Government regulations and growing concerns regarding global warming has lead to an increasing number of passenger vehicles on the roads today that are not powered by the conventional internal combustion (IC) engine. Automotive manufacturers have introduced electric powertrains over the last 10 years which have introduced new challenges regarding powering accessory loads historically reliant on the mechanical energy of the IC engine. High density batteries are used to store the electrical energy required by an electric powertrain and due to their relatively narrow acceptable temperature range, require liquid cooling. The cooling system in place currently utilizes the A/C compressor for cooling and a separate electric element for heating which is energy expensive when the source of energy is electricity. The proposed solution is a thermoelectric heat pump for both heating and cooling. A model predictive controller (MPC) is designed, implemented and tested to optimize the operation of the thermoelectric heat pump. The model predictive controller is chosen due to its ability to accept multiple constrained inputs and outputs as well as optimize the system according to a cost function which may consist of any parameters the designer chooses. The system is highly non-linear and complex therefore both physical modelling and system identi cation were used to derive an accurate model of the system. A steepest descent algorithm was used for optimization of the cost function. The controller was tested in a test bench environment. The results show the thermoelectric heat pump does hold the battery at the speci ed set point however more optimization was expected from the controller. The controller fell short of expectation due to operational restriction enforced during design meant to simplify the problem. The MPC controller is capable of much better performance through adding more detail to the model, an improved optimization algorithm and allowing more flexibility in set point selection.

In this thesis, a cognitive neural predictive controller system has been designed to guide a nonholonomic wheeled mobile robot during continuous and non-continuous trajectory tracking and to navigate through static obstacles with collision-free and minimum tracking error. The structure of the controller consists of two layers; the first layer is a neural network system that controls the mobile robot actuators in order to track a desired path. The second layer of the controller is cognitive layer that collects information from the environment and plans the optimal path. In addition to this, it detects if there is any obstacle in the path so it can be avoided by re-planning the trajectory using particle swarm optimisation (PSO) technique. Two neural networks models are used: the first model is modified Elman recurrent neural network model that describes the kinematic and dynamic model of the mobile robot and it is trained off-line and on-line stages to guarantee that the outputs of the model will accurately represent the actual outputs of the mobile robot system. The trained neural model acts as the position and orientation identifier. The second model is feedforward multi-layer perceptron neural network that describes a feedforward neural controller and it is trained off-line and its weights are adapted on-line to find the reference torques, which controls the steady-state outputs of the mobile robot system. The feedback neural controller is based on the posture neural identifier and quadratic performance index predictive optimisation algorithm for N step-ahead prediction in order to find the optimal torque action in the transient to stabilise the tracking error of the mobile robot system when the trajectory of the robot is drifted from the desired path during transient state. Three controller methodologies were developed: the first is the feedback neural controller; the second is the nonlinear PID neural feedback controller and the third is nonlinear inverse dynamic neural feedback controller, based on the back-stepping method and Lyapunov criterion. The main advantages of the presented approaches are to plan an optimal path for itself avoiding obstructions by using intelligent (PSO) technique as well as the analytically derived control law, which has significantly high computational accuracy with predictive optimisation technique to obtain the optimal torques control action and lead to minimum tracking error of the mobile robot for different types of trajectories. The proposed control algorithm has been applied to monitor a nonholonomic wheeled mobile robot, has demonstrated the capability of tracking different trajectories with continuous gradients (lemniscates and circular) or non-continuous gradients (square) with bounded external disturbances and static obstacles. Simulations results and experimental work showed the effectiveness of the proposed cognitive neural predictive control algorithm; this is demonstrated by the minimised tracking error to less than (1 cm) and obtained smoothness of the torque control signal less than maximum torque (0.236 N.m), especially when external disturbances are applied and navigating through static obstacles. Results show that the five steps-ahead prediction algorithm has better performance compared to one step-ahead for all the control methodologies because of a more complex control structure and taking into account future values of the desired one, not only the current value, as with one step-ahead method. The mean-square error method is used for each component of the state error vector to compare between each of the performance control methodologies in order to give better control results.

