Session: CIE-24-01 - AMS-CAPPD-SEIKEM: Artificial Intelligence and Machine Learning in Design and Manufacturing

Paper Number: 89073

89073 - Reinforcement Learning As an Alternative for Parameter Prediction In Design for Sheet Bulk Metal Forming 

Machine Learning (ML) and Multi-Objective Optimization (MOO) provide different points of view for the optimization of design parameters in manufacturing processes. Although, both have been proven capable for many applications, they have inherent limitations. While supervised learning requires sufficient training data, MOO, e. g. Genetic Algorithms (GA), lack the ability to explain how sufficient design parameters were achieved. Reinforcement Learning (RL) could help to overcome both issues. First, RL is independent from training data. Second, RL learns a policy leading to suitable design parameter combinations in the solution space as a sequence of parameter adaptions, each of which expresses the change of the design parameters according to the objectives. Therefore, relations between design parameters and objectives become accessible and might help to understand their effects on the manufacturing process. To probe RL, this contribution presents an approach and a case study to compare RL and GA for design parameter optimization in Sheet Bulk Metal Forming (SBMF). The feasibility of RL to optimize design parameters and necessary training effort are investigated. An approach to visualize the learned policy is provided to demonstrate the relation between design parameters and objectives. The results of a Reinforcement Learner and a GA are compared and discussed to answer the question under which circumstances can RL provide an alternative for parameter optimization.

Presenting Author: Christopher Sauer Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Engineering Design

Presenting Author Biography: Mr. Christopher Sauer has been a research associate at the Department of Mechanical Engineering, Engineering Design, since 2017. He is a machine learning expert in the manufacturing domain. His main research is the self-learning expert system (SLASSY), which predicts process characteristics and optimizes design parameters in Sheet Bulk Metal Forming.

Authors:

Fabian Dworschak Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Engineering Design
Christopher Sauer Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Engineering Design
Benjamin Schleich Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Engineering Design
Sandro Wartzack Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Engineering Design