with learning by doing and learning by searching". InternationalJ ournal of Global Energy Issues 14.1-4 (2000), pp. 104-115. [20] Goksin Kavlak, James McNerney, and Jessika E Trancik. "Evaluating the causes of cost reduction in photovoltaic modules". Energy policy 123 (2018), pp. 700-710. [21] Jessika E Trancik and Daniel Cross-Call. "Energy technologies evaluated against climate targets using a cost and carbon trade-off curve". Environmental science & technology 47.12 (2013), pp. 6673-6680. 110 [221 Jessika E Trancik. "Renewable energy: Back the renewables boom". Nature News 507.7492 (2014), p. 300. [23] Ran Fu, Robert M Margolis, and David J Feldman. US Solar PhotovoltaicS ystem Cost Benchmark: Q1 2018. Tech. rep. National Renewable Energy Lab.(NREL), Golden, CO (United States), 2018. [24] Magdalena M Klemun, Goksin Kavlak, James McNerney, and Jessika E Trancik. "Evolution of hard and soft costs in technologies and the case of photovoltaic systems" (2019). [25] Ran Fu, David Feldman, Robert Margolis, Mike Woodhouse, and Kristen Ardani. US solar photovoltaic system cost benchmark: Q1 2017. Tech. rep. EERE Publication and Product Library, 2017. [26] Amro M Elshurafa, Shahad R Albardi, Simona Bigerna, and Carlo Andrea Bollino. "Estimating the learning curve of solar PV balance-of-system for over 20 countries: Implications and policy recommendations". Journalo f Cleaner Production (2018). [27] Kristen Ardani and Robert Margolis. DecreasingS oft Costs for Solar Photovoltaics by Improving the InterconnectionP rocess. A Case Study of PacificG as and Electric. Tech. rep. National Renewable Energy Lab. (NREL), Golden, CO (United States), 2015. [28] International Renewable Energy Agency. Solar Energy. https : / / www . irena . org / solar. [Online; accessed 15-April-20191. 2019. [291 Steven Hegedus and A. Luque. Handbook of Photovoltaic Science and Engineering. Vol. 2nd ed. Wiley, 2011. ISBN: 9780470721698. URL: http://tibproxy.mit. edu /login ?url=http: //search. ebscohost. com/login. aspx?direct=true&db=nlebk&AN=522082&site=ehost- live&scope=site. [30] Michael Boxwell. The Solar Electricity Handbook-2017 Edition: A simple, practical guide to solar energy-designing and installings olar photovoltaic systems. Greenstream Publishing, 2017. [31] U.S. Department of Energy. DOE Request for Information (RFI) - Plug and Play Technologies and Systems. https: //eere-exchange.energy.gov/FiLeContent.aspx?FileID=c481c344-7ela- 4643-bb8O-a7dO7abecbc2. [Online; accessed 22-April-2019]. 2011. [32] John J Bzura. "The AC module: An overview and update on self-contained modular PV systems". IEEE PES General Meeting. IEEE. 2010, pp. 1-3. [33] Aishwarya S Mundada, Yuenyong Nilsiam, and Joshua M Pearce. "A review of technical require- ments for plug-and-play solar photovoltaic microinverter systems in the United States". Solar Energy 135 (2016), pp. 455-470. [341 Christian Hoepfner. Plug and Play PV Systems for American Homes. Tech. rep. Fraunhofer USA, Inc., Boston, MA (United States), 2016. 111 [35] Matthew Kromer, Christian Hoepfner, and Jacqueline Ashmore. "Reducing the cost of residential- scale PV through "Plug & Play PV" systems and standardized electronic workflows". 2016 IEEE 43rd Photovoltaic Specialists Conference (PVSC). IEEE. 2016, pp. 3481-3486. [36] Matthew Kromer and Christian Hoepfner. Plug and Play PV Standards Portfolio. Tech. rep. Fraunhofer Center for Sustainable Energy Systems, 2016. [37] R Mark Lawson, Ray G Ogden, and Rory Bergin. "Application of modular construction in high- rise buildings". Journal of architecturale ngineering 18.2 (2011), pp. 148-154. [38] Richard M Swanson. "A vision for crystalline silicon photovoltaics". Progress in photovoltaics: Research and Applications 14.5 (2006), pp. 443-453. [39] Carlos Balaguer and Mohamed Abderrahim. "Trends in robotics and automation in construc- tion". Robotics and Automation in Construction. IntechOpen, 2008. [40] Martin Junginger, Wilfried Van Sark, and Andre Faaij. Technological learning in the energy sector: lessons for policy, industry and science. Edward Elgar Publishing, 2010. [41] Plug and Play Solar. Plug and Play Solar Kits. https: //plugandplaysolarkits . com. [Online; accessed 24-OCtober-2018]. 