Efficient Physics-Based System Level Thermal Management for Electric Drive Units using Reduced Order Modeling Techniques Assisted by Neural Networks

Abstract

<div class="section abstract"><div class="htmlview paragraph">Efficient thermal management is essential in high power density electric drive units (EDUs) due to limited space and working environment. Major heat sources in EDUs are from the inverter, motor and gearbox. System level thermal response prediction models comprising various components within the EDU are of interest from both product performance and software controls standpoint.</div><div class="htmlview paragraph">A system level physics-based lumped parameter thermal network (LPTN) model is built in a one-dimensional (1D) framework using inputs from empirical, electromagnetic, three-dimensional conjugate fluid/heat transfer analysis and test data to predict the component temperature within the EDU. Empirical models were used to calculate heating due to efficiency loss from the gearbox. The thermal loses from the motor are estimated as outputs from electromagnetic simulations. Three-dimensional computational fluid dynamics (CFD) conjugate heat transfer (CHT) simulations were also used at both system and component level to determine heat transfer within the gearbox, motor and inverter. Later due to memory constraints of the microcontroller unit (MCU) the LPTN model is further reduced using a novel temperature estimation based deep neural network to predict component temperatures of interest within the EDU for a given duty cycle.</div></div>