Aeronautics and Astronautics - Ph.D. / Sc.D.
http://hdl.handle.net/1721.1/7766
Fri, 18 Apr 2014 16:47:56 GMT2014-04-18T16:47:56ZA robust simplex cut-cell method for adaptive high-order discretizations of aerodynamics and multi-physics problems
http://hdl.handle.net/1721.1/85764
A robust simplex cut-cell method for adaptive high-order discretizations of aerodynamics and multi-physics problems
Sun, Huafei
Despite the wide use of partial differential equation (PDE) solvers, lack of automation still hinders realizing their full potential in assisting engineering analysis and design. In particular, the process of establishing a suitable mesh for a given problem often requires heavy person-in-the-loop involvement. This thesis presents work toward the development of a robust PDE solution framework that provides a reliable output prediction in a fully-automated manner. The framework consists of: a simplex cut-cell technique which allows the mesh generation process to be independent of the geometry of interest; a discontinuous Galerkin (DG) discretization which permits an easy extension to high-order accuracy; and an anisotropic output-based adaptation which improves the discretization mesh for an accurate output prediction in a fully-automated manner. Two issues are addressed that limit the automation and robustness of the existing simplex cut-cell technique in three dimensions. The first is the intersection ambiguity due to numerical precision. We introduce adaptive precision arithmetic that guarantees intersection correctness, and develop various techniques to improve the efficiency of using this arithmetic. The second is the poor quadrature quality for arbitrarily shaped elements. We propose a high-quality and efficient cut-cell quadrature rule that satisfies a quality measure we define, and demonstrate the improvement in nonlinear solver robustness using this quadrature rule. The robustness and automation of the solution framework is then demonstrated through a range of aerodynamics problems, including inviscid and laminar flows. We develop a high-order DG method with a dual-consistent output evaluation for elliptic interface problems, and extend the simplex cut-cell technique for these problems, together with a metric-optimization adaptation algorithm to handle cut elements. This solution strategy is further extended for multi-physics problems, governed by different PDEs across the interfaces. Through numerical examples, including elliptic interface problems and a conjugate heat transfer problem, high-order accuracy is demonstrated on non-interface-conforming meshes constructed by the cut-cell technique, and mesh element size and shape on each material are automatically adjusted for an accurate output prediction.
Thesis: Ph. D., Massachusetts Institute of Technology, Department of Aeronautics and Astronautics, 2013.; Cataloged from PDF version of thesis.; Includes bibliographical references (pages 189-199).
Tue, 01 Jan 2013 00:00:00 GMThttp://hdl.handle.net/1721.1/857642013-01-01T00:00:00ZComputational modeling of size-dependent superelasticity in shape memory alloys
http://hdl.handle.net/1721.1/85763
Computational modeling of size-dependent superelasticity in shape memory alloys
Qiao, Lei, Ph. D. Massachusetts Institute of Technology
The superelastic effect in shape memory alloys (SMAs) is attributed to the stress-induced reversible austenitic-martensitic phase transformations. It is characterized by the development of significant strains which are fully recoverable upon unloading, and also characterized by the stress-hysteresis in the loading and unloading cycle which corresponds to the energy dissipated during phase transformations. Recently, experiments have revealed size-dependent effects in the superelastic responses of SMAs at micro- and nanoscales. For instance, the CuAlNi microwires and submicron pillars show a substantially higher capacity for the energy dissipation than that of bulk samples, which offers a significant promise for the applications in protective materials. In this thesis, a continuum model is developed in order to improve our understanding of size effects in SMAs at small scales. The modeling approach combines classic superelastic models, which use the volume fraction as an internal variable to represent the martensitic phase transformation, with strain gradient plasticity theories. Size effects are incorporated through two internal length scales, an energetic length scale and a dissipative length scale, which correspond to the martensitic volume fraction gradient and its time rate of change, respectively. Introducing the gradient of the martensitic volume fraction leads to coupled macro- and microforce balance equations, where the displacements and the martensitic volume fraction are both independent fields. A variational formulation for the temporally-discretized coupled macro- and microforce balance equations is proposed, as well as a computational framework based on this formulation. A robust and scalable parallel algorithm is implemented within this computational framework, which enables the large-scale three-dimensional study of size effects in SMAs with unprecedented resolution. This modeling and computational framework furnishes, in effect, a versatile tool to analyze a broad range of problems involving size effects in superelasticity with the potential to guide microstructure design and optimization. In particular, the model captures the increase of the stress hysteresis and strain hardening in bulk polycrystalline SMAs for decreasing grain size, as well as the increase of the residual strain for decreasing pillar size in NiTi pillars. The model confirms that constraints like grain boundaries and the surface Ti oxide layer are responsible for the size-dependent superelasticity in SMAs.
Thesis: Ph. D., Massachusetts Institute of Technology, Department of Aeronautics and Astronautics, 2013.; Cataloged from PDF version of thesis.; Includes bibliographical references (pages 145-155).
Tue, 01 Jan 2013 00:00:00 GMThttp://hdl.handle.net/1721.1/857632013-01-01T00:00:00ZBayesian nonparametric reward learning from demonstration
http://hdl.handle.net/1721.1/85762
Bayesian nonparametric reward learning from demonstration
Michini, Bernard (Bernard J.)
