基于神经网络和证据融合理论的航空发动机气路故障诊断

Neural Network and Dempster-Shafter Theory Based Fault Diagnosis for Aeroengine Gas Path

为解决航空发动机这一复杂系统的故障诊断问题,提高智能化诊断方法的准确率,使用了改良的D-S证据理论,对基于自组织竞争网络和BP神经网络的2个诊断子系统的诊断结果进行决策级融合。结果显示,经过融合整个系统降低了误诊率,改善了诊断性能。文章还针对噪声干扰的情况,通过调整对于2个子系统的权重,在保证高准确率的同时提高了系统的抗噪声性能。研究结果表明D-S证据理论的使用可以达到比单独运用2个神经网络子系统都要好的诊断效能。

Aeroengine is a very complex nonlinear system, which brings a big challenge for its fault diagnosis. In the past decade, intelligent techniques have been utilized, such as self-organizing competitive neural network and BP neural network, to solve the problem. These two techniques have their own advantages and disadvantages. It seems that the best diagnosis result cannot be got with only one of the two techniques. In this paper, Dempster-Shafter theory is used to build a new diagnosis system by blending and tuning the output of self-organizing competitive neural network based diagnosis sub-system and that of BP neural network based sub-system. Test results show that the system can diagnose and detect the faults of aeroengine gas path with more precision and efficiency than either single sub-system. In addition, this new D-S theory and two neural networks based diagnosis system can be used to obtain better ability of rejecting noise than the two sub-systems, with adjusting the weights of the diagnosis decisions of two sub-systems.