Theses (Mechanical Engineering)

Permanent URI for this collection

https://dspace.library.uvic.ca/handle/1828/89

Browse

Now showing 1 - 20 of 577

Developing an impedimetric glucose sensor using multi-layer molecularly imprinting technique
(2025) Abbaspour, Koorosh; Hoorfar, Mina
A sensitive and cost-effective measurement of glucose has always been a priority in clinical and quality control arena. In this article, molecularly imprinted and non-imprinted polymer (MIP & NIP) based on conducting polymers and functionalized composites was suggested to achieve repeatable and stable determination of glucose in POC analyses. Glucose was introduced into the imprinted layer composed of polypyrrole (PPy) along with (APBA), which was formed on the carboxylayed multiwalled carbon nanotubes (COO-MWCNT) as the step to immobilize glucose. Further polymerization of an imprinting layer for glucose, in a basic medium, was performed to cage the glucose molecule. A layer-by-layer analysis of the sensor was performed with electrochemical and surface analyzing methods. Also, both NIP and MIP were characterized before and after the glucose removal by phosphate buffer. The glucose rebinding to form glucoboronate ester in sensor was tested using impedance technique and a linear range between 1 µM and 20 µM with the detection limit of 6 µM was achieved. The reusability and stability of the imprinted sensor were determined to be 10 times and 96% of the beginning current was maintained after 15 days. In overall, such a sensor demonstrates promising future for developing cheap, reusable and non-invasive glucose sensors.
On switched control systems and model predictive control under uncertainty: Theory and applications
(2025) Shang, Xinxin; Yang, Shi
Hybrid systems are a widely applied class of dynamic systems, leveraging both continuous and discrete variables to characterize practical physical processes, including discrete variables like switches and logic, as well as continuous variables like position and velocity. As a powerful tool to model a variety of control systems, it has been widely applied in control system design and utilized in a large number of practical applications, such as aerospace, industrial electronics, and biomedical engineering. Since the seminal work published in Automatica in 1999 by Prof. Bemporad (Professor of Control Systems, IMT School for Advanced Studies Lucca, Italy), more and more control scientists and engineers have been increasingly devoted to studying the fundamental theories and applications of hybrid systems. Along with this historical research road, this dissertation focuses on: • Theories: Stability and stabilization (Chapter 3), robustness (Chapter 4), data-driven model predictive control (MPC) (Chapter 5) • Applications: Path planning and obstacle avoidance (Chapter 6), mobile communication networks, and hydrogen refueling station optimization (Chapter 7) In the following, a brief introduction will be given. In Chapter 1, a brief introduction to a class of hybrid systems is provided, including their stability and stabilization methods, as well as a typical case, i.e., switched systems. Moreover, a comprehensive review of MPC variants designed for handling uncertain hybrid systems is also presented. In Chapter 2, preliminary concepts and notations are introduced, providing the foundational understanding required for the subsequent chapters. In Chapter 3, an asynchronous stabilization of discrete-time switched linear systems under dwell-time constraints is presented. This research investigates the stability and control of systems that switch between different modes in an asynchronous manner, and a novel convex stability criterion is developed, facilitating efficient control design. Following that, in Chapter 4, from a more practical perspective for stabilizing switched systems, a new control strategy is provided to minimize the error between nominal and disturbed states by employing ellipsoidal techniques and demonstrates how system stability can be maintained despite disturbances. In Chapter 5, a lightweight data-driven approach is developed to construct a novel data-driven MPC framework to control an unknown linear system. The proposed theories ensure two of the most critical properties in MPC frameworks: system stability and recursive feasibility, even under significant uncertainties. Then, Chapter 6 is devoted to the scenario-based MPC for path planning and obstacle avoidance with chance constraints. This work provides solutions for dealing with uncertainties in real-time decision-making under safety-critical conditions. In Chapter 7, a representative application is presented, demonstrating how the fundamental theories developed in previous chapters can address practical requirements. The application involves modeling the hydrogen refueling processes as hybrid systems and leveraging MPC to optimize energy costs while satisfying safety constraints (e.g., temperature and pressure). Finally, the conclusion and future works of the dissertation are presented in Chapter 8.
