A novel CNN-TCN-TAM classification model based method for fault diagnosis of chiller sensors

Accurate measurement of sensors plays a crucial role in achieving optimal control and energy management of air-cooled chillers. In order to improve the recognition rate of chiller sensor bias faults, a network model CNN-TCN-TAM which combines convolutional neural network (CNN), time convolutional network (TCN) and time attention mechanism (TAM) is proposed. Firstly, the model uses CNN to extract the data correlation between the physical quantities of the chiller, and then uses TCN to realize the memory of the different time correlations of the chiller sensor, and finally introduces TAM to retain global information and strengthen useful information to further capture relevant features. It is verified by the datasets collected from a real air-cooled chiller platform that the CNN-TCN-TAM based sensor fault diagnosis method has achieved a good fault diagnosis effect. Comparing with different network models, the experimental results show that each improvement step can effectively improve the sensor fault recognition rate.