Describes the use of a two-stage artificial neural network for fault diagnosis in a simulated air-handling unit. The stage one neural network is trained to identify the subsystem in which a fault occurs. The stage two neural network is trained to diagnose the specific cause of a fault at the subsystem level. Regression equations for the supply and mixed-air temperatures are obtained from simulation data and are used to compute parameters to the neural networks. Simulation results are presented that demonstrate that, after a successful diagnosis of a supply air temperature sensor fault, the recovered estimate of the supply air temperature obtained from the regression equation can be used in a feedback control loop to bring air temperature back to the setpoint value. Results are presented that illustrate the evolution of the diagnosis of the two-stage artificial neural network from normal operation to various fault modes of operation.

KEYWORDS: year 1997, Failure, detecting, temperature, sensors, expert systems, air handling units, air temperature, air temperature control, measuring, monitoring, calculating