Chemical Equipment Fault Diagnosis via Zero-Shot Learning

Chemical equipment has high reliability requirement and high fault risk, so the real-time fault diagnosis of chemical equipment is of great significance. In fact, the fault sample data of chemical equipment is not easy to collect, and it is common to have no sample. To solve the problem of chemical equipment fault diagnosis under zero sample condition, a fault identification method based on adversarial generation network is proposed. Firstly, an adaptive generative adversarial network with dual discrimination ability is built, which is trained with some known fault data. Secondly, taking the attributes of unknown faults and features extracted from test samples as input, the trained generator is used to generate simulation fault samples. Finally, the KNN algorithm is used to identify the fault of the test samples by comparing with the simulation samples. Experimental results of Tennessee-Eastman process show that the proposed method has better prediction performance than other Zero-Shot learning methods.