Generative AI for Interpretable Satisficing Solution Design in Manufacturing

Abstract In multi-objective manufacturing designers face increasing difficulty in visualizing and interpreting trade-offs as the number of goals and decision variables grows. Traditional visualization methods, offer intuitive insights for three-goal problems but do not scale effectively to higher dimensions. In this paper we introduce a Generative Artificial Intelligence (GenAI) framework to support trade-off analysis in manufacturing decision-making which can further be generalized. The framework is designed to generate interpretable satisficing solutions, that satisfy all key requirements to an acceptable level rather than optimizing one metric at the expense of others. In the proposed framework we employ Large Language Models (LLMs) to produce narrative explanations and adaptive interpretations of high-dimensional trade-offs, to help designers to explore feasible solution regions and understand conflicts among goals. Using the Hot Rod Rolling (HRR) steel manufacturing process chain problem as a test problem, we show how GenAI can synthesize the complex relationships among six goals and overcome the interpretability limitations of static methods which is further scalable to higher numbers of goals. By using LLM's we propose a structured method to summarize trade-offs and suggest balanced weight allocations to restore feasibility when strict constraints lead to an empty solution set. Additionally, in this paper we incorporate an LLM-based knowledge graph implementation, enhanced from an open-source version to structure extracted insights and support scalable decision analysis. Cumulatively, in this paper we provide a pathway for scalable, explainable, and human-centered decision support in complex manufacturing systems.