Valve fault detection and diagnosis based on CMAC neural networks

This paper presents a method for monitoring and diagnosing the degradation in the performance of heating/cooling coil valves, which might result in serious energy waste, without requiring the valves to be demounted or adding additional sensors. A recurrent cerebellar model articulation controller (RCMAC) is developed to learn the normal characteristics of the valve. When degradation in the performance of the valve occurs, the response of the RCMAC deviates from the normal. Two characteristic variables are defined as the degradation index and the waveform index for analysing the residual errors. A strategy is developed to identify the type of degradation and estimate the severity of the degradation. Tests on a typical valve with five faulty cases in an AHU demonstrate the effectiveness and robustness of the strategy.