2018. [42] Plug-In-Solar Ltd. Plug In Solar - Your DIY Solar Solution. http: //www. pluginsolar. co. uk/. [Online; accessed 24-October-2018]. 2018. [43] Steven D Eppinger and Tyson R Browning. Design structure matrix methods and applications. MIT press, 2012. [44] Petri T Helo. "Product configuration analysis with design structure matrix". IndustrialM anage- ment & Data Systems 106.7 (2006), pp. 997-1011. [45] Tyson R Browning. "Design structure matrix extensions and innovations: a survey and new opportunities". IEEE Transactions on Engineering Management 63.1 (2016), pp. 27-52. [46] Maqsood Sandhu. "Project logistics with the dependency structure matrix approach-an analysis of a power plant delivery". InternationalJ ournalo f Logistics Systems and Management 2.4 (2006), pp. 387-403. [47] IdentifyingA rchitecturalM odularityi n the Smart Grid. Innovative Architectural Models. Phoenix, Arizona USA, 2011. [48] Peter Lang. "Nuclear Power Learning and Deployment Rates; Disruption and Global Benefits Forgone". Energies 10.12 (2017), p. 2169. [49] John M Marshall and Peter Navarro. "Costs of nuclear power plant construction: theory and new evidence". The Rand journal of economics (1991), pp. 148-154. [50] Nathan E Hultman, Jonathan G Koomey, and Daniel M Kammen. What history can teach us about the future costs of US nuclear power. 2007. 112 [51] U.S. Energy Information Administration. Analysis of nuclear power plant construction costs. Tech. rep. U.S. Department of Energy, Jan. 1986. DOI: 10.2172/6071600. URL: https://www.osti. gov/servlets/purl/6071600%20http://www.osti.gov/servlets/purt/6071600-oqRQro/. [52] Russell Gold. "Tab Swells to $25 Billion for Nuclear-Power Plant in Georgia". The Wall Street Journal (2017). [53] Gavin Bade. "Vogtle nuke cost could top $25B as decision time looms". Utility Drive (2017). [54] Alexander Gilbert, Benjamin K Sovacool, Phil Johnstone, and Andy Stirling. "Cost overruns and financial risk in the construction of nuclear power reactors: A critical appraisal". Energy Policy 102 (2017), pp. 644-649. [S5] U.S. Energy Information Administration. Assumptions to the Annual Energy Outlook 2018: Elec- tricity Market Module. 2018. [56] Wesley Cole, Trieu Mai, Jeffrey Logan, Daniel Steinberg, James McCall McCall, James Richards, Benjamin Sigrin, and Gian Porro. "2016 Standard Scenarios Report: A U.S. Electricity Sector Outlook". National Renewable Energy Laboratory Technical Report TP-6A20-66939 (2016). [57] William Rasin, Kiyoshi Ono, Evelyne Bertel, Pierre Berbey, Romney Duffey, Hussein Khalil, Izumi Kinoshita, Sermet Kuran, ManKi Lee, Eugene Onopko, Geoffrey Rothwell, and Kent Williams. Cost Estimating Guidelinesf or Generation IV Nuclear Energy Systems. Tech. rep. Generation IV International Forum, 2007, p. 181. [58] International Atomic Energy Agency. Guidance for the evaluation of innovative nuclear reactors and fuel cycles. Tech. rep. Department of Nuclear Energy, 2003. [59] A. Abdulla, I. L. Azevedo, and M. G. Morgan. "Expert assessments of the cost of light water small modular reactors". Proceedingso f the NationalA cademy of Sciences 110.24 (2013), pP. 9686- 9691. DOI: 10.1073/pnas.1300195110. URL: http: //www. pnas.org/cgi/doi/10.1073/pnas. 1300195110. [60] Terry L Schulz. "Westinghouse APlooo advanced passive plant". Nuclear