A Novel SVD-SDP-CNN Fault Diagnosis Method

The one-dimensional (1D) data analysis of stress wave data typically involves the time domain, frequency domain, and wavelet domain. However, the 1D format can be limited in capturing complex feature information, leading to potential loss of diagnostic detail. To address this, the paper introduces an innovative fault diagnosis method that combines optimized Singular Value Decomposition (SVD), Symmetrized Dot Patterns (SDP), and Convolutional Neural Networks (CNN)-the SVD-SDP-CNN approach. This method transforms 1D stress wave data into a two-dimensional (2D) format using SDP, while enhancing data quality through a high-order transformation. The process begins with the application of SVD to decompose the stress wave data. Subsequently, the SVD decomposition matrix is optimized with Network Embedding Update (NUE), resulting in a high-order decomposition matrix that enriches the characteristic information of the signal’ s image representation. The SDP then maps the 1D signal into a 2D image, upon which the CNN performs gear fault diagnosis. Experimental results demonstrate that the fault diagnosis accuracy of the image data refined by NUE has improved from 87.5% to 90%.