Many of large-scale systems have embraced a number of highly nonlinear dynamic behaviours and thus pose an operational challenge for fault diagnosis schemes based on a linear perturbation model. In this paper, we proposed a novel Fault detection and identification (FDI) scheme for a class of nonlinear systems. The scheme seeks to design an on-line approximator with an adjustable parameter, which is the so-called Fault tracking approximator (FTA). Motivated by the success of predictive control and iterative learning control theory, a stable iterative algorithm is also exploited to update the adjustable parameter in the FTA. The main progress made in this work is that the proposed approach has enabled a leading-edge feature of detecting and identifying the fault shape and fault magnitude simultaneously. The proposed approach is illustrated to be very robust and effective in numerical simulation experiment of this paper.