Engineering and Design 236.14-16 (2006), pp. 1547-1557. [61] Energy Innovation Portfolio Plan FY2018-2022. Tech. rep. U.S. Department of Energy, 2017. [62] David Petti, Jacopo Buongiorno, Michael Corradini, and John Parsons. "The Future of Nuclear Energy in a Carbon-Constrained World". Massachusetts Institute of Technology Energy Initiative (MITEI) (2018). [63] M. D. Carelli, P. Garrone, G. Locatelli, M. Mancini, C. Mycoff, P. Trucco, and M. E. Ricotti. "Economic features of integral, modular, small-to-medium size reactors". Progress in Nuclear Energy 52.4 (2010), pp. 403-414. ISSN: 01491970. DOI: 10. 101 6/j . pnucene. 2009.09.003. URL: http://dx.doi.org/10. 1016/j.pnucene.2009.09.003. 113 [64] J Delene and C Hudson. Cost Estimate Guidelinesf or Advance Nuclear Power Technologies (R3). Tech. rep. Oak Ridge National Laboratory, 1993. [65] Robert Rosner and Stephen Goldberg. "Small Modular Reactors-Key to Future Nuclear Power Generation in the US". Energy Policy Institute at Chicago, The University of Chicago, Chicago (2011). [66] Lauren M Boldon and Piyush Sabharwall. "Small modular reactor: First-of-a-Kind (FOAK) and Nth-of-a-Kind (NOAK) Economic Analysis". Idaho NationalL ab. (INL): Idaho Falls, ID, USA (2014). [67] U.S. Energy Information Administration. Electric Power Monthly. 2018. [68] World Nuclear Association. Country Profile: Nuclear Power in the USA. 2018. [69] Michael Buchdahl Roth and Paulina Jaramillo. "Going nuclear for climate mitigation: An analysis of the cost effectiveness of preserving existing US nuclear power plants as a carbon avoidance strategy". Energy 131 (2017), pp. 67-77. [70] Gokul Iyer, C Ledna, L Clarke, H McJeon, J Edmonds, and M Wise. "GCAM-USA analysis of US electric power sector transitions". Pacific Northwest National Laboratory, Richland, Washington (2017). [71] Intergovernmental Panel on Climate Change. Emissions Scenarios. A Special Report of Working Group III of the Intergovernmental Panel on Climate Change. Tech. rep. 2000, p. 20. [72] Jonathan Koomey and Nathan E Hultman. "A reactor-level analysis of busbar costs for US nuclear plants, 1970-2005". Energy Policy 35.11 (2007), pp. 5630-5642. [73] Jessica R. Lovering, Arthur Yip, and Ted Nordhaus. "Historical construction costs of global nuclear power reactors". Energy Policy 91 (2016), pp. 371-382. ISSN: 03014215. DOI: 10. 1 01 6/j. enpol.2016.01.011.URL: http://dx.doi.org/10.1016/j.enpol.2016.01.011. [74) Yuhji Matsuo and Hisanori Nei. "An analysis of the historical trends in nuclear power plant construction costs: The Japanese experience". Energy policy 124 (2019), pp. 180-198. [75] Arnulf Grubler. "The costs of the French nuclear scale-up: A case of negative learning by doing". Energy Policy 38.9 (2010), pp. 5174-5188. [76] Michel Berthelemy and Lina Escobar Rangel. "Nuclear reactors' construction costs: The role of lead-time, standardization and technological progress". Energy Policy 82 (2015), pp. 118-130. [77] Gordon MacKerron. "Nuclear Costs: Why do they keep rising?" Energy Policy 20.7 (1992), pp. 641- 652. [78] Mark Cooper. "Policy Challenges of Nuclear Reactor Construction, Cost Escalation and Crowding Out Alternatives". Institutef or Energy and the Environment, Vermont Law ... 