Signal-based parameter and fault identification in roller bearings using adaptive neuro-fuzzy inference systems

The rolling element bearings are used in high load-bearing, high stiffness, and high-speed applications. They have wide applications in aero-engine and automobile rotors. In practice, major rotor failures occur with bearing faults. Therefore, it is required to identify the location and intensity of such bearing faults from time to time. In recent times, several signal-based fault identification approaches were proposed for the condition monitoring of ball and roller bearing systems in rotors. In the present work, an experimental framework of the rotor-bearing system is established to study the dynamics of the system under different operating conditions including the faults on the inner race, roller, and outer race. Experiments are conducted under different operating conditions with these faults. The experimental results are compared initially with finite element analysis as a means of validation. Using the empirical mode decomposition (EMD) method, the intrinsic modal functions are estimated for the time response signals. An inverse identification approach is proposed for the identification of the operating parameters from the vibration response using a counter propagation neural network (CPNN) model. Later, an adaptive neuro-fuzzy inference system (ANFIS) is proposed for the classification and identification of faults by analyzing the operating conditions from CPNN and statistical parameters from EMD. The proposed CPNN- and ANFIS-based methodology could predict the faults in roller by 100%, inner race by 87.5%, and outer race by 96%.