A novel approach integrating dimensional analysis and neural networks for the detection of localized faults in roller bearings

The detection of the defective/worn out bearing components used in rotating machines is one of the main concerns in various applications. To improve the computational efficiency in the nonlinear dynamic analysis for the rolling contact bearings, a new methodology based on dimensional analysis (DA) theory is proposed in this paper. The developed model is used to predict the vibration responses due to artificially spalled bearing components to quantify the level of structural damages into these components. The use of a back propagation neural network (BPNN) has been made that also predicted responses from the network trained by developed algorithm using the experimental data obtained from the defective bearing components on the developed test rig. A comparison between the responses predicted by proposed DA method and the BPNN showed a fair amount of the agreement between the two approaches and validated the proposed model and proved outstanding tool for identification of spalled/damaged bearing components.