FDD method for a variable-speed heat pump with natural refrigerants

Heat pumps are one of the most efficient devices to provide heat and cool. The number of heat pumps sold in Europe increases every year as European Legislation moves towards the use of natural refrigerants that have negligible global warming potential compared to synthetic refrigerants. Variable-speed domestic heat pumps may have hard-to-detect faults that increase energy consumption while the demand is still covered. These faults could worsen and take down the equipment. Fault detection and diagnosis (FDD) systems aim to detect these types of soft faults, reducing operating and maintenance costs. The present study is the result of developing an FDD system for variable-speed heat pumps. The FDD system has been tested with a 10 kW water-to-water variable-speed heat pump charged with propane. Some of the most common faults were emulated for 10 kW and 12 kW heating loads. These faults were evaporator fouling, compressor valve leakage, liquid line restriction and refrigerant overcharge. The present paper presents the overall structure of the developed FDD, each of its different modules and the performance indicators during tests. The FDD developed consists of different modules: a steady-state detector, the input space module, the no-fault regression models and the diagnosis module. The steady-state detector filters the measurements to select only the steady-state data. The input space classifies the data in clusters defined by the heat pump driving variables. For each of these clusters, a regression model is trained. Once trained, the deviation between the models and the real data will indicate a fault occurrence. The diagnosis module analyses the trends of different features to diagnose the fault. The FDD was able to monitor in real time the heat pump performance during the fault tests. The results showed fault detection before 10 minutes with COP drifts above 7 %. Each fault could be diagnosed correctly, except evaporator fouling, which was detected as a fault but could not be distinguished from the others.