5.6 2 (2010), pp. 1-5. URL: http://www.ises.org.il/assets/fies/News/20100909%7B%5C_%7DcooperStudy.pdf. 114 [79] Edward S Rubin, Ines M L Azevedo, Paulina Jaramillo, and Sonia Yeh. "Review article A review of learning rates for electricity supply technologies". Energy Policy 86 (2015), pp. 198-218. IsSN: 0301-4215. DOI: 10.1016/j.enpol.2015.06.011. URL: http://dx.doi.org/10.1016/j. enpol.2015.06.011. [80] Amory B Lovins. "The origins of the nuclear power fiasco". The Politics of Energy Research and Development, Energy Policy Studies vol 3 (1986), pp. 7-34. [81] Richard K Lester and Mark J McCabe. "The effect of industrial structure on learning by doing in nuclear power plant operation". The Rand Journal of Economics (1993), pp. 418-438. [82] United Engineers & Constructors. Capital Cost: Pressurized Water Reactor Plant. Tech. rep. Nuclear Regulatory Commission & Energy Research and Development Administration, 1977. DOI: 10.2172/6033498. [83] United Engineers & Constructors. Energy Economic Data Base (EEDB) Program: Phase I. Tech. rep. U.S. Department of Energy, 1979. [84] United Engineers & Constructors. Energy Economic Data Base (EEDB) Program: Phase II. Tech. rep. U.S. Department of Energy, 1981. DOI: 10. 2172/6477534. [85] United Engineers & Constructors. Energy Economic Data Base (EEDB) Program: Phase III. Tech. rep. U.S. Department of Energy, 1981. [86] United Engineers & Constructors. Energy Economic Data Base (EEDB) Program:P hase IV. Tech. rep. U.S Department of Energy, 1981. DOI: 10. 2172/5388083. [87] United Engineers & Constructors. Energy Economic Data Base (EEDB) Program:P hase VI. Tech. rep. U.S. Department of Energy, 1984. DOI: 10. 2172/6504693. [88] United Engineers & Constructors. Energy Economic Data Base (EEDB) Program: Phase VII. Tech. rep. U.S. Department of Energy, 1985. DOI: 10. 2172/5237914. [89] United Engineers & Constructors. Energy Economic Data Base (EEDB) Program:P hase IX. Tech. rep. U.S. Department of Energy, 1988. [90] Grant Harris, Phil Heptonstall, Robert Gross, and David Handley. "Cost estimates for nuclear power in the UK". Energy Policy 62 (2013), pp. 431-442. [91] Giovanni Maronati, Bojan Petrovic, Jurie J. Van Wyk, Matthew H. Kelley, and Chelsea C. White. "EVAL: A methodological approach to identify NPP total capital investment cost drivers and sensitivities". Progress in Nuclear Energy 104 (2018), pp. 190-202. ISSN: 01491970. DOI: 10. 1016/ j.pnucene.2017.09.014. URL: https://doi.org/10.1016/j.pnucene.2017.09.014. [92] J Delene and C Hudson. Cost Estimate Guidelines for Advance Nuclear Power Technologies (R2). Tech. rep. Oak Ridge National Laboratory, 1990. 115 [93] International Panel on Climate Change. Climate Change 2007 Synthesis Report. Tech. rep. 2007, p. 104. DOI: 10.1256/004316502320517344. arXiv: 9809069v [arXiv:gr-qcl. [94] Philip Eash-Gates, Ajinkya Shrish Kamat, Magdalena M. Klemun, Goksin Kavlak, and Jessika E. Trancik. "Effects of plug-and-play photovoltaic design on balance of system costs". In preparation for submission (2019). [95] Philip Eash-Gates, Magdalena M. Klemun, Goksin Kavlak, James McNerney, Jacopo Buongiorno, David A. Petti, and Jessika E. Trancik. "Sources of cost overruns in nuclear power plant con- struction". In preparationf or submission (2019). [96] Peter Brownell and Kenneth A Merchant. "The budgetary and performance influences of product standardization and manufacturing process automation". Journal ofAccounting Research (1990), pp. 388-397. [97] Kristen Ardani, Carolyn Davidson, Robert Margolis, and Erin Nobler. State-Level Comparison of Processes and Timelines for Distributed Photovoltaic Interconnection in the United States. Tech. rep. National Renewable Energy Lab.(NREL), Golden, CO (United States), 2015. [98] Chiara Candelise, Mark Winskel, and Robert JK Gross. "The dynamics of solar PV costs and prices as a challenge for technology forecasting". Renewable and Sustainable Energy Reviews 26 (2013), pp. 96-107. [99] National Fire Protection Association, National Board of Fire Underwriters, and National Fire Protection Association. National Electrical Code Committee. National electrical code, 2017 ed. Vol. 70. National Fire Protection Association., 2016. [ioo] Meter Socket Specification Sheet. E361188. Simple ConnectDER. ConnectDER. Aug. 2018. [lol] Fraunhofer Center for Sustainable Energy Systems. Plug and Play PV Systems. https: //www. cse. fraunhofer. org/pnp. [Online; accessed 26-October-2018]. 