Fragment prediction of reinforced concrete wall under close-in explosion using Fragment Graph Network (FGN)

Reinforced concrete (RC) walls are vulnerable to severe damage under high-intensity, close-in TNT explosions. Substantial secondary fragments at high ejecting velocities could be generated from the damaged wall, posing serious threats to people, facilities and structures in the area. Predicting the blast-induced secondary fragments remains a great challenge. Traditional computational methods, such as the finite element method (FEM) or meshfree methods, are often used to predict the fragment characteristics despite their inherent problems, such as the application of erosion and predefining the weak sections in the simulation. They also require high computational power to perform the simulation, thus limiting their use in creating an adequate dataset to thoroughly analyse the characteristics of secondary fragments and the associated threats. This study employs a recently developed machine learning-based approach named Fragment Graph Network (FGN), a variant of Graph Neural Networks (GNNs), to generate a large dataset of fragment characteristics. This FGN model can efficiently predict the fragment mass, size, and velocity with a significantly reduced computational cost. Intensive predictions of fragments from different wall configurations and explosion intensities are carried out. The results are used to develop analytical formulae for predicting secondary fragments of RC walls subjected to close-in explosions.