基于证据理论的集成神经网络故障诊断方法

Fault diagnosis method of integration of neural networks based on Dempster-Shafer evidential theory

以Dempster Shafer证据理论为基础,给出了基于神经网络的基本概率分配构造方法和诊断决策规则,提出了一种神经网络初步诊断和证据理论融合决策诊断相结合的集成神经网络故障诊断方法,建立了相应的功能模型。并以变速箱轴承故障诊断为例,详细说明了该方法的具体步骤。结果表明,经过多故障特征信息,诊断结论的可信度明显提高,不确定性明显减小,充分显示了该诊断方法的有效性。

The basic probability assignment method and the rules of diagnosis decision-making are given, and then a new method of fault diagnosis based on integration of neural networks is presented, based on Dempster-Shafer evidential theory. Taking the gearbox bearing fault diagnosis for example, the implementing process of this method is elaborated in details. The results indicate that through information fusion of multi-fault features the reliablilty of diagnosis result is improved evidently, and the uncertainty decreases markedly. The effectiveness of the method is proved fully.