Fault Diagnosis Method for Oil Pump Bearing Under Variable Operating Conditions Based on GADF and ConvNeXt Improved by SE Module

Aiming at the serious consequences that may be triggered by oil pump bearing failures as well as the problems of traditional methods in terms of feature extraction, diagnostic accuracy, and model structure, this study proposes a variable operating condition fault diagnosis based on the combination of Gramian Angular Difference Fields (GADF) and ConvNeXt model improved by SE module. The model takes advantage of GADF's ability to capture the changing trends in a bearing vibration sequence through angular and relative variations, and the lightweight residual convolution structure in the ConvNeXt deep neural network, which is capable of extracting and learning the features in the vibration signals of the oil pump bearings end-to-end while consuming fewer parameters and arithmetic power. By randomly adding noise to 12kHz sampling frequency data of the Case Western Reserve University (CWRU) bearing dataset to simulate real-world working conditions, and in turn, using model training, a 90.1% accuracy was achieved on the test set, significantly better than previously published results. Moreover, the optimal model ideally diagnoses faulty conditions on the CWRU bearing dataset under different operating conditions and JNU bearing dataset, possessing strong generalization ability and effectiveness.