Computational fluid dynamics investigation on radon concentration in residential buildings
(2025) Wilson, Titu; Valeo, Caterina; Mukhopadhyaya, Phalguni
Radon, a naturally occurring radioactive gas formed from uranium decay, is a significant health hazard and the leading cause of lung cancer among the general population. Radon gas can infiltrate building structures through cracks and openings in the building envelope. The concentration inside the building is influenced by factors such as indoor temperature, humidity, and ventilation rates. This study focuses on a case study house in Victoria, BC, Canada, where radon measurements were taken over a six-week period during the heating season to capture real-world indoor radon behaviour. A combination of real-world data collection, analytical calculations, and computational fluid dynamics (CFD) simulations was used to analyze radon concentration patterns under varying temperature, humidity, and ventilation conditions. The CFD model was developed using ANSYS Fluent, with the indoor environment simulated under different temperatures (18°C, 21°C, and 24°C) and humidity (20%, 40%, and 60%) conditions at an air change rate of 0.5 ACH (air change per hour) from the field experiment. Numerical and experimental (field) observations revealed that indoor temperature significantly influences radon concentration, with higher temperatures enhancing the stack effect, leading to increased radon levels. Numerical simulations showed that humidity also played a critical role, where higher humidity levels acted as a barrier to radon infiltration, reducing its accumulation. The study validated the CFD model by comparing it with measured field data and analytical results, demonstrating less than a 2% difference, confirming its reliability. This research contributes valuable insights into indoor radon concentration and behaviour, emphasizing the importance of maintaining optimal indoor environmental conditions to manage radon exposure. The findings highlight the need for integrated radon mitigation strategies, considering temperature, humidity, and ventilation to ensure safer indoor air quality in residential buildings in Victoria, BC, Canada. These insights can inform building design, public health policies, and radon management practices, helping to reduce radon-related health risks.
Nonlinear heat and mass transfer resistivities for liquid-vapor interfaces
(2025) Feyzi Oskouei, Pouria; Struchtrup, Henning
Nonlinear heat and mass transfer at liquid-vapor interfaces is studied, focusing on how interface resistivities vary with the intensity of non-equilibrium. Two distinct experimental approaches are considered: conventional experiments with relatively small mass and heat fluxes [G. Fang and C. A. Ward, Phys. Rev. E 59, 419 (1999)], and Molecular Dynamics (MD) experiments with relatively large fluxes [Homes, Simon and Vrabec, Jadran, Physics of Fluids 36, 2 (2024)]. This contrast leads us to the question: whether the strength of non-equilibrium impacts interface resistivities [Henning Struchtrup and Hans Christian Öttinger. Phys. Rev. E, 108(6):064801, 2023]. Based on a kinetic interface model, nonlinear resistivities are assessed in relation to interface temperature and the fluxes of mass and heat. The results show that for smaller fluxes, resistivities depend solely on local temperature, as is typically assumed in Linear Irreversible Thermodynamics. However, for larger fluxes, resistivities are influenced by the fluxes themselves as well.
Data-driven real-time model identification of UAS for adaptive control
(2025) Bazzocchi, Sean; Suleman, Afzal
This dissertation presents a comprehensive investigation into the development, modeling, and control of novel unmanned aerial vehicles (UAVs) within the Eusphyra project. Structured as a thesis-by-publication, the work delivers significant advancements in UAV design, flight dynamics modeling, autopilot tuning, and adaptive control, offering innovative methodologies to enhance performance and autonomy. The research begins with the design and airworthiness assessment of the Eusphyra UAV, detailing an iterative development process that culminates in the validation of an innovative tri-rotor VTOL configuration. A high-fidelity flight dynamics model is then developed using limited onboard sensor data and state-of-the-art system identification techniques to capture the complex aero-propulsive coupling inherent in the system. This model is rigorously validated against out-of-sample flight data, confirming its reliability and predictive capability. Building on these foundational insights, an automated offline autopilot tuning framework is introduced that leverages a simplified system identification process in conjunction with genetic algorithms. This approach minimizes human oversight and enables rapid retuning in response to design modifications. Further extending the scope of the work, the dissertation explores real-time system identification by integrating unsupervised learning techniques to dynamically update UAV models during flight. This capability is advanced into the development of a Model Identification Adaptive Controller (MIAC), which combines Sparse Identification of Nonlinear Dynamics (SINDy) with Model Predictive Control (MPC) for adaptive, online control under varying flight conditions. Comprehensive hardware-in-the-loop simulations and flight tests confirm the feasibility and performance of MIAC, marking a significant step forward in UAV autonomy and adaptability, and laying the groundwork for future research in advanced adaptive control for complex aerial systems.
