Bayesian Global Optimization applied to the design of shape-memory alloys

The method of Bayesian Global Optimization, using a surrogate model and a utility function, is reviewed and its application toward finding alloys with targeted properties, such as high transition temperature and very small thermal hysteresis, is discussed. We also address the calculation of estimates of uncertainties from data and compare the estimates obtained by using standard deviation and the Infinitesimal Jackknife on two experimental datasets, one for piezoelectrics and one for shape-memory alloys (SMAs). We also discuss the importance of the utility function for selection and ranking next candidate data points in minimizing the number of evaluations to nd optima. Finally, we illustrate how these ideas can be applied to discovering new SMAs with high transition temperatures and very small thermal hysteresis by reviewing the results of the synthesis and characterization of alloys carried out previously.