A fault reconstruction approach based on the filter is proposed for nonlinear systems with actuator and sensor fault. To make the algorithm applicable to both state and output sides, a new state equation is constructed by the system equation to transform and extend the system, which can convert nonlinear terms and fault of original system output to the state equation of the extended system. Afterwards, a fault diagnosis filter based on the iterative learning algorithm is chosen to update virtual fault to make it approximate to actual fault, and the algorithm can detect and estimate the system faults of different types adaptively. Simulation results of a single-joint robot show the feasibility and effectiveness of the proposed algorithm.