To improve the accuracy of fault diagnosis for chillers

a fault diagnosis strategy based on an exponentially weighted moving average (EWMA) and Bayesian network (BN) is proposed in this study. The EWMA-BN method used an EWMA control chart to detect faults

and its control limits classified the fault data into the three states of higher

lower

and normal. The conditional probability table was obtained through probability statistics

and the prior probabilities were obtained from expert knowledge. The conditional probabilities were input to BN for fault diagnosis. With respect to number

input order

and completeness of evidence nodes

experimental data were used to analyze the characteristics of the method for fault diagnosis. The results showed that the EWMA-BN method had a significant effect on fault diagnosis for chillers

and the posterior probability values (fault diagnosis results) were all higher than 85%. The results also showed that the increase of evidence nodes could improve the accuracy of fault diagnosis results

but the order of the input nodes had no effect on the final results. The use of uncertain and incomplete information further improved the fault diagnosis capability of the method. The EWMA-BN method was validated using the data provided by the ASHRAE Project

which revealed that this strategy is robust and effective.