Development of an intelligent design and simulation aid system for heat treatment processes based on LLM

Abstract

Heat treatment of steel is a multi-physics coupled process. Designing programs that meet the desired results is challenging. The current design of processes relies on experience and experimentation, leading to high costs in developing processes and challenges in training practitioners. To reduce research and development costs in the industry and enable novices to reach expert levels, we propose an intelligent heat treatment process design and simulation assistant system based on large language models (LLMs), named Chat-IMSHT. Chat-IMSHT can impart knowledge and recommend processes. Additionally, Chat-IMSHT optimizes the interaction between humans and Computer Aided Engineering (CAE) software. To achieve knowledge impartation and process recommendation, a dialogue model based on Retrieval Augmented Generation (RAG) and LLMs was designed. It characterizes and compresses a massive amount of heat treatment knowledge and process data. The system designs a new CAE software interaction paradigm, using LLMs to map parameters from natural language into formatted text for the CAE software COSMAP. A Steel Heat Treatment Knowledge Understanding (SHTKU) evaluation method was designed. The improved model significantly increased the accuracy of knowledge responses, with a maximum accuracy of 94.54 %. Experimental results show that Chat-IMSHT effectively imparts knowledge and generates formatted text, completing the task of process recommendation.

Keywords

• Metal material heat treatment
• Expert knowledge system
• Large language model
• Knowledge embedding
• Intelligent simulation system