Application of a New Model Based on Convolutional Autoencoder Denoising Technique And Few-Shot Learning In Bearing Fault Diagnosis  
Author Ying Chen

 
Co-Author(s) Mengshu Yang; Xufeng Zhao

 
Abstract Bearing fault diagnosis is essential for maintaining the reliability and safety of machinery. However, current research faces significant challenges, including the lack of labeled samples and the interference of noise during signal acquisition, which can adversely affect diagnostic accuracy. These issues have often been overlooked in existing studies. This paper aims to address these challenges by developing a model that combines denoising techniques with few-shot learning for effective fault prediction. We establish a framework that enhances the denoising process and improves modeling capabilities with limited data. In order to reduce the noise effect, we use encoders and decoders to compress and reconstruct the feature information to obtain the denoised samples. Then, the siamese network is used to distinguish the similarity to achieve fault classification. Experimental results demonstrate that the proposed model shows promising performance in denoising and accurately predicting faults on small sample datasets.

 
Keywords Fault diagnosis, rolling bearing, few-shot learning, convolutional denoising autoencoder, limited data
   
    Article #:  DSBFI25-23
 
Proceedings of 3rd ISSAT International Conference on Data Science in Business, Finance and Industry
January 6-8, 2025 - Da Nang, Vietnam