O escopo desta dissertação é o desenvolvimento de uma metodologia para identificar os modelos de canais da matriz dinâmica que estejam degradando o desempenho de controladores preditivos, ou MPC (Model Predictive Control), baseado nas técnicas de auditoria e diagnóstico deste tipo de controlador propostas por BOTELHO et al. (2015) e BOTELHO; TRIERWEILER; FARENZENA (2016) e CLARO (2016). A metodologia desenvolvida contempla dois métodos distintos. O primeiro, chamado método direto compensado, tem como base o método direto de identificação em malha fechada (LJUNG, 1987)e compensa cada saída medida do processo de modo a se reter apenas a contribuição do canal que se deseja identificar. O segundo, chamado método do erro nominal, utiliza a definição de saída nominal do processo, proposta por BOTELHO et al. (2015), como métrica para se quantificar o quão próximo o modelo está do comportamento da planta através da minimização do erro nominal. Os métodos foram aplicados ao sistema de quatro tanques cilíndricos (JOHANSSON, 2000) para dois cenários distintos, sendo o primeiro um sistema 2x2 em fase não mínima contendo um MPC trabalhando com setpoint e o segundo um sistema 4x4 em fase mínima com o MPC atuando por faixas. Para o sistema 2x2, se avaliou a influência da localização do canal discrepante (dentro ou fora da diagonal principal da matriz dinâmica de transferência) na eficácia dos métodos. Para o sistema 4x4, o estudo foi voltado para a eficácia dos métodos frentes a controladores que atuam dentro de limites para as variáveis. Os modelos identificados foram comparados pela capacidade de identificar um modelo que capturasse o zero de transmissão da planta e o RGA dinâmico, par ao sistema 2x2, e pelas respostas degrau e diagrama de Bode para o sistema 4x4. O método direto compensado resultou em baixo erro relativo no valor do zero para a discrepância na diagonal principal da matriz dinâmica e alto valor quando a discrepância se encontrava fora da diagonal principal. O método do erro nominal, por sua vez, foi capaz de identificar um modelo cujo zero de transmissão possuía baixo erro relativo frente ao zero da planta em ambos os cenários. No cenário do controlador atuando por faixas, os métodos propostos obtiveram melhores estimativas dos modelos quando comparados com o método concorrente, uma vez que apresentou alto percentual de aderência das saídas simuladas com as saídas medidas. Em todos os cenários estudados, o método do erro nominal se mostrou capaz de identificar um modelo mais robusto, pois este apresentou RGA dinâmico compatível com a planta em todo o range de frequências analisado. / The objective of this dissertation is to develop a method to identify the model for the channel of the dynamic matrix that are affecting the performance of model predictive controllers (MPC), based on the assessment and diagnosis techniques for this type of controller proposed by BOTELHO et al. (2015) e BOTELHO; TRIERWEILER; FARENZENA (2016) and CLARO (2016). The proposed methodology includes two different methods. The first, called the compensated direct method, is based on the closed-loop direct identification method (LJUNG, 1987) and compensates each process measured output in order to retain only the contribution of the channel being identified. The second, called nominal error method, uses the definition of the process nominal output, proposed by BOTELHO et al. (2015), as a metric to quantify how close the model is to the actual plant behavior by minimizing the nominal error. The proposed methods were applied to the quadruple-tank system (JOHANSSON, 2000) for two distinct scenarios, the first being a nonminimum-phase 2x2 system containing a MPC working with setpoint and the second a minimum-phase 4x4 system with the MPC working by ranges. For the 2x2 system, the influence of the model mismatch location (inside or outside the main diagonal of the dynamic transfer matrix) on the effectiveness of the methods was evaluated. For the 4x4 system, the study was focused on the effectiveness of the methods with controllers that operate within limits for the variables. The identified models were compared by the capability of identifying a model with accurate plant transmission zero and dynamic RGA, for the 2x2 system, and by the step responses and Bode diagram for the 4x4 system. The compensated direct method resulted in low relative error in the value of the transmission zero for the model mismatch located in the main diagonal of the dynamic matrix and high relative error when the mismatch was outside the main diagonal. On the other hand, the nominal error method was able to identify a model whose transmission zero had low relative error against the plant zero in both scenarios. In the scenario of a controller working by range, the proposed methods obtained better estimates of the models when compared to the concurrent method, since it presented a high percentage of adherence of the simulated outputs with the measured outputs. In all the studied scenarios, the nominal error method was able to identify a more robust model, since it presented dynamic RGA compatible with the plant in the entire range of analyzed frequencies.

