基于粗糙集-BP神经网络的发动机故障诊断

由于发动机光谱分析监控数据中磨损微粒种类过多,如果将这些微粒信息直接作为神经网络的输入,则存在输入层神经元过多、网络结构复杂等诸多问题。本文将粗糙集引入到发动机故障诊断中来,利用粗糙集在属性约简方面的优势,删除冗余磨损微粒,提取出重要磨损微粒,并将其作为BP神经网络的输入,建立发动机故障诊断模型。该方法降低输入层的神经元个数,简化了网络结构,缩短网络训练时间,并且由于剔除了冗余磨损微粒,减少了由该部分微粒信息不准确而带来的误差,有效提高了故障诊断的精确度。最后通过算例分析验证了相关算法和诊断模型的准确性和有效性。

Owing to massive types of wear particles,there exist problems of massive neurons and complex network structure,if all types is used directly as input of fault diagnosis of engine.To solve these problems,rough set is introduced and rough set & BP neural network is presented in this paper,which makes use of advantages in attributes reduction.Rough set can get rid of redundant factors and preserve important factors.The important factors are used as the inputs of BP network.The method can not only decrease number of neurons and simplify network structure,but also condense train time.The inaccurate information of redundant factors is eliminated,and the method also improve the accuracy of fault diagnosis.An example test is used to validate that the method is effective for fault diagnosis of engine with spectrum analysis data.