Abstract: A fault detection method based on varying-length scanning model is proposed.This method performs minus-operation on the time series data to be detected and then gets the corresponding difference minus sequence.The statistic characters of two neighboring difference minus sequences are affected due to the change point,so the proposed method estimates the value interval at next time through current difference minus sequence,and then realizes fault detection by estimating whether the change point exists at next time.Considering the time complexity of the algorithm,a scanning window size mechanism is designed to bring a balance between the detection speed and precision.The question to select the confidence interval and scanning window size is discussed by a simulation experiment.This fault detection method is applied to integrated GPS/INS(global positioning system / inertial navigation system) navigation system.The experiment shows that it can detect the change point in the data effectively and enhance the precision of filter.