This research deals with the modeling and control of a plasticating twin screw extruder (TSE) that will be used to obtain consistent product quality. The TSE is a widely used process technology for compounding raw polymers. Compounding creates a polymer with improved properties that satisfy the demand of modern plastic applications. Modeling and control of a TSE is challenging because of its high nonlinearity, inherent time delay, and multiple interactive dynamic behavior. A complete methodology is proposed in this thesis to design an advanced control scheme for a TSE. This methodology was used to develop a model predictive control scheme for a laboratory scale plasticating TSE and to implement the control scheme in real-time. The TSE has a processing length of 925 mm and a length to screw diameter ratio (L/D) of 37. High density polyethylenes with different melt indices were used as processing materials. Manipulated variables and disturbance variables were selected based on knowledge of the process. Controlled variables were selected using a selection method that includes a steady state correlation between process output variables and product quality variables, and dynamic considerations. Two process output variables, melt temperature (Tm) at the die and melt pressure (Pm) at the die, were selected as controlled variables. A new modeling approach was proposed to develop grey box models based on excitation in the extruder screw speed (N), one of the manipulated variables. The extruder was excited using a predesigned random binary sequence (RBS) type excitation in N and nonlinear models relating Tm and Pm to N were developed using this approach. System identification techniques were used to obtain model parameters. The obtained models have an autoregressive moving average with exogenous (ARMAX) input structure and the models explain the physics of the extrusion process successfully. The TSE was also excited using a predesigned RBS in the feed rate (F) as a manipulated variable. Models relating Tm and Pm to F were developed using a classical system identification technique; both models have ARMAX structures. The model between Pm and F was found to give excellent prediction for data obtained from a stair type excitation, indicating that the obtained models provide a good representation of the dynamics of the twin screw extruder. Analysis of the TSE open loop process indicated two manipulated variables, N and F, and two controlled variables, Tm and Pm. Thus, a model predictive controller (MPC) was designed using the developed models for this 2X2 system and implemented in real-time. The performance of the MPC was studied by checking its set-point tracking ability. The robustness of the MPC was also examined by imposing external disturbances. Finally, a multimodel operating regime was used to model Tm and N. The operating regime was divided based on the screw speed, N. Local models were developed using system identification techniques. The global model was developed by combining local models using fuzzy logic methodology. Simulated results showed excellent response of Tm for a wide operating range. A similar approach was used to design a global nonlinear proportional-integral controller (n-PI) and a nonlinear MPC (n-MPC). Both the controllers showed good set-points tracking ability over the operating range. The multiple model-based MPC showed smooth transitions from one operating regime to another operating regime. / Process Control

No description available.

