Nonlinear Dynamic System Based on SAE-LDS Model for Fault Diagnosis

Dynamic, uncertainty and nonlinearity are three very important characteristics of complex industrial process data. In this paper, a method of fault detection and classification based on stacked autoencoder-linear dynamic system (SAE-LDS) model is proposed. This method aims to simultaneously consider multiple characteristics of industrial process data for fault detection and classification. Firstly, the deep learning algorithm SAE is used to mine the nonlinear characteristic of process data. On this basis, the SAE model is embedded into the LDS model. The LDS model is used to further extract the dynamics and uncertainty, so as to establish a SAE-LDS model that can comprehensively consider multiple characteristics of data. Finally, the SAE-LDS model is applied to the TE process, compared with the traditional method for fault detection and fault classification. The experimental results show that SAE- LDS model has good effectiveness.