MultOpt++: a fast regression-based model for the constraint violation fraction due to composition uncertainties

Abstract

MultOpt++ is a an alloy design tool based on numerical optimization. The concept is similar to approaches in the literature termed ‘Alloy by design’ and usually dealing with multi-objective optimization problem resulting in a Pareto front with a set of optimal compositions. Unpreventable scattering of element concentrations during the alloy production process causes property deviations of an optimal alloy composition resulting in unfeasible solutions. The violation fractions of such compositions should be taken into account during the optimization process to be able to determine optimal and feasible alloys. Established models for violation fractions require a high computational effort due to a high number of necessary calculations. In this work, we derive a fast model for the constraint violation fraction based on a regression analysis of the mean variation width of alloy properties. We apply this model to nickel-base superalloy properties predicted with the CALPHAD approach. The model allows to select alloys with a lower violation fraction of targeted constraints.

Numerical Design of CoNi-Base Superalloys With Improved Casting Structure

“…[16,17] In this study, we employ the latter approach by coupling CALPHAD calculations to genetic multi-criteria optimization algorithms. We employ the numerical alloy design tool MultOpt++ [18][19][20] and its Python port PyMultOpt for the design process. In this framework, the alloy design task is defined as an optimization problem consisting of optimization goals that should be minimized or maximized and constraints that may not be violated.…”