Neste trabalho é estudada a convergência no método dos elementos finitos (MEF) quando da sua aplicação a dispositivos eletromagnéticos com características não lineares, ou seja, constituídos de materiais ferromagnéticos com curva de magnetização BxH que apresenta o fenômeno da saturação magnética. O método de Newton-Raphson é o método numérico utilizado para a resolução do sistema de equações não linear gerado. São abordadas versões desse método que utilizam fator de relaxação, bem como técnicas para a otimização desse fator, com o intuito de acelerar a sua convergência, mostrando sua efetividade. Aborda-se também a influência da escolha do modelo para a característica B(H) dos materiais ferromagnéticos, tanto em termos de desempenho computacional quanto de precisão. Duas classes de modelos para a curva vxB², bem como para sua derivada dv/dB², são analisados, quais sejam: modelos analíticos e modelos interpoladores. Dentre os modelos analíticos destacam-se os modelos de O\'Connor, Marrocco e de Brauer (exponencial). Os modelos interpoladores utilizados foram de Akima e Spline cúbica. Além do tratamento de materiais com características não lineares, também se realizou um breve estudo da aplicação de materiais do tipo ímã permanente e sua formulação pelo MEF. Outro aspecto avaliado foi a influência de diferentes métodos de resolução do sistema linear de equações algébricas na convergência do método de Newton-Raphson, como o ICCG, BiCGStab, decomposição LU e Wavelet-based Multigrid. Os resultados obtidos mostraram a potencialidade desta última técnica em termos de ganho de desempenho global da solução. / This work presents a study on the convergence of non-linear finite analysis applied to the modeling of electromagnetic devices with plane symmetry. Three aspects have been thoroughly investigated which have influence on the convergence performance, namely: the relaxation factor in the Newton-Raphson algorithm, the model of the magnetization curve and the choice of the linear solver. It is shown that convergence acceleration of the Newton-Raphson algorithm can be achieved with the use of a relaxation factor, along with numerical techniques for optimizing this factor, such as Line Search or the Brent method. Several models for the magnetization curve, both analytical (O\'Connor, Marrocco, Brauer) and interpolation-based (Akima and Cubic Splines), have been analised and comparisons in terms of their computational efficiency, as well as accuracy, are presented in order to give some guidance to the choice of the model. The influence of several linear solvers in the convergence of the Newton-Raphson algorithm has finally been investigated. The following linear solvers have been compared: LU decomposition , ICCG, BiCGStab and Wavelet-based Multigrid. The results show the superiority of the latter to improve the overall computational performance of the solution.

Neste trabalho é estudada a convergência no método dos elementos finitos (MEF) quando da sua aplicação a dispositivos eletromagnéticos com características não lineares, ou seja, constituídos de materiais ferromagnéticos com curva de magnetização BxH que apresenta o fenômeno da saturação magnética. O método de Newton-Raphson é o método numérico utilizado para a resolução do sistema de equações não linear gerado. São abordadas versões desse método que utilizam fator de relaxação, bem como técnicas para a otimização desse fator, com o intuito de acelerar a sua convergência, mostrando sua efetividade. Aborda-se também a influência da escolha do modelo para a característica B(H) dos materiais ferromagnéticos, tanto em termos de desempenho computacional quanto de precisão. Duas classes de modelos para a curva vxB², bem como para sua derivada dv/dB², são analisados, quais sejam: modelos analíticos e modelos interpoladores. Dentre os modelos analíticos destacam-se os modelos de O\'Connor, Marrocco e de Brauer (exponencial). Os modelos interpoladores utilizados foram de Akima e Spline cúbica. Além do tratamento de materiais com características não lineares, também se realizou um breve estudo da aplicação de materiais do tipo ímã permanente e sua formulação pelo MEF. Outro aspecto avaliado foi a influência de diferentes métodos de resolução do sistema linear de equações algébricas na convergência do método de Newton-Raphson, como o ICCG, BiCGStab, decomposição LU e Wavelet-based Multigrid. Os resultados obtidos mostraram a potencialidade desta última técnica em termos de ganho de desempenho global da solução. / This work presents a study on the convergence of non-linear finite analysis applied to the modeling of electromagnetic devices with plane symmetry. Three aspects have been thoroughly investigated which have influence on the convergence performance, namely: the relaxation factor in the Newton-Raphson algorithm, the model of the magnetization curve and the choice of the linear solver. It is shown that convergence acceleration of the Newton-Raphson algorithm can be achieved with the use of a relaxation factor, along with numerical techniques for optimizing this factor, such as Line Search or the Brent method. Several models for the magnetization curve, both analytical (O\'Connor, Marrocco, Brauer) and interpolation-based (Akima and Cubic Splines), have been analised and comparisons in terms of their computational efficiency, as well as accuracy, are presented in order to give some guidance to the choice of the model. The influence of several linear solvers in the convergence of the Newton-Raphson algorithm has finally been investigated. The following linear solvers have been compared: LU decomposition , ICCG, BiCGStab and Wavelet-based Multigrid. The results show the superiority of the latter to improve the overall computational performance of the solution.

The dissertation studies the nonlinear resonance problems of power systems. Generally speaking, ferroresonance has usually occurred in low-voltage distribution system, especially for potential transformers. Due to the considerable increase of power consumption, the power system is more complex than before. Besides, a number of under-ground cables are used, and transformers¡¦ loss reduce due to improvement of core iron material. These factors could probably result in ferroresnance occurring in extra-high-voltage power system. The dissertation proposed three-phase representation method to analyze unbalance and non-linear system. This method employ magnetically coupled electrical circuit techniques and the original voltage equations can be used without the need for any transformations, which improves significantly computation accuracy. Consequently, it is quite suitable for power system design and incident investigation. Since the traditional d-q-0 model is not well suited for the study of unbalanced faults and requires further transformation, the analytical solution becomes rather complicated and the solutions are still inaccurate. At last we simulated the ferroresonant overvoltages occurring at a nuclear power station in Taiwan in order to investigate the causes and afford mitigation. The simulation results were enough to prove accuracy and practicability of this method.

A pulse hysteresisgraph system was constructed capable outputting current source and voltages source waveforms. MATLAB scripts were created to analyze the collected data. Three toroidal samples of soft magnetic materials were prepared. Theoretical modeling was done to predict the variation of effective applied magnetic fields inside the toroids from ideal assumptions due to three effects: wire spacing, cylindrical spreading, and eddy current generated fields. Data was collected under sinusoidal voltage source and sinusoidal current source excitation at 1 kHz. Large differences in core loss were noted especially at higher field excitations. Core loss under sinusoidal current source excitation was found to always be greater than or equal to core loss under sinusoidal voltage source. Normal magnetization curves under sinusoidal current and voltage source excitation were also compared. Significant differences were apparent in the magnetization curves of one sample toroid, and slight differences noted in the curves of the other two samples. Eddy currents were offered as a primary mechanism for the difference in core loss between sinusoidal current source and sinusoidal voltage source. A formula to predict the relative eddy current losses to be expected from an arbitrary, periodic voltage waveform shape is given. / text