基于改进证据理论的齿轮泵故障诊断方法研究

摘要

Aiming at uncertainty of multi-sensor,a new fault diagnosis approach based on improved evidence is presented theory.Firstly,cosine between evidence vectors is defined,and conflict evidence criterion is proposed to determine the conflict.Secondly,tow-class robust fusion strategy is built up.Characteristic level fusion proceeds by way of RBF neural network to generate original evidence,then conflict evidence is found out by criterion and modified by similarity.Finally,the modified evidence is fused in combination formula.The gear pump vibration test proves the validity of the new method and fusion strategy by contrast with neural network,classical evidence theory and other improved methods.%针对多源传感器信息的不确定性,提出了一种基于改进证据理论的故障诊断方法.首先定义了证据向量的夹角余弦,提出了冲突证据判定法则,对证据进行冲突性判定;然后建立二级鲁棒融合策略,通过RBF神经网络进行特征层融合,经过训练产生初始证据,应用冲突证据判定法则找出冲突证据并利用相似度对其进行局部修正;最后对证据进行融合和诊断.通过齿轮泵振动试验,将此方法与神经网络、传统证据理论和其他代表性改进方法的诊断结果进行对比,验证了新方法和融合策略的有效性.