Based on a coventional Extended Kalman Filter (EKF), a sub-optimal fading factor EKF is proposed in this paper, which can be used for the joint estimation of states and parameters of nonlinear time-varying stochastic systems. It is used for health monitoring in such a complex system as liquid rocket engine. Numerical simulation result shows the proposed estimator has better properties such as convergence, real time, and dynamic tracking ability etc. . In addition, some problems connected with the joint estimation and the applicability for real plants are also discussed.