Automation of Manufacturing Optimization : Can LLMs Help Non-Experts Solve Manufacturing Trade-offs using Simulation Based Optimization?

Sammanfattning

Manufacturing companies face many challenges in production due to dynamic environments and complex systems, leading to internal trade-offs between production objectives. Traditionally, these production disputes requires significant expertise and resource consuming simulation modelling and optimization to find viable solutions. To make simulation based optimization more accessible, the purpose of this thesis was to investigate how to leverage Large Language Models (LLMs) to generate and optimize Discrete Event Simulation (DES) models. This to let non-expert of simulation optimization find trade-off solutions to manufacturing problems. In the study a LLM-driven, human-in-the-loop, Python based framework was developed for simulation model generation and evolutionary multi objective optimization. The results demonstrated that the LLM-driven framework could find system bottlenecks and optimize various objectives by finding the optimal decision variable values. However, the study also identified some drawbacks of the framework regarding reliability, due to LLM hallucinations causing erroneous code generation and inaccurate reasonings. While the framework successfully lets non-expert build simulation models and find trade-off solutions, the findings emphasize human oversight for validation and error checking. This work exemplifies that LLMs can be well suited to help operators with finding production trade-off solutions. Furthermore, it contributes to the paradigm of Industry 5.0 by providing an example of human-LLM interaction for industrial optimization.