Abstract
In material science, models are derived to describe emergent properties (e.g.elasticity, strength, conductivity, ...) and their relations to the materialand processing conditions. Constitutive models are models that describe thebehaviour of materials for instance deformation processes through appliedforces. We describe a general method for the extension of constitutivewhite-box models using genetic programming (GP). The method is demonstrated fora constitutive model which is used in computational mechanics to describe thematerial response to various thermo-mechanical loading conditions. This modeloperates with internal material variables such a dislocation density andcontains a number of parameters. Among them there are three calibrationparameters, which are usually fit to measured data since the relations to theprocessing conditions (e.g. deformation temperature, strain rate) are not fullyunderstood. Our proposed GP-based method identifies these relations andgenerates short expressions which can be plugged into the constitutive modelinstead of the calibration parameters. This would enable interpolation as wellas extrapolation which is crucial for constitutive modelling. Our results show that evolving extensions for the constitutive modelimplicitly with GP leads to better predictive accuracy than explicitly learningindependent formulas for predicting calibration parameters with symbolicregression.