Published 8 issues per year
ISSN Print: 2150-3621
ISSN Online: 2150-363X
Indexed in
UNIVERSAL STATE SPACE MODEL OF THERMAL TRANSIENT RESPONSE OF POWER ELECTRONICS DEVICES
ABSTRACT
Power electronics devices based on IGBTs are destined to play a critical role in the sustainable energy economy by providing the efficient interface between lower emissions and renewable primary movers such as wind turbines, microturbines, photovoltaic, fuel cells, etc and local power consumers or electrical distribution system such as microgrids. The designers of Thermal Management Systems (TMS) for power electronics are faced with ever increasing demand of improving efficiency, increased reliability and life time as well as reducing the design cycle time, which requires the capability to quickly analyze and optimize enhanced thermal solutions though accurate prediction of three-dimensional transient temperature fields.
A universal training strategy for a state space model is suggested that reproduces temperature responses of power electronic devices with any transient power dissipation profile.
Full CFD and FEA transient models of IGBT modules of various structural elements cooled by natural convection and conduction have been created using the ANSYS IcePak software package. It is extremely time-consuming to use this full model for multiple-mission studies therefore a reduced order model in state space form has been developed to significantly reduce design time. The reduced order model has been trained using a single finite element model run with a universal power dissipation profile. This profile consisted of band limited white noise with a magnitude and frequency limit based on the combination of given devices and set of mission profiles.
It has been shown that the state space model trained with such a profile reproduces the response to a wide variety of different power dissipation patterns typical for our application with an accuracy of about 2°C.
Several different approaches for the choice of the state space model have been studied as well as proper selection of state space size and metrics for estimation of modeling error.
It is suggested that the proposed method can be a universal way of extracting a reduced order model from the high fidelity combination of CFD and FEA models typical for power electronics applications, and possibly for a broader range of applications as well.