A methodology to systematically design a model-based nonlinear model predictive controller is presented. The controller is referred to as hybrid since it uses the first-principles model to calculate the value of the controlled variables along the prediction and control horizons whereas uses the empirical model to ensure a terminal condition that accounts for model errors. The empirical Volterra series model was split into nominal and uncertain parts that were then used to formulate a structured singular value based robustness test. The proposed hybrid controller was compared against a robust empirical that uses solely an empirical model and to a nonrobust first principles model based nonlinear model predictive controller. To show the benefits of considering robustness in the controller formulation, extensive simulation studies were conducted that considered mismatch between the real process parameters and the model parameters. It is shown that in some case the performance of the hybrid controller can be superior to the purely empirical and to the first principles based controllers.