Intelligent machine learning-based leakage detection and localization in vacuum-assisted composite manufacturing
(2025) Esmaeili Shahmiri, Yussuf Reza; Najjaran, Homayoun
Vacuum-assisted composite manufacturing methods, such as vacuum bag prepreg layup and vacuum-assisted resin transfer molding (VARTM), utilize atmospheric pressure as a uniform external force to consolidate and saturate fabric components. However, vacuum bag leakages can result in defects such as air bubbles, resin traps, voids, non-uniform surface finishes, and ultimately, inferior mechanical properties. Detecting and repairing these leakages before the autoclave curing stage is therefore essential. The leakage localization method used in this study relies on volumetric flow rate measurements of air evacuation lines. Multiple air evacuation channels, known as vacuum ports, are strategically placed at different locations in the production layup. Each port is equipped with sensors capable of independently measuring the volumetric flow rates of air during the process. In the presence of a leakage, the measured flow rate values will not stabilize at zero because air continuously enters the vacuum bag through the leak. The flow rate values correlate with the location of the leak, the overall layup configuration, and the positions of the vacuum ports. We introduced an intelligent machine learning-based framework for leakage detection and localization, designed to learn the complex relationships between flow rate values and leak locations. To generate sufficient training data, an electric circuit analogy was developed to simulate the vacuum process. This approach provides a fast and reliable alternative to complex analytical simulations and extensive physical experiments. The proposed method has been validated and compared across various experimental configurations, demonstrating its effectiveness. Using the available and synthesized data, we employed various machine learning models, including regression models, a Grid neural network, a physics-informed Grid neural network, leakage classification models, and physical parameter training algorithms for leakage prediction. Our methods not only predict leakage locations with acceptable accuracy but also generalize well across different configurations. Additionally, we addressed challenges associated with complex, non-uniform layups featuring regions of varying permeability. For the first time, our framework also tackled scenarios involving multiple simultaneous leakages, successfully localizing all leaks on the layup. Our results demonstrate significant advancements over state-of-the-art methods. These improvements go beyond higher prediction accuracy, focusing on enhanced generalizability across various layups, reduced data requirements for training, and the ability to tackle complex scenarios, such as non-uniform permeability and multiple leakages, which were previously unaddressed. Notably, the novel PI-GNN framework outperforms regression models in both generalizability and data efficiency. By integrating physical knowledge with data science, the PI-GNN framework establishes a robust foundation for addressing layups of varying sizes and geometries. Furthermore, our proposed physical parameter training algorithm effectively learns the permeability of different regions within the layup, enabling the development of a more accurate and robust simulation tool for model training. Optimizing the placement of vacuum ports to improve leakage location prediction is another challenge addressed in our work. Each layup offers numerous possible configurations for positioning vacuum ports to enhance leakage localization. We tackled this optimization problem by maximizing flow rate variance among the vacuum ports. Given the problem’s large state space, a hierarchical optimization approach was employed to identify the optimal configuration. Experimental validation confirmed that optimizing the port configuration significantly reduces leakage prediction errors.
