摘要
针对传感器证据信息全局修正针对性不强及冲突证据无法判别等问题,提出基于改进证据理论的故障诊断方法。利用冲突证据判据判断可信证据与冲突证据,保留可信证据,通过可信度对冲突证据进行针对性修正。该判断及局部修正可降低传感器证据信息的不确定性,并减弱冲突信息对合成结果影响。结合神经网络建立特征空间到证据空间的映射,有效利用网络输出结果,通过信息熵构建原始证据。将所有证据用改进D-S公式合成。通过齿轮泵早期故障试验,与神经网络及其它证据合成方法对比表明,该方法诊断精度较高,从而验证了融合方法的有效性。
Aiming at problems that full modification of sensor's evidence information has infirm pertinence and fails to distinguish conflict evidences, a fault diagnosis approach based on improved evidence theory was proposed. Similar evidences and conflict evidences were recognized by means of a conflict evidence criterion. Similar evidences were reserved, while conflict evidences were modified in virtue of a similitude. With this method, the uncertainty of evidence information of sensors was reduced and the effect of conflict evidences on combination results was weakened. Then, combined with a neural network the mapping from a characteristic space to an evidence space was built, and the network output results were used to construct the original evidence with information entropy. Finally, all the evidences were combined with the improved D-S formula. A gear pump's early fault test proved that this new approach has a higher diagnosis precision compared with the neural network method and other combination methods.
出处
《振动与冲击》
EI
CSCD
北大核心
2013年第18期54-58,共5页
Journal of Vibration and Shock
关键词
证据理论
冲突证据
信息融合
故障诊断
evidence theory
conflict evidence
information fusion
fault diagnosis