Research On Fault Transfer Diagnosis Method Of Different Specs Rolling Bearings

Rolling bearings are widely used in industrial production,and there are many rolling bearing specs in actual production.There are great differences between the failure frequencies of the vibration signals of different specs of rolling bearings,when the traditional fault diagnosis method under a single spec is directly applied to the different specs,the error rate may be higher.Therefore,it is of great significance to realize fault diagnosis of bearings of different specs with high accuracy.Two methods are proposed to solve the problem of fault diagnosis of rolling bearing of different specs.A fault diagnosis method for rolling bearing of different specs based on a deep conditional confrontation network is proposed.This method first converts the one-dimensional vibration signal of the rolling bearing into the two-dimensional images,and then selects a rolling bearing of a certain spec with status information as the source domain,and the target domain is a rolling bearing of other spec without status information.The source domain and target domain data are jointly input into the deep conditional adversarial network improved by random linear combination,the deep convolutional network is used to extract the deep features of different domain data,and the transfer learning is combined to further reduce the distribution difference between bearing data of different specs,so as to identify the failure states of rolling bearings of different specs.The experimental results show that the proposed method solves the problem of fault diagnosis of rolling bearing of different specs with high diagnostic accuracy.In the above method,the source domain and target domain have the same number of categories,and it is easy to miss characteristic information when using the same network in different domains.Based on the above problems,another fault diagnosis method is proposed.A faults diagnosis method for rolling bearing of different specs based on semi-supervised heterogeneous model transfer is proposed.The input sample data of this method are also two-dimensional images,and the source domain and target domain data belong to different specs of rolling bearings.Among them,the source domain data all have labels,and the target domain data contain a few labels.First,input the source domain data into the Res Net-34 network for training,and get the pre-training model.Then the improved parameter transfer strategy adaptively determines the knowledge level and content of the transfer pre-training model,and introduces it into the target domain Res Net-152 network data training process to assist target domain data to construct fault state recognition model and obtain fault diagnosis results.The experimental results show that the proposed method can achieve higher diagnosis accuracy when the specs of source domain and target domain and the number of fault states are different.Compared with the method(1),the scope of application of this method is wider,and it is not required that the source domain and the target domain have the same number of data categories.