To address helicopter rotor health monitoring issues and get a kind of the damage sensitivity but disturbance insensitive metrics, the evolving properties of damaged rotor behaviors are investigated in the reconstructed phase space, then a new damage tracking method is developed. First, an aeroelastic model of the rotor system is derived using the finite element method, and the simulated measurements are reconstructed in a higher state space according to the embedding theory. A globally nonlinear reference model to predict the rotor state is formulated using the Volterra series. The difference between the model-estimated state and measured results is used as the state prediction error, the average value of which is evaluated in some disjoint regions of the reconstructed phase space and combined into a damage tracking feature vector. Next, the time series of the damage tracking feature vectors are used directly to extract the dimension fact and trending information about the blade damage by solving an eigenvalue problem. In the case of fault to failure time prediction, the double exponential smoothing method is employed to establish damage trending prognosis models. The feasibility and effectiveness of the proposed method are verified using the data from the blade damage model and the rotor aeroelastic model simulations. The results show that this method can provide fault pattern auto-recognition capabilities and is suitable for tracking the hidden damage in situations in which no pre-knowledge about damage dimension or evolution models is available. The method can also reconstruct the dynamic nature of the underlying system in the phase space using the nonlinear property of the single monitoring signal, which provides a new way to study the system degeneration process in different dimensional spaces with a proper time scale.