Advancing Vehicle Diagnostic: Exploring the Application of Large Language Models in the Automotive Industry

Abstract

The integration of Artificial Intelligence (AI) in the automotive industry has enhanced vehicle functionality, making them smarter, safer, and more efficient. However, the potential of AI in vehicle fault diagnostics and troubleshooting remains untapped. This thesis explores the application of Large Language Models (LLMs) to improve vehicle diagnostics, a field that traditionally relies on time consuming manual data interpretation, which can be inefficient and error-prone. The study aims to showcase and assess the capabilities of Claude v2 by Anthropic in rapidly processing and analyzing a large vehicle failure dataset. Our research uses a comparative analysis methodology, evaluating the performance of a LLM with traditional machine learning models in diagnosing and classifying vehicle faults. The open source data consists of maintenance records, problem descriptions, and customer feedback, entered into National Highway Traffic Safety Administration (NHTSA) vehicle owner’s complaint database. The findings suggest that the LLM can, up to a certain level of complexity, accurately analyze fault descriptions and predict the category of failures. It can also extract useful information from complaint descriptions that helps in the diagnostic process and decision-making. However, the accuracy of the LLM’s decisions decreases as the task complexity increases and approaches real-world scenarios. Additionally, the results indicate that traditional supervised machine learning classifiers generally perform better in text classification tasks within our automotive faults context. This study wants to contribute to academic knowledge in AI applications and offer some insights for automotive industry professionals. It introduces a methodology that promises to serve as a foundation for advancements in vehicle diagnostics, aligning with the ongoing shift towards automation and efficiency.