Learning from demonstration provides an attractive solution to the problem of teaching autonomous systems how to perform complex tasks. Demonstration opens autonomy development to non-experts and is an intuitive means of communication for humans, who naturally use demonstration to teach others. This thesis focuses on a specific form of learning from demonstration, namely inverse reinforcement learning, whereby the reward of the demonstrator is inferred. Formally, inverse reinforcement learning (IRL) is the task of learning the reward function of a Markov Decision Process (MDP) given knowledge of the transition function and a set of observed demonstrations. While reward learning is a promising method of inferring a rich and transferable representation of the demonstrator's intents, current algorithms suffer from intractability and inefficiency in large, real-world domains. This thesis presents a reward learning framework that infers multiple reward functions from a single, unsegmented demonstration, provides several key approximations which enable scalability to large real-world domains, and generalizes to fully continuous demonstration domains without the need for discretization of the state space, all of which are not handled by previous methods. In the thesis, modifications are proposed to an existing Bayesian IRL algorithm to improve its efficiency and tractability in situations where the state space is large and the demonstrations span only a small portion of it. A modified algorithm is presented and simulation results show substantially faster convergence while maintaining the solution quality of the original method. Even with the proposed efficiency improvements, a key limitation of Bayesian IRL (and most current IRL methods) is the assumption that the demonstrator is maximizing a single reward function. This presents problems when dealing with unsegmented demonstrations containing multiple distinct tasks, common in robot learning from demonstration (e.g. in large tasks that may require multiple subtasks to complete). A key contribution of this thesis is the development of a method that learns multiple reward functions from a single demonstration. The proposed method, termed Bayesian nonparametric inverse reinforcement learning (BNIRL), uses a Bayesian nonparametric mixture model to automatically partition the data and find a set of simple reward functions corresponding to each partition. The simple rewards are interpreted intuitively as subgoals, which can be used to predict actions or analyze which states are important to the demonstrator. Simulation results demonstrate the ability of BNIRL to handle cyclic tasks that break existing algorithms due to the existence of multiple subgoal rewards in the demonstration. The BNIRL algorithm is easily parallelized, and several approximations to the demonstrator likelihood function are offered to further improve computational tractability in large domains. Since BNIRL is only applicable to discrete domains, the Bayesian nonparametric reward learning framework is extended to general continuous demonstration domains using Gaussian process reward representations. The resulting algorithm, termed Gaussian process subgoal reward learning (GPSRL), is the only learning from demonstration method that is able to learn multiple reward functions from unsegmented demonstration in general continuous domains. GPSRL does not require discretization of the continuous state space and focuses computation efficiently around the demonstration itself. Learned subgoal rewards are cast as Markov decision process options to enable execution of the learned behaviors by the robotic system and provide a principled basis for future learning and skill refinement. Experiments conducted in the MIT RAVEN indoor test facility demonstrate the ability of both BNIRL and GPSRL to learn challenging maneuvers from demonstration on a quadrotor helicopter and a remote-controlled car.
Thesis: Ph. D., Massachusetts Institute of Technology, Department of Aeronautics and Astronautics, 2013.; Cataloged from PDF version of thesis.; Includes bibliographical references (pages 123-132).
Tue, 01 Jan 2013 00:00:00 GMThttp://hdl.handle.net/1721.1/857622013-01-01T00:00:00ZNetworked control of aircraft operations at airports and in terminal areas
http://hdl.handle.net/1721.1/85761
Networked control of aircraft operations at airports and in terminal areas
Khadilkar, Harshad (Harshad Dilip)
The goal of this thesis is to develop a control strategy for airport operations that integrates the management of arrivals and departures. The strategy is based on four central ideas: (1) the objective of reducing aircraft flight times, taxi times and fuel burn, (2) the emphasis on developing models using data from actual aircraft operations, (3) the need to be compatible with current air traffic control procedures, and (4) the requirement to not adversely affect airport performance. The scope of this work covers the airport surface and arrival airspace, which are two of the most congested regions of the air transportation network. A new approach is proposed for modeling airport surface operations. Drawing an analogy from the field of network congestion control, the airport surface is assumed to be a network consisting of major taxiways and their intersections. Posing the problem in this framework relaxes the requirement of precisely predicting the taxi time of each aircraft, instead emphasizing the accurate representation of the underlying stochastic processes. At the same time, it allows one to address the issues of network stability and performance through analytical approaches. Based on this model for surface operations, a control algorithm is developed for regulating the time of entry of aircraft into the network. Simulations show that this strategy can significantly reduce surface congestion and aircraft fuel burn without hampering airport performance. The arrival airspace control algorithm presented in this thesis proposes a hybrid centralized / distributed algorithm for conflict detection and resolution. It combines distributed control in low-density airspace with centralized control in high-density terminal areas. This approach has the advantage of reducing ground infrastructure cost due to decentralization, while still operating at an efficiency level close to that of a fully centralized control strategy. The arrival and departure control algorithms are then combined to formulate an integrated strategy for managing airport operations, significantly improving the separate gains that can be obtained from each component.
Thesis: Ph. D., Massachusetts Institute of Technology, Department of Aeronautics and Astronautics, 2013.; Cataloged from PDF version of thesis.; Includes bibliographical references (pages 149-157).
Tue, 01 Jan 2013 00:00:00 GMThttp://hdl.handle.net/1721.1/857612013-01-01T00:00:00Z