关键词: 往复泵液阀, 格兰杰因果关系, 故障预警, 振动信号

Abstract: To tackle the challenges of low accuracy and poor real-time performance in early warning of liquid valve faults in reciprocating pumps, a novel method based on Granger causality analysis was proposed. Initially,vibration signals collected under normal operating conditions were processed in the angle domain for feature extraction,and a feature matrix was constructed accordingly. This matrix was partitioned into reference data and training data. Granger causality tests were applied to the reference data to derive a baseline causality matrix. Subsequently,a sliding window with an appropriate size and step was applied to the training data to perform Granger causality analysis,resulting in a series of causality matrices. These matrices were compared with the baseline to calculate differences,and the maximum difference was selected as the fault detection threshold. During real-time monitoring,a causality matrix was continuously computed and compared against the baseline; if the computed difference exceeds the threshold,an early warning for a potential liquid valve fault was implemented. Finally,comparative experiments were conducted on a reciprocating pump test platform using the proposed method,as well as LSTM,MSET,and DBN models. Experimental results demonstrate that the proposed method improves early warning accuracy by 7.34 %,3.67 %,and 9.34 % compared to LSTM,MSET,and DBN,respectively,while achieving reductions in response time of 18.27 s,2.86 s,10.38 s.

Key words: reciprocating pump liquid valve, Granger causality, fault warning, vibration signal