基于流形结构约束的堆栈自编码机模拟电路故障诊断

Analog Circuit Fault Diagnosis Based on Manifold Structure Stacked Autoencoder

针对传统智能故障诊断方法过度依赖特征提取算法和诊断模型泛化能力不足的问题,提出将堆栈自编码机应用于模拟电路故障诊断;通过分析堆栈自编码机在模拟电路故障诊断应用中的不足,提出基于流形结构约束的堆栈自编码机。通过给自编码机的隐层输出加上流形结构特征约束,增强堆栈自编码机的特征提取能力。在3个不同的数据集上,用不同的分类器做故障诊断仿真实验,结果表明,该方法在3个数据集上较之其他分类器都表现出更高的识别率,并且在训练数据量很小时仍然有很好的诊断结果。

Aiming at the problem that traditional intelligent fault diagnosis methods over-depend on feature extraction algorithm and have insufficient generalization,stacked autoencoders are applied to analog circuit fault diagnosis;then,by analyzing the disadvantage when apply stacked autoencoders to analog circuit fault diagnosis,stacked autoencoders based on manifold structure is put forward. By adding manifold structure restriction to the output of hidden layer in autoencoder,the feature extraction ability of stacked autoencoders is enhanced. Do fault diagnosis experiments with different classifiers in three different datasets,the results of which shows that the method proposed gets better accuracy than other classifiers in three different datasets,and even when the training dataset is small.