Abstarct: With the daily increasing demand of industrial large and small customers for production the efficiency and quality product, the kind of supervision controller as fault detection and diagnosis is made. Advanced methods of fault detection are baxsed on mathematical signal and process models, which is obtained by analytically theory and simulations that is done with expert people. In this thesis, we introduce the different fault, fault diagnosis methods and introduce and implementation the sliding mode observer for fault diagnosis respectively. The main idea is implementation the optimization algorithm on the sliding mode observer baxse on neural network in order to fault detection, without linearization. For examine the accuracy and efficiently idea, the comparison is done between the state estimation fault, fault estimation fault, states estimation and fault estimation.