Fault Diagnosis Method For Rotating Machinery Driven By Fault Dynamics And Domain Adaptation

In recent years,intelligent fault diagnosis methods for mechanical equipment have become increasingly abundant with the development of artificial intelligence technology.The failure of rotating machinery has harmful influence,which can affect the efficiency of machinery operation,even cause whole equipment failure,resulting in loss of time and property,and even endangering people’s lives and health.In the field of fault diagnosis of rotating machinery,the most methods require a large amount of labeled data for training,and many models are usually validated with test data that is similar to the training data,resulting in a lack of generalization ability of large-scale models.Therefore,traditional methods have to retrain on new scenarios when be used.However,a large amount of marked fault data is difficult to obtain in real application scenarios,while labeled healthy operation data is easy to obtain.Such task characteristics make it very difficult to deploy traditional methods in practical scenarios.In order to solve this problem,the paper proposed an intelligent fault diagnosis method driven by fault dynamics model and domain adaptation learning.Taking rolling bearings,an important part of rotating machinery,as an example to validate the proposed method.We used the simulation data generated by the bearing fault dynamics model and the Bearing data in real scenes.And the domain adversarial transfer learning method is introduced to complete the fault diagnosis task under unlabeled and few samples.First,the general form of bearing fault is introduced,and the fault conditions of the bearing are analyzed,which corresponding to different vibration characteristic frequencies,such as fault,unbalance,and misalignment,and then a 2-DOF fault dynamic model of bearing with fault characterization ability is constructed based on bearing-rotor dynamics.Second,the geometric parameters and shape parameters of the bearing under a specific instance are introduced.Then,the vibration simulation data of the rolling bearing under different health states and different conditions are obtained by solving the dynamic equation use the numerical integration method,and then combined with the real situation of the bearing signal in the healthy condition.In order to combine the working conditions and background noise information in the real scene.The dynamic model of the rolling bearing contains the key characteristics of the bearing in different health states,which can be used to assist the training of the fault diagnosis model.Finally,fault identification is performed on three groups of real cases through a few-shot domain adversarial fault diagnosis model.The results show that the fault diagnosis framework based on dynamics and domain adaptation algorithm has good recognition performance for few-shot bearing fault recognition scenarios.