Sarcouncil Journal of Engineering and Computer Sciences

Abstract: The present study introduces a machine learning–based predictive maintenance framework that fuses vibration sensor data and thermal imaging to accurately forecast bearing failure in rotating machinery. The research aims to overcome the limitations of traditional single-sensor monitoring systems by integrating mechanical and thermal features for comprehensive fault detection and life prediction. Experimental data were collected from a bearing test rig under varying speeds, loads, and fault conditions; healthy, inner race, outer race, ball defect, and combined faults. After preprocessing, critical time-domain, frequency-domain, and texture-based features were extracted and fused using Principal Component Analysis (PCA) for dimensionality reduction. Four supervised machine learning models; Support Vector Machine (SVM), Random Forest (RF), Artificial Neural Network (ANN), and Convolutional Neural Network (CNN) were developed and evaluated. The CNN outperformed all other models, achieving 98.2% classification accuracy and an AUC of 0.99, confirming its superior capability in capturing nonlinear and spatial-temporal relationships in fused data. Additionally, an ANN regression model was implemented to predict the Remaining Useful Life (RUL), achieving R² values between 0.95 and 0.97 with minimal mean absolute error. The strong positive correlation (r = 0.81) between vibration and thermal parameters validated the synergistic nature of multimodal data. The proposed framework thus demonstrates a significant advancement in predictive maintenance by enabling early fault detection, accurate classification, and reliable RUL estimation, providing a scalable solution for Industry 4.0–driven intelligent maintenance systems.