2018. [102] International Code Council. 2018 Internationale nergy conservation code. Vol. 1. Cengage Learn- ing, 2017. [103] Montgomery County Maryland. Residential Photovoltaic (Solar) Inspections. https : / / www . montgomerycountymd . gov/DPS/Resources/Files/RCI/ResidentialPhotovoltaic(Solar) Inspections. pdf. [Online; accessed 03-April-2019]. 2019. [104] City of Milwaukee Wisconsin. Solar Permitting Information. https : //city. milwaukee. gov/ MilwaukeeShines/Solar-Professionals/Permitting. htm. [Online; accessed lo-April-2019]. 2019. [105] Interconnection Requirements For Distributed Generation. Rev 7.1. Arizona Public Service Com- pany. July 2012. 116 [1061 Consolidated Edison Company of New York. Verification Testing and Inspection Checklist for Solar Projects. https: / /www. coned. com/ -/media/ files /coned/documents /save - energy- money / using- private - generation /solar - verification - testing- and - inspection - checklist. pdf. [Online; accessed 03-April-2019]. 2017. [107] NC Clean Energy Technology Center. Plug and Play PV. https: //nccleantech . ncsu . edu/ technology/renewable-energy/solar/plug-and-play-pv/. [Online; accessed u-October- 2018]. 2018. [1o8] Electrical Three-Line Easy Plug Roof/Ground Mount. 2018-PVO112. Plug & Play Solar. Jan. 2018. [109] Distribution Interconnection Guide for Customer-Owned Facilities less than io MW. Simplified Diagrams for Solar PV: Figures i to 5. Revision 9.o. Austin Energy. Apr. 2018. [no] Alan Goodrich, Ted James, and Michael Woodhouse. Residential, commercial, and utility-scale photovoltaic (PV) system prices in the United States: current drivers and cost-reduction opportuni- ties. Tech. rep. National Renewable Energy Lab.(NREL), Golden, CO (United States), 2012. [ill] Kristen Ardani, Galen Barbose, Robert Margolis, Ryan Wiser, David Feldman, and Sean Ong. Benchmarking Non-Hardware Balance of System (Soft) Costs for US Photovoltaic Systems Using a Data-DrivenA nalysis from PV Installer Survey Results. Tech. rep. National Renewable Energy Lab.(NREL), Golden, CO (United States), 2012. [112] SMA Sunny Boy Specification Sheet: 3.0-US / 3.8-US / 5.0-US / 6.o-US / 7.0-US / 7.7-US. SB3.o- 7.7-US-DUS184327. Version: 2.7. SMA Solar Technology AG. [113] Unirac Code-Compliant Installation Manual. Pub 15JANoi-ucc. UNIRAC. 2015. [114] MNPV6-25o AC Disconnect Installation Instructions. 10-183-1. Rev: B. Midnite Solar. [115] Fraunhofer USA. Plug and Play PV. https: //www. cse. f raunhof er. org/pnp. [Online; accessed 12-March-2019]. 2016. [116] Plug & Play Solar. Easy Plug Roof Mount Solar Panel Unboxing and Installation. https :/ / plugandplaysolarkits . com / collections / plug - in - roof - mount / products / plug - and - play-roof-mount-solar-unit?variant=1 6876437176433. [Online; accessed 12-March-2019]. 2018. [117] U.S. Bureau of Economic Analysis. Table 1.1.4. Price Indexes for Gross Domestic Product. https: //www. bea . gov. [Online; accessed 20-March-2019]. 2019. [118] Q Cable Specification Sheet. Q-12-XX-240. Connectorized cables. Enphase Energy. May 2018. [119] Richard Schmalensee. The future of solar energy: An interdisciplinaryM IT study. Energy Initiative, Massachusetts Institute of Technology, 2015. 117 [120] Rebecca Jones-Albertus, David Feldman, Ran Fu, Kelsey Horowitz, and Michael Woodhouse. "Technology advances needed for photovoltaics to achieve widespread grid price parity". Progress in photovoltaics: research and applications 24.9 (2016), pp. 1272-1283. [121] Michael H Coddington. "Evaluating the rationale for the utility-accessible external disconnect switch". 