Research On Fault Diagnosis Methods Of Reciprocating Compressor Valve

Reciprocating compressor is one core equipment of the pillar industry of the national economy such as petrochemical industry, natural gas, which possesses the advantages of high exhaust pressure, high thermal efficiency, stable displacement and etc., with extensive application and key function. Especially for petrochemical industry, its running state directly affects the enterprise’s economic benefit. Due to the complexity of reciprocating compressor structure and many vulnerable parts, the safety and reliability in the operation demands high. Especially many key parts’structure is complex, with long-term work under the environment of high temperature and high pressure, and under the effect of cyclic loading, the reciprocating compressor often fails in use process, not only affecting the safe production of the enterprise, also bringing a certain economic losses to country.The fault type of reciprocating compressor is complicated. Compared with other parts, valve is the most vulnerable and fault occurs frequently. According to statistics, at least40%of reciprocating compressor fault occurs on the valve. The stand or fall of air valve working condition, therefore, is one of the core issue of reciprocating compressor’s normal operation. Discover and diagnose the cause of the air valve’s malfunction timely, and take reasonable measures to prevent and control, it is of great significance in the aspect of safety production and economics improving.Because of the complexity of the reciprocating compressor fault diagnosis, the difficulty of its fault diagnosis is increased. It is difficult to use a single parameter to determine fault, so it often requires a variety of means and the combination of common diagnosis. In this paper, the wavelet threshold denoising and empirical mode decomposition (EMD) method are combined, decomposing signal and getting rid of the noise composition and other interference information according to the frequency, at the same time statistical "normalization complexity " is introduced as the fault diagnosis factor of reciprocating compressor exhaust recognition valve, the factor can express the complexity of the signal as a number between0and1, in order to realize the distinction between the different fault signals. In addition, this article also uses the wavelet packet decomposition method to extract the characteristic frequency band of signal energy, and construct the energy ratio parameters, through the analysis of the change of the frequency band energy ratio under different fault conditions, to identify the fault type. Integrated application of normalized complexity and power ratio, in combination with other time-domain characteristic parameters, we proposed a reciprocating machinery fault intelligent diagnosis method based on neural network, and realized the typical fault intelligent diagnosis of reciprocating compressor exhaust valve, and also compared the recognition effect between the two kinds of neural network in valve fault diagnosis application.