Automated Fault Detection and Diagnosis for an Air Handling Unit Based on a GA-Trained RBF Network

The objective of this study is to describe the application of a radius basis function (RBF) network to the problem of automated fault detection and diagnosis (FDD) in the air handling unit (AHU) of a heating ventilation air-conditioning (HVAC) system. First, we analyze the common AHU faults and their dominant symptoms. Next, An FDD strategy for the AHU is proposed which adopts an RBF network to model the causation of symptoms and faults. Gaussian functions are selected as the basis functions of the hidden layer neurons. The parameters of the Gaussian functions and the weights of the network are obtained by using a novel network training method which combines genetic algorithm (GA) and pseudo-inverse matrix algorithm. Finally, an automated FDD program in C language based on the proposed strategy is developed. The FDD program is tested with the HVAC system installed in an artificial environment laboratory and successfully identifies each of the seven faults artificially introduced at the test site