基于神经网络规则提取的航空发动机磨损故障诊断知识获取

Knowledge acquisition for aero-engine wear fault diagnosis based on rule extraction from neural networks

针对神经网络智能诊断与专家系统中知识难于理解和诊断解释能力差等问题,研究了一种新的基于功能性观点的神经网络规则提取方法,介绍了方法流程及关键算法.并用UCI(加利福尼亚大学埃尔文分校)机器学习数据对方法进行了分析和验证.最后,将方法应用于实际航空发动机磨损故障诊断中,采集了某型航空发动机实测油样光谱数据237个样本,利用神经网络规则提取方法提取了发动机磨损故障诊断知识规则,并对其进行了解释,结果表明了方法的正确有效性.

In view of the problem that it is difficult to understand the knowledge and diagnosis process in intelligent and expert systems based on neural network, a new rule extraction method from neural network based on the functional point of view was studied, and the flow and the key algorithms of the new method were introduced. The UCI （University of California Irvine）machine learning data were used to analyze and verify the rule extraction method. Finally, this method was applied to aero-engine wear faults diagnosis. 237 spectral oil analysis samples were acquired from practical aero-engine, the rules extraction from NN （Neural netwoks） method was used to extract the diagnosis knowledge rules, the extracted rules were explained and analyzed. The results fully show the correctness and rationality of the new method.