Research On Fault Diagnosis Of Rotor System Based On Multiple Sample Sources

Fault diagnosis of rotor system is the key to improving the safety and reliability of rotating machinery.With the development of machine learning theory,intelligent diagnosis technology is widely applied to the fault diagnosis of rotor system.In practical engineering application,the class-imbalance problem often exists in the intelligent fault diagnosis due to insufficient fault data,which leads to low fault diagnosis accuracy.To solve the problem,a fault diagnosis method of rotor system based on multiple sample sources is proposed.The sample sources include monitoring data,simulation data and the generator of generative adversarial network(GAN).The main research contents are as follows:1.In order to obtain accurate simulation data,the research on high precision modeling of rotor system with fault is carried out.A dynamic model of rotor system with vibration fault is established based on the finite element method.The fault types include unbalance fault,misalignment fault and rub impact fault.Each fault includes three severity levels: slight,medium and severe.To improve the modeling accuracy,an identification method based on long short-term memory(LSTM)neural network is proposed to identify the support stiffness and damping of the rotor system.The deep learning neural network with LSTM as the core layer is applied to construct the nonlinear relationship between the support parameters and the displacement responses.The support parameters are directly identified with the generalization of the neural network.The effectiveness of the dynamic model is verified by the vibration fault tests,the simulation data highly similar to the experimental data are obtained.2.Aiming at the fault diagnosis of rotor system with various fault types and fault severity,an intelligent fault diagnosis method based on parameter optimized symmetrized dot pattern(SDP)analysis is proposed.In the method,the fault signal characteristics are extracted through SDP analysis.The parameters of SDP analysis are optimized based on beetle antennae search(BAS)algorithm.The fault diagnosis model is constructed by convolutional neural network(CNN).Firstly,the effectiveness and superiority of the proposed method is verified based on the monitoring data.Then,the proposed method is applied to study the influence of class-imbalance problem on fault diagnosis accuracy and prove the effectiveness of the simulation samples.3.Aiming at the class-imbalance problem in the fault diagnosis of rotor system,an intelligent fault diagnosis method based on multiple sample sources is proposed.The initial sample space is constructed by the monitoring samples.To achieve the completeness of the sample space,the simulation samples and generated samples are applied to supplement the sample space.Firstly,the research on image sample generation based on GAN is carried out to obtain the generated samples that are highly similar to the monitoring samples.Then,the effectiveness of the proposed method is verified by the study on fault diagnosis with class-imbalance problem on the rotor system test-platform.