On the constitutive modeling using meta-models and their deployment for finite element analysis to evaluate the high temperature deformation behaviour of Al 2014 alloy

Isothermal hot compression tests were carried out on Al 2014 alloy over a range of deformation temperatures (300-500 & DEG;C) and strain rates (0.0003-1 s(-1)). The flow stress data obtained from the experiment as a function of temperature, strain rate, and strain were used to develop constitutive models using three popularly used meta-models viz., genetic algorithm (GA), response surface methodology (RSM), and artificial neural network (ANN). The prediction accuracy of meta-models was evaluated using standard statistical parameters such as correlation coefficient (R) and average absolute relative error (AARE). RSM and ANN showed better prediction at high temperature and strain rates when compared to GA. A quartic regression equation was suggested by RSM, while a 3-15-1 neural network architecture provided a better correlation. To check the reliability of the meta-models, the established meta-model constitutive equations were deployed in a commercial finite element analysis software through user subroutines. The results obtained were then compared and discussed here.

Automated summary

Isothermal hot compression tests were conducted on Al 2014 alloy to investigate the effects of deformation temperature (300-500°C) and strain rate (0.0003-1 s(-1)). Three commonly used meta-models, namely genetic algorithm (GA), response surface methodology (RSM), and artificial neural network (ANN), were employed to develop constitutive models using the flow stress data obtained from the experiments. Evaluation of the meta-models' prediction accuracy was carried out using standard statistical parameters such as correlation coefficient (R) and average absolute relative error (AARE). RSM and ANN performed better than GA, especially at high temperature and strain rates. RSM suggested a quartic regression equation, while a 3-15-1 neural network architecture provided a better correlation. The reliability of the meta-models was verified by implementing the established meta-model constitutive equations in a commercial finite element analysis software and comparing the results.