Fault identification in air handling units using physical models and neural networks

The main purpose of this paper is to develop fault detection modules for BEMS (Building Energy Management Systems), a software to aid building operators in detecting and diagnosing faults in HVAC systems. The fault detection modules proposed are based on two fundamentally different approaches based on component models: physical models and neural net-works. These modules using these mo approaches are illustrated for a cooling coil of an Air Handling Unit. The parameters to be observed, the threshold (the limit over which the fault is considered), the sampling interval and the values of energy overconsumptions are established. The modules are then tested using data bases (no fault, fault) obtained from simulations of an HVAC system.