Monitor And Distinguish The Low Flow Rate Conditions Of Centrifugal Pumps According To The Radiated Acoustic Signal

Centrifugal pumps are widely used in the field of heating,ventilation and air conditioning.Long-term low flow conditions will seriously affect the life of the centrifugal pump and the stability and safety of the system.The existing variable frequency signal and flow rate monitoring can’t be combined with the actual characteristic curve of the pump and the characteristic curve of the pipe network to monitor its working state points,so that it is impossible to judge whether the pump is in a low flow condition.In addition,in theory,there are countless low-flow operating conditions of the pump,and it is difficult to conduct experiments or simulation analysis on every unacceptable operating condition.In response to these problems,this study verified the EMD statistical characteristics of white noise,established an acoustic signal feature extraction method,combined with artificial intelligence algorithms,and proposed a method of using radiated acoustic signals to identify the low-flow operating conditions of centrifugal pumps.This paper extracts and analyzes the characteristics of digital simulation analog signals,computer room and laboratory environmental noise,and centrifugal pump radiated acoustic signals.Using the centrifugal pump radiated acoustic signals in the laboratory acoustic environment,a flow condition discrimination model is established,and the acoustic signals are analyzed.The feature extraction results and the two algorithms are compared and analyzed.By extracting the characteristics of environmental noise,it is proved that the environmental noise of laboratory and computer room is similar to white noise.In addition,the characteristics of the acoustic signal radiated by the centrifugal pump extracted by this method are consistent with the existing research results.In terms of algorithms,when the types of working conditions in the training set samples are limited,the BP neural network and SVM algorithm can still distinguish unknown working conditions,and the accuracy of the discrimination can reach 100%.Among them,the SVM algorithm has obvious advantages in small sample problems.With the reduction of the training set,the average discrimination accuracy rate of the SVM model remained above 98%,while the average discrimination accuracy rate of the BP neural network model dropped sharply to 86%.The equipment condition monitoring method proposed in this paper is universal.When using acoustic signals for equipment monitoring,feature extraction can be performed through the statistical characteristics of white noise EMD,and then the extracted features are used as input vectors to establish an artificial intelligence model to determine the operating conditions of the equipment.