Optimization of dilated convolution networks with application in remaining useful life prediction of induction motors.

• A novel neural network was proposed, which combined the advantages of two kinds of convolution. • A supervised feature selection strategy based on grey relation analysis was proposed. • Multi-sensor data were considered in the remaining useful life prediction of induction motors. • The proposed method was validated by remaining useful life prediction of induction motors. A novel remaining useful life prediction (RUL) method of induction motors based on hybrid dilated convolution networks (HDCN) and grey relation analysis (GRA) was proposed. To test the performance of the proposed method, the RUL prediction experiment was performed on a data set containing 8 motors. Firstly, the features of time domain, frequency domain, and entropy were extracted from the original signals. Secondly, GRA was used for feature selection, and a feature selection strategy was proposed. Finally, the data set after feature selection was imported into HDCN for training and testing. The results show that the predicted RUL and real RUL have the same trend and similar values. Seven comparison methods were designed and the same experiments were carried out. The results show that most of the root mean square error and mean absolute error of the proposed method are smaller than those of the other seven methods. [ABSTRACT FROM AUTHOR]