Reduced order modeling and experimental validation of steady turbulent convection in connected domains

Abstract

The characterization and design of complex thermo/fluid electronics systems such as electronic racks requires attention to multiple length scales and transport phenomena, including conduction at the chip package level and convection at rack level. Detailed numerical calculations or experimental measurements are often time consuming and computationally expensive or simply infeasible. An efficient strategy to bridge length scales in multi-scale characterization is presented. Reduced order models for various system components obtained using proper orthogonal decomposition (POD) are assembled to model the complete system through flow network modeling (FNM). Boundary profile capturing capability at the interfaces of components is incorporated into the flux matching based POD approach, without introducing extra computation. The simulation errors of the component reduced order models are within 7.4% for temperature, velocity and pressure fields, compared to full-field computational simulations. Assembled component models result in a 12% error norm at the system level. Experimental validation with a representative electronic cabinet shows a difference of less than 10% in chip junction temperatures prediction.