Fault detection and diagnosis algorithm for multiple simultaneous faults in residential air-conditioning systems: Development, validation study and critical analysis

• Model generates realistic datasets for single and combined faults in AC systems. • Study bridges simulation-based training with experimental validation for FDD. • SVM algorithm predicts fault-free, single, and combined faults in residential ACs. • Study demonstrates feasibility of simulation-based FDD for low-instrumentation ACs. Detecting and diagnosing faults in residential air conditioning (AC) systems is crucial for preventing critical failures and ensuring efficient operation. Some of these soft faults are refrigerant undercharge (UC), overcharge (OC), inadequate evaporator airflow (EA), or liquid line restrictions (LL). Data-driven methodologies have proven effective in achieving reliable detection but require a large amount of data. This paper proposes utilizing a model based on vapor compression cycles capable of simulating a wide range of indoor/outdoor operating conditions and faults imposed individually and several co-occurring (multiple combinations) for a residential AC unit. This simulated dataset is used to train a Support Vector Machine (SVM) classification algorithm to distinguish between fault, non-fault, and multiple fault states. The SVM uses features obtained from a small number of sensors, making this methodology adequate for residential AC units that typically have limited instrumentation. The proposed method is applied to the same simulated unit, which is experimentally tested in the laboratory with imposed faults, resulting in a balanced accuracy of 82% and a weighted F1-score of 84%. These results demonstrate the feasibility of training a fault detection and diagnosis algorithm with simulation data without requiring an extensive experimental campaign.