2008 33rd IEEE Photovoltaic Specialists Conference. IEEE. 2008, pp. 1-5. [122] Thomas Udo Pimmler and Steven D Eppinger. "Integration analysis of product decompositions" (1994). [123] David Feldman, Barry Friedman, and Robert Margolis. Financing, overhead, and profit: an in-depth discussion of costs associated with third-partyf inancing of residential and commercial photovoltaic systems. Tech. rep. National Renewable Energy Lab.(NREL), Golden, CO (United States), 2013. [124] David M Scholten, N Ertugrul, and WL Soong. "Micro-inverters in small scale PV systems: A review and future directions". 2013 Australasian Universities Power Engineering Conference (AUPEC). IEEE. 2013, pp. 1-6. [125] D Pal, H Koniki, and P Bajpai. "Central and micro inverters for solar photovoltaic integration in AC grid". 2016 National Power Systems Conference (NPSC). IEEE. 2016, pp. 1-6. [126] Jessika E Trancik, Joel Jean, Goksin Kavlak, Magdalena M Klemun, Morgan R Edwards, James McNerney, Marco Miotti, Patrick R Brown, Joshua M Mueller, and Zachary A Needell. Technology improvement and emissions reductions as mutually reinforcing efforts: Observationsf rom the global development of solar and wind energy. Tech. rep. MIT, 2015. [127] Ruben Laleman, Johan Albrecht, and Jo Dewulf. "Life cycle analysis to estimate the environmen- tal impact of residential photovoltaic systems in regions with a low solar irradiation". Renewable and SustainableE nergy Reviews 15.1 (2011), pp. 267-281. [128] International Code Council. 2015 Internationale nergy conservation code. Vol. 1. Cengage Learn- ing, 2014. [129] Cityof Austin Texas. Commercial SolarR eady Guidelines. http: //www. austintexas. gov/sites/ default/file f iles/DevelopmentServices/Applications/commercial/Commercial_ Sola rReadyGui del ines . pdf. [Online; accessed lo-April-2019]. 2017. [130] The State of California. 2016 Nonresidential Compliance Manual: Solar Ready. https: //www. energy. ca . gov/2015pubtications/CEC-400-2015-033/chapters/chapter 09 solar_ ready. pdf. [Online; accessed 1o-April-2019]. 2017. [131] Secretary of Energy Advisory Board. Report of the Task Force on the Future of Nuclear Power. Tech. rep. U.S. Department of Energy, 2016. 118 [132] John M Deutch, Charles W Forsberg, Andrew C Kadak, Mujid S Kazimi, Ernest J Moniz, John E Parsons, et al. "Update of the MIT 2003 Future of Nuclear Power". Cambridge, Mass.: Reportfor Massachusetts Institute of Technology. Retrieved September 17 (2009), p. 2009. [133] International Atomic Energy Agency. Power Reactor Information System: History of Electricity Production. Tech. rep. 2018. URL: https: //pris. iaea.org. [134] World Nuclear Association. Energy Subsidies. 2018. URL: http: / /www . worldenergyoutlook. org/resources/energysubsidies/ (visited on 2018). [1351 J Samuel Walker and Thomas R. Wellock. "A Short History of Nuclear Regulation, 1946-2009". United States Nuclear Regulatory Commission (2010), pp. 1-116. [136] International Atomic Energy Agency. "Power Reactor Information System: Country Details, United States of America" (2018). [137] U.S. Energy Information Administration. Nuclear Power Plant Construction Activity, 1984. Tech. rep. U.S. Department of Energy, 1985. [138] Leo Sveikauskas, Samuel Rowe, James Mildenberger, Jennifer Price, and Arthur Young. "Measur- ing productivity growth in construction". Monthly Labor Review, U.S. Bureau of Labor Statistics (2014). DOI: https://doi.org/10.21916/mlr.2018.1. [139] "Phone interview". David Jones, Southern Company (2018). [140] Charles Komanoff. Power plant cost escalation: nuclear and coal capital costs, regulation, and economics. Vol. 12. Van Nostrand Reinhold Company, 1981. [141] Charles Komanoff. "Assessing the high costs of new nuclear power plants". Public Utilities Fortnightly 114.8 (1984), pp. 