Fault Diagnosis Model of Semi-supervised Rolling Bearings Based on Pseudo-label Deep Learning
In view of the difficulty of collecting the marked rolling bearing fault samples and low accuracy of traditional diagnosis models in practical engineering applications,a semi-supervised rolling bearing fault diagnosis model based on a pseudo-label learning fused parameter migration deep learning network was proposed.Firstly,the parameters of the pre-trained residual network(ResNet)model on the ImageNet dataset were transferred into this model as the initial parameters,and the network layer parameters were finely tuned using different learning rates to accelerate the model convergence.Then,the model was trained with labeled data and predicted with unlabeled data using pseudo-label semi-supervised learning.Finally,a ResNet model with migrated parameters was trained using labeled and pseudo-labeled data,and the diagnostic effect was evaluated.The semi-supervised fault diagnosis experiments and cross-domain fault diagnosis experiments were carried out on the two kinds of rolling bearing fault data.It is shown that the proposed model can be migrated to various devices in order to complete the diagnosis with a large set of unlabeled samples.It has high robustness and can be used to solve fault diagnosis problems in complex industrial settings.
fault diagnosisrolling bearingsemi-supervised learningdeep transfer learning
国家科技期刊平台
NSTL
万方数据
