A data-driven rate-dependent constitutive model for ultra-high strength alkali-activated concrete under impact load

This study presents a rate-dependent constitutive model for ultra-high strength alkali-activated concrete (AA-UHSC) based on a data-driven approach and fundamental plasticity theory. First, the overall frame of data-driven rate-dependent constitutive modeling was proposed. Based on the proposed frame, a data-driven rate-dependent constitutive model (DD-RDCM) for AA-UHSC was developed, and the corresponding parameters influencing the accuracy of model were comprehensively studied. A compatible numerical implementation algorithm was subsequently proposed to integrate the data-driven constitutive model into a finite element calculation. Finally, the proposed model was compared with the calibrated Holmquist-Johnson-Cook (HJC) model in terms of predictability and computation time of dynamic impact compression tests of AA-UHSC. The results indicated that using a data-driven approach to establish a constitutive model for rate-dependent materials is feasible, but considering that the stress-strain relationship from the dynamic impact compression of AA-UHSC is path- and strain rate dependent, the incremental constitutive model should be selected. The activation function type, number of hidden layers and neurons in each hidden layer have a significant effect on the predictability of the model. The network model with two hidden layers and each hidden layer with ten neurons performs best when the sigmoid function and purelin function were adopted in hidden layers and output layers, respectively. Compared with the HJC model, the predicting accuracy of DD-RDCM can be improved by maximum value of 18.4 %; the computation time increases by maximum of 105 % and minimum of 6.9 %; the reduction of computation efficiency of DD-RDCM was not significant and worthwhile owing to its high prediction accuracy.