A. James Clark School of Engineering
Permanent URI for this communityhttp://hdl.handle.net/1903/1654
The collections in this community comprise faculty research works, as well as graduate theses and dissertations.
Browse
50 results
Search Results
Item OPTIMAL PROBING OF BATTERY CYCLES FOR MACHINE LEARNING-BASED MODEL DEVELOPMENT(2024) Nozarijouybari, Zahra; Fathy, Hosam HF; Mechanical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This dissertation examines the problems of optimizing the selection of the datasets and experiments used for parameterizing machine learning-based electrochemical battery models. The key idea is that data selection, or “probing” can empower such models to achieve greater fidelity levels. The dissertation is motivated by the potential of battery models to enable theprediction and optimization of battery performance and control strategies. The literature presents multiple battery modeling approaches, including equivalent circuit, physics-based, and machine learning models. Machine learning is particularly attractive in the battery systems domain, thanks to its flexibility and ability to model battery performance and aging dynamics. Moreover, there is a growing interest in the literature in hybrid models that combine the benefits of machine learning with either the simplicity of equivalent circuit models or the predictiveness of physics-based models or both. The focus of this dissertation is on both hybrid and purely data-driven battery models. Moreover, the overarching question guiding the dissertation is: how does the selection of the datasets and experiments used for parameterizing these models affect their fidelity and parameter identifiability? Parameter identifiability is a fundamental concept from information theory that refers to the degree to which one can accurately estimate a given model’s parameters from input-output data. There is substantial existing research in the literature on battery parameter identifiability. An important lesson from this literature is that the design of a battery experiment can affect parameter identifiability significantly. Hence, test trajectory optimization has the potential to substantially improve model parameter identifiability. The literature explores this lesson for equivalent circuit and physics-based battery models. However, there is a noticeable gap in the literature regarding identifiability analysis and optimization for both machine learning-based and hybrid battery models. To address the above gap, the dissertation makes four novel contributions to the literature. The first contribution is an extensive survey of the machine learning-based battery modeling literature, highlighting the critical need for information-rich and clean datasets for parameterizing data-driven battery models. The second contribution is a K-means clustering-based algorithm for detecting outlier patterns in experimental battery cycling data. This algorithm is used for pre-cleaning the experimental cycling datasets for laboratory-fabricated lithium-sulfur (Li-S) batteries, thereby enabling the higher-fidelity fitting of a neural network model to these datasets. The third contribution is a novel algorithm for optimizing the cycling of a lithium iron phosphate (LFP) to maximize the parameter identifiability of a hybrid model of this battery. This algorithm succeeds in improving the resulting model’s Fisher identifiability significantly in simulation. The final contribution focuses on the application of such test trajectory optimization to the experimental cycling of commercial LFP cells. This work shows that test trajectory optimization is s effective not just at improving parameter identifiability, but also at probing and uncovering higher-order battery dynamics not incorporated in the initial baseline model. Collectively, all four of these contributions show the degree to which the selection of battery cycling datasets and experiments for richness and cleanness can enable higher-fidelity data-driven and hybrid modeling, for multiple battery chemistries.Item A Framework for Remaining Useful Life Prediction and Optimization for Complex Engineering Systems(2024) Weiner, Matthew Joesph; Azarm, Shapour; Groth, Katrina M; Reliability Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Remaining useful life (RUL) prediction plays a crucial role in maintaining the operational efficiency, reliability, and performance of complex engineering systems. Recent efforts have primarily focused on individual components or subsystems, neglecting the intricate relationships between components and their impact on system-level RUL (SRUL). The existing gap in predictive methodologies has prompted the need for an integrated approach to address the complex nature of these systems, while optimizing the performance with respect to these predictive indicators. This thesis introduces a novel methodology for predicting and optimizing SRUL, and demonstrates how the predicted SRUL can be used to optimize system operation. The approach incorporates various types of data, including condition monitoring sensor data and component reliability data. The methodology leverages probabilistic deep learning (PDL) techniques to predict component RUL distributions based on sensor data and component reliability data when sensor data is not available. Furthermore, an equation