With systems engineering and artificial intelligent methods, an earlywarning system of dam health (EWSDH) is developed. This system consists of integration control module, intelligent inference engine (IIE), support base cluster, information management and input/output modules. As a central processing unit of EWSDH, IIE is a decision support system for monitoring the operation characteristics and diagnosing unexpected behaviour of dam health. With the time-frequency domain localization properties and self-learning ability of wavelet networks based on wavelet frames, IIE builds some new monitoring models of dam health. The models are used to approximate and forecast the operation characteristics of dam. The methods of attributions reduction in rough sets theory are presented to diagnose adaptively the unexpected behaviour. The proposed system has been used to monitor dam health successfully.