FAULT DIAGNOSIS OF ROLLING BEARING UNDER MARINE NOISY ENVIRONMENTS AND VARYING WORKING CONDITIONS

Rolling bearings are the core components of ship rotating mechanism, and accurate fault diagnosis has great significance in preventing serious failures of them and performing effective maintenance strategy during navigation. However, varying working conditions and cabin noisy environment would affect the monitored signals of bearings, resulting in the degradation of diagnosis performance. To address this problem, a new fault diagnosis method combining the Denoising Convolutional Auto-Encoder (DCAE) and the multi-scale Residual Net (MSRNet) is proposed in this paper. The DCAE model is designed for denoising. The fully-connected layers of Denoising Auto-Encoder (DAE) are replaced with one-dimensional convolutional layers. In the MSRNet, multiple-size convolutional kernels are utilized in each residual block for extracting multi-scale features of the signals. Typical bearing failure dataset is used for validation. The experimental results show superior performance on noise resistance and domain adaptation of the proposed method under noisy environments and varying working conditions.