A novel transfer learning method based on multi-kernel MMD and domain adversarial for fault diagnosis of bearings in robots

When diagnosing faults of different types of bearings in robots, there are not only differences in edge distribution between the source domain and the target domain, but also differences in internal classification. Considering the high requirements for stability and accuracy in robot bearing fault diagnosis, this paper proposes a neural network model named DAOCM based on multi-kernel MMD and domain adversarial for fault diagnosis of different types of bearings in robots. As a domain difference measurement method, multi-kernel MMD can simultaneously reduce the edge distribution distance and conditional distribution distance of the source and target domains. Domain confrontation can help transfer knowledge and patterns learned in the source domain to the target domain through domain adaptation, improving the generalization of the model. The experiment shows that DAOCM can achieve fault diagnosis of different types of bearings under variable working conditions. The accuracy of bearing fault diagnosis is above 96%.