Machine learning based predictive maintenance framework for belt-pulley systems

Abstract

Belt-pulley systems, widely used in industrial power-transmission applications, depend on predictive maintenance to maintain dependability and operational effectiveness. A machine-learning-based system for fault identification and classification based on vibration, slip ratio, and belt-tension measurements obtained from a controlled experimental setup is presented in this paper. Four different operating conditions—Normal, Misalignment, Belt Slip, and Insufficient Tension—were used to create a comprehensive dataset. The dataset comprised 600 samples: Normal (188 instances), Misalignment (153 instances), Belt Slip (150 instances), and Insufficient Tension (109 instances). The model performance was analyzed using Receiver Operating Characteristic (ROC) curves, confusion matrices, and standard metrics. To guarantee high-fidelity ground truth, each condition was labelled using experimentally determined threshold-based rules. Using a single pipeline that included feature extraction, supervised learning, and multi-metric performance evaluation, three classification models—Logistic Regression (LR), Support Vector Machine (SVM), and Random Forest (RF) —were created and assessed. Quantitative analysis demonstrates that the Random Forest classifier achieved the highest performance, with 99.5% accuracy, 99.0% precision, 99.5% recall, and an Area Under the Curve (AUC) of 1.00, indicating near-perfect separability of fault states. In comparison, LR and SVM yielded moderate accuracies of 78.5% and 74.0%, respectively, with lower discriminative ability reflected in their ROC curves. Furthermore, ensemble learning successfully lowers misclassification among closely linked defects and resolves nonlinear decision boundaries, according to confusion matrix analysis. By achieving strong fault detection and high classification accuracy, the suggested framework makes it possible to identify degradation patterns early on that are challenging to detect using conventional monitoring techniques. Overall, the findings demonstrate machine learning’s potential for scaled implementation in industrial predictive maintenance settings and establish its applicability for belt-pulley systems’ real-time health evaluation. The outcomes confirm the potential of machine learning for scalable industrial predictive maintenance and real-time health monitoring of belt-pulley systems.