O escopo desta dissertação é o desenvolvimento de uma metodologia para identificar os modelos de canais da matriz dinâmica que estejam degradando o desempenho de controladores preditivos, ou MPC (Model Predictive Control), baseado nas técnicas de auditoria e diagnóstico deste tipo de controlador propostas por BOTELHO et al. (2015) e BOTELHO; TRIERWEILER; FARENZENA (2016) e CLARO (2016). A metodologia desenvolvida contempla dois métodos distintos. O primeiro, chamado método direto compensado, tem como base o método direto de identificação em malha fechada (LJUNG, 1987)e compensa cada saída medida do processo de modo a se reter apenas a contribuição do canal que se deseja identificar. O segundo, chamado método do erro nominal, utiliza a definição de saída nominal do processo, proposta por BOTELHO et al. (2015), como métrica para se quantificar o quão próximo o modelo está do comportamento da planta através da minimização do erro nominal. Os métodos foram aplicados ao sistema de quatro tanques cilíndricos (JOHANSSON, 2000) para dois cenários distintos, sendo o primeiro um sistema 2x2 em fase não mínima contendo um MPC trabalhando com setpoint e o segundo um sistema 4x4 em fase mínima com o MPC atuando por faixas. Para o sistema 2x2, se avaliou a influência da localização do canal discrepante (dentro ou fora da diagonal principal da matriz dinâmica de transferência) na eficácia dos métodos. Para o sistema 4x4, o estudo foi voltado para a eficácia dos métodos frentes a controladores que atuam dentro de limites para as variáveis. Os modelos identificados foram comparados pela capacidade de identificar um modelo que capturasse o zero de transmissão da planta e o RGA dinâmico, par ao sistema 2x2, e pelas respostas degrau e diagrama de Bode para o sistema 4x4. O método direto compensado resultou em baixo erro relativo no valor do zero para a discrepância na diagonal principal da matriz dinâmica e alto valor quando a discrepância se encontrava fora da diagonal principal. O método do erro nominal, por sua vez, foi capaz de identificar um modelo cujo zero de transmissão possuía baixo erro relativo frente ao zero da planta em ambos os cenários. No cenário do controlador atuando por faixas, os métodos propostos obtiveram melhores estimativas dos modelos quando comparados com o método concorrente, uma vez que apresentou alto percentual de aderência das saídas simuladas com as saídas medidas. Em todos os cenários estudados, o método do erro nominal se mostrou capaz de identificar um modelo mais robusto, pois este apresentou RGA dinâmico compatível com a planta em todo o range de frequências analisado. / The objective of this dissertation is to develop a method to identify the model for the channel of the dynamic matrix that are affecting the performance of model predictive controllers (MPC), based on the assessment and diagnosis techniques for this type of controller proposed by BOTELHO et al. (2015) e BOTELHO; TRIERWEILER; FARENZENA (2016) and CLARO (2016). The proposed methodology includes two different methods. The first, called the compensated direct method, is based on the closed-loop direct identification method (LJUNG, 1987) and compensates each process measured output in order to retain only the contribution of the channel being identified. The second, called nominal error method, uses the definition of the process nominal output, proposed by BOTELHO et al. (2015), as a metric to quantify how close the model is to the actual plant behavior by minimizing the nominal error. The proposed methods were applied to the quadruple-tank system (JOHANSSON, 2000) for two distinct scenarios, the first being a nonminimum-phase 2x2 system containing a MPC working with setpoint and the second a minimum-phase 4x4 system with the MPC working by ranges. For the 2x2 system, the influence of the model mismatch location (inside or outside the main diagonal of the dynamic transfer matrix) on the effectiveness of the methods was evaluated. For the 4x4 system, the study was focused on the effectiveness of the methods with controllers that operate within limits for the variables. The identified models were compared by the capability of identifying a model with accurate plant transmission zero and dynamic RGA, for the 2x2 system, and by the step responses and Bode diagram for the 4x4 system. The compensated direct method resulted in low relative error in the value of the transmission zero for the model mismatch located in the main diagonal of the dynamic matrix and high relative error when the mismatch was outside the main diagonal. On the other hand, the nominal error method was able to identify a model whose transmission zero had low relative error against the plant zero in both scenarios. In the scenario of a controller working by range, the proposed methods obtained better estimates of the models when compared to the concurrent method, since it presented a high percentage of adherence of the simulated outputs with the measured outputs. In all the studied scenarios, the nominal error method was able to identify a more robust model, since it presented dynamic RGA compatible with the plant in the entire range of analyzed frequencies.

