A Robust Rolling Bearing Fault Diagnosis Approach Using Hybrid Red Deer Optimization-Based 1D-CNN

Rolling bearings serve as one of the critical elements in rotating machines, and ensuring its fault-free operation is essential for the overall safety and efficiency of the equipment. In recent years, data driven fault diagnosis techniques are extensively proposed for detecting faults in rolling bearing due to rising demands from industry. This work introduces a model for identifying the faults in rolling bearings using variational mode decomposition (VMD), multifractal detrended fluctuation analysis (MFDFA) and optimized one dimensional convolutional neural network (O-1D-CNN). Firstly, with the prevalent noise present in the original vibration signal, a wavelet threshold denoising (WTD) model is employed to effectively denoise the raw vibration signals. Secondly by recognizing the inherent nonlinearity of rolling bearing vibration signals, the VMD technique and MFDFA are exploited for extracting meaningful features. Finally, the features are classified using the O-1D-CNN where the network is an integration of the hybrid red deer algorithm (HRDA) and 1D-CNN. The proposed model is applied to the datasets from experiments and from the Case Western Reserve University (CWRU) for comparison and verification. The results demonstrate that the proposed method has a maximum fault classification accuracy of 99.5% and 99.7% based on the experimental and CWRU dataset respectively.