Research On Fault Diagnosis Method Of Rotor System Based On Deep Transfer Learning Under Variable Working Conditions

The working conditions of rotating machine equipment in complex working conditions often depend on its key component rotor system.With the improvement of technology,the modern industrial level is constantly showing new heights,and the rotating machines equipment is constantly evolving and developing.The operating conditions are complex and changing,and the rotor system with a high fault rate causes that machinery equipment fail frequently.After rotating machinery generate faults,fault factors spread rapidly.If the fault cannot be precisely discovered and the cause of the fault cannot be addressed,it will not only lead to economic loss,but could also lead to casualties.Therefore,it is of great practical importance and application value to analyze rotor system fault and conduct extensive research on the method of diagnosing rotor system fault under varying working conditions,and discover rotor system fault quickly and accurately.Rotor system is taken as the research topic.This paper integrates deep learning and transfer learning into fault diagnosis,and builds a fault diagnosis model of rotor system under varying working conditions based on deep learning and transfer learning.Aiming at the influence of Deep Belief Network(DBN)parameters on the accuracy of fault diagnosis,Sparrow Search Algorithm is introduced and CWTSSA algorithm based on chaotic map and variation of Tdistribution is proposed to optimize the DBN model parameters.In this way,a DBN parameter optimization method based on CWTSSA is implemented,an optimized DBN model is obtained,and a fault diagnosis model with good generalization ability and classification accuracy is built.Aiming at the problem that the distribution difference of the data characteristics caused by the change of the operating conditions of the rotor system affects the diagnosis accuracy of the fault diagnosis model.The JDA(Joint Distribution Adaption)method is used to measure the source domain data and target domain data in order to reduce the joint distribution difference between the samples.The fault characteristics matrix is constructed and a variable condition rotor system failure diagnosis(JACADN)method based on the adaptive distribution joint and on the optimized deep trust network is proposed,which can realize the fault diagnosis of the rotor system under variable conditions.Accurate diagnosis of rotor system faults under variable operating conditions is realized.The effectiveness of rotor system fault diagnosis method based on CWTSSA optimized DBN and variable working condition rotor system fault diagnosis method based on transfer learning and deep belief network optimized is verified by experimental data on the vibrations of the QPZZ-II datasets.Experimental results show that CWTSSA can efficiently optimize DBN parameters and obtain a fault diagnosis model with good generalization and classification accuracy.The JDA algorithm was introduced to approximate the distribution between different data.meanwhile,the JACADN method is also employed to achieve better accuracy in the fault diagnosis of the complex working condition changing system.Thus,the fault diagnosis method proposed in this document brings a new idea for fault diagnosing of other rotating machines.