33-38. [142] Lena Neij. "Cost development of future technologies for power generation-A study based on experience curves and complementary bottom-up assessments". Energy policy 36.6 (2008), pp. 2200-2211. [143] Martin Junginger, Erika de Visser, Kurt Hjort-Gregersen, Joris Koornneef, Rob Raven, Andre Faaij, and Wim Turkenburg. "Technological learning in bioenergy systems". Energy Policy 34.18 (2006), pp. 4024-4041. [144] Martin Weiss, Martin Junginger, Martin K Patel, and Kornelis Blok. "A review of experience curve analyses for energy demand technologies". Technological forecasting and social change 77.3 (2010), pp. 411-428. [145] Ines Azevedo, Paulina Jaramillio, Edward S. Rubin, and Sonia Yeh. "Technology Learning Curves and the Future Cost of Electric Power Generation Technology". EPRI 18th Annual Energy and Climate Change Research Seminar. Washington, DC, 2013. [146] Gunnar A Harstead. Component nuclear containment structure. US Patent 4,175,005. Nov. 1979. 119 [147] Lawrence E Conway and William A Stewart. Passive containment cooling system. US Patent 5,049,353. Sept. 1991. [148] Frank T Vereb, William L Brown, and Forrest T Johnson. Passive containment air cooling for nuclearp ower plants. US Patent 9,177,675. Nov. 2015. [149] Fluor Corporation. Basis of "Estimate to Complete": Plant Vogtle Units 3 & 4, VC Summer Units 2 & 3. Tech. rep. Greenville, SC, 2016. [150] Bechtel Power Corporation. Cost and Schedule Assessment for the Completion of Construction for Southern Nuclear Operating Company's Vogtle Units 3 & 4. Tech. rep. 2017. URL: http: //www.psc.state.ga.us/factsv2/Document.aspx?documentNumber=171748. [151] John Borcherding and Scott Sebastian. "Major factors influencing craft productivity in nuclear power plant construction". Transactions of the American Association of Cost Engineers (1980), pp. 1.1.1-1.1.5. [152] Amit H Varma, Sanjeev R Malushte, Kadir C Sener, and Zhichao Lai. "Steel-plate composite (SC) walls for safety related nuclear facilities: Design for in-plane forces and out-of-plane moments". Nuclear Engineering and Design 269 (2014), pp. 240-249. [153] John Deutch, Ernest Moniz, S Ansolabehere, Michael Driscoll, Paul Gray, John Holdren, Paul Joskow, Richard Lester, and Neil Todreas. "The Future of Nuclear Power". an MIT Interdisci- plinary Study, http://web. mit. edu/nuclearpower (2003). [154] Robynne Boyd. Safety ConcernsD elay Approval of the First U.S. Nuclear Reactor in Decades. 2010. [iss] Kenneth J Arrow. "The economic implications of learning by doing". The review of economic studies 29.3 (1962), pp. 155-173. [156] Yuichiro Anzai and Herbert A Simon. "The theory of learning by doing." Psychological review 86.2 (1979), p. 124. [157] U.S. Nuclear Regulatory Commission. Backgrounder on NRC Resident Inspectors Program. 2018. URL: %5Curl%7Bhttps :/ /www. nrc. gov/reading- rm/doc-collections/fact-sheets/ resident-inspectors-bg.html%7D. [158] U.S. Nuclear Regulatory Commission. Inspections, Tests, Analyses, and Acceptance Criteria( ITAAC). 2018. URL: %5Curl%7Bhttps: //www.nrc.gov/reactors/new-reactors/oversight/itaac. html%7D. [159] CK Paulson. "Westinghouse APiooo Advanced Plant Simplification Results, Measures, and Ben- efits". loth International Conference on Nuclear Engineering. American Society of Mechanical Engineers. 2002, pp. 1o65-1o68. 