Comparison of several optimisation strategies for the determination of material constants in the Chaboche visco-plasticity model

Abstract

Determining representative material constant sets for models that can accurately predict the complex plasticity and creep behaviour of components undergoing cyclic loading is of great interest to many industries. The Chaboche unified visco-plasticity model is an example of a model that, with the correct modifications, shows much promise for this particular application. Methods to approximate material constant values in the Chaboche model have been well established; however, the need for optimisation of these parameters is vital due to assumptions made in the initial estimation process. Optimisation of a material constant set is conducted by fitting the predicted response to the experimental results of cyclic tests. It is expected that any experimental data set (found using the same values of test parameters such as temperature; the dependency of which is not accounted for in the original Chaboche model) should yield a single set of optimised material parameters for a given material. In practice, this may not be the case. Experimental test programs usually include multiple loading waveforms; therefore, it is often possible to optimise for separate, different sets of material constants for the same material operating under comparable conditions. Several optimisation strategies that utilise multiple sets of experimental data to form the objective functions in optimisation programs have been applied and critiqued. A procedure that evaluates objective functions derived from the multiple experimental data types simultaneously (i.e. in the same optimisation iteration) was found to give the most consistently high-quality fitting. In the present work, this is demonstrated using cyclic experimental data for a P91 steel at 600 °C. Similar strategies may be applied to many constitutive laws that require some form of optimisation to determine material constant values.