O escopo desta dissertação é o desenvolvimento de uma metodologia para identificar os modelos de canais da matriz dinâmica que estejam degradando o desempenho de controladores preditivos, ou MPC (Model Predictive Control), baseado nas técnicas de auditoria e diagnóstico deste tipo de controlador propostas por BOTELHO et al. (2015) e BOTELHO; TRIERWEILER; FARENZENA (2016) e CLARO (2016). A metodologia desenvolvida contempla dois métodos distintos. O primeiro, chamado método direto compensado, tem como base o método direto de identificação em malha fechada (LJUNG, 1987)e compensa cada saída medida do processo de modo a se reter apenas a contribuição do canal que se deseja identificar. O segundo, chamado método do erro nominal, utiliza a definição de saída nominal do processo, proposta por BOTELHO et al. (2015), como métrica para se quantificar o quão próximo o modelo está do comportamento da planta através da minimização do erro nominal. Os métodos foram aplicados ao sistema de quatro tanques cilíndricos (JOHANSSON, 2000) para dois cenários distintos, sendo o primeiro um sistema 2x2 em fase não mínima contendo um MPC trabalhando com setpoint e o segundo um sistema 4x4 em fase mínima com o MPC atuando por faixas. Para o sistema 2x2, se avaliou a influência da localização do canal discrepante (dentro ou fora da diagonal principal da matriz dinâmica de transferência) na eficácia dos métodos. Para o sistema 4x4, o estudo foi voltado para a eficácia dos métodos frentes a controladores que atuam dentro de limites para as variáveis. Os modelos identificados foram comparados pela capacidade de identificar um modelo que capturasse o zero de transmissão da planta e o RGA dinâmico, par ao sistema 2x2, e pelas respostas degrau e diagrama de Bode para o sistema 4x4. O método direto compensado resultou em baixo erro relativo no valor do zero para a discrepância na diagonal principal da matriz dinâmica e alto valor quando a discrepância se encontrava fora da diagonal principal. O método do erro nominal, por sua vez, foi capaz de identificar um modelo cujo zero de transmissão possuía baixo erro relativo frente ao zero da planta em ambos os cenários. No cenário do controlador atuando por faixas, os métodos propostos obtiveram melhores estimativas dos modelos quando comparados com o método concorrente, uma vez que apresentou alto percentual de aderência das saídas simuladas com as saídas medidas. Em todos os cenários estudados, o método do erro nominal se mostrou capaz de identificar um modelo mais robusto, pois este apresentou RGA dinâmico compatível com a planta em todo o range de frequências analisado. / The objective of this dissertation is to develop a method to identify the model for the channel of the dynamic matrix that are affecting the performance of model predictive controllers (MPC), based on the assessment and diagnosis techniques for this type of controller proposed by BOTELHO et al. (2015) e BOTELHO; TRIERWEILER; FARENZENA (2016) and CLARO (2016). The proposed methodology includes two different methods. The first, called the compensated direct method, is based on the closed-loop direct identification method (LJUNG, 1987) and compensates each process measured output in order to retain only the contribution of the channel being identified. The second, called nominal error method, uses the definition of the process nominal output, proposed by BOTELHO et al. (2015), as a metric to quantify how close the model is to the actual plant behavior by minimizing the nominal error. The proposed methods were applied to the quadruple-tank system (JOHANSSON, 2000) for two distinct scenarios, the first being a nonminimum-phase 2x2 system containing a MPC working with setpoint and the second a minimum-phase 4x4 system with the MPC working by ranges. For the 2x2 system, the influence of the model mismatch location (inside or outside the main diagonal of the dynamic transfer matrix) on the effectiveness of the methods was evaluated. For the 4x4 system, the study was focused on the effectiveness of the methods with controllers that operate within limits for the variables. The identified models were compared by the capability of identifying a model with accurate plant transmission zero and dynamic RGA, for the 2x2 system, and by the step responses and Bode diagram for the 4x4 system. The compensated direct method resulted in low relative error in the value of the transmission zero for the model mismatch located in the main diagonal of the dynamic matrix and high relative error when the mismatch was outside the main diagonal. On the other hand, the nominal error method was able to identify a model whose transmission zero had low relative error against the plant zero in both scenarios. In the scenario of a controller working by range, the proposed methods obtained better estimates of the models when compared to the concurrent method, since it presented a high percentage of adherence of the simulated outputs with the measured outputs. In all the studied scenarios, the nominal error method was able to identify a more robust model, since it presented dynamic RGA compatible with the plant in the entire range of analyzed frequencies.

