Abstract: Artificial neural network(ANN)and Multiple Model Adaptive Filter(MMAF)are combined in a new failure detection method-ANN-based MMAF,which uses ANN's self-learning ability instead of analytical computation of the filter design.Numerical simulations of an areoengine have been performed to detect the sensor failure of the control system.The results show that this method not only can detect the failure of the system with an uncertain model,but also has self-learning ability for the unknown failure pattern.