发动机磨损故障的集成神经网络融合诊断

针对发动机试车过程中的磨损故障诊断问题．本文运用了四种最常用的润滑油分析技术——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，提出运用集成神经网络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，通过建立合适的权重矩阵．利用模糊综合决策理论，时集成神经网络的诊断结果进行综合，从而得到最终的融合诊断结果；最后，运用一个算例表明了本文方法的有效性。

Fusion Diagnosis for Engine Wear Fault Based on Integrated Neural Network

The fault diagnosis problem of engine wear during test-driving is studied. Four common oil analysis techniques, namely ferrography analysis, spectrometric analysis, particle count analysis, and oil quality testing, are applied, and the test-driving data are combined with them at the same time. The fusion diagnosis method of engine wear fault based on integrated neural network (INN) is put forward. Firstly, according to a standard wear limit, original data are transformed into BOOL value; Secondly each sub-NN structure is established and their training samples are obtained by the expert experience. After each sub-NN is successfully trained, the diagnosis results are obtained by each sub-NN; Thirdly, with a proper weight matrix, using fuzzy integration decision-making theory, the final fusion results are obtained; Finally, a practical example verifies that the method is effective.