Probabilistic Design and Analysis for Robust Design of Advanced Thermoelectric Conversion Systems

Recent research investigated the impacts of single- and multi-variable stochasticity on optimum thermoelectric (TE) system design for automotive and industrial energy recovery because many critical design and environmental parameters used in design optimization can be randomly variable. Analysis tools and techniques have been developed to investigate a variety of stochastic behaviors in critical input parameters, including Gaussian, Log-Normal, Weibull, Gamma, or any type of user-defined probability distribution. Recent accomplishments discussed herein show that: 1) Gaussian input probability distributions can create non-Gaussian outcome distributions for optimum TE areas, required cold-side mass flow rates, and expected power generation; 2) optimum deterministically-derived designs (TE areas and cold-side mass flow rates) should be significantly modified in response to stochastically variable inputs; and 3) outcome parameter standard deviations can be significant and magnified relative to input parameter standard deviations. Multiple variable stochastic inputs tend to significantly increase the output design parameter variability (i.e., standard deviations). Interactive effects of multiple stochastic input parameters have demonstrated that reductions of optimum TE areas by 9–10% relative to deterministic optimum values was warranted in key stochastic analysis cases. Reductions in required cold-side mass flow rates may also be justified. Optimum system power output also was characterized by relatively high variability (i.e., standard deviation) resulting from stochastic input effects on the TE design optimization process. This is an important consideration when integrating the overall power system design with power management electronics and energy storage subsystems.