A model-based strategy for self-correction of sensor faults in variable air volume air handling units

Abstract

This article presents an on-line model-based self-correction strategy intended to handle problems associated with faulty sensors in variable air volume air-handling units. Data-driven gray-box models were used to create “virtual sensors” aimed at substituting for defective supply air temperature and static pressure sensors. Two types of models were used: fixed-parameter models, based on data collected over a long term, and “self-tuning” models, based on data collected on the previous day. The parameters of the self-tuning models were adjusted with a genetic algorithm to reduce the residual between predictions and measurements. The proposed self-correction strategy was tested in a real variable air volume air-conditioning system. Test results show that virtual sensors can be an effective tool to temporarily replace missing sensor data in air-handling units.