Study On Fault Diagnosis Of Multistage Reciprocating Compressor Valve Based On LM-BP Neural Network

This topic "fault diagnosis study of multistage reciprocating compressor valvebased on LM-BP neural network" comes from project item of Jinchang ChemicalIndustry Group Co., Ltd.—development of fault diagnosis system of reciprocatingcompressor air inlet and outlet valve. Currently running6M25reciprocatingcompressor of the enterprises regarded as the research object, it fully researches thescene situation. According to the operation data of the equipments and experience ofthe field engineers, it analyzes the relationship between the changes ofthermodynamic parameters and valve fault.In this paper, mainly through the study of theoretical basis of BP (Back Propagation)algorithm and the analysis of the mechanism of BP neural network algorithm, theperformance of BP neural network in solving the problem in mechanical faultdiagnosis is systematically studied, and a fault diagnosis method for multistagereciprocating compressor valve is put forward which is based on LM(Levenberg-Marquardt) learning-algorithm optimized BP neural network. Six-levelpressure difference and six-level temperature difference of6M25-185/314hydrogennitrogen compressor regarded as the input vector of the network, the influence on thediagnostic results by the structure of three-layer LM-BP neural network, the settingof network parameters and the selection of the training mode is analyzed in detail.Finally, a LM-BP neural network model for online monitoring and fault diagnosisof one-to-six level valve fault of reciprocating compressor is built.100groups offault data regarded as the network training samples,30sets of data regarded as thenetwork test samples for fault diagnosis, fault diagnosis of reciprocating compressorvalve is realized in Matlab environment. The simulation results show that the resultsof fault diagnosis of LM-BP neural network agree well with the measured values, andcompared to the diagnosis of the gradient BP neural network and RBF neural network,it is more rapid and stable. Designed by using MATLAB software platform and basedon LM-BP neural network, the fault diagnosis system of the reciprocating compressor valve shows the diagnosis results by way of data and images, through a simple andconvenient GUI user interface. The model is simple and easy to be applied in practicalengineering.