Demonstration of a statistical, rule-based fault detection and diagnostic method on a rooftop air conditioning unit

and diagnostic method on a rooftop air conditioning unit. When faults occur in the unit, the measured thermodynamic states differ from the states which are predicted by a model for normal system behavior, generating residuals. The magnitude and statistical uncertainty of the residuals determine the detection sensitivity of the technique. The directional changes in the residuals are statistically compared with a set of rules in order to diagnose a fault. By experimentally introducing faults to the air conditioning unit at five different operating conditions and recording the changes in output states, the detection sensitivity of the technique was quantified and the robustness of the diagnostic rules was verified. Results show that the technique is able to reliably detect refrigerant leakage, condenser fouling, evaporator fouling, liquid line restriction, and compressor valve leakage over a wide range of operating conditions, before a significant decrease in the capacity and COP has occurred. However, the sensitivity of the technique is influenced by the operating conditions of the unit.