Fault Diagnosis for Rolling Bearing of Road Heading Machine via SVDS-ICNN

Abstract

Abstract A novel fault diagnosis method based on Improved Singular Value Decomposition (SVD), S-transformation and Improved Convolutional Neural Networks (ICNN) is proposed for the non-stationary, nonlinear, interfered by strong background noise and difficult feature extraction problems of rolling bearing vibration signal of the road heading machine. Firstly, the original signal is constructed into a Hankel matrix which was decomposed by SVD. The effective singular values are selected according to the curvature spectrum of the singular values for signal recon-struction, and the reconstructed signals are transformed by S to generate the feature map, which is input into ICNN adaptive feature extraction for the fault identification. Secondly, the im-proved convolutional neural network uses VGG16 as a Bottleneck structure, introduces the bot-tleneck structure, selects input data with different sizes for feature extraction, adds Fine Tune on the basis of ICNN, and finally realizes fault classification and recognition through network pa-rameter adjustment. The proposed method is applied to the fault diagnosis of road heading ma-chine rolling bearings, and the accuracy rate is 98.2%, which is 9.55% higher than the classic VGG16 model.