Abstarct: In this thesis we present research on fault detection in dynamic systems particularly observer-baxsed sensor fault detection. We select Kalman filter as the observer which is improved using fuzzy logic. Kalman filter is very effective to state estimation in noisy conditions. This filter is a model-baxsed algorithm, therefore if the parameters of the model is not selected properly results of the Estimation is not valid. The main object in this thesis is to present the effect of incorrect parameter selection on Kalman filter estimations. In addition we find an approach to compensate it. We use filter adaptation by fuzzy logic to overcome parameter uncertainties. In other word, incorrect parameter in an adaptation process converges to their desired values. We give statistic information of one of the filter characteristic as an input to a fuzzy system. Then, the output of such a fuzzy system is used to adapt a parameter of filter. In the end fault detection scheme uses this adapted filter.