A novel adaptive controller, suitable for linear and non-linear systems was developed. The controller is a discrete algorithm suitable for computer implementation and is based on gradient descent adaptation rules. Traditional recursive least squares based algorithms suffer from performance deterioration due to the continuous reduction of a covariance matrix used for adaptation. When this covariance matrix becomes too small, recursive least squares algorithms respond slow to changes in model parameters. Gradient descent adaptation was used to avoid the performance deterioration with time associated with regression based adaptation such as Recursive Least Squares methods. Stability was proven with Lyapunov stability theory, using an error filter designed to fulfill stability requirements. Similarities between the proposed controller with PI control have been found. A framework for on-line tuning was developed using the concept of estimation tracks. Estimation tracks allow the estimation gains to be selected from a finite set of possible values, while meeting Lyapunov stability requirements. The trade-off between sufficient excitation for learning and controller performance, typical for dual adaptive control techniques, are met by properly tuning the adaptation and filter gains to drive the rate of adaptation in response to a fixed excitation signal. Two methods for selecting the estimation track were developed. The first method uses simulations to predict the value of the bicriteria cost function that is a combination of prediction and feedback errors, to generate a performance score for each estimation track. The second method uses a linear matrix inequality formulation to find an upper bound on feedback error within the range of uncertainty of the plant parameters and acceptable reference signals. The linear matrix inequality approach was derived from a robust control approach. Numerical simulations were performed to systematically evaluate the performance and computational burden of configuration parameters, such as the number of estimation tracks used for tuning. Comparisons were performed for both tuning methods with an arbitrarily tuned adaptive controller, with arbitrarily selected tuning parameters as well as a common adaptive control algorithm.

A novel adaptive controller, suitable for linear and non-linear systems was developed. The controller is a discrete algorithm suitable for computer implementation and is based on gradient descent adaptation rules. Traditional recursive least squares based algorithms suffer from performance deterioration due to the continuous reduction of a covariance matrix used for adaptation. When this covariance matrix becomes too small, recursive least squares algorithms respond slow to changes in model parameters. Gradient descent adaptation was used to avoid the performance deterioration with time associated with regression based adaptation such as Recursive Least Squares methods. Stability was proven with Lyapunov stability theory, using an error filter designed to fulfill stability requirements. Similarities between the proposed controller with PI control have been found. A framework for on-line tuning was developed using the concept of estimation tracks. Estimation tracks allow the estimation gains to be selected from a finite set of possible values, while meeting Lyapunov stability requirements. The trade-off between sufficient excitation for learning and controller performance, typical for dual adaptive control techniques, are met by properly tuning the adaptation and filter gains to drive the rate of adaptation in response to a fixed excitation signal. Two methods for selecting the estimation track were developed. The first method uses simulations to predict the value of the bicriteria cost function that is a combination of prediction and feedback errors, to generate a performance score for each estimation track. The second method uses a linear matrix inequality formulation to find an upper bound on feedback error within the range of uncertainty of the plant parameters and acceptable reference signals. The linear matrix inequality approach was derived from a robust control approach. Numerical simulations were performed to systematically evaluate the performance and computational burden of configuration parameters, such as the number of estimation tracks used for tuning. Comparisons were performed for both tuning methods with an arbitrarily tuned adaptive controller, with arbitrarily selected tuning parameters as well as a common adaptive control algorithm.

The internet revolution has led to the demand for high speed, low cost solutions for providing high bandwidth to the consumers. Cable and DSL systems address these requirements through sophisticated analog and digital signal processing schemes. A key element of the analog front end of such systems is the line driver which interfaces with the transmission medium such as co-axial cable or twisted pair. The line driver is an amplifier that provides the necessary output current to drive the low impedance of the line. The main requirements for design are high output swing, high linearity, matched impedance to the line and power efficiency. These requirements are addressed by a class AB amplifier whose output impedance can be controlled through feedback. The property of this topology is that when the gain is unity, the output resistance of the driver is matched to the line resistance. Unity gain is achieved for varying line conditions through a tuning loop consisting of peak-to-peak detectors and differential difference amplifier. The design is fabricated in 0.5 micron AMI CMOS process technology. For line variations from 65 to 170 ohms, the gain is unity with an error of 3 % and the impedance matching error is 20 % at the worst-case. The linearity is better than 50 dB for a 1.2 V peak-to-peak signal over the signal bandwidth from 10 kHz to 5 MHz and the line resitance range from 65 to 160 ohms.

