| This research is carried out under the support of the National Key Research and Development Program(No.2016YFF0203301)and the National Natural Science Foundation(No.51779106).Cavitation is the "cancer" of hydraulic machinery,which will not only affect the performance of hydraulic machinery,induce vibration and noise,but also lead to corrosion damage of overflow components.In order to prevent further damage caused by the development of cavitation,the cavitation status of centrifugal pumps should be identified and warned in time.Cavitation condition monitoring can not only prevent centrifugal pumps from running in cavitation condition for a long time and affect the performance of centrifugal pumps,but also help to achieve predictive maintenance.In this paper,centrifugal pump is taken as the research object,and the changing rules of different signals in different stages of cavitation are studied through experiments.A method of cavitation state recognition based on given threshold of single eigenvalue and multi-resolution multi-point information fusion is proposed,and a centrifugal pump cavitation state recognition system is developed.The main contents and achievements of this paper are as follows:1.Introduced the mechanism and harm of cavitation in centrifugal pump.The research status of cavitation state recognition methods at home and abroad is systematically summarized.The advantages and disadvantages of different cavitation monitoring methods are compared.The signal feature extraction and pattern recognition methods suitable for the identification of cavitation state of centrifugal pump are analyzed.The current centrifugal pump cavitation state identification system has been comprehensively summarized.2.Simultaneously collect and analyze the changes of inlet cavitation distribution,external characteristics,pressure pulsation,liquid-borne noise and vibration signals under different cavitation states of centrifugal pump.A cavitationstate recognition method based on pressure pulsation,liquid-borne noise,vibration signal time domain and frequency domain signal statistical eigenvalues is proposed,and the sensitivity of different cavitation methods is compared.The research shows that the high-speed photography method has the highest sensitivity for cavitation judgment;the performance parameter method has the worst sensitivity;the pressure pulsation method has poor anti-interference;the acoustic method and the vibration method can identify the cavitation state earlier,which is more suitable for online monitoring of cavitation.And identification.3.The occurrence of most faults or changes in working conditions of the centrifugal pump will cause changes in the characteristic values such as head,sound level and vibration level.Therefore,for the problem that the cavitation is easy to be misjudged by using a single eigenvalue threshold,based on the liquid-borne noise and vibration signal of the centrifugal pump,a single-point multi-resolution cavitation state identification of centrifugal pump is proposed.method:The wavelet packet decomposition is used to extract the eigenvalues such as the rms value and energy entropy of the multi-scale time-varying moment of the cavitation signal after noise reduction.The feature matrix is reduced by dimension reduction and the input is used as input.Construct an RBF neural network.The results show that the overall recognition rate of cavitation states of non-cavitation,cavitation and severe cavitation is over 96%,and the recognition rate of cavitation nascent state is above70%.4.Aiming at the problem that the recognition rate of cavitation primary state is low based on the single-point signal multi-resolution analysis method,a cavitation state recognition method based on multi-measurement feature level information fusion is proposed:The characteristic values of each node of the single-point signal are arranged into a new feature matrix according to the number of different measuring points and the combination method.After the dimensionality reduction by principal component analysis,a new different RBF neural network is trained torealize the feature-level fusion.The analysis results show that the recognition rate of the cavitation primary state of the centrifugal pump is more than 82% based on the two-point signal characteristic level fusion method,and the recognition rate of the three cavitation states is 100%.5.Due to external harsh environment or changes in incentives and sudden changes in operating conditions,sensor acquisition data often suffers from unpredictable disturbances.In order to improve the anti-jamming capability of the cavitation state recognition method,a cavitation state recognition method based on multi-measurement decision-level information fusion is proposed:Through the DS evidence theory,the decision-level fusion of the single-point signal cavitation state results is carried out.Under the premise of ensuring that the cavitation initial state recognition rate reaches 98% or more when ensuring the integration of the two measurement points and above,the resistance is greatly improved.Interfering.6.The multi-point feature level information fusion method is applied to the single fault state identification of centrifugal pump blade breakage and ring ring wear,and the recognition result is good.The failure of a centrifugal pump often does not occur separately.In order to study the identification of multi-fault conditions,the axial frequency and its multiplier are used as the basis of frequency band division,and a multi-fault concurrency state recognition method based on octave band is proposed.The analysis results show that the single-point multi-fault concurrent state recognition rate reaches over 97%,and the four-point and above signal feature level fusion all fault state recognition rate reaches 100%.7.Based on the above cavitation state identification method,a matching centrifugal pump cavitation state identification system was designed and developed and verified:Construct a TCP/USB/serial communication network,design and manufacture an electric control box,and complete the design of the centrifugal pump cavitation state recognition system hardware;The software display interface of the centrifugal pump cavitation state recognition system was developed by usingLabVIEW.The MatlabScript node is used to write complex algorithms such as wavelet packet and RBF neural network,which realizes system software and hardware communication,centrifugal pump performance/cavitation/high frequency/no load test,time-frequency domain signal processing,online parameter monitoring/alarm/recording of operating parameters,cavitation status recognition,Office report generation and Access database management functions;Using the second chapter test bench and test pump as the research object to test and verify the developed system,The results show that the system measurement accuracy meets the national acceptance criteria.The overall rate of single-point cavitation recognition is over 97.2%.The multi-measurement feature level and decision-level fusion cavitation recognition rate are better than single measurement points. |
