This thesis studies the analysis and design of predictor based robust controllers for processes with dead time. the main objectives are: (i) to analyze the effect of the predictor structure in the closed-loop behaviour and robustness of advanced controllers, lineaprocesses, mainly for the petroleum industry.r and nonlinear; (ii) to propose better predictor structures to improve robustness and performance of control loops; (iii) to apply the results in simulated and real industrial the results of this thesis are: an improvement on the well-known dynamic matrix control algorithm, from the model predictive control family, and a predictor for nonlinear systems with time delay based on the smith predictor. concerning the model predictive control, in this work, an improved industrial mpc controller based on the widely used dynamic matrix control (dmc) approach is presented. the proposed controller is especially oriented to control mimo plants with multiple delays that are the most common case in process industry. a mimo filter is included in the prediction model of the controller in order to achieve two important advantages when compared to traditional industrial dmcs: (i) disturbance rejection response can be speeded up and (ii) robustness can be improved, mainly when errors in the estimation of the delays are considered. the filter properties are demonstrated by means of an equivalent analysis of the unconstrained dmc using a dead-time compensation (dtc) approach, namely the filtered smith predictor. moreover implementation and tuning of the filter is simple and intuitive. simulation results using a water-methanol distillation column are presented to illustrate the advantages of the proposed approach. for the case of nonlinear processes with time delay, a nonlinear filtered smith predictor (nlfsp) structure is proposed for nonlinear systems. it will be shown that the nlfsp maintains the characteristics of the linear smith predictor and that, with appropriate tuning, it can increase the robustness of the closed-loop system. the nlfsp is applied to various examples and case studies to demonstrate these characteristics.