A practical chiller fault diagnosis method based on discrete Bayesian network

On site application of the fault diagnosis (FD) techniques is beneficial to reduce energy use and to extend life of the equipment. Considering the following aspects, a practical chiller FD method is proposed by introducing discretization to Bayesian network (BN) in this study. Firstly, most real-world domains involve continuous variables which are not easy to handle, and the gaussian hypothesis is not always realistic. Secondly, BN is easier to be dealt with discrete variables, but the traditional discrete FD method based on chiller experts is time-consuming and inefficient. The proposed method makes no assumptions concerning the distribution of the input features, and can quickly determine the parameters of BN without experts, thus it is more efficient and has strong robustness in practical applications of FD. Using the experimental data from ASHRAE RP-1043 to evaluate the proposed method, the results show that the proposed method is very effective for chiller FD. (C) 2019 Elsevier Ltd and IIR. All rights reserved.