Explainable AI-based Prediction of Seatbelt BSR Noise via Design Parameter-based Vibration Analysis

This study proposes an eXplainable AI (XAI)-based vibration evaluation method to assess the buzz, squeak, and rattle (BSR) performance of vehicle seatbelt retractors, focusing on noise-induced anomalies. The proposed method enables the prediction of vibration characteristics from seatbelt retractor design parameters, enabling the pre-screening of potential BSR noise occurrence at the design stage. Compliance measurements were collected from 15 different vehicles, and vibration evaluation indicators (i.e., system stiffness and Q-factor) were derived to quantify dynamic response characteristics related to BSR noise. These indicators serve as key metrics for BSR-related performance evaluation, enabling defect identification based on predefined thresholds. Furthermore, an XAI-integrated machine learning model was developed to predict the system stiffness and Q-factor directly from seatbelt retractor design parameters. The predicted indicators were validated against experimental thresholds, and the Shapley additive explanations algorithm was employed to analyze the influence of design parameters on noise-induced vibration performance. By leveraging the XAI approach with the compliance data from multiple vehicle models, this paper presents a systematic methodology for BSR-focused vibration analysis and quality control of seatbelt retractors. The proposed approach enables manufacturers to predict vibration characteristics without direct measurement, enabling earlier detection of potential BSR issues and reducing reliance on costly experimental testing while ensuring product reliability.