High-Fidelity Reduced Order Modeling Approach for Medium Voltage Drives and Artificial Intelligence Capable Systems

The design of thermal management for medium voltage (MV) drives is an important subject that requires computationally intensive and time consuming simulations. This paper presents an innovative way of leveraging the power of CFD simulations by using Reduced Order Model (ROM) technology. A physics-aware non-intrusive ROM is introduced that leverages the linear and time-invariant properties of the system to predict the thermal behavior of MV drive components under varying flow rates of the cooling medium to generate a linear parameter varying (LPV) ROM. The developed ROM creation technology is tested on a power converter that is part of a MV drive and its results are compared with test measurements. Compared to a full-scale CFD simulation, the ROM approach results in significant reduction in time and computing resources to obtain thermal responses. The availability of such ROMs opens the possibility of drive controllers implementing accurate thermal models in real time thereby allowing further development of AI systems.