Inverse model-based virtual sensors for detection of hard faults in air handling units

Abstract

Faults in air handling units (AHUs) in commercial buildings often waste energy and/or cause discomfort. Intermediate AHU sensors are dedicated merely for diagnostics and are essential to isolate faults rather than for controls; However, in many AHUs, such sensors are either missing, misplaced, or uncalibrated. This paper investigates three model-based methods to make up for the lack of reliable intermediate sensor data. To this end, trend data from five AHUs in five different buildings in Ottawa, Canada, are extracted for 2019. The accuracy of three different model forms is compared - artificial neural network (ANN), genetic algorithm (GA), and multiple linear regression (MLR) - are applied to model the supply air temperature of the AHUs. The behaviour of AHU heating and cooling coil valves and outside air dampers with and without the intermediate sensors is studied in this paper. Although installing temperature sensors before and after the heating and cooling coils facilitates detection of the faults occurring in AHUs, the authors showed the generated inverse models can act as virtual temperature sensors to estimate the intermediate measurements and isolate hard faults in AHU outside air dampers in addition to heating/cooling coil valves.