基于神经网络和D-S证据理论的发动机磨损故障融合诊断

Fusion Diagnosis of Engine Wearing Fault Based on Neural Networks and D-S Evidence Theory

运用了4种最常用的滑油分析技术——铁谱分析、光谱分析、颗粒计数分析及理化指标分析,同时结合发动机试车台监测数据,提出运用神经网络和D-S证据理论对发动机试车状态进行融合诊断的方法。首先依据各种分析方法的标准磨损界限值,将原始数据进行了预处理,统一转换成故障征兆的布尔值;其次,建立各子神经网络的拓扑结构,并依据专家经验建立各子系统的输入征兆与故障论域的映射关系,从而得到各子神经网络的训练样本,对各网络进行成功训练后,利用神经网络实现各子网络的诊断并得到中间诊断结果;然后,将每种方法的神经网络诊断结果作为对各种故障模式的基本概率分配值,利用D-S证据理论,实现对神经网络的诊断结果的融合,从而得到最终的融合诊断结果;最后,运用算例表明了本文方法的有效性。

Four common oil analysis techniques, namely Ferrography analysis, Spectrometric analysis, Particle count analysis, and Oil chemical-physics analysis, were used together with the engine test data to develop the fusion diagnosis method of engine wearing fault based on Neural Networks (NN) and D-S evidence theory. Firstly, according to standard wear limit, original data were transformed into BOOL value. Then, each sub-NN structure was established, and their training samples were obtained based on expert experience. After each sub-NN was trained successfully, the intermediate diagnosis results were obtained through each sub-NN. Finally, the NN diagnosis results are used as the basic probability distribution value to each fault mode, and the D-S evidence theory is applied, and the final fusion diagnosis results are obtained. An example was used to verify the method presented in this paper.