Artificial intelligence assisted false alarm detection and diagnosis system development for reducing maintenance cost of chillers at the data centre

According to a literature survey, artificial intelligence (AI)-assisted fault detection and diagnosis (FDD) for heat, ventilation, and air conditioning equipment can only reach a correct diagnosis rate of 97%-98.8% Yet, for industrial application, even 1% uncertainty may cause serious problems. Therefore, with AI-assisted FDD (AI-FDD), it is difficult to satisfy high-reliability requirements. This study proposed an innovative AI-assisted false alarm detection and diagnosis (AI-FADD) system. The proposed AI-FADD system combines signal preprocessing, a fault-free classifier, and a fault classifier to meet the requirement of Six Sigma and 3.4 defective parts per million opportunities (DPMO); the developed system achieved 14.48 DPMO, approaching the Six Sigma standard; for 1 million opportunities for alarm, the system only guaranteed 14.48 incorrect false alarm rejections. The developed system can be used for chillers at data centres, which have stringent reliability requirements. Operational data for 14 chillers distributed across north, central, and south Taiwan in 2019 indicated that chiller malfunction alarms were triggered 122 times, but the AI-FADD system determined that only 57 were actual malfunctions; the others were false alarms. According to data verification, the system reached a 100% correct rejection rate, yielding a maintenance savings of up to 260 person-hours. These substantial labour cost savings indicate that commercial application of the AI-FADD system can result in financial benefits.

Automated summary

The study proposed an innovative AI-assisted false alarm detection and diagnosis system, which can effectively reduce false alarms and improve the accuracy of fault detection. The system can meet high-reliability requirements and achieve significant maintenance cost savings.