多特征信息融合的贝叶斯网络故障诊断方法研究  

Fault Diagnosis Approach Study of Bayesian Networks Based on Multi-characteristic Information Fusion

作　　者：姜万录[1] 刘思远[1] 

机构地区：[1]燕山大学,秦皇岛066004

出　　处：《中国机械工程》2010年第8期940-945,967,共7页China Mechanical Engineering

基　　金：国家自然科学基金资助项目(50775198);河北省自然科学基金资助项目(E2008000812)

摘　　要：针对轴向柱塞泵故障特征的模糊性和不完备性特点,提出一种多特征信息融合与贝叶斯网络相结合的故障诊断方法。该方法从柱塞泵采集的振动信号中提取出频域和幅域的多个故障特征,并将这些特征当作来自多个不同传感器的多源信息。利用贝叶斯参数估计算法进行多特征信息融合。通过构造贝叶斯网络并建立贝叶斯分类器来简化融合后的结果,通过最大后验概率估计值的计算进行故障识别。经过轴向柱塞泵多故障模式的诊断实验,验证了该方法能够有效地实现柱塞泵柱塞松靴和脱靴故障的诊断。Aiming at the fuzzy and incomplete nature of fault characteristics of axial piston pump,a method of Bayesian networks and multi-characteristic information fusion was proposed.Firstly,multi-fault characteristics in the frequency domain and amplitude domain were extracted from vibration signals and regarded,as multi-source informations coming from different sensors.Then,Bayesian parameter estimation algorithm was applied to fuse multi-characteristic information.Next,the fusion result was simplified by constructing a Bayesian network and establishing Bayesian classifier.Finally,through calculating the maximum posterior probability estimation and the fault patterns were identified.The validity of this method was verified through experiments of multi-fault patterns on an axial piston pump.

关 键 词：柱塞泵 故障诊断 多特征信息融合 贝叶斯网络 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]