Integrating Machine Learning with FMEA for failure prioritization and process risk analysis in manufacturing

Abstract

This study proposes an integrated framework combining Machine Learning (ML) with Failure Mode and Effects Analysis (FMEA) to enhance failure prioritization and process risk assessment in manufacturing systems. Historical FMEA data from the lightbulb forming and vacuum processes were used to model and analyze 50 documented failure modes. The Extreme Gradient Boosting (XGBoost) algorithm was employed to capture nonlinear relationships among the FMEA parameters—Severity (S), Occurrence (O), and Detection (D)—and predict Risk Priority Numbers (RPNs) with high accuracy. The model achieved a coefficient of determination (R²) of 0.985 and a Mean Absolute Percentage Error (MAPE) of 2.84 %, validating its predictive robustness. To improve interpretability, SHapley Additive exPlanations (SHAP) were applied to identify the most influential parameters, revealing that Severity had the most significant contribution to process risk, followed by Occurrence and Detection. Unsupervised clustering techniques, including K-Means, Agglomerative Hierarchical Clustering, and Gaussian Mixture Models (GMM), were used to classify failure modes into distinct risk-based groups. The results demonstrated that integrating ML and FMEA provides a transparent, data-driven, and adaptive framework that improves risk prediction, feature interpretability, and prioritization stability. This study contributes theoretically by integrating explainable AI into FMEA and, practically, by offering a reliable, intelligent decision-support tool for data-driven quality management and Industry 4.0 applications.