Feature selection based on Bayesian network for chiller fault diagnosis from the perspective of field applications

Abstract From the perspective of field applications, a feature selection (FS) method is proposed in this study for chiller fault diagnosis (FD). First, the candidate existing features that can be retained (CRE-features) are nominated through the following criteria: high existent frequency of sensors installed on the field chillers, high sensitivity to faults and small amount of calculation. Then these features are evaluated by using an FD method based on a Bayesian network merged distance rejection (DR-BN) technique to remove redundant features. Second, when the expected performance cannot be obtained by only using these specifically retained existing features (RE-features), additional features need be included. Supplemental features (S-features) are nominated through the following criteria: low cost of measurement and high sensitivity to faults. Then these S-features are evaluated together with RE-features by using the same method to determine the specific S-features. The experimental data from ASHRAE RP-1043 are used to validate the FS method. Results show that the proposed FS method is effective for chiller FD, and can make the matched FD method perform well by selecting the commonly available features in the field and by supplementing a few features with low cost of measurement to indicate faults.