TKEO-Enhanced Machine Learning for Classification of Bearing Faults in Predictive Maintenance

Predictive maintenance is essential for improving the efficiency of equipment and reducing downtime in industrial operations. This study investigates the application of machine learning in predictive maintenance, specifically emphasizing data preprocessing and classification techniques using the Teager–Kaiser Energy Operator (TKEO) method, which captures dynamic variation in signals. The effectiveness of TKEO was compared against conventional methods, using the Case Western Reserve University (CWRU) dataset, with vibration data collected from bearings operating under different load conditions. Different data segmentation lengths (2400 and 12,000 samples) were evaluated to assess the impact of segment size on classification accuracy. The study also investigated the effects of various feature selection strategies by comparing four- and six-feature combinations. Advanced classifiers, including support vector machines and random forests, demonstrated that TKEO effectively improved model accuracy in the capture of fault-related signal dynamics. These findings offer new insights to support reliable predictive maintenance in industrial settings and provide a new perspective for future research into active vibration control, where vibration signal analysis, feature extraction, and mathematical modeling play key roles in optimizing control algorithms and enhancing the efficiency of adaptive control systems.