Knowledge Mining using Generative AI for Causal Discovery in Electronics Production

This paper investigates an innovative approach for incorporating unstructured domain knowledge into the causal discovery process, focusing on the electronics manufacturing industry. The goal is to reduce the effort of setting up a causal graph, subsequently allowing the data-driven analysis of process influences to reduce defect rates and improve the product quality. For this purpose, a Large Language Model (LLM) is enabled to serve as a proxy for human process experts via retrieval of information from unstructured domain knowledge. The study analyzes the capability of LLMs to determine the causal structure of an industrial process, and the likelihood of individual Cause-and-Effect Relations (CERs), to obtain a causal graph. The analysis is conducted for two real-world use cases in electronics production. The investigation showcases the ability of LLMs to derive an understanding of process-specific CERs and their potential to allow causal discovery beyond covariance-based methods. The results indicate that generative AI can significantly alleviate human involvement in initiating causal analysis, a key obstacle to the widespread adoption of causal inference in the manufacturing industry.