A common practice that I do when building models in Abaqus, is to fit the material property. For example, I try out all the possible material properties and look at the surface deflection given by the model, and then find out the one that matches our experimental observation the best. Practically, I compare the value with model output and experimental data, get an R-squared value and try to minimize the value of -1. * R2.
I have been using the scipy optimization toolbox to do this in Abaqus. However, there is one question: there are cases where the model would not converge with certain given parameters that the optimizer try. In these cases, what values should I set to R2? Should I set it to -1. * numpy.inf, or -1. * numpy.nan (assuming import numpy as np)?
Moreover, there are situations where I use optimization functions that does not support general constraints, for example modulus_1 > modulus_2; if it asks me to submit a job where modulus_1 <= modulus_2, can I just return a -1. * np.inf or -1. * np.nan as penalty?
The problem happens because there is no way to know where the model would fail to converge in the parameter space apriori. Any help would really be appreciated. Thank you so much!