A machine learning based sensitivity analysis of the GTN damage parameters for dynamic fracture propagation in X70 pipeline steel

The Gurson-Tvergaard-Needleman (GTN) damage model is widely used to predict ductile failure initiation and propagation. However, the material-dependent parameters can show a significant spread when determined for the same steel grade material. Different calibration procedures and optimisation processes cause a significant variation in the obtained parameter values. Furthermore, there is no clear consensus on which parameters require calibration. In this study, the influence of the materialdependent parameters used to model the dynamic ductile fracture behaviour of X70 grade pipeline steel is investigated. A sensitivity analysis is performed on a finite element model of a Charpy V-Notch (CVN) specimen. Seven GTN model parameters are considered in a total of 70 simulations. A feedforward backpropagating artificial neural network (ANN) is constructed and trained using data obtained through the numerical simulations. A connected weights (CW) algorithm allows to determine the relative influence of each parameter on the fracture energy. It was observed that the void growth acceleration factor plays an important role with respect to the parameter influences. Remarkably, the mean nucleation strain, eN has the highest relative importance whilst the critical void volume fraction, fc-which is considered as a crucial damage parameter-showed the smallest influence when the acceleration factor is low. On the contrary, when considering a high acceleration factor, fc becomes the most influential parameter. Based on the obtained importance for each parameter, it is suggested that parameters f0, fc, fF, and fN should be selected for calibration in each individual application. Finally, the applied machine learning approach is used to predict the fracture energy for a given set of damage parameters for X70 grade steel. It is observed that the trained neural network is able to provide a satisfactory approximation of the CVN fracture energy.