Images based fault diagnosis of air handling unit via combining kernel slow feature analysis and deep learning method

Fault diagnosis of the air handling unit (AHU) is essential for the normal operation of the heating ventilating and air conditioning (HVAC) system. In the AHU's actual running, its various process variables are time-varying, highly nonlinear and coupled, with strong temporal and spatial characteristics . To tackle the AHU's temporal and spatial characteristics and to enhance the fault diagnosis performance, this paper proposes a novel images based deep learning model for its fault diagnosis. Firstly, the proposed method adopts the kernel slow feature analysis method to extract and rank the features according to the slow varying degrees of the process variables. Then, to enhance the neighborhood information and the spatial characteristics of the extracted features, they are transformed into two-dimensional grayscale images , and are augmented by the sliding window technique. Furthermore, the convolutional neural network (CNN) is adopted to deal with the slow feature images and to diagnose the corresponding fault patterns from the data set. Detailed experiments and comparisons are made using the ASHRAE RP-1312 data set. Experimental results verified that the proposed method has high fault diagnosis accuracy under different fault types. Compared with some other popular algorithms, the proposed images based deep diagnosis model has the best diagnosis performance. • Kernel slow feature analysis is adopted to address the temporal dynamics and nonlinearity of air handling unit. • The time series datasets are transformed into images to tackle the spatial characteristic. • The convolutional neural network is employed to generate the images based fault diagnosis model. • Detailed experiments and comparisons are made to verify the superiority of the algorithm.