Research on the Fault Diagnosis Method of Reciprocating Compressors Based on the Probability Diagram Model

Abstract

A reciprocating compressor is one of the key pieces of equipment in the petrochemical production industry. Its structure is complex and there are many sources of excitation causing vibration, making the analysis of the causes of failure challenging. Depending upon fault types, the failures in reciprocating compressors can be classified as leakage, wear, fracture, loosening, impact, blockage, etc.; depending upon where the faults occur, the failures can be classified by components affected such as a valve type, a transmission component type, a sealing component type, etc. This paper proposes a parameter self-learning method for diagnosing mechanical faults. With a probability diagram model for the “fault-sign” relationship on the equipment established according to Bayesian networks, the paper proposes a parameter self-learning method in which, based on the failure rate curve, the prior probability is determined subject to the double-parameter Weibull distribution while the conditional probability can be adjusted automatically according to the diagnosis results. After the probability diagram diagnosis model is built, it collects real-time operating data of the equipment, extracts eigenvalues, sets alarm thresholds, and enters into the diagnosis model the comparison results between the eigenvalues and alarm thresholds as the node states in the sign layer. The changes of the node states in the sign layer will change the posterior probabilities of the corresponding nodes in the fault layer so as to deliver the set of diagnosis results with the highest probability of fault occurrence, thus achieving the purpose of diagnosing the equipment fault.