基于改进正余弦算法优化堆叠降噪自动编码器的电机轴承故障诊断

作者：Li Bing^*; Liang Shuqi; Shan Wanning; Zeng Wenbo; He Yigang

来源：Transactions of China Electrotechnical Society, 2022, 37(16): 4084-4093.

DOI：10.19595/j.cnki.1000-6753.tces.210306

• adaptive
• fault diagnosis
• improved sine cosine algorithm
• motor bearing
• Stacked denoising auto encoders

The bearing is an important part of motor, but its fault vibration signal has noise interference, which makes feature extraction difficult. Stacked denoising auto encoders (SDAE) can effectively suppress the noise interference by setting the input data to zero and training the network randomly. In addition, the unsatisfactory combination of hyperparameters is likely to cause poor diagnostic performance of SDAE. Therefore, an improved sine cosine algorithm (ISCA) was proposed to optimize SDAE for motor bearing fault diagnosis. Firstly, the nonlinear inertia weight was introduced into the particle value update formula of sine cosine algorithm (SCA), and the control parameters were added with cosine change to construct ISCA. The hyperparameters of SDAE were adaptively selected by ISCA. Secondly, the unsupervised self-learning feature extraction method of SDAE model with optimal network structure was used to extract the characteristic parameters of vibration signals, so as to achieve better fault diagnosis effect. Simulation and field experiment results show that the proposed method has high convergence speed, high diagnosis accuracy and strong robustness, and has a good application prospect in motor bearing fault diagnosis. ? 2022 Chinese Machine Press.

• 单位
  合肥工业大学