Integration of dynamic model and classification methods for fault detection and diagnosis in a chiller

Fault detection and diagnosis techniques can contribute to the reduction of operational costs, downtime periods and increase of component lifetime in vapour compression systems. In this study, a dynamic simulation model was developed to represent different operational points of an experimental chiller, namely fault-free operation, condenser fouling, refrigerant leakage and reduced condenser water flow rate. Simulated operational data allowed the use of four classification methods, namely logistic regression, naïve Bayes, decision tree and random forest. The results showed that the proposed framework was able to predict 94 % of faulty and fault-free operational points of the chiller when logistic regression was used. Fault prevention from the implementation of this framework was estimated to increase the average COP of the chiller by nearly 4 %. This study indicated the possibility to induce faults in dynamic simulation models combined with classification algorithms for fault detection and diagnosis in vapour compression systems.