This thesis considers the problem of control of nonlinear systems subject to limited availability of measurements and uncertainty in model parameters. To address this problem, first a linear offset free MPC is designed. Subsequently, a Lyapunov-based offset free MPC design is presented to handle structured uncertainty subject to constant disturbances. The controller's ability to handle unstructured uncertainty and measurement noise is demonstrated through simulation examples. Next, the problem of handling lack of state measurements as well as uncertainty is considered. To achieve simultaneous state and disturbance parameter estimation, a Lyapunov-based model predictive controller (MPC) is integrated with a moving horizon based mechanism, to achieve (where possible) offset elimination in the unmeasured states as well. A chemical reaction process example is presented to illustrate the key points. Finally its efficacy is demonstrated through a polymerization process example. / Thesis / Doctor of Philosophy (PhD)

A retroalimentação de controladores preditivos que utilizam modelos em espaço de estado pode ser realizada de duas formas: (a) correção por bias, em que as saídas preditas são corrigidas adicionando-se um valor proporcional a discrepância encontrada entre o valor medido atual e sua respectiva predição e por (b) retroalimentação dos estados, onde se determinam as condições iniciais através da estimação dos estados, e a partir de uma melhor condição inicial se realizam as predições futuras usadas no cálculo das ações de controle. Nesta dissertação estas duas abordagens são comparadas utilizando a Planta Laboratorial de Seis Tanques Esféricos. As técnicas de Filtro de Kalman Estendido (EKF) e Filtro de Kalman Estendido com Restrições (CEKF) foram empregadas para estimar os estados não medidos. Inicialmente foram feitos testes off-line destes algoritmos de estimação. Para estes testes são utilizados uma série de dados da planta laboratorial do estudo de caso, na qual são estudadas as influências de diversos fatores de ajuste que determinam a qualidade final de estimação. Estes ajustes serviram de base para a aplicação destes algoritmos em tempo real, quando então, estimadores de estados estão associados ao sistema de controle do processo baseado em um algoritmo de controle preditivo. Após se ter certificado a qualidade das estimações de estado, partiu-se para sua utilização como uma alternativa de retroalimentação de controladores preditivos. Estes resultados foram comparados com os obtidos através da correção simples por bias. Os resultados experimentais apontam para uma marginal piora devido à retroalimentação por estimadores de estados frente à correção por bias, pelo menos para o caso do controlador preditivo linear utilizado na comparação. Entretanto, espera-se que resultados melhores sejam obtidos no caso de modelos preditivos não-lineares, uma vez que nestes casos o modelo é bem mais sensível à qualidade da condição inicial. / The feedback of controllers that use predictive models in state space can be accomplished in two ways: (a) bias correction, where the predicted outputs are corrected by adding a value proportional to the discrepancy found between the current measurement and its respective prediction; and by (b) state feedback, which establishes the initial conditions through the states estimation, and from a better initial condition are carried out the future predictions used in the calculation of control. In this thesis these two approaches are compared using a Laboratorial Plant of Six Spherical Tanks. The techniques of Extended Kalman Filter (EKF) and Constraint Extended Kalman Filter (CEKF) were used to estimate the unmeasured states. Initially, tests were carried out off-line for theses estimation algorithms. For such testing are used a dataset of the plant in case study, in which are studied the influences of several adjustment factors that they determine the final quality of estimation. These adjustments were used of base for the application of these algorithms in real time, when then state estimators are associated with the system of process control based on a predictive control algorithm. After having ascertained the quality of the state estimates, begins its use as an alternative for feedback of predictive controllers. These results were compared with those obtained by the simple correction of bias. The experimental results show a marginal worsening due to feedback from state estimated compared with bias correction, at least for the case of linear predictive controller used in the comparison. However, one expects that better results will be obtained in the case of non-linear predictive models, since in these cases the model is much more sensitive to the quality of the initial condition.

