Liquid-cooled heat sink design for a multilevel inverter switch with considerations for heat spreading and manufacturability

Abstract

Multilevel inverter switches (MIS) present a thermal management challenge by virtue of their small size yet high power operation. In this contribution, a water-cooled heat sink for an MIS was designed and optimized through a series of computational fluid dynamics simulations and surrogate modeling. The parameters under consideration were technical metrics (thermal resistance, pressure loss, and temperature uniformity) and manufacturing costs. The stark difference between air-cooled and liquid-cooled heat sinks was demonstrated in parallel modeling efforts. The heat sink consisted of a metallic baseplate with pin-fins and a separate resin manifold containing an array of nozzles for flow impingement and fluid extraction. The pin–fin baseplate was parametrically designed using three different fin densities: restricted array (RA), variable array semifull (VAS), and variable array (VA), where fin diameter and pitch were correlated. The manufacturability of each design iteration was determined by obtaining quotes from a vendor. Since all heat sinks had the same inlet manifold, the pressure drop was similar across heat sink designs, indicating that the fin distribution had a minimum effect on the overall pumping power required for circulating the flow through the heat sinks. Effective thermal resistance calculations illustrated a complex interplay between fin diameter and pitch spacing. The complexity was better visualized using a feedforward neural network (FNN) and an overall performance metric (PPTR). A sensibility analysis using the FNN indicated that the fin pitch was the dominant variable in the determination of the overall heat sink performance. The PPTR allowed the creation of a ranking system that matched the FFN outcome and indicated that the VA2D1 design was the most technically promising; nonetheless, once manufacturing cost was considered in the analysis, as PPTR-$, the RA1D1 was the most commercially viable as it satisfied technical cooling goals per the lowest unit cost.