node-based Bayesian network (BN) is employed to capture the complex causal relationships between components and predict the SRUL. Finally, the system operation is optimized using a multi-objective genetic algorithm (MOGA), where SRUL is treated as a constraint and also as an objective function, and the other objective relates to mission completion time. The validation process includes a thorough examination of the component-level methodology using the C-MAPSS data set. The practical application of the proposed methodology in this thesis is through a case study involving an unmanned surface vessel (USV), which incorporates all aspects of the methodology, including system-level validation through qualitative metrics. Evaluation metrics are employed to quantify and qualify both component and system-level results, as well as the results from the optimizer, providing a comprehensive understanding of the proposed approach’s performance. There are several main contributions of this thesis. These include a new deep learning structure for component-level PHM, one that utilizes a hybrid-loss function for a multi-layer long short-term memory (LSTM) regression model to predict RUL with a given confidence interval while also considering the complex interactions among components. Another contribution is the development of a new framework for computing SRUL from these predicted component RULs, in which a Bayesian network is used to perform logic operations and determine the SRUL. These contributions advance the field of PHM, but also provide a practical application in engineering. The ability to accurately predict and manage the RUL of components within a system has profound implications for maintenance scheduling, cost reduction, and overall system reliability. The integration of the proposed method with an optimization algorithm closes the loop, offering a comprehensive solution for offline planning and SRUL prediction and optimization. The results of this research can be used to enhance the efficiency and reliability of engineering systems, leading to more informed decision-making.Item Aero Database Development and Two-Dimensional Hypersonic Trajectory Optization for the High-speed Army Reference Vehicle(2023) James, Brendan; Brehm, Christoph; Larsson, Johan; Aerospace Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Steady-flow inviscid and simulations of the High-Speed Army Reference Vehicle geometry were performed within the CHAMPS solver framework at Mach numbers of 4, 6, and 8, and an integrated streamline method was used to apply viscous corrections for Reynolds numbers up to 2x10^8. For each flow Mach, angle of attack sweeps from -10° to +10° were used to determine baseline drag, lift, and moment coefficient alpha dependencies. Coefficient values were then interpolated across Mach, alpha, and Reynolds number parameter spaces to construct an aerodynamic force coefficient database for use in two-dimensional flight simulation and trajectory optimization. By simulating flight with a maximum lift-to-drag control input, sample trajectories for determining maximum vehicle range were produced. A proportional-navigation (PN) controller was implemented which allowed for the targeting of specific altitudes throughout the progression of a trajectory. The PN controller and simulation schemes were then utilized in genetic-algorithm optimization to produce trajectory profiles for achieving minimum time-to-target for gliding flight in standard atmospheric conditions. Over the examined range of initial altitudes, Mach numbers, and release angles, the fastest trajectories were consistently shown to be those which achieved or maintained stratospheric altitudes and consequently benefited from significantly reduced drag before performing a nose-over maneuver for an accurate ground strike.Item Safe Navigation of Autonomous Vehicles in Structured Mixed-Traffic Environments(2023) Tariq, Faizan Muhammad; Baras, John S; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The primary driving force behind autonomous vehicle (AV) research is the prospect of enhancing road safety by preventing accidents caused by human errors. To that end, it seems rather improbable that AVs will replace all human-driven vehicles in the near future. The more plausible scenario is that AVs will gradually be introduced on public roads and highways in the presence of human-driven vehicles, leading to mixed-traffic scenarios. In addition to the existing challenges associated with autonomous driving stemming from various uncertainty factors associated with sensing, prediction, control, and computation, these situations pose further difficulties pertaining to the variability in human driving patterns. Therefore, to ensure widespread public acceptance of AVs, it is crucial to develop planning and decision-making algorithms, while benefiting from modern sensing, computation, and control methods, that can deliver safe, efficient, and reliable performance in mixed-traffic situations. Considering the need to cater to the behavior variability of human drivers, we address the joint decision-making and motion planning problem in structured environments with a multi-timescale navigation architecture. Specifically, we design algorithms for commonly encountered highway driving scenarios that require effective real-time decision-making, reliable