基于改进CAE-KPCA特征融合的轴承故障诊断OA

To address the issues that traditional rolling bearing feature extraction methods have diffi-culty fully extracting fault pulses and exhibit limited feature extraction capabilities,this paper proposes a feature extraction method for rolling bearings based on improved CAE-KPCA framework.The original vi-bration signals are first converted into SDP images using the point symmetry analysis method.Abstract features are then extracted from these SDP images by an improved Convolutional Autoencoder(CAE),with the model trained efficiently using the Adam optimization algorithm for multi-parameter tuning.Subsequently,the histogram equalization is applied to process SDP images and calculate their statistical fea-ture set.A fault response gain index is used to screen for statistical features that demonstrate significant fault responses,so as to improve their quality.Finally,the Kernel Principal Component Analysis(KPCA)method is employed to reduce dimensionality and fuse the abstract features with the selected statistical fea-tures,eliminating redundancy among them.Experimental results show that compared to using either ab-stract features or statistical features alone,the fused features obtained by the improved CAE-KPCA meth-od achieve the highest fault diagnosis accuracy across several commonly used fault diagnosis models.