This research work aims at developping fault-detection methods easily fit, either in decision assistance tools for real time control of sewer network, or in computer tools for the elaboration of systematic and synthetic balances for estimation the pollution produced from the sewer network. This methodological study then produces a synthesis of fault detection and fault diagnosis technics applied to sewer networks, and develops three model-based fault detection methods. One of these methods consists in building forecasting sewer network's behaviours synthesized in a qualitative model and compared with the measurements. The two others approachs are based on a simplified rainfall-runoff model integrated in a Kalman-filter. The detection relies either on a Kalman-filter bank with a multiple-hypothesis test, or on one Kalman-filter with several sequential probability ratio tests. Sensitivity studies of these detection methods are performed. Moreover several rainy events were analysed with these detection methods. Besides, a dry weather data validation method is developped. It's based on dry weather scenarios compared with hydraulic measurements thanks to hypothesis tests. This approach allows to identify the hydrologic context and the hydraulic configuration of the measurement node.
