The Detection And Analysis Technology Of Pressure Signal Collected From Reciprocating Compressor Cylinder

Reciprocating compressors are used widely in production. Many stimulating sourcesexist in the compressor, and the structure of reciprocating compressor is very complicated. Iftheir faults arise in the practical production, it would not only affect industrial process ofproduction, but also bring a series of potential problems of safety. Thus, in order to ensure thecompressor unit working safely, an economic and reliable reciprocating compressor faultdiagnosis system is badly needed by factories,the system can be used to forecast, monitor anddiagnose the faults of reciprocating compressors. 2D12 reciprocating compressors arecommon equipment to transport natural gas, and faults occurred frequently in its cylinder. P-Vindicating diagram of reciprocating compressor can reflect the change of pressure incompressor cylinder, and it is a truthful record of working condition in the cylinder. Thispaper would study the detection and analysis technology of pressure signal collected fromreciprocating compressor cylinder. At the beginning of this paper, the fundamental theory and development situation ofmechanical equipment fault diagnosis technology were analyzed synthetically;the mainproblems and development situation of compressor fault diagnosis technology wereintroduced. This paper presented two kind of fault recognition method—Artificial NeuralNetwork and Support Vector Machine on the basis of the character of compressor cylinderfaults, and explained the fundamental theory of them. Secondly, the structure of 2D12reciprocating compressors, simulation experiment of cylinder faults and detection technologyof cylinder pressure signal were introduced, the fault mechanism of principal parts wasanalyzed, and the indicating diagrams of reciprocating compressor were normalized bysegmentation reconfiguration method. Thirdly, this paper explained the design and trainingprocess of BP network and the determination of original weight, the number of network layers,the number of neuron node and target error of system, then a briefly BP network wasconstructed. This paper also introduced the choice of kernel function and optimization ofparameter about Support Vector Machine, and a Support Vector Machine multi-classifier wasconstructed based on them. Indicating diagrams which had been normalized were recognizedby Artificial Neural Network and Support Vector Machine respectively, and a satisfactoryresult was concluded. Finally, cylinder fault diagnosis system software was developed byMATLAB according to the function requirement of fault diagnosis system.