Optimal energy management of electrified propulsion systems with data-driven li-ion battery and PEM fuel cell performance and degradation predictions
(2025) Pang, Bo; Zuomin, Dong
Electrified propulsion systems for vehicles and marine vessels, including engine-battery hybrid, fuel cell-battery hybrid, and battery electric propulsion systems, present clean propulsion solutions for improving performance, energy efficiency, emissions and lifecycle costs (LCC). Battery energy storage systems (BESSs) are essential in these systems, storing and delivering electrical energy to complement or replace the onboard energy converter. With the rapid development of Lithium-ion (Li-ion) battery technology, battery electric vehicles (BEVs) are becoming increasingly popular for personal transportation. Fuel cell electric vehicles (FCEVs), powered by a proton exchange membrane fuel cell (PEMFC) system and a BESS supplement, offer a zero-emission propulsion solution for heavy-duty applications, overcoming some of the limitations of BEVs. Liquefied natural gas (LNG), a low-cost cleaner fuel, is a viable replacement for diesel in compression ignition (CI) engines for heavy-duty engine-battery hybrid electric propulsions to reduce fuel consumption, air pollutants and GHG emissions. However, BESSs and PEMFC systems suffer relatively short service lives and high replacement costs. A better understanding and modelling of their degradation patterns and corresponding optimal system design and energy management are vital to extending their service lives to reduce the vehicles' LCCs. Ultracapacitors (UCs), known for their high-power density and insensitivity to operating temperatures, if properly designed and controlled, can be combined with the BESS to form hybrid energy storage systems (HESS) to extend battery life, system performance and energy efficiency. The performance of BESSs and PEMFCs depends on their degradation levels. Usage patterns and temperature conditions influence their degradation rate. This study collected performance and degradation data for Li-ion batteries and PEMFCs under various usage conditions to develop advanced battery and PEMFC performance, degradation, and thermal models. These models aid in optimizing the hybrid electric propulsion and HESS design and energy management across different operating scenarios. The research also introduced new methods for dynamically updating the BESS and PEMFC degradation models using real-time operational data to improve the optimal energy management of electrified vehicles and marine vessels. This work developed new methods for generating integrated optimal system design and energy management strategy using nested global optimizations to satisfy system design requirements and achieve maximum energy efficiency and minimum life cycle costs. Dynamic programming (DP) was used to search for the optimal energy management solutions for each system design, and a simulation-based, top-level global optimization was formulated to identify the best system design solution and solved using a very efficient metamodel global optimization algorithm. Methods for generating real-time optimal energy management and control for the hybrid electric propulsion system and HESS, segment by segment, based on an extended model prediction control (MPC), were introduced. New methods for using real-time operation data to dynamically update the Li-ion battery degradation model using identified equivalent total charge/discharge cycle number and the PEMFC degradation model using identified actual active area were introduced. Jointly considering the measured vehicle speed and benchmark test cycle, as well as the BESS and PEMFC operating data and updated degradation models using the Extended Kalman filter (EKF), these approaches provided improved real-time optimal control for the hybrid propulsion system and HESS. Case study 1 involves a fuel cell electric ferry ship powered by a PEMFC system and BESS. The approach minimizes LCC by balancing system performance, fuel economy, and degradations of the PEMFC and BESS. An optimal EMS is developed for the ship based on real-time operational data by introducing accurate performance and degradation models. This approach significantly reduces LCC, promoting clean ship propulsion technologies. Case study 2 focuses on a global optimal propulsion system design for an LNG-fueled hybrid electric ferry ship. The system addresses the increased CO2 equivalent emissions due to methane leakage from LNG engines and the high costs associated with BESS replacements. The optimal integration of the LNG engine, BESS, and EMS is achieved using DP, while an EKF models real-time changes in ship propulsion power. MPC is used to develop an optimal control strategy that optimizes fuel consumption, BESS degradation, and emissions. This case highlights the advantages of global optimization and real-time control. Case study 3 introduces a new approach to optimizing the design and EMS of a battery-UC HESS. The approach improves Li-ion battery operation under high current charge/discharge, mitigates low-temperature impact, and significantly extends battery life. The combination of battery and UC improves overall performance, while adding an active UC-based battery thermal management strategy (TMS) in the optimal EMS reduces the LCC of the BEVs. Updating the battery performance and degradation models in real time continuously enabled more precise optimal control through MPC. These case studies demonstrate the feasibility, advantages and benefits of the newly introduced integrated modelling, design and control optimization methods.
The health assessment of lithium-ion batteries using machine learning
(2024) Murphy, Lucas; Crawford, Curran
Lithium-ion batteries are emerging as a crucial technology in the world’s clean energy transition. These batteries face challenges as they degrade with use due to unwanted chemical side reactions. In this thesis, we propose two methods of using relatively accessible battery data to predict important health metrics. These health metrics improve battery safety, control, and decision-making. In the first method, we leverage battery charging times to decipher measures of internal chemical degradation. Using machine learning, different modes of degradation can be attributed to segments of the constant current and constant voltage charging curves. This model is trained and tested using cells cycled under varying depths of discharge and C-rate conditions inducing an array of degradation pathways. We can gather insights into the model’s learning through input feature analysis to determine key areas within the charging regime. At the end of the battery’s first life, we can analyze its degradation modes to determine its viability in second-life applications. This is conducted by using features extracted from electrochemical impedance spectroscopy as input data to a binary classifier. This determines whether a battery should be reused or recycled. The distinction is made based on a metric that includes the current state of health of the battery, and the slope of capacity degradation to the end of second life. These contributions look to quantify variance and non-linearity in Lithium-ion battery degradation to inform economic and safety-based decision-making. These contributions also address challenges in data-driven battery modelling regarding model explainability and data scarcity.