120 [160] Yi Wei Daniel Tay, Biranchi Panda, Suvash Chandra Paul, Nisar Ahamed Noor Mohamed, Ming Jen Tan, and Kah Fai Leong. "3D printing trends in building and construction industry: a review". Virtual and Physical Prototyping 12.3 (2017), pp. 261-276. [161] Steven J Keating, Julian C Leland, Levi Cai, and Neri Oxman. "Toward site-specific and self- sufficient robotic fabrication on architectural scales". Science Robotics 2.5 (2017), eaam8986. [162] Md Salim and Leonhard E Bernold. "Effects of design-integrated process planning on produc- tivity in rebar placement". Journal of Construction Engineering and Management 120.4 (1994), pp. 720-738. [163] Development of Non-ProprietaryU ltra-High Performance Concretef or Use in The Highway Bridge Sector. Tech. rep. U.S. Department of Transportation, 2017. [164] Cost and Ecological Feasibilityo f Using Ultra- High Performance Concrete in Highway Bridge Piers. Tech. rep. Nevada Department of Transportation, 2017. [165] Robert Devine, Yahya C Kurama, Ashley P Thrall, Scott Edward Sanborn, Mathew Van Liew, Steven M Barbachyn, Max Ducey, and Madalyn Sower. PrefabricatedH igh-Strength Rebar Sys- tems with High-PerformanceC oncretef or Accelerated Construction of Nuclear Concrete Structures. Tech. rep. Sandia National Lab. (SNL-NM), Albuquerque, NM (United States), 2015. [166] Use of High-Strength Reinforcement in Earthquake-Resistant Concrete Structures. Tech. rep. Na- tional Institutes of Standards and Technology, 2017. [167] Ana Spasojevic. "Structural Implications of Ultra-High Performance Fiber-Reinforced Concrete in Bridge Design". Acole Polytechnique Fidedrale de Lausanne (EPFL), Diploma Thesis (2008). [168] Caroline Schlaseman. Application of Advanced Construction Technologies to New Nuclear Power Plants. Tech. rep. Nuclear Regulatory Commission, 2004. [169] Bridge Design and DraftingM anual. Tech. rep. Oregon Department of Transportation, 2017. [170] Raul B Rebak and Xiaoyuan Lou. Environmental Cracking and IrradiationR esistant Stainless Steels by Additive Manufacturing. Tech. rep. General Electric Company (GE), Schenectady, NY (United States), 2018. [171] Ambuj D Sagar and Bob Van der Zwaan. "Technological innovation in the energy sector: R&D, deployment, and learning-by-doing". Energy Policy 34-17 (2006), pp. 2601-2608. [172] Arnulf Grubler. "An assessment of the Costs of the French Nuclear PWR Program 1970-2000" (2009). [173] Westinghouse Electric Company. APiooo Design Control Document. Tech. rep. 2011. [174] T.D. Kelly and G.R. Matos. Historical statistics for mineral and material commodities in the United States (2016 version). Tech. rep. US Geological Survey Data Series, 2016. URL: http: //minerals. usgs. gov/minerals/pubs/historical-statistics/. 121 [175] Southern Company PlantV ogtle 3 and 4: Milestones 2017. 2017. URL: https: //www. southern company. com/innovation/nuclear-energy/plant-vogtle-3-and-4.html. [176] Patrick Champlin. "Techno-Economic Evaluation of Cross-Cutting Technologies for Cost Re- duction in Nuclear Power Plants (thesis)". Doctoral dissertation. Massachusetts Institute of Technology, 2018. URL: ADD%20URL%20WHEN%20POSTED. [177] WE Cummins, RF Wright, and TL Schulz. "APiooo status overview" (2001). [178] MD Carelli, JW Winters, WE Cummins, and HJ Bruschi. "Safety features and research needs of westinghouse advanced reactors". Advanced Nuclear (2002), p. 119. [179] Regis A Matzie. "APiooo will meet the challenges of near-term deployment". Nuclear engineering and design 238.8 (2008), pp. 1856-1862. [180] Balendra Sutharshan, Meena Mutyala, Ronald P Vijuk, and Alok Mishra. "The APioooTM reactor: passive safety and modular design". Energy Procedia7 (2011), pp. 293-302. [181] David L. Chandler. "Thin coating on condensers could make power plants more efficient". MIT News Office (2015). [182] Bahman Zohuri, Patrick J McDaniel, and Cassiano RR De Oliveira. "Advanced nuclear open air- brayton cycles for highly efficient power conversion". Nuclear Technology 192.1 (2015), pp. 48- 60. [183] Daniel J Preston, Daniela L Mafra, Nenad Miljkovic, Jing Kong, and Evelyn N Wang. "Scalable graphene coatings for enhanced condensation heat transfer". Nano letters 15.5 (2015), pp. 2902- 2909. 122