Research On Pressure Pulsation And Vibration Characteristics Of High-speed Centrifugal Pumps And Online Fault Diagnosis Based On Unsupervised Learning

In the production process of continuous large-scale production of petrochemical and other industrial processes,high-speed centrifugal pumps,as the core equipment in the process,must have good stability and reliability during operation under the requirement of performance.In the actual operation process,the failure of the highspeed centrifugal pump due to the failure to find the problem in time will seriously affect the continuous large-scale production and cause huge economic losses.At present,the fault diagnosis of high-speed centrifugal pumps in the industry is still based on the API610 standard for regular inspections,and a small number of online monitoring are only threshold alarms.In terms of research,although there are many researches on fault diagnosis,they are rarely used in the actual fault diagnosis of high-speed centrifugal pumps.The main reason is the lack of complete fault data samples.In practical engineering,it is difficult to obtain complete and universal fault samples.Therefore,it is difficult to carry out effective application of supervised artificial intelligence methods based on sample classification.Moreover,most of the research is only carried out from the aspects of rotating machinery and signals,and less related to the flow field characteristics of high-speed centrifugal pumps.Therefore,through the research on the characteristics of the internal flow field of the high-speed centrifugal pump,the pressure pulsation,the fault diagnosis model in the absence of fault data samples,etc.,this paper finds a method suitable for the online fault diagnosis and monitoring of the high-speed centrifugal pump,and establishes the fault diagnosis of the high-speed centrifugal pump.The diagnostic platform realizes online monitoring,operation judgment and safety warning of high-speed centrifugal pumps,which has certain value for long-term reliable operation and predictive maintenance of high-speed centrifugal pumps.In this paper,two self-made experimental pumps are used as the research objects,and numerical modeling and simulation are carried out.The external characteristic experiment is carried out on the experimental platform,and the comparison with the numerical simulation is carried out.The effectiveness of the numerical simulation method is obtained by comparison and accuracy.Based on numerical simulation and experimental data,the characteristics of the flow field and pressure pulsation in the high-speed centrifugal pump are analyzed,and the correlation between the internal flow characteristics,pressure pulsation and vibration of the high-speed centrifugal pump under different operating conditions such as rotational speed and flow rate is analyzed.Analyze the characteristics of unstable flow of high-speed centrifugal pump under different flow conditions,as well as the resulting pressure pulsation and its spectral characteristics.Based on these conclusions,the difficulties of fault analysis in the special working conditions of high-speed centrifugal pump are analyzed.In order to solve these problems,a fault diagnosis method based on unsupervised deep learning is introduced,and the characteristics of vibration data of high-speed centrifugal pump are analyzed.A fault diagnosis model of high-speed centrifugal pump with variational autoencoder is introduced.The variational autoencoder extracts its high-dimensional features according to the vibration signal,obtains the statistical distribution model of its high-dimensional features through training,and obtains the diagnostic model according to the training.Based on the obtained model,the research and design of the online monitoring platform for fault diagnosis of high-speed centrifugal pump is carried out,and the application of the fault monitoring platform and engineering pump is realized under the cloud platform.