Deep Learning-Based Bearing Fault Diagnosis Using a Trusted Multiscale Quadratic Attention-Embedded Convolutional Neural Network

Bearing fault diagnosis is essential for ensuring the safety and reliability of industrial systems. Recently, deep learning approaches, especially the convolutional neural network (CNN), have demonstrated exceptional performance in bearing fault diagnosis. However, the limited availability of training samples has been a persistent issue, leading to a significant reduction in diagnostic accuracy. Additionally, noise interference or load variation during bearing operation poses significant challenges for fault diagnosis. To tackle the above issues, this article explores the application of quadratic neurons with attention-embedded for fault diagnosis networks and introduces a trusted multiscale learning strategy that fully considers the characteristics of bearing vibration signals. Building upon these concepts, a trusted multiscale quadratic attention-embedded CNN (TMQACNN) is proposed for bearing faults diagnosis. Experimental results indicate that the proposed network outperforms six state-of-the-art networks under noise interference or load variation superimposed on small samples.