关键词: 经验模态分解（EMD）, 邻域粗糙集, 驱动端电流, C4.5决策树, 能量熵, 故障诊断, 属性约简

Abstract: Fault diagnosis of the liquid solenoid valve is the effective measure to ensure the normal operation of aircraft power system and achieve rapid fault location. A new method based on Empirical Mode Decomposition （EMD） and neighborhood rough set is proposed for the liquid solenoid valve detection and diagnosis. Firstly, the structure, fault mode, fault mechanism of the liquid solenoid valve are analyzed, then the five conditions of the solenoid valve driving current signal are collected, including normal state, spring failure state, valve spool stuck state, coil anomaly state and electrical short circuit. Afterwards, the current characteristics of different states are analyzed. To resolve the problem of controlling difficulty that exists in the current steady state length, and the incontinuity of energy-entropy of the Intrinsic Mode Function （IMF） which is obtained by the EMD, the current rate of change is chosen as the characteristic to carry out EMD decomposition. Introducing the concept of data mining, a greedy attribute reduction algorithm is constructed using neighborhood rough sets to reduce attribution, then a diagnosis classifier is designed based on C4.5 decision tree by which the samples are trained. As a result, the diagnosis accuracy rate has reached 98%. The results show that this method can realize the fault diagnosis of the solenoid valve and have high application value.

Key words: Empirical Mode Decomposition（EMD）, neighborhood rough set, driving current, C4.5 decision tree, energy-entropy, fault diagnosis, attribute reduction