The work undertaken in this research aims to develop a mathematical model which can replicate the behaviour of a racing driver controlling a vehicle at its handling limit. Most of the models proposed in the literature assume a perfect driver. A formulation taking human limitations into account would serve as a design and simulation tool for the automotive sector. A nonlinear vehicle model with five degrees of freedom under the action of external disturbances controlled by a Linear Quadratic Regulator (LQR) is first proposed to assess the validity of state variances as stability metrics. Comparison to existing stability and controllability criteria indicates that this novel metric can provide meaningful insights into vehicle performance. The LQR however, fails to stabilise the vehicle as tyres saturate. The formulation is extended to improve its robustness. Full nonlinear optimisation with direct transcription is used to derive a controller that can stabilise a vehicle at the handling limit under the action of disturbances. The careful choice of discretisation method and track description allow for reduced computing times. The performance of the controller is assessed using two vehicle configurations, Understeered and Oversteered, in scenarios characterised by increasing levels of non- linearity and geometrical complexity. All tests confirm that vehicles can be stabilised at the handling limit. Parameter studies are also carried out to reveal key aspects of the driving strategy. The driver model is validated against Driver In The Loop simulations for simple and complex manoeuvres. The analysis of experimental data led to the proposal of a novel driving strategy. Driver randomness is modelled as an external disturbance in the driver Neuromuscular System. The statistics of states and controls are found to be in good agreement. The prediction capabilities of the controller can be considered satisfactory.

Este trabalho visa o desenvolvimento de uma metodologia para a re-identificação do modelo usado em controladores preditivos (MPC) desenvolvidos em uma estrutura em duas camadas: uma camada estática que calcula os targets para as variáveis manipuladas e uma dinâmica que implementa os targets para as entradas. Espera-se que esse procedimento de reidenticação seja acionado sempre que for observada uma significativa degradação do modelo de controle do processo. Neste trabalho assume-se que a re-identicação do modelo deve ser realizada em malha fechada. No método aqui proposto, admite-se que o código fonte do programa do controlador preditivo não está disponível, e conseqüentemente, o método proposto não deve requerer qualquer modificação no código fonte. No método aqui proposto, o sinal de excitação é introduzido através dos coeficientes da função objetivo da camada estática que calcula os targets para as entradas. O método proposto é testado por simulação em dois processos diferentes. O primeiro processo é uma coluna de destilação para a qual estão disponíveis vários modelos lineares obtidos em diferentes condições operacionais. O segundo processo aqui estudado é um reator químico não linear que deve ser representado localmente por um modelo linear. / This work aims at the development of a methodology to the re-identification of the model to be used in a MPC, which is developed in a two layers structure: a target calculation layer and a dynamic layer where the targets to the inputs are implemented. It is expected that the reidentification procedure should be started whenever it is observed a significant degradation of the process model. Here, it is assumed that the model re-identification is to be performed in closed-loop. In the method proposed here, it is assumed that the source code of the MPC controller is not available, and consequently, the proposed method should not require any modification the source code. In the method proposed here, the excitation signal is introduced through the coefficients of the objective function of the target calculation layer. The proposed method is tested by simulation in two different processes. The first one is a distillation column where several linear models obtained at different operating conditions are available. The second process studied here is a nonlinear chemical reactor that is locally represented by a linear model.

Este trabalho visa o desenvolvimento de uma metodologia para a re-identificação do modelo usado em controladores preditivos (MPC) desenvolvidos em uma estrutura em duas camadas: uma camada estática que calcula os targets para as variáveis manipuladas e uma dinâmica que implementa os targets para as entradas. Espera-se que esse procedimento de reidenticação seja acionado sempre que for observada uma significativa degradação do modelo de controle do processo. Neste trabalho assume-se que a re-identicação do modelo deve ser realizada em malha fechada. No método aqui proposto, admite-se que o código fonte do programa do controlador preditivo não está disponível, e conseqüentemente, o método proposto não deve requerer qualquer modificação no código fonte. No método aqui proposto, o sinal de excitação é introduzido através dos coeficientes da função objetivo da camada estática que calcula os targets para as entradas. O método proposto é testado por simulação em dois processos diferentes. O primeiro processo é uma coluna de destilação para a qual estão disponíveis vários modelos lineares obtidos em diferentes condições operacionais. O segundo processo aqui estudado é um reator químico não linear que deve ser representado localmente por um modelo linear. / This work aims at the development of a methodology to the re-identification of the model to be used in a MPC, which is developed in a two layers structure: a target calculation layer and a dynamic layer where the targets to the inputs are implemented. It is expected that the reidentification procedure should be started whenever it is observed a significant degradation of the process model. Here, it is assumed that the model re-identification is to be performed in closed-loop. In the method proposed here, it is assumed that the source code of the MPC controller is not available, and consequently, the proposed method should not require any modification the source code. In the method proposed here, the excitation signal is introduced through the coefficients of the objective function of the target calculation layer. The proposed method is tested by simulation in two different processes. The first one is a distillation column where several linear models obtained at different operating conditions are available. The second process studied here is a nonlinear chemical reactor that is locally represented by a linear model.

<p>A container crane of STS-type, Ship To Shore, consists of a spreader hanging underneath a railrunning trolly. As the container is under the influence of wind, it is likely that it starts to turn in a torsional pendulum. This report handles how the torsional pendulum of a container crane can be damped. </p><p>A number of different models have been developed to analyze how different placement of the actuators affects the system. Two differens types of controllers, LQG and MPC, have been developed and applied to these models. The different models and controlers were evaluated and compared by studying simulation results in timedomain. Moreover in order to make the simulations more realistic, a wind model has been developed and applied. </p><p>The models and controllers have been analyzed with bodediagrams and sensitivity functions. </p><p>The analyses shows clearly that the best placement of the actuators for control of the torsional pendulum on an STS-crane is in the trolly, pulling and relaxing the wires. This control is best handled by a state feedback control (LQG). Furthermore, the control should in this way, with addition of in the horizontalplane movable suspensions in the trolly, work acceptably in the whole operational area of a STS-crane.</p>

A container crane of STS-type, Ship To Shore, consists of a spreader hanging underneath a railrunning trolly. As the container is under the influence of wind, it is likely that it starts to turn in a torsional pendulum. This report handles how the torsional pendulum of a container crane can be damped. A number of different models have been developed to analyze how different placement of the actuators affects the system. Two differens types of controllers, LQG and MPC, have been developed and applied to these models. The different models and controlers were evaluated and compared by studying simulation results in timedomain. Moreover in order to make the simulations more realistic, a wind model has been developed and applied. The models and controllers have been analyzed with bodediagrams and sensitivity functions. The analyses shows clearly that the best placement of the actuators for control of the torsional pendulum on an STS-crane is in the trolly, pulling and relaxing the wires. This control is best handled by a state feedback control (LQG). Furthermore, the control should in this way, with addition of in the horizontalplane movable suspensions in the trolly, work acceptably in the whole operational area of a STS-crane.