Gearbox Fault Diagnosis Based on Multi-source Dynamic Weighting and Improved Domain Adversarial Network
Existing gearbox multi-source migration diagnosis models have limited generalization capabilities in different migration tasks.In addition,the impact of feature distribution differences between the target domain and each source domain on the model is ignored.To solve these problems,a fault diagnosis method based on multi-source dynamically weighted and improved domain adversarial networks is proposed.First,high-dimensional features of all samples are extracted through one-dimensional convolutional neural network;then,the kernel function in the maximum metric difference is improved into an adaptive hybrid kernel function,and a mapping with learnable parameters is introduced.The function is used as a hybrid weight to adaptively measure the difference between domains;finally,the corresponding metric difference and domain adversarial loss of the target domain and each source domain are used as a joint weight to assign to the domain adversarial strategy.The data sets of gearboxes at different shaft speeds verify the generalization ability and effectiveness of this method across working conditions,providing a reference for the research and application of fault diagnosis based on big data.
fault diagnosismulti-source transfer learningcross-working conditionsmulti-source dynamic weightingimproved domain adversarial network
NETL
NSTL
万方数据
