Chiller fault diagnosis with field sensors using the technology of imbalanced data

Data-driven diagnostic models for refrigeration systems are often used exclusively for a dedicated object. For other types of chillers, a new model must be trained using normal and faulty data, which is both time consuming and heavily data dependent, and accordingly, curbs its application. In this study, the technology for handling imbalanced data is introduced and combined with support vector machine (SVM) to probe the possibility of transferring the FDD (fault detection and diagnosis) knowledge of a centrifugal chiller to a screw chiller by using just a small amount of new data. Principal component analysis (PCA) and the synthetic minority oversampling technique (SMOTE) were used to oversample the faulty sample set with an imbalance ratio of 5%, and a support vector machine (SVM) was employed for fault diagnosis. The experimental results indicate that by using PCA-SMOET-SVM technology, the overall diagnostic performance of the screw chiller with much less data/information is improved with the aid of the prior knowledge transferred from the centrifugal chiller. By investigating the oversampling ratios between 100% and 400%, it was found that the ratio of 100% was the best with the average diagnostic accuracy reaching 96.70% for the faults of the screw chiller.