motion prediction of on-road entities, behavior consideration of on-road agents, and attention to safety as well as passenger comfort. The specific problems addressed in this dissertation include bidirectional highway overtaking, highway maneuvering in traffic, and crash mitigation on highways. In the proposed framework, we first identify and exploit the different timescales involved in the navigation architecture. Then, we propose algorithmic modules while pursuing systematic complexity (data and computation) reduction at different timescales to gain immediate performance improvements in inference and action/response delay minimization. This leads to real-time situation assessment, computation, and action/control, allowing us to satisfy some of the key requirements for autonomous driving algorithms. Notably, the algorithms proposed in this dissertation ensure that the safety of the overall system is a fundamental constraint built into the system. Distinctive features of the proposed approaches include real-time operation capability, consideration for behavior variability of on-road agents, modularity in design, and optimality with respect to various metrics. The algorithms developed and implemented as part of this dissertation fundamentally rely upon the application of optimization techniques in a receding horizon fashion which allows for optimality in performance while explicitly accounting for actuation limits, vehicle dynamics, and safety. Even though the modularity of the proposed navigation framework allows for the incorporation of modern prediction and control methods, we develop various prediction modules for the trajectory prediction of on-road agents. We further benefit from risk evaluation methodologies to ensure robustness to behavior variability of human drivers on the road and handle collision-prone situations. In the spirit of emulating real-world situations, we place special emphasis on utilizing realistic driving simulations that capture the effects of communication delays between different modules, limitations in computation resources, and randomization of scenarios. For the developed algorithms, we evaluate the performance in comparative singular case studies as well as randomized Monte Carlo simulations with respect to several metrics to assess the efficacy of the developed methods.Item Ordering Non-Linear Subspaces for Airfoil Design and Optimization via Rotation Augmented Gradients(2023) Van Slooten, Alec; Fuge, Mark; Mechanical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Airfoil optimization is critical to the design of turbine blades and aerial vehicle wings, among other aerodynamic applications. This design process is often constrained by the computational time required to perform CFD simulations on different design options, or the availability of adjoint solvers. A common method to mitigate some of this computational expense in nongradient optimization is to perform dimensionality reduction on the data and optimize the design within this smaller subspace. Although learning these low-dimensional airfoil manifolds often facilitates aerodynamic optimization, these subspaces are often still computationally expensive to explore. Moreover, the complex data organization of many current nonlinear models make it difficult to reduce dimensionality without model retraining. Inducing orderings of latent components restructures the data, reduces dimensionality reduction information loss, and shows promise in providing near-optimal representations in various dimensions while only requiring the model to be trained once. Exploring the response of airfoil manifolds to data and model selection and inducing latent component orderings have potential to expedite airfoil design and optimization processes. This thesis first investigates airfoil manifolds by testing the performance of linear and nonlinear dimensionality reduction models, examining if optimized geometries occupy lower dimensional manifolds than non-optimized geometries, and by testing if the learned representation can be improved by using target optimization conditions as data set features. We find that autoencoders, although often suffering from stability issues, have increased performance over linear methods such as PCA in low dimensional representations of airfoil databases. We also find that the use of optimized geometry and the addition of performance parameters have little effect on the intrinsic dimensionality of the data. This thesis then explores a recently proposed approach for inducing latent space orderings called Rotation Augmented Gradient (RAG) [1]. We extend their algorithm to nonlinear models to evaluate its efficacy at creating easily-navigable latent spaces with reduced training, increased stability, and improved design space preconditioning. Our extension of the RAG algorithm to nonlinear models has potential to expedite dimensional analyses in cases with near-zero gradients and long training times by eliminating the need to retrain the model for different dimensional subspacesItem Single- and Multi-Objective Feasibility Robust Optimization under Interval Uncertainty with Surrogate Modeling(2022) Kania, Randall Joseph; Azarm, Shapour; Mechanical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This dissertation presents new methods for solving single- and multi-objective