Fault Diagnosis of Gearbox Based on Improved Transformer

Health services for the gearbox are essential to ensure the safety of industrial production. However, gearbox fault diagnosis remains a challenge due to complex responses caused by multiple gears. In recent years, there has been an increasing number of scholars who have embraced deep learning techniques for gearbox fault diagnosis. Compared with traditional deep networks, transformers exhibit outstanding pattern recognition capabilities, but their ability to process local information is lacking. Therefore, we propose an improved Transformer network that applies multi-scale perception layers and linear embedding to enhance the ability to capture local feature information, analyze fault features on multiple time scales, and retain the position information of the original signal. The validation was conducted on the Southeast University (SEU) gearbox dataset. The model's classification accuracy can reach 99.4% when detecting five different fault signals with a signal-to-noise ratio of 10. Experimental results have shown that the method has excellent diagnostic and anti-interference capabilities, which would be applied in engineering practice.