Improved deep extreme learning machine learning for early fault diagnosis of rolling bearings

Rolling bearing wear is a common equipment failure problem in the process of bearing use. In this paper, an effective method for fault diagnosis of rolling bearings is proposed by combining squirrel algorithm and deep extreme learning. First, Fast Fourier Transform (FFT) was used to extract the fault bearing vibration signal and reconstruct the time domain features as the input data of the model. Then, the Extreme Learning Machine (ELM) was combined with the Auto encoder (AE) to form the Extreme Learning Machine-Autoencoder (ELM-AE). ELM-AE is used as the basic unit of unsupervised learning to perform greedy layer-by-layer unsupervised pre-training on the input data and save the output weight matrix obtained by ELM-AE by the least square method. The output layer weights are initialized for ELM-AE. Finally, the weights of the output layer of DELM were learned by combining the label information, and parameters such as the input weights of DELM were optimized by SSA to obtain the experimental model.