Research on inter-shaft bearing fault diagnosis method based on STFT-CNN modeling

Abstract

The fault signals of inter-shaft bearings exhibit nonlinear and non-stationary characteristics, making it difficult for traditional feature extraction methods to effectively capture their intrinsic features, which in turn affects the accuracy of fault identification. Consequently, fault type diagnosis has become a significant challenge in this field. To improve the accuracy of fault type prediction for inter-shaft bearings, this study proposes an STFT-CNN model based on the AlexNet architecture, which integrates short-time fourier transform (STFT) with convolutional neural networks (CNN). This approach addresses the common issues in conventional inter-shaft bearing fault diagnosis for aero-engines, such as strong reliance on expert knowledge and limited feature extraction capability. First, real-world vibration fault signals of inter-shaft bearings were collected through experiments, and the STFT method was employed to enrich the feature representation of non-stationary signals. Second, by leveraging the strong feature representation and certain visualization capabilities of the STFT-CNN model in the time-frequency domain, fault features under various operating conditions were effectively extracted, enhancing the analysis of the intrinsic characteristics of non-stationary signals. Finally, thanks to the advantages of the STFT-CNN model in feature extraction and generalization, it demonstrated high accuracy and stability in fault type classification and prediction tasks. The training process involved comparative analysis of different pooling algorithms, time-frequency analysis methods, and various deep learning network models. Experimental results show that the STFT-CNN model with max pooling outperformed other models in inter-shaft bearing fault prediction, achieving an average fault prediction accuracy of 98.8%.