A method for rolling bearing life state recognition by combining health indicator and anti-noise deep residual shrinkage network

Recognizing the life states of rolling bearing accurately is not only convenient for experts to make maintenance strategy in advance, but also to avoid the direct economic loss caused by equipment downtime and damage. Aiming at the difficulty in assessing bearings degradation performance accurately and the low accuracy of recognizing bearing life states, a new method for recognizing the life state of rolling bearing combining health indicator and anti-noise deep residual shrinkage network (ADRSN) was proposed. Firstly, the envelope spectrum data samples can clearly reveal the fault characteristic frequencies, so multipliers of the signal were input into deep convolutional autoencoder to realize the automatic extraction and expression of bearing life state features. Secondly, the bearing health indicator is constructed in the low-dimensional feature space based on multidimensional scaling and Euclidean distance, which is used to assess the bearing degradation performance. After that, the bearing life states were classified and the state samples were labeled according to the bearing health indicator curves. Subsequently, to further enhance the noise resistance of the recognition model, dropblock regularization and dropout regularization were introduced into the deep residual shrinkage network to construct the ADRSN model, and the labeled samples were used to train the ADRSN to obtain the bearing life state recognition model. Finally, the proposed bearing health indicator and ADRSN model were applied to the bearing full-life experimental data. The results proved that the proposed health assessment method has more advantages than other conventional method. In addition, the ADRSN improves recognition accuracy by about 20% in noisy environment, which shows that ADRSN has more resistance and generalization in noisy environment.