Engineering Design Automation via Imitation Learning and Reinforcement Learning
| dc.contributor.advisor | Thekinen, Joseph | |
| dc.contributor.author | Bozorgmehry Boozarjomehry, Ghazal | |
| dc.contributor.committeemember | Bisheban, Mahdis | |
| dc.contributor.committeemember | Yanushkevich, Svetlana | |
| dc.date | 2025-06 | |
| dc.date.accessioned | 2025-02-05T23:28:14Z | |
| dc.date.available | 2025-02-05T23:28:14Z | |
| dc.date.issued | 2025-01-31 | |
| dc.description.abstract | Reinforcement Learning (RL) has achieved notable success in robotics and gaming, yet its application to automating engineering design faces significant challenges, including slow training times and poor generalization. Traditional RL methods require exploring millions of design states, which is computationally expensive, especially when dealing with complex physics models. In contrast, behavioral cloning, which enables RL agents to mimic human designers based on their decision data, presents a more resource-efficient alternative. This thesis investigates the use of both RL and imitation learning to automate engineering design, using aircraft design as a surrogate task to model engineering design. We evaluate the performance of a behavioral cloning agent trained on human design decision data, employing recurrent neural networks such as GRU, LSTM, and simple-RNN. We define a metric Q-score, which quantifies design quality on a scale between 0 and 1, with higher values indicating better design quality. Our findings demonstrate that the GRU architecture outperforms both LSTM and simple-RNN in terms of accuracy, achieving a Q-score of 0.8 after training on a relatively small dataset. The GRU model strikes an optimal balance between accuracy, simplicity, and computational efficiency, making it particularly suitable for surrogate design tasks like aircraft design optimization. Additionally, we assess the performance of RL agents, specifically Proximal Policy Optimization and Advantage Actor-Critic ,in the same design task. Both RL approaches achieved higher Q-scores (up to 0.99) but incurred significant computational costs and required extensive training time. In contrast, behavioral cloning provided a faster, more computationally efficient approach, though its performance was constrained by the availability of labeled human decision data. The results suggest that while RL methods excel in exploration and optimization, imitation learning offers a faster and more resource-efficient solution, albeit with reduced exploration and adaptability. | |
| dc.identifier.citation | Bozorgmehry Boozarjomehry, G. (2025). Engineering design automation via imitation learning and reinforcement learning (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. | |
| dc.identifier.uri | https://hdl.handle.net/1880/120726 | |
| dc.language.iso | en | |
| dc.publisher.faculty | Schulich School of Engineering | |
| dc.publisher.institution | University of Calgary | |
| dc.rights | University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission. | |
| dc.subject.classification | Engineering--Mechanical | |
| dc.subject.classification | Artificial Intelligence | |
| dc.title | Engineering Design Automation via Imitation Learning and Reinforcement Learning | |
| dc.type | master thesis | |
| thesis.degree.discipline | Engineering – Mechanical & Manufacturing | |
| thesis.degree.grantor | University of Calgary | |
| thesis.degree.name | Master of Science (MSc) | |
| ucalgary.thesis.accesssetbystudent | I do not require a thesis withhold – my thesis will have open access and can be viewed and downloaded publicly as soon as possible. |