Enhancing Automated Warranty Repair Reports Analysis with AI Generative Technologies: Failure Modes Detection in Industrial Vehicles for Reliability Improvement

Warranty claims provide valuable information for identifying recurring failures and improving industrial vehicles. This work presents a practical framework that combines community-detection-based clustering and AI-assisted semantic analysis to process large, heterogeneous maintenance reports. The method integrates human-defined labels with AI-generated insights, handling inconsistent terminology and sparse taxonomies to identify robust failure patterns. Thousands of warranty reports per job run were analyzed, producing structured labels and clusters that highlight recurring failure modes, related symptoms, and interactions across different vehicle systems. The approach uses community detection techniques to group similar issues, making patterns easier to interpret than manual review alone. The system is designed for scalability and reproducibility, allowing experts to refine clusters and verify results iteratively. Interactive dashboards enable hierarchical exploration of failures, helping engineers track issues over time and understand cross-system effects. Results show that combining basic statistical signals with semantic information improves the clarity and practical usability of warranty data. The framework helps convert unstructured maintenance reports into structured evidence that engineers can interpret and validate directly. It supports reliability engineering, design improvements, and after-sales operations while remaining adaptable to different datasets and industrial contexts.

Keywords: Warranty data analysis; industrial vehicles; failure modes detection; Jaccard similarity; Louvain clustering; generative AI; reliability engineering; dashboards; label harmonization.