Impacts of curtailment costs on optimal generation and storage capacity
(2024) Herrera Ibarra, Luis Antonio; Rowe, Andrew Michael; Wild, Peter Martin
This study examines the effects of curtailment costs on cost-minimized energy capacity for Metro Vancouver, focusing on electrification and Renewable Gas (RG) pathways. Using Calliope, we assess the impact of curtailment costs on storage capacity, renewable generation, and system costs. Results show that curtailment costs significantly affect the electrification pathway, driving increased battery storage activity and selective deployment of renewable generation to limit curtailment. In contrast, the RG pathway adjusts only gas storage capacity in response to curtailment costs, relying solely on wind technology as its Variable Renewable Energy source without the need of an electric storage. These findings highlight the importance of tailored curtailment cost strategies for efficient renewable integration, enhancing resilience and cost-effectiveness across energy transition pathways.
Modelling, experimental validation and evaluation of a parallel hybrid-electric propulsion system for small-scale UAVs
(2024) Wilson, Jamal; Suleman, Afzal
The aviation industry is currently focused on research and development of propulsion systems that produce less emissions, are more efficient, and can provide better range/endurance. Hybrid-electric systems have shown a promising potential in reducing emissions. Current battery technologies do not have the energy density required to meet most application requirements. While combustion technology has improved in efficiency over the decades; hybrid technology is required to take the next big step. In addition to the environmental benefits parallel hybrid technology provides other benefits such as operating mode variety, redundancy, and higher endurance. In this thesis research, the primary goal is to model, evaluate and validate a mathematical model developed for a parallel hybrid-electric propulsion system. This model will be used in Model Based Design (MBD) to predict system performance, improve component selection, and optimize operation. For this thesis the modelling and design was completed for small-scale unmanned aerial vehicles (UAVs). A test bench was redesigned to handle the power produced by the combustion engine and electric motor (EM). For the experimental configuration a 50cc Corvid-50 combustion engine was used which produces 2.8kW at 7000RPM. This engine was combined with a SKP 6485 electric motor capable of 4.12kW continuously with a maximum speed of 8364RPM. Both power units are coupled together with a Mayr 500.301.0 type 4 electromagnetic clutch rated for 40Nm at 7000RPM. The propulsion system is connected to an electric dynamometer with programmable load capable of simulating any power profile. Telemetry for the system is collected through National Instruments hardware, electronic speed controller, and engine control unit. During operation the test bench is able to operate in five different modes. Combustion and electric-only operation are capable by disabling the clutch for electric-only and running the electric motor at zero current for combustion. When running the system in a hybrid configuration there are three command modes: dual speed, throttle and speed, speed and current. Each of these command modes dictate which power unit governs speed while the other has direct torque control. Each of these hybrid modes provides the opportunity for regeneration and boost modes when requested. The test bench generates the experimental data required to model the propulsion system with accuracy. Using datasheets, unloaded runs, governing equations, and controller values component level models were created to complete initial simulations. The model and test iterative process repeated until the system level model responded well. To further develop the simulation model a virtual flight mission was created. The test bench and Simulink model ran the virtual flight mission in combustion, electric, and hybrid modes. These different runs provided the data required to assess the model’s accuracy and demonstrate the difference between each propulsion technology. For these tests the simulation was able to predict speed and torque within a range of 1-12\% for steady-state operation between flight segments. The starting torque of the electric motor to initiate combustion was modelled to represent cold starts where the torque range was between 2-3Nm. Over the length of the fifteen-minute flight mission run the simulation predicted battery charge and fuel consumption within 5\%. Energy density of each propulsion type was analyzed for the components used on the test bench. This showed that combustion power has the highest available energy density at 1.17MJ/kg; electric power is substantially lower at 0.35MJ/kg. The initial energy density of the hybrid system is 0.71MJ/kg but can be further optimized. By optimizing the energy masses for the hybrid-electric system an energy density equal to the combustion engine was accomplished with a 0.77kg mass reduction. This optimization process can be taken further by improving the command sequence of the system to incorporate regeneration, clutch disengagement, and throttle curve modification to reduce fuel flow. The results of this research project created a simulation model and test bench capable of high-power flight tests. Both the model and test bench will continue to develop; further increasing the ability to design, optimize, and test parallel hybrid systems. This will provide the experience and knowledge to design, build, and integrate a power unit ready for flight testing.