A retroalimentação de controladores preditivos que utilizam modelos em espaço de estado pode ser realizada de duas formas: (a) correção por bias, em que as saídas preditas são corrigidas adicionando-se um valor proporcional a discrepância encontrada entre o valor medido atual e sua respectiva predição e por (b) retroalimentação dos estados, onde se determinam as condições iniciais através da estimação dos estados, e a partir de uma melhor condição inicial se realizam as predições futuras usadas no cálculo das ações de controle. Nesta dissertação estas duas abordagens são comparadas utilizando a Planta Laboratorial de Seis Tanques Esféricos. As técnicas de Filtro de Kalman Estendido (EKF) e Filtro de Kalman Estendido com Restrições (CEKF) foram empregadas para estimar os estados não medidos. Inicialmente foram feitos testes off-line destes algoritmos de estimação. Para estes testes são utilizados uma série de dados da planta laboratorial do estudo de caso, na qual são estudadas as influências de diversos fatores de ajuste que determinam a qualidade final de estimação. Estes ajustes serviram de base para a aplicação destes algoritmos em tempo real, quando então, estimadores de estados estão associados ao sistema de controle do processo baseado em um algoritmo de controle preditivo. Após se ter certificado a qualidade das estimações de estado, partiu-se para sua utilização como uma alternativa de retroalimentação de controladores preditivos. Estes resultados foram comparados com os obtidos através da correção simples por bias. Os resultados experimentais apontam para uma marginal piora devido à retroalimentação por estimadores de estados frente à correção por bias, pelo menos para o caso do controlador preditivo linear utilizado na comparação. Entretanto, espera-se que resultados melhores sejam obtidos no caso de modelos preditivos não-lineares, uma vez que nestes casos o modelo é bem mais sensível à qualidade da condição inicial. / The feedback of controllers that use predictive models in state space can be accomplished in two ways: (a) bias correction, where the predicted outputs are corrected by adding a value proportional to the discrepancy found between the current measurement and its respective prediction; and by (b) state feedback, which establishes the initial conditions through the states estimation, and from a better initial condition are carried out the future predictions used in the calculation of control. In this thesis these two approaches are compared using a Laboratorial Plant of Six Spherical Tanks. The techniques of Extended Kalman Filter (EKF) and Constraint Extended Kalman Filter (CEKF) were used to estimate the unmeasured states. Initially, tests were carried out off-line for theses estimation algorithms. For such testing are used a dataset of the plant in case study, in which are studied the influences of several adjustment factors that they determine the final quality of estimation. These adjustments were used of base for the application of these algorithms in real time, when then state estimators are associated with the system of process control based on a predictive control algorithm. After having ascertained the quality of the state estimates, begins its use as an alternative for feedback of predictive controllers. These results were compared with those obtained by the simple correction of bias. The experimental results show a marginal worsening due to feedback from state estimated compared with bias correction, at least for the case of linear predictive controller used in the comparison. However, one expects that better results will be obtained in the case of non-linear predictive models, since in these cases the model is much more sensitive to the quality of the initial condition.