Bayesian surrogates for integrating numerical, analytical and experimental data: application to inverse heat transfer in wearable computers

The increasing thermal challenges facing compact systems have motivated new cooling strategies. Building models to assess design parameter effects is critical for effective incorporation of these thermal strategies into products. System models to enable design space exploration are built from various information sources, e.g. numerical simulations, experiments, analytical solutions and heuristics. These models, called surrogates, are nonlinear and adaptive and thus suitable for system responses where limited information is available and few realizations are feasible. In this paper, the surrogate framework is applied to estimate physical parameter values of an embedded electronics system. For this purpose, experiments and simulations are performed on a prototype TIA (technical information assistant) wearable computer. Numerical models are studied which use five and three unknown parameters, with and without thermal contact resistances, respectively. Using orthogonal arrays and optimal sampling, the parameter space exploration is performed to determine system parameters such as thermal conductivities, thermal contact resistances and heat transfer. Surrogate models are built that combine data from numerical simulations, experimental measurements, and a thermal resistance network simplified model. The integration of several information sources reduces the number of numerical simulations needed to find reliable system parameter estimates and allows identification of the best numerical model. For the embedded electronics case, use of the thermal resistance network model data greatly reduces computational effort required.