Integration of model predictive control and reinforcement learning for dynamic systems with application to robot manipulators
(2024) Hu, Pengcheng; Shi, Yang
The last decade has witnessed great progress in the development of reinforcement learning (RL) across many applications, such as games and autonomous driving. RL is effective in solving control problems for complex systems whose dynamics are intractable to be accurately modeled. In an RL algorithm, the agent learns the optimal policy in terms of the maximum reward based on measurement samples from the interactions with the environment. To obtain the optimal policy, RL requires collecting sufficiently large number of samples, which is challenging in real-world applications, e.g., robotics, manufacturing, and so on. To tackle this problem, model predictive control-based RL (MPC-based RL) is proposed to improve the sample efficiency. In the MPC-based RL algorithm, a model is learned from collected samples, the learned model and MPC are utilized to predict trajectories over a specified prediction horizon, and an action is obtained through the RL algorithm by maximizing the cumulative reward. This thesis is devoted to the investigation of the MPC-based RL design and its application to robot manipulators. In Chapter 2, an MPC-based deep RL framework for constrained linear systems with bounded disturbances is proposed. In the proposed framework, a rigid tube-based MPC (RTMPC) method is employed to predict a trajectory by solving the corresponding optimization problem. Then, the predicted trajectory is stored in a replay buffer as the form of data pairs. Further, the soft actor-critic (SAC) algorithm is applied to modify the loss function and update the policy online, based on the predicted data pairs. Numerical simulations validate the effectiveness of the proposed method. In addition, comparison results demonstrate the advantages of the proposed method including requirement of fewer real samples and providing better control performance with comparable computational complexity to RTMPC. In Chapter 3, we investigate the application of three methods for manipulators. Firstly, we apply an MPC-based RL algorithm, a nonlinear MPC (NMPC) method, and two model-free RL algorithms to tackle the regulation problem for a 2-degree-of-freedom manipulator system, and compare their training control performance. Secondly, the training and control performance evaluation for the model-free RL algorithm and the MPC-based RL algorithm are provided. The MPC-based RL algorithm shows better training performance in terms of sample efficiency and total return but poorer control performance. Thirdly, simulation studies are provided to compare the training performance of the MPC-based RL algorithm and two model-free RL algorithms. From the simulation results, the MPC-based RL algorithm presents poorer training performance compared with model-free RL algorithms for the twelve-dimensional system. In Chapter 4, conclusions and future work are summarized.
Meta-optimization in safe reinforcement learning: Enhancing safety at training and deployment with fewer hyperparameters
(2024) Honari, Homayoun; Najjaran, Homayoun
Reinforcement learning (RL) is a trial-and-error framework for enabling intelligent systems to learn the optimal behaviour based on the feedback from the environment. In recent years, successful application of RL in controlling various embodied systems have been observed. However, the real-world deployment and training of RL methods require paying attention to certain limitations imposed by the robot and its surroundings. To address these limitations, safe RL algorithms aim to define safety constraints based on the physics of the system and modify the training regime of the RL methods to satisfy them during training and inference. While safe RL offers a promising direction for achieving real-world deployability, challenges such as sample efficiency and hyperparameter tuning hinders its applicability in real-world scenarios. To address these challenges, this thesis proposes various approaches. First, a metagradient-based training pipeline called Meta Soft Actor-Critic Lagrangian (Meta SAC-Lag) is proposed which aims to optimize the aforementioned safety-related hyperparameters under the conventional Lagrangian framework. To study the performance, the proposed method is evaluated in various safety-critical simulated environments. In addition, a real-world task is designed, and the algorithm is successfully deployed on a Kinova Gen3 robotic arm to showcase its real-world deployability with minimal hyperparameter tuning requirements. Furthermore, a multi-objective policy optimization framework is proposed which specifies the trade-off between optimality and safety directly and optimizes both of them simultaneously. The competitive performance of the proposed algorithm compared to the state-of-the-art safe RL methods with fewer hyperparameters showcases its potential in providing a powerful alternative framework for safe RL.
