This report describes a methodology for efficient calibration of an internal state variable (ISV) material model that includes uncertainty. The model was developed to represent uncertainty in the material performance due to variations and inaccuracies in the model parameters. The model parameters comprise of constants collected from the literature (melting temperature, bulk modulus, shear modulus, etc.), constants derived from microscope images of microstructural features (voids, cracks, inclusion particles) in the material sample, and constants calibrated from experimental stress-strain data. The uncertainty in the model arises from inaccuracies in literature data, from variations in the measurements of the microstructures, and from inaccuracies in the stress-strain calibration data. The model calibration process is very computationally-intensive since variability distributions need to be calculated for the parameters in the model. The process involves Monte Carlo simulation with a large number of samples (). For each sample, a function minimization problem is solved using a derivative-free method, requiring up to twenty seconds of solution time. As an estimate of the magnitude of the calculations, a ten-second solution time for each sample translates into more than eleven days of total process runtime on a serial machine. Thus, the process must be executed on a high performance parallel environment in order to obtain results within a reasonable period. The codes used in the process include Fortran routines for a material point simulator and an ABAQUS UMAT, MATLAB scripts for optimization and MatlabMPI for parallel execution on a Linux cluster. The computational methodology we describe enables the model calibration with uncertainty to complete over a weekend on 16 processors.
Crack P Code Matlab Codel
2ff7e9595c
Comments