Are current turbulence modeling practices addressing industry's needs for electronics thermal design?

Abstract

Since the 1990's, computational fluid dynamics (CFD) has been widely adopted in the electronics industry for the thermal design of electronic products. Its advantages in terms of product improvements and enhanced productivity of design analysis, are undisputed. However, the industry has also experienced that incorrect product design decisions can be taken as a result of inaccurate CFD predictions, with consequences ranging from reduced product performance and reliability, to catastrophic field failure. Consequently, understanding and minimizing CFD prediction errors is a major concern to ensure a return on capital, software and human resource investments in the thermal design process. Sources of CFD inaccuracy can be categorized as either (i) errors of a numerical (e.g., round-offs, convergence, discretization), coding-, or user nature, or (ii) uncertainties in model inputs (e.g., limited information or approximations in the representation of geometry, material properties, and boundary conditions such as fan flows and screen losses) and in physical models (e.g., representation of physical processes such as turbulence, simplifying assumptions such as steady-state analysis or adiabatic heat transfer boundary). Assuming that the CFD code is correct and that user errors are negligible, this Talk focuses on two unvoidable sources of CFD prediction errors, and candidate solutions to minimize them: (a) physical model uncertainties associated with current Reynolds-averaged Navier-Stokes (RANS) turbulence modeling, and (b) input uncertainties associated with boundary conditions such as fan flows. In relation to topic (a), an overview of the current state-of-the-art in CFD for electronics cooling applications is presented, as well as published CFD benchmarks relating to air- and liquid cooling applications. The challenges for improved predictive accuracy are outlined in the context of two opposing arguments relating to physical model uncertainties: (i) the development and optimization of turbulence models for limited categories of flows, versus (ii) the search for a comprehensive, general-purpose turbulent flow model. With respect to topic (b), the fact that detailed modeling inputs are generally not available during the design phase, may no longer justify the view that standard turbulence models applied on simple grids are satisfactory, offer efficient analysis and solution stability. This argument may become outdated with increases in computational power, which will facilitate the application of more sophisticated turbulence models to electronic system thermal design. To illustrate the difficulties typically encountered in the industry to predict heat transfer and fluid flow in electronic systems, a benchmark study of an air-cooled mock-up telecommunication unit is presented. The case study highlights substantial CFD prediction discrepandies with experimental measurements of temperature and fluid flow, which are found to be associated with uncertainties in fan performance characteristic curve, and physical modeling of turbulent flows, including fan-induced flows. Finally, the lack of benchmark studies reporting large but actual CFD prediction discrepancies is emphasized, highlighting the need for realistic benchmarks.