Study on Chiller Fault Detection and Diagnosis Method Based on KNN Algorithm and ANOVA

As the economy, population, and industry have grown in recent years, more and more water chiller systems have been installed in many buildings throughout the world. However, faults can appear during operation, leading to a reduction in the life of a system and increased energy consumption. As a result, it is necessary to identify and overcome these faults. This paper proposes a chiller fault detection and diagnosis (FDD) method based on the K-nearest neighbors (KNN) algorithm and an analysis of variance (ANOVA) to reduce the number of sensors in a real system and to improve the performance of chiller FDD. A Python program based on the KNN and ANOVA models was developed to simulate and validate the chiller fault detection and diagnosis. The results showed that the correct rates (CRs) of stages 1 and 2 in Case 1 were 99.53% and 99.60%, respectively, whereas the CRs of stages 1 and 2 in Case 2 were 99.08% and 99.48%, respectively. The highest performance of the proposed chiller FDD method was achieved when compared to the CBA method, the EBD-DBN method, and the GDW-SVDD method for Case 2 with slight-severity levels 1 and 2. Furthermore, this method was validated using real data under normal operating conditions and the condenser fouling fault of a centrifugal water-cooled chiller from the Saigon Center building in Vietnam. The results showed that the overall performance of chiller FDD was 97.61%, and the hit rate of the condenser fouling fault was 93.46%. This has high reliability and can be used in industry.