Application of CNN-1d based on feature fusion in bearing fault diagnosis

摘要

Aiming at the problems of small training sample capacity of ordinary deep learning models, single feature extractor size, insufficient feature extraction of bearing faults, and low recognition rate of bearing health status under variable loads, a feature fusion one-dimensional convolutional neural network algorithm model was proposed. It took the raw vibration signal as input, used the feature fusion module to extract multi-scale time information and the convolution-pooling alternating layer adaptively overcome the time-dependent characteristics. An adaptive batch normalization was introduced to reduce the difference in sample distribution between source and target domains and enhance model generalization capabilities. It combined the softmax classification layer to construct a feature extraction-feature classification dual intelligent fault diagnostic algorithms. The experiments were performed on the rolling bearing fault data set. The results show that the method has a rather high generalization ability and a high fault recognition rate under single load and cross load conditions. Simultaneously, the introduction of t-SNE dimensionality reduction visualization revealed the deep network model structure has strong feature extraction ability for large-volume samples.