Nonlinear oscillations in high power systems

The main topic of this work is to investigate on nonlinear phenomena affecting high power systems and on the strategies adopted to model them. In the first chapter the attention is focused on two big areas of high power systems: power electronics and systems/devices used to sustain plasma fusion. Although it is common that System Engineers tend to associate high power systems with power electronics, it is worth noting that power systems related to nuclear fusion represent a challenging area rich in nonlinearities. Specifically, while nonlinear oscillations in power electronics are due to oscillations of electrical nature, the ones present in nuclear fusion can also refer to other physical quantities. We will refer to the latter taking into account macroscopic plasma instabilities affecting JET plasmas, and proposing both theoretical approaches and experimental ones to describe their dynamic. The former rely on nonlinear mathematical equations able to mimic the nonlinear behavior of the system under certain conditions while the latter are based on a physical realization of the system starting from its mathematical model. High power systems related to power electronics are investigated in Chapter 2 where the importance of thermal modeling for the power electronics modules is pointed out and a new modeling strategy which starts from a distributed parameter analysis to obtain a lumped parameter model is introduced. In this case, the proposed methodology is based on the assumption that the heat transfer problem can be assumed to be linear and the thermal impedances approaches can be therefore used. In this relevant case study nonlinearities in modeling high power systems can also be neglected under certain conditions. In particular, concerning high power modules, it is well-known how the geometry of the device and the proper choice of the cooling system can play a key role for these simplifications. A data-driven approach based on neural networks to model plasma instabilities is presented in Chapter 3. This approach is introduced because physical models often require a deep knowledge of the system parameters that sometimes is difficult to obtain. In Chapter 4 considerations and results on new identification methodologies based on parallel identification models for discrete-time systems are presented.