World model based multi-agent proximal policy optimization framework for multi-agent pathfinding
(2024) Chung, Jaehoon; Najjaran, Homayoun
Multi-agent pathfinding plays a crucial role in various robot applications. Recently, deep reinforcement learning methods have been adopted to solve large-scale planning problems in a decentralized manner. Nonetheless, such approaches pose challenges such as non-stationarity and partial observability. This thesis addresses these challenges by introducing a centralized communication block into a multi-agent proximal policy optimization framework. The evaluation is conducted in a simulation based environment, featuring continuous state and action spaces. The simulator consists of a vectorized 2D physics engine where agents are bound by the laws of physics. Within the framework, a World model is utilized to extract and abstract representation features from the global map, leveraging the global context to enhance the training process. This approach involves decoupling the feature extractor from the agent training process, enabling a more accurate representation of the global state that remains unbiased by the actions of the agents. Furthermore, the modularized approach offers the flexibility to replace the representation model with another model or modify tasks within the global map without the retraining of the agents. The empirical study demonstrates the effectiveness of the proposed approach by comparing three proximal policy optimization-based multi-agent pathfinding frameworks. The results indicate that utilizing an autoencoder-based state representation model as the centralized communication model sufficiently provides the global context. Additionally, introducing centralized communication block improves performance and the generalization capability of agent policies.
Dynamic wave-soil-structure interaction analysis : with applications to tall buildings
(2003) Yao, Ming Ming
This thesis presents a study on Soil-Structure Interaction (SSI) by analyzing the variance of fundamental frequencies and vibration modes of tall buildings through the application of parametric studies. There are two methods available for modeling the SSI: the direct method and substructure method. In this thesis, the substructure method is employed to model the subsystems of unbounded soil and structure separately. The unbounded soil is modeled by using the Scaled Boundary Finite Element Method (SBFEM), an infinitesimal finite-element cell method, [ 61] which naturally satisfies the radiation condition for the wave propagation problem. The structure is modeled using the standard finite element method. The SBFEM results in less degree of freedoms of the soil than the direct method by only modeling the interface of the interested area. In this work, a Dynamic Soil Structure Interaction Analysis program (DSSIA) [65] is used and modified to investigate the SSI effect on tall buildings in the frequency and time domains. A parametric study is carried out where the buildings are subjected to external impulse loadings and also earth quake loading. The fundamental frequency and the vibration and peak displacement along the height of the buildings are obtained in the time domain analysis. The coupling be tween the building's height, hysteretic damping ratio, soil dynamics and SSI effect are also investigated. The results are compared with building codes, field measurements and other numerical methods.
Some linear viscoelastic wave propagation problems
(1993) Zhang, Miqin
The calculation of flow and heat transfer over surface mounted ribs using a domain decomposition method
(1993) Zapach, Trevor George
Dynamic traffic signal control using a self-learning, fuzzy-neural intelligent system
(1995) Wu, Jian
Optimal system performance in a traffic system is achieved by planning, control, and scheduling of the system's many traffic movements. This task of optimizing the time and facility conflicting activities is challenging. In the last a few decades, extensive research has been carried out on the planning and control of manufacturing processes to improve productivity and reduce manufacturing costs. In this work, we apply the quantitative intelligent system concept, developed in the optimal planning of manufacturing activities, to the dynamic signal control problem of a corridor traffic system to minimize traffic delays. Today, the ever-increasing demand for individual mobility and reduced traffic delays, coupled with economic and environmental restrictions on increases in physical road capacity, requires efficient control of signalized intersection traffic. Present traffic signal control is based upon static traffic control strategy using several fixed timing plans. These timing plans contain optimal control parameters for representative traffic patterns manually identified by a traffic engineer. Several timing plans are applied during different time periods according to the traffic demand statistics , regardless the actual traffic flow condition at the instant. To improve the quality of signalized traffic control, and to reduce traffic delay and congestions caused by the inaccurate estimation of traffic demands, the development of a dynamic control strategy based upon the on-line acquired traffic flow condition becomes necessary. In this work, an approach for dynamic traffic signal control, based upon a fuzzy-neural intelligent system and traffic delay minimization, is introduced. This approach consists of three major parts: (a) automated traffic flow pattern identification using fuzzy pattern clustering, (b) optimization of traffic control parameters (timing plan design) for identified traffic flow patterns, and (c) dynamic