Agentic fault detection and diagnosis for building HVAC systems using AI agent, ontology, and diagnosis rules

Intelligent detection and diagnosis methods have shown great potential in identifying faults of building energy systems. However, practical applications remain limited due to its reliance on prompt engineering, the developed models, and a lack of interpretability. To address these challenges, this study develops an agentic fault detection and diagnosis (AFDD) framework that integrates the AI agent, ontology, and generalized diagnosis rules. The ontology can describe the knowledge of buildings and control logic. The generalized diagnosis rules can diagnose the fault of the actuator, sensor, load, plant in other control loops, and setpoint. And the AI agent combined with ontology and diagnosis rules generates interpretable diagnosis reports and suggestions. The AFDD framework is validated on a real-world heating, ventilation, and air conditioning (HVAC) system. The result shows that the framework achieves information extraction for building energy system and supply air temperature control loop and realizes ontology extension for the supply chilled water temperature. All faults for the supply air sensor, temperature control valve, cooling capacity, supply chilled water temperature and unreasonable setpoint can be automatically diagnosed without misjudgment using data form one hour, one day, to one week. The diagnosis report including the fault number, breakdown, and detailed analysis and operation and maintenance suggestions are generated. In summary, the AFDD framework enhances automation, interpretability, and decision-making for building energy systems. It provides a new perspective for the next-generation intelligent diagnosis method.