Enhancing Manufacturing Performance to Plan with Predictive Analytics
Name
Weisberg_Thesis_v2_3_Sig_Page_Removal[48].pdf
Description
Thesis PDF
Size
1.08 MB
Format
Adobe PDF
Checksum (MD5)
46cdf62e0d1c0ce70dde8bf1edad7435
Author(s)
Weisberg, Joshua
Advisor(s)
Perakis, Georgia
Amin, Saurabh
Date Issued
June 2023
Publisher
Massachusetts Institute of Technology
Abstract
Modern manufacturing requires meticulous planning to coordinate tightly wound supply chain activities in the face of disruption. This is especially true for automotive companies, which produce complex products at high rates. Their production planning process involves estimating the demand for various vehicles, determining the most profitable mix of products to meet that demand, and then selecting the production parameters which provide maximum efficiency. All this is done while balancing the the short term demands of a volatile market with the long term implications of capital equipment purchases, staffing changes, and supplier management. From the time of each decision to the day of production, demand may change, supply may be disrupted, and manufacturing performance may fall short of expectations. These uncertainties lead to high error in production plans, which propagates to suppliers, other areas of the business, and future periods. Changes harm stability, efficiency, and thus profitability for all stakeholders. This study shows how predictions of performance can be used to revise a plan, using predictive analytics models trained on the characteristics of the plan. To this end, 480+ features are developed to describe plan characteristics and recent manufacturing performance. Several algorithms are utilized to evaluate the relationship between these features and manufacturing performance to plan, measured by ratios of actual production rate to that planned, and hours actually worked to planned. Results of the best performing features, algorithms, and modeling architectures on out-of-sample manufacturing days in the Post-Covid Era showed Median Absolute Error improvements of 40%-60% over a 3-month lead time and 10%-40% over a 1-month lead time across several production lines. These reductions in error can improve stability such that better decisions can be made. Interpretation of the predictive models can lead to improvements in the factory’s ability to meet demand. Benefactors include customers looking to purchase and receive their desired products, employees needing more day-to-day consistency, and suppliers aiming to maintain a healthy business. The only certainty in operations is uncertainty, making it critical for operations companies to improve their understanding and estimation of their performance to plan.
MIT Department
Sloan School of Management
Massachusetts Institute of Technology. Department of Civil and Environmental Engineering
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