traffic signal control by real-time traffic flow monitoring, traffic flow pattern matching using the fuzzy-neural system, and execution of stored optimal control parameters. The traffic flow pattern identification is carried out using the fuzzy pattern clustering and matching techniques. A mathematical model is first introduced to quantify the fuzzy traffic conditions. The traffic condition at a moment is expressed as a hyper point in a m-dimensional traffic parameter space. Similar traffic conditions show as clouds of hyperpoints. The quantified traffic condition description allows the "characteristic groups" of traffic flow conditions being recognized as traffic patterns, using the fuzzy clustering methods. These traffic patterns are closely studied. The optimal timing plans of these traffic patterns, which contain the optimal signal control parameters, are generated using commercial software through extensive optimization. Dynamic traffic control is accomplished using fuzzy traffic pattern clustering/matching methods and traffic plan optimization. The task is carried out in two steps: off-line learning and on-line control. The off-line learning part identifies all representative traffic patterns based upon previously collected traffic data, designs a timing plan for each identified traffic pattern using traffic delay minimization, and trains a fuzzy-neural system using the traffic pattern - optimal timing plan pairs generated. The on-line control part senses traffic flow in real-time, matches the sensed traffic flow condition with the best fitted traffic pattern, assigns the optimal timing plan of the matched traffic pattern to related traffic controllers dynamically. A method for short-term traffic flow condition prediction is also developed to offset the short delay in traffic flow condition sensing, and quasi-optimal traffic signal parameter updating at the controller. The approach makes dynamic traffic signal control of a corridor traffic system with quasi-optimal performance possible. The system is self-adaptive and capable of carrying out self-learning to varying traffic demands. Computer simulation and prototype testing using the real traffic data have demonstrated significant traffic delay reduction. The research directly contributes to static and dynamic traffic control research and practice. It also extends the research and applications of the quantitative intelligent system approach, and benefits the research on intelligent scheduling and planning for time and facility conflict activities. The research on developing a hybrid fuzzy-neural system combines the reasoning ability of a fuzzy system and the learning ability of a neural network, which is critical for a self learning and self-adaptive, intelligent system.
Developments on the entropy of thermal radiation
(1998) Wright, Seanr
The objectives of this work are to improve the understanding and also to simplify the calculation of the entropy transferred by thermal radiation (TR). Thermal radiation plays an important role in the thermodynamic analysis of many systems. For example consider the Earth. Thermal radiation exchange is frequently the dominant form of energy and entropy transfer within the Earth system and is the only significant form of exchange between the Earth and the universe ([15], p. 38,39). Moreover, radiation-matter interaction is responsible for most of the entropy produced by the Earth system. With respect to the energy of TR there has been extensive research and application in engineering analysis. The entropy of TR with an arbitrary spectrum can be calculated by numerical integration using Planck's [1] fundamental spectral entropy radiance (LJ expression. This requires the knowledge of the spectral energy radiance (KJ spectrum with position and direction. Exclusive use of numerical integration is straightforward but laborious and from an analytical perspective the entropy of non-blackbody radiation (NBR) is not well understood. Heat transfer textbooks usually exclude entropy altogether. Furthermore, most thermodynamic texts are misleading because they state that the entropy flux of 'heat' transfer is the ratio of the energy flux to the local temperature ( q/T) with no restriction for TR (e.g. Moran and Shapiro [2], p. 220 and 230; Reynolds and Perkins [3], p. 223; and McGovern [4], p. 177).
Meeting national targets : carbon dioxide and the future development of the Canadian energy system
(1993) Wells, John David
Wrap-around B-spline surface fitting
(1995) Weir, Donald James
Reverse engineering involves digitizing a 3-D model or part, by means of a tactile or non-contact optical sensor, converting the data to a computer aided design (CAD) database description, and manufacturing the part by computer numeric controlled (CNC) machines. A common reverse engineering approach has been to scan the physical model from a single viewpoint to obtain points on a single valued surface and then fit relatively flat B-spline surface patches. This thesis presents a new approach which allows wrap-around B-spline surfaces to be fitted. The approach is demonstrated in the reverse engineering of physical models by employing a 3-D laser scanning system in conjunction with the surface fitting software developed by the author. Accurate surface data is collected by the laser scanner and then input to the surface fitting software. Surface entities such as B-spline and quadric functions are employed to build the CAD model. The CAD model is compatible with commonly used design and manufacturing software packages. The reverse engineering of a bicycle seat and a telephone receiver are used to illustrate the efficiency of the process.

Browse

Recent Submissions