基于神经网络和证据理论的液压系统故障诊断

Hydraulic System Fault Diagnosis Based on Neural Network and Evidence Theory

针对液压系统故障多样性和复杂性等特点,基于信息融合原理,提出了一种基于神经网络和D-S(Dempster-Shafer)证据理论相结合的液压系统故障诊断方法。该方法通过构建多子神经网络分类模块进行局部诊断,利用各子神经网络的输出值作为证据理论中的基本可信度,经过证据理论的再次融合得出最终的诊断结果。实例表明,该方法通过简化神经网络结构,提高了局部诊断网络的诊断能力,通过对多源多特征参数的融合,充分利用各传感器的冗余和互补的故障信息,与单一故障特征的诊断相比,显著提高了故障诊断的准确率,降低了决策的不确定性。

According to the diversity and complexity features of hydraulic system fault,a hydraulic system failure diagnosis method combing neural networks and D-S evidence theory was presented by means of information fusion theory.This method conducts local diagnostic by building multi-neural network classification module,using the output of each neural networks as the evidence′s basis belief assignment,then through D-S evidence combination to get the final result.The example verifies that this method simplifies the neural network structure and improves the diagnostic capabilities of diagnostic networks,through combing the multi-source and multi-feature,the accuracy of fault diagnosis is improved significantly and the uncertainty of decision-making is reduced,when comparing with diagnostic based on single fault characteristic by making full use of various redundant and complementary information from multi-sensor.