Fault diagnosis of rolling bearings with noise signal based on modified kernel principal component analysis and DC‐ResNet

• Published in: CAAI Transactions on Intelligence Technology Vol. 8; no. 3; pp. 1014 - 1028
• Main Authors: Zhao, Yunji, Zhou, Menglin, Xu, Xiaozhuo, Zhang, Nannan
• Format: Journal Article
• Language: English
• Published: Beijing John Wiley & Sons, Inc 01.09.2023; Wiley
• Subjects: Accuracy; Algorithms; artificial intelligence; Bearings; Classification; Data acquisition; Decoupling method; Deep learning; Deformation effects; Failure; Fault diagnosis; fault feature extraction; Formability; Fourier transforms; Industrial production; Machinery; Methods; Neural networks; Noise; Noise levels; Pattern recognition; Principal components analysis; Random noise; Random variables; Roller bearings; Signal processing; Support vector machines; Wavelet transforms
• ISSN: 2468-6557; 2468-2322; 2468-2322
• DOI: 10.1049/cit2.12173

In view of the influence of aliasing noise on the effectiveness and accuracy of bearing fault diagnosis, a bearing fault diagnosis algorithm based on the spatial decoupling method of modified kernel principal component analysis (MKPCA) and the residual network with deformable convolution (DC‐ResNet) is innovatively proposed. Firstly, the Gaussian noise with different signal‐to‐noise ratios (SNRs) is added to the data to simulate the different degrees of noise in the actual data acquisition process. The MKPCA is used to project the fault signal with different SNRs in the kernel space to reduce the data dimension and eliminate some noise effects. Finally, the DC‐ResNet model is used to further filter the noise effects and fully extract the fault features through the training of the preprocessed data. The proposed algorithm is tested on the Case Western Reserve University (CWRU) and Xi'an Jiaotong University and Changxing Sumyoung Technology Co., Ltd. (XJTU‐SY) bearing data sets with different SNR noise. The fault diagnosis accuracy can reach 100% within 30 min, which has better performance than most of the existing methods. The experimental results show that the algorithm has an excellent effect on accuracy and computation complexity under different noise levels.