Least squares support vector machine (LS-SVM)-based chiller fault diagnosis using fault indicative features

Abstract

Fault detection and diagnosis (FDD) of chillers, generally the single most energy consuming piece of building equipment, is an important but hard task where many parameters are involved, and the problem is always quite non-linear. This study proposes a least squares support vector machine (LS-SVM) model optimized by cross validation to implement FDD on a 90-ton centrifugal chiller. Four component-level and three system-level faults were investigated. The effectiveness and efficiency of the eight fault-indicative features extracted from the original 64 parameters have been validated and employed in the detailed discussions. The results indicated that as compared with the other two machine learning methods, the proposed LS-SVM model with optimization showed a better FDD performance in terms of the overall correct rate for all the samples, the individual correct rate for each fault, the diagnostic efficiency, the detection rate and the false alarm rate, etc., especially when it comes to the faults of system level where the detection and diagnosis become rather more difficult due to the system-level effect (SLE)—widespread symptoms caused throughout the system. The correct rates for system-level faults of refrigerant leak/undercharge, refrigerant overcharge and excessive oil were as high as 99.59%, 99.26% and 99.38%, respectively, and the running time was only 36.7% of that of the SVM model.