optimization problems when there are uncertain parameter values. The uncertainty in these problems is considered to come from sources with no known or assumed probability distribution, bounded only by an interval. The goal is to obtain a single solution (for single-objective optimization problems) or multiple solutions (for multi-objective optimization problems) that are optimal and “feasibly robust”. A feasibly robust solution is one that remains feasible for all values of uncertain parameters within the uncertainty interval. Obtaining such a solution can become computationally costly and require many function calls. To reduce the computational cost, the presented methods use surrogate modeling to approximate the functions in the optimization problem.This dissertation aims at addressing several key research questions. The first Research Question (RQ1) is: How can the computational cost for solving single-objective robust optimization problems be reduced with surrogate modelling when compared to previous work? RQ2 is: How can the computational cost of solving bi-objective robust optimization problems be improved by using surrogates in concert with a Bayesian optimization technique when compared to previous work? And RQ3 is: How can surrogate modeling be leveraged to make multi-objective robust optimization computationally less expensive when compared to previous work? In addressing RQ1, a new single-objective robust optimization method has been developed with improvements over an existing method from the literature. This method uses a deterministic, local solver, paired with a surrogate modelling technique for finding worst-case scenario of parameter configurations. Using this single-objective robust optimization method, improved large-scale performance and robust feasibility were demonstrated. The second method presented solves bi-objective robust optimization problems under interval uncertainty by introducing a relaxation technique to facilitate combining iterative robust optimization and Bayesian optimization techniques. This method showed improved feasibility robustness and performance at larger problem sizes over existing methods. The third method presented in this dissertation extends the current literature by considering multiple (beyond two) competing objectives for surrogate robust optimization. Increasing the number of objectives adds more dimensions and complexity to the search for solutions and can greatly increase the computational costs. In the third method, the robust optimization strategy from the bi-objective second method was combined with a new Monte Carlo approximated method. The key contributions in this dissertation are 1) a new single-objective robust optimization method combining a local optimization solver and surrogate modelling for robustness, 2) a bi-objective robust optimization method that employs iterative Bayesian optimization technique in tandem with iterative robust optimization techniques, and 3) a new acquisition function for robust optimization in problems of more than two objectives.Item COMPUTATIONAL ANALYSIS OF METABOLIC NETWORKS AND ISOTOPE TRACER EXPERIMENTS FOR METABOLIC FLUX EVALUATION IN COMPLEX SYSTEMS(2021) Lugar, Daniel James; Sriram, Ganesh; Chemical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Metabolic engineering endeavors seek to develop microorganisms as feedstocks for biofuels and commodity chemicals. Towards this, quantifying metabolic fluxes is an important step for characterizing an organism’s metabolism and designing effective engineering strategies. Metabolic fluxes are quantified using sophisticated techniques, namely flux balance analysis (FBA), an in silico technique, and isotope-assisted metabolic flux analysis (MFA), a hybrid experimental and computational technique. FBA uses a network’s stoichiometry with linear programming techniques to generate in silico flux predictions for genome-scale networks. MFA uses measurements from stable isotope (typically 13C) tracer experiments to estimate fluxes of central carbon metabolism. In MFA, fluxes are parameters to a model developed from the network’s carbon atom rearrangements, which is fit to isotope labeling data, typically acquired using mass spectrometry.We developed novel mathematical and computational techniques for quantifying and analyzing flux predictions obtained using MFA and FBA. FBA applications typically generate flux predictions for networks with on the order of 1000 [O(1000)] reactions and metabolites. We developed a network reduction algorithm that uses matrix algebra to reduce a large network and flux prediction to a smaller representation. From this reduced representation, a researcher may quickly gain holistic insights from the FBA model. In isotopically nonstationary MFA, time-series labeling measurements are acquired on the approach to steady state. A model consisting of a large system of typically O(1000) ordinary differential equations is fit to the measurements to estimate fluxes and pool sizes. For detailed networks, the number of parameters may be large. We developed a computationally effective framework for solving this problem having robust convergence and efficient scalability to large networks. In this approach, we formulate the problem as an equality-constrained nonlinear program (NLP), solved efficiently using a solver implemented on