Search

According to the statistical results of the reciprocating compressor maintenance in chemical enterprises, the probability of faults caused by wear or damage of vulnerable parts inside the cylinder is close to 80%. Now, fault diagnosis of vulnerable parts inside the cylinder is relied on vibration signal, acoustic emission signal, or thermal parameters. However, it is difficult to extract eigenvalues from vibration signals and acoustic emission signals, which can be disturbed by noise easily. Thermal parameters are relatively stable and less affected by noise. The measurement of thermal parameters requires drilling testing holes through the cylinder wall, which will decrease the strength of the cylinder. According to the working principle of the compressor, fault of the vulnerable parts inside the cylinder would change the pressure and temperature distribution at the suction and the discharge port of the cylinder, while these state parameters are monitored by the parameter monitoring system on most of the compressors. For these reasons, a fault prediction model based on state parameters learning is proposed in this paper aiming to fault diagnosis of vulnerable parts inside the cylinder, which could make fault prediction without adding new hardware cost, such as sensors. Optimized back propagation neural network method by genetic algorithm (GA-BP) is applied to establish the fault prediction model to describe normal working process of the compressor. When fault of vulnerable parts inside the cylinder occurs, monitored pressure and temperature would deviate from the predicted value by the fault prediction. The degree of the deviation is adopted to the fault diagnosis. Then, experiments simulating the fault condition were carried out, and it shows that the accuracy of fault diagnosis using this method could exceed 95%. Verification test shows that the proposed method could successfully predict the fault before the unplanned shutdown.

The fault of vulnerable parts in the cylinder accounts for a large proportion of the fault of the reciprocating compressor. In order to find the fault of the compressor in time, parameter monitoring is usually applied to give forewarning or warning of the compressor fault. Because the setting of forewarning value relies on experiences, over protection and delayed warning events often occur. Recently, the research of fault diagnosis methods combined with artificial intelligence technology is gradually rising. However, most of the fault diagnosis methods based on artificial intelligence technology rely on large number of fault data for fault learning. The problem is that it is almost impossible to obtain the sample data including all of the fault types. To solve above problems, a fault diagnosis method for reciprocating compressor based on the prediction of comprehensive index extracted from the expansion process in indicator diagram is proposed. In this method, a model predicting the comprehensive index of expansion process under normal working conditions is established. The expansion process will change significantly when faults occur, which could result in a deviation between the predicted value and the actual value of the comprehensive index. This deviation is utilized in this paper for fault diagnosis. Fault experiments are carried out to verify the effectiveness of the proposed method. An obvious advantage of this method is that no fault data is needed to establish the fault diagnosis model. It would relatively save the cost of the fault diagnosis and be conducive to application. The accuracy of this method for discriminating fault conditions of an individual vulnerable part is more than 94%. The combined fault conditions with multiple parts under fault state could also be successfully distinguished from normal conditions by this method.