Enhancing Rolling Bearing Fault Diagnosis in Motors Using the OCSSA-VMD-CNN-BiLSTM Model: A Novel Approach for Fast and Accurate Identification

Abstract

This study addresses the challenges posed by the strong noise and nonstationary characteristics of vibration signals to enhance the efficiency and accuracy of rolling-bearing fault diagnosis in electric motors. A fault diagnosis model is proposed based on improved variational mode decomposition (VMD) and a convolutional neural network bidirectional long short-term memory (CNN-BiLSTM). In the feature extraction stage, the Osprey-Cauchy-Sparrow search algorithm (OCSSA) optimizes the modal number K and penalty coefficient $\alpha $ of the VMD, facilitating the decomposition and reconstruction of the original vibration signals to extract fault features based on the minimum envelope entropy criterion. In the fault diagnosis stage, the mean, variance, peak value, kurtosis, RMS value, peak-to-average ratio (PAR), impulse factors, form factor, and clearance factor were computed from the reconstructed signals. These indicators were used to construct a feature vector for each sample, serving as the input for the OCSSA-VMD-CNN-BiLSTM fault diagnosis model, which quickly and accurately identifies the fault types. Experimental verification confirms that this method enhances the speed and accuracy of rolling-bearing fault identification compared to traditional approaches.

Keywords

• Fault diagnosis
• variational mode decomposition (VMD)
• osprey optimization algorithm (OOA)
• sparrow search algorithm (SSA)
• bidirectional long short-term memory (BiLSTM)