基于深度学习的故障诊断方法研究

Fault Diagnosis Method Based on Deep Learning

针对传统故障诊断方法在处理大数据量、样本结构复杂的工业过程中诊断效果不理想问题,提出一种深度学习与softmax分类器相结合的故障诊断方法。该方法首先采用深度学习方法最大限度地挖掘数据中的隐含特征,充分体现样本的表现力,实现有效的特征提取。然后应用预训练和微调相结合的策略对故障诊断模型训练。最后应用softmax分类器输出故障结果。为了可以提高故障诊断模型的稳定性,简化训练过程,深度学习网络选择栈式编码器深度网络。仿真中将该故障诊断模型与简单softmax分类器诊断模型进行比较分析,结果显示该方法的诊断精度得到了显著提高,能够满足复杂工业过程故障诊断的需求。

A method of fault diagnosis combining depth learning with softmax classifier is proposed to solve the problem that the diagnosis effect is not satisfactory in the process of dealing with large amount of data and complex sample structure. Firstly, this method can maximize the hidden features of mining data by using deep learning pro- cessing which fully reflects the expression of the samples and achieves feature extraction effectively. Then the strategy of combining the pre-training and fine-tuning was used to train fault diagnosis models. Finally, the softmax classifier was used to output the fault results. In order to improve the stability of the fault diagnosis model and simplify the training process, the deep learning network method used stacked encoder depth network. Also, in the simulation, the fault diagnosis model was compared with the simple softmax classifier diagnosis model. The results show that the diag- nostic accuracy of this method has been significantly improved, and it is able to meet the demand of complex industri- al process fault diagnosis.