WPT-SVMs based approach for fault detection of valves in reciprocating pumps

This paper presents a novel approach for fault detection of valves in three-cylinder reciprocating pumps. Since the vibration signals collected from pumps apparently show the existence of non-stationary signals and the interference of neighboring valves, the wavelet packet transform (WPT) is introduced as a preprocessing means of extracting time-frequency information from vibration signals to obtain the fault characteristics of the valves. To classify multiple fault modes of valves, a support vector machines (SVMs) based multi-class classifier is constructed and used in the valve faults detection. The results in experiments prove that fault types and positions of faulty valves can be identified and diagnosed by the above method. Furthermore, compared with the results using an artificial neural network, more excellent diagnosis accuracy indicates the potential of the SVMs techniques in machinery fault detection.