AI-Based Predictive Maintenance for Industrial Machinery: A Machine Learning Approach to Failure Prediction

Abstract

Unplanned equipment failure is one of the most expensive setbacks in modern manufacturing, and data-driven predictive maintenance is one way to predict equipment failures before they happen. In this study, a secondary dataset of operating parameters such as temperature, vibration, rotation speed, pressure, and torque were developed for the prediction of industrial machine failure and machine models were developed using machine learning techniques. After data cleaning, standardization, descriptive and correlation analysis, three classifiers were trained on a stratified training set: Random Forest, Support Vector Machine and XGBoost classifiers, which were then tested on a held-out test set using accuracy, precision, recall, F1-score and the area under the receiver operating characteristic curve (ROC-AUC). The problem of class imbalance was overcome by cost-sensitive learning while the reasons for failure were analyzed by means of gain-based feature importance and SHapley Additive exPlanations. The highest overall scores (accuracy = .985, recall = .838, F1-score = .797, ROC-AUC = .972) were obtained by XGBoost, while Random Forest and Support Vector Machine performed in a similar manner. Vibration and torque were the most significant factors leading to failure. The results suggest that gradient-boosted tree models have the potential to be accurate and interpretable early-warning models for industrial machinery and the study ends with useful suggestions on how these models can be incorporated into maintenance planning.