The use of deep reinforcement learning for aiding in the solution of mechanical engineering problems

Abstract

As the digitization of the industry advances, the skillset required for mechanical engineers to tackle contemporary challenges expands correspondingly. This thesis presents a comprehensive overview of Reinforcement Learning (RL) and explores its potential applications in resolving mechanical engineering problems. The work initiates with a discussion on the importance of RL applications for mechanical engineers. Subsequently, a detailed summary of the fundamental aspects of RL is provided, acquainting readers with the field's nomenclature, primary algorithms, and core concepts. A methodology is then introduced for translating mechanical engineering problems into RL problems. As part of this study, we also developed an open-source software to establish a framework for creating and solving RL problems. Finally, three distinct mechanical problems were formulated and resolved using RL algorithms, with the results compared against traditional solutions. This endeavor illuminates the potential of RL as a viable tool for advancing mechanical engineering solutions.