UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Extended reality-based training and adaptive machine learning-based optimization for thermoforming process Jalilvand, Iman

Abstract

Extended reality (XR) and machine learning (ML) are becoming pivotal enabling technologies in smart manufacturing, particularly for optimizing complex/multi-step processes. XR platforms provide immersive, interactive training environments for operators, while ML enables precise control and optimization of manufacturing processes. However, integrating these technologies into dynamic, real-time environments still presents challenges, particularly when faced with process variability and uncertainty. This dissertation addresses this technological trend in the thermoforming manufacturing process by developing a set of new XR systems, including virtual reality (VR) and mixed reality (MR) platforms integrated with real-time heat transfer simulations, for operator training. These immersive systems offer hands-on interaction with critical process variables such as temperature distribution, allowing operators to gain practical, real-world experience in a simulated environment. It is shown that the enhanced user engagement and precision achieved through the developed XR platforms significantly improve training performance metrics, as opposed to traditional training methods which often lack interactivity and real-time feedback. In the second part of this research, deep reinforcement learning (DRL) models are trained, using the same simulation models, and utilized to optimize the thermoforming process. Algorithms such as Proximal Policy Optimization (PPO) and Deep Q-Networks (DQN) are employed and compared to control temperature distribution across thermoplastic sheet, resulting in better energy efficiency and product quality. The study also explores a multi-agent reinforcement learning (MARL) to optimize multi-heater systems in the process, providing a decentralized control and improving the model training process efficiency. Finally, a novel transfer learning (TL) application is introduced to improve the adaptability of the above RL models, ensuring a robust performance despite practicable uncertainties like diverse material properties and varying environmental conditions that are faced in practice.

Item Citations and Data

Rights

Attribution-NonCommercial-NoDerivatives 4.0 International