Based on a open network topology of Bayesian network (BN)

on-site observed information is fused into BN to improve the fault diagnostic performances. A mechanism of distance rejection is introduced to determine the probability distribution of sensor measurement parameters. A chiller fault diagnosis method based on fusional BN is proposed. This method is able to detect new types of chiller fault and update its fault library dynamically. Use the experimental data from ASHRAE RP-1043 to evaluate the performances of the proposed method. The results show that the accuracy of the new type of fault (NF1) is 99.8%

and fusing on-site observed information increases the detection accuracies of the new types of fault (NF2) by 32.6% and the diagnostic accuracies of known fault rl and ro by 4.8% and 11.2% respectively.