Research On Intelligent Detection Methods For Rotor System Operation Anomalies

With the continuous development of science and technology and manufacturing technology,mechanical equipment is developing in the direction of large and intelligent,and the reliable operation of equipment is of great importance.As one of the core industrial machinery,rotary machine has a wide range of applications in many fields.As the core component of rotary machine,the running stability of rotor system plays a decisive role in the running stability and safety of the entire equipment.If the rotor system failure,it will lead to the whole equipment can not operate normally,and there may even be a major safety accident,causing serious losses of property at the same time also threatened the personal safety of staff.Therefore,in order to find the abnormal operation of the rotor system in time due to faults and other reasons,this paper proposes an intelligent detection method for the abnormal operation of the rotor system.The main research contents are as follows:(1)Analysis of rotor system operation characteristics.This paper briefly describes the experimental data--blower operation data collection and data resources,On the basis of the data resources,the operation characteristics of the subsystem are analyzed from the perspectives of time domain and frequency domain;The paper focuses on the analysis of rotor system operation characteristics based on holographic spectrum technology and the method and process of extraction of axis center trajectory of rotor system by using harmonic wavelet transform.(2)Research on anomaly detection method based on machine learning.The method and process of intelligent detection of abnormal operation of rotor system are studied from the perspective of supervised learning.It mainly includes data preprocessing,feature engineering,model construction and model assessment.Considering the actual industrial application scenarios,two research objectives are proposed: anomaly detection and operating state recognition.Abnormaly detection is mainly to judge whether abnormal operation occurs in the unit,which is abstracted as a binary classification problem and solved by support vector machine algorithm.Running state recognition is mainly to judge the running state of the unit with abnormal operation,which is abstracted as a multiclassification problem,and solved by splitting strategy combined with Light GBM algorithm.(3)Research on anomaly detection method based on generative adversarial network.Aiming at the problem that fault data is difficult to obtain,the abnormal intelligent detection method and process of rotor system operation are studied from the perspective of unsupervised learning.The model training stage only uses the data under normal operation state.After some processing,the axis trajectory data after purification and filtering is taken as the input of the network,and then the loss value of the network output is calculated as the index to measure whether the unit has abnormal occurrence.The loss value consists of two parts,residual loss and discriminant loss.Residual loss can measure the degree of visual similarity between generated data and real data,while discriminant loss measures the degree of similarity between the feature representation of generated data and that of real data.According to the above research contents,an example analysis of abnormal intelligent detection of blower operation is completed,and the validity of the method in this paper is verified through the real data collected from the actual industry.