Submitted by Rosivalda Pereira (mrs.pereira@ufma.br) on 2017-07-18T19:13:43Z No. of bitstreams: 1 HiltonSantos.pdf: 3137200 bytes, checksum: a7b77b12eeb29959ab49e7ef675229d9 (MD5) / Made available in DSpace on 2017-07-18T19:13:43Z (GMT). No. of bitstreams: 1 HiltonSantos.pdf: 3137200 bytes, checksum: a7b77b12eeb29959ab49e7ef675229d9 (MD5) Previous issue date: 2017-06-27 / The emergence of new industrial plants with great complexity and the need to improve the operation of existing plants has fostered the development of high performance control systems, these systems must not only meet the design specifications, such as merit figures, but also operate at minimal cost and impacts at environment. Motivated by this demand, it is presented in this dissertation the development of methods for on-line tuning of control system parameters, ie, a methodology is presented for the on-line tuning of adaptive and optimal PID controllers via Artificial Neural Networks(ANNs). The approach developed in this dissertation is based on three PID controllers parameters. [Artificial neural networks with radial base functions and Model Predictive Control (MPC). From the union of these approaches a general formulation of an Adaptive-optimal PID controller via artificial neural networks with on-line tuning was presented. The on-line tuning methodology for the ANN parameters is presented in the context of MPC, predicting plant output. For the PID controller, we proposed a modification of the standard structure in order to adapt the error function. The adjustment of the PID controller parameters and the prediction of the optimally plant output, are performed by the ANN-RBF weights adjustments. In addition, an indoor implementation of the control system were proposed for the positioning of a photovoltaic panel. The performance evaluations of the proposed system were obtained from computational experiments results that were based on mathematical models and hardware experiments, that were obtained from a reduced model of a photovoltaic panel. Finally, a comparison between the proposed methodology with the classical PID controller were performed and the proposed methodology presented to be more flexible to the insertion of new performance metrics and the results achieved from the ANN, were better than the ones obtained by the classical PID tuning, such as: Ziegler-Nichols or trial and error. / O surgimento de novas plantas industriais com grande complexidade e a necessidade de melhorar a operação das plantas já existentes tem fomentado o desenvolvimento de sistemas de controle de alto desempenho, estes sistemas devem atender não só as especificações de projeto, tal como: figuras de mérito, mas também devem operar com um custo mínimo e sem causar impactos desastrosos para o meio ambiente. Motivados por esta demanda, apresenta-se nesta dissertação o desenvolvimento de métodos para sintonia online dos parâmetros dos sistemas de controle, ie, apresenta-se uma metodologia para a sintonia online de controladores PID adaptativo e ótimo via Redes Neurais Artificiais (RNAs). A abordagem desenvolvida nesta dissertação tem base as ações dos controladores PID de três termos, redes neurais artificiais com funções de base radial e Controle preditivo baseado em modelo (MPC - Model Predictive Control), a partir da união destas abordagens elabora-se a formulação geral do controlador PID Adaptativo-Ótimo via redes neurais artificiais, com sintonia online. A metodologia de ajuste online dos parâmetros da RNA está no contexto do MPC para predição de saída da planta. Para o caso do controlador PID, tem-se a modificação da estrutura padrão com o objetivo de adaptação em função do erro. O ajuste dos termos do controlador PID e da predição da saída na planta, de forma ótima, é realizada pelo ajustes dos pesos da RNA-RBF. Além disso, apresenta-se a implementação indoor do sistema de controle desenvolvido para o posicionamento de um painel fotovoltaico. As avaliações de desempenho do sistema proposto são obtidos de resultados de experimentos computacionais que são baseados em modelos matemáticos e experimentos em hardware que são obtidos de um modelo reduzido de um painel fotovoltaico. Por fim, comparando o PID clássico com o controlador desenvolvido constatou-se que este último apresenta mais flexibilidade para inserir novas métricas de desempenho e os resultados atingidos são melhores do que os parâmetros obtidos por meio da sintonia do PID clássica, tais como: métodos de Ziegler-Nichols ou tentativa e erro