Reducing false alarms by two-stage fault detection using moving average filter and time-ratio deadbands

Principal component analysis (PCA) techniques are seldom applied to fault symptom detection in real-world air conditioning systems, primarily due to high rates of false alarms and missed detections. Although the feeding data are typically pre-processed with noise reduction procedures, these measures do not seem to sufficiently alleviate the issues. This study proposes a novel two-stage strategy to balance both false and missed alarm rates, thereby enhancing the practicality of PCA-based methods in air conditioning systems. Stage 1 employs a moving average filter to adapt the lagged responses to time-varying dynamic loads. Stage 2 utilises time-ratio deadbands to suppress false alarms triggered by transient fluctuations. Compared to the benchmark baseline, Stage 1 demonstrates similar effectiveness in reducing false alarms and maintaining true alarms. For Stage 2, the false alarm rate decreases to 4.23% and lower when the time-ratio threshold exceeds 40%. The recommended time-ratio thresholds are between 40% and 70%.