Machine-learning based temperature- and rate-dependent plasticity model: Application to analysis of fracture experiments on DP steel

Slow, intermediate and high strain rate experiments are carried out on flat smooth and notched tensile specimens extracted from dual phase steel sheets. A split Hopkinson pressure bar testing system with load inversion device is utilized to achieve high strain rates. Temperature dependent experiments ranging from 20 °C to 300 °C are performed at quasi-static strain rates. The experimental results reveal a non-monotonic effect of the temperature on the stress-strain curve for DP800 steel. For the temperatures considered, the lowest and highest curves are obtained for 120 °C and 300 °C, respectively. A modified Johnson-Cook plasticity model is developed to capture the observed unconventional effect of the strain rate- and temperature on the hardening response. It includes a von Mises yield surface, a non-associated Hill’48 flow rule, a Swift-Voce reference stress-strain curve describing the rate-independent behavior at room temperature, and a neural network function depending on the plastic strain, strain rate and temperature. This model is implemented into a material user subroutine and identified using a combination of analytical formulas along with a resilient back propagation algorithm. It is found that the obtained machine-learning based Johnson-Cook plasticity model can describe all experimental data with high accuracy, including both force-displacement and local surface strain measurements. Using a hybrid experimental-numerical approach, the loading paths to fracture are determined. It is found that the temperature not only affects the plasticity of the DP800 steel in a non-monotonic manner, but also its strain to fracture, with ductility loss of up to 30% when increasing the temperature from 20 to 120 °C, followed by a gain in ductility when increasing the temperature further to 300 °C.