State estimation has been favored in many studies. In many control design methods such as state feedback and optimal regulators state variables are needed but very often they are not available at the output of the plant and should be estimated. Also many Fault Detection and Isolation (FDI) techniques use state variables to determine fault kind and location. There exist varies observers for using in linear systems. In this paper a nonlinear observer for state estimation of a nonlinear system is proposed. It is based on the gradient method and its convergence is proved using lyapunov direct method. The method is implemented for boiler which is a nonlinear MIMO system. Also a method to restrict boiler control efforts in optimal control design is suggested. The proposed method in this paper for restriction control effort signals relives us needing saturation function. This is done simply with using time varying weighing functions in optimal control design.