an algebraic modeling language. Finally, we apply this approach to a detailed model of Phaeodactylum tricornutum photoautotrophic and mixotrophic (on acetate) metabolism. Using the flux estimates, we characterized this organism’s metabolism under disparate growth conditions, which may inform future endeavors to engineer P. tricornutum as a chemical feedstock.Item Selected Problems in Many-Revolution Trajectory Optimization Using Q-Law(2021) Shannon, Jackson; Hartzell, Christine; Aerospace Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Q-Law is a Lyapunov guidance law for low-thrust trajectory design. Most prior implementations of Q-Law were limited to relatively simple low-thrust transfers. This work aims to improve the optimality, usability, and efficiency of Q-Law for better application to the mission design process. To accomplish this, Q-Law is combined with direct collocation to form an efficient hybrid method for high-fidelity, many-revolution trajectory design. Additionally, forward and backward Q-Law propagation are combined to form a novel method for Lunar transfer trajectories. This technique rapidly produces spiral trajectories to the Moon and provides mission designers with a means for efficient trade space exploration. Additionally, backward propagated Q-Law is combined with heritage trajectory design software to produce spiral escape trajectories as well as single and double Lunar swingby trajectories for interplanetary rideshare mission scenarios. Lastly, analytical partial derivatives of the Q-Law thrust vector calculation are derived, and the Q-Law algorithm is wrapped in a nonlinear programming problem. When these derivatives are used to generate the trajectory state transition matrix, the efficiency and accuracy of the optimization is superior to finite difference solutions. Using this approach, a novel Q-Law multiple shooting method is formulated and tested on various low-thrust transfer problems. These enhancements to the standard Q-Law algorithm enable efficient trade space exploration for more complex low-thrust trajectories, with a specific emphasis on the needs of SmallSat rideshare missions.Item AUTOMATIC OPTIMIZATION METHODS FOR PATIENT-SPECIFIC TISSUE-ENGINEERED VASCULAR GRAFTS(2020) Hess, Rachel; Fuge, Mark; Mechanical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Surgical intervention is sometimes necessary in cases of Coarctation of the Aorta (CoA). The post-repair geometry of the aorta can result in sub-optimal hemodynamics and can have long-term health impacts. Patient-specific designs for tissue-engineered vascular grafts (TEVGs) allow greater control over post-repair geometry. This thesis proposes a method for automatically optimizing patient-specific TEVGs using computational fluid dynamics (CFD) simulations and the ANSYS Fluent adjoint solver. Our method decreases power loss in the graft by 25-60% compared to the native geometry. As patient-specific graft design can be challenging due to incomplete or uncertain flow and geometry data, this thesis also quantifies the robustness of the optimal designs with respect to CFD boundary conditions derived from imaging data. We show that using velocity conditions that deviate by more than 20% of the measured peak systolic velocity, our method produces grafts with deviations on the order of 5% in predicted power loss performance. Lastly, as one way to accelerate the optimization process, we demonstrate and compare how some established machine learning models (K Nearest Neighbors and Kernel Ridge Regression) predict reasonable starting points for an optimizer on a 2D bifurcated pipe dataset.Item IMPROVING INVERSE ANALYSIS OF PYROLYSIS MODEL PARAMETERIZATION USING HILL CLIMBING ALGORITHMS(2019) Fiola, Gregory; Stoliarov, Stanislav I; Fire Protection Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Pyrolysis models are valuable tools for understanding material flammability and modeling fire growth. However, the development of comprehensive pyrolysis models is difficult and time-consuming due to the sheer number of material parameters required. Previous parameterization attempts employ massively parallel optimization problems using heuristic search algorithms to extract parameters from experimental data, but have been criticized for lacking physical significance and having reduced accuracy outside of calibrated ranges. This work sought to improve upon a previously developed manual methodology wherein the experimental results of both milligram- and bench-scale tests are inversely analyzed in a hierarchical approach. Three steps in the hierarchical process are automated using simple steepest ascent hill climbing optimization algorithms. The novelty of this approach lies in the custom fitness criteria and highly constrained and physical significant search space resulting from well-defined experiments. Two distinct materials were studied to evaluate the methodology: poly(methyl methacrylate) and rigid polyisocyanurate foam. The optimization programs were able to consistently fit both mass loss rate (MLR) from thermogravimetry (TGA) experiments and back surface temperature histories from Controlled Atmosphere Gasification Apparatus (CAPA II) experiments within experimental uncertainty. Models were validated against independent MLR histories from CAPA II experiments under low and high heat fluxes.