In this paper, a model-based prognosis method using a particle filter that takes model uncertainty, measurement uncertainty and future loading uncertainty into account is proposed. A nonlinear analytical model of the degradation that depends on loading parameters is established, and then a particle filter is used to estimate and forecast these unknown inputs at the same time as the degradation state. Moreover, adding to this joint input-state estimation, a two-sided CUSUM algorithm is implemented to detect load variations. This would help the prognosis module to adapt to a change in the degradation state evolution, in order to correct the remaining useful life prediction. Real data from fatigue tests on fiber-reinforced metal matrix composite materials are used to demonstrate the efficiency of the proposed methodology for crack growth prognosis.