In this work, a model order reduction (MOR) technique for system-level modeling for motor drive dynamic system is presented. As the power electronics continues towards higher switching speed, higher packaging density, and power dissipation, it is critical to have a compact electro-thermal model, easily transferred into a micro-processor, to predict the online temperature dynamics, for self-production, especially at low output-speed conditions when the power chip could potentially operate at over temperature conditions.
The compact model generation process starts with electro-thermal modeling with the commercial finite element package to generate full-order system matrices. The corresponding reduced model is then constructed by retrieving the original full-order system matrices then projecting these matrices into a low-order Krylov subspace. This subspace is spanned by orthogonal basis vectors, obtained by an Arnoldi algorithm. The generated compact model is then incorporated into system-level modeling simulator such as Simplorer or Simulink for fast transient and frequencies simulation. It is demonstrated that the reduced order models are at two orders of magnitude smaller than the full model and still preserve the dynamics characteristics and required accuracy for industrial applications. Due to their compact size, the compact models give significantly speed-up and allow product design engineers to accurately predict temperature response of motor drive at product prototype stage, not otherwise obtainable without this technology.