Fault classification in rotor-bearing system using advanced signal processing and machine learning techniques

Tapered roller bearings (TRBs) can handle radial and axial loads effectively, making them suitable for various applications. However, faults can develop over time, thus affecting the dynamic behavior of the rotor-bearing system. Hence, periodic monitoring is essential to ensure the reliability and performance of these systems. This study presents a novel approach to TRB fault classification that makes use of machine learning and signal processing methods. Experiments with both healthy and faulty bearings under various operating circumstances are part of the approach. Vibration data is gathered from bearings with specified fault conditions, such as inner and outer races, cage, roller, and transverse cracks in shaft. Healthy bearings are found to have frequencies of 52.34 Hz and 125.68 Hz. Ensemble empirical mode decomposition (EEMD) is used to process gathered data to find operating parameters that pointed out particular fault types. These parameters are used as input to train machine-learning models that diagnose bearing faults, such as radial basis function neural networks (RBFNN), back propagation neural networks (BPNN), and counter propagation neural networks (CPNN). An adaptive neuro-fuzzy inference system (ANFIS) is presented in this study to improve fault classification accuracy based on these characteristics. After training, the system has achieved a remarkable accuracy rate of 90.9% in fault classification. • Developing an experimental rotor-bearing model to classify the faults. • EEMD is used to decompose the raw signals under different operating conditions. • RBFNN shows a faster convergence rate than BPNN and CPNN models. • Fault classification accuracy using ANFIS is achieved more than 90.9%. • An experimental frequency response compared with the FE model.