Reduced-order electro-thermal models for computationally efficient thermal analysis of power electronics modules

Silicon and Silicon Carbide-based power module are common in power electronic systems used in a wide range of applications, including renewable energy, industrial drives and transportation. Reliability of power electronics converters is very important in many applications. It is well known that reliability and ultimately the lifetime of power modules is affected by the running temperature during power cycles. Although accurate thermal models of power electronics assemblies are widely available, based e.g. on computational fluid dynamics (CFD) solvers, their computational complexity hinders the application in real-time temperature monitoring applications. In the thesis, geometry-based numerical thermal models and compact thermal models will be developed to address the fast thermal simulation in the electronic design process and real-time temperature monitoring, respectively. Accurate geometry-based mathematical models for dynamic thermal analyses can be established with the help of finite difference methods (FDM). However, the computational complexity result from the fine mesh and large dimension of ordinary differential equations (ODE) system matrix makes a drawback on the analysis in parametric studies. In this thesis, a novel multi-parameter order reduction technique is proposed, which can significantly improve the simulation efficiency without having a significant impact on the prediction accuracy. Based on the block Arnoldi method, this method is illustrated by referring to the multi-chip power module connected with air-force cooling system including plate-fin heatsink. In real-time temperature monitoring, more compact tools might be preferable, especially if operating and boundary conditions such as losses and cooling are now known accurately, as it’s often the case in practical applications. Compared with geometry-based model which is more suitable in the design of power modules, lumped parameter thermal compact model is simpler and can be applied in real-time temperature prediction during the power cycles of power modules. This thesis proposes a reduced order state space observer to minimize the error caused by air temperature and air flow rate. Additionally, a novel feedback mechanism for disturbance estimation is introduced to compensate the effect result from the error of input power loss, air flow and changes of other nonlinearities