Research On Fault Diagnosis Method Of Centrifugal Pump Based On Mixed Domain Feature And CNN

Centrifugal pump is a key rotating machine in petrochemical,metallurgy,electric power and national defense industries.The complex operating environment makes it easy to have all kinds of mechanical faults.To determine the type of faults in time is the most important to ensure the reliable operation and production safety of the equipment.The traditional fault diagnosis based on signal analysis has higher requirements for people.If we want to achieve accurate fault analysis,we need relevant personnel to have comprehensive fault diagnosis expertise,signal processing analysis theory and rich practical field experience.In order to solve the above problems,this paper simulates the typical faults of centrifugal pump.Based on the experimental and field vibration data,an intelligent identification method of centrifugal pump vibration fault is proposed.The following research results have been formed:(1)In order to get rid of the redundant features in the feature set and reduce the influence of the calculation and redundant features on the model training results,the features are selected based on the compensation distance evaluation technology;(2)This paper proposes a fast feature construction algorithm based on spectrum component,which can quickly construct CNN two-dimensional feature input,greatly improve the speed of converting vibration signal to two-dimensional feature,and realize fast and accurate fault type identification;(3)A fault identification method based on compensation distance evaluation and one-dimensional CNN is proposed,which can realize the accurate and fast identification of different equipment,different working conditions and non fault.(4)In view of the difficulty in acquiring fault data and the complexity of environment noise,the anti noise ability,robustness and model training ability under the background of small samples are tested.The model based on compensation distance evaluation and one-dimensional CNN has the best overall performance,which can realize fast and accurate field fault identification.