AI model predicts seismic vulnerability of concrete buildings in milliseconds

"Thousands of reinforced concrete apartment blocks across southern Europe were built before modern seismic codes existed, and engineers have long struggled to assess how quickly each one would fail under a major earthquake. A new study published in Neural Computing and Applications by Gianni D'Angelo, Francesco Nigro and Enzo Martinelli of the University of Salerno describes an artificial intelligence system that can perform this assessment in roughly seventy milliseconds, a speed that contrasts sharply with the hour-long nonlinear finite element simulations traditionally required for a single building."

"The core problem the researchers tackled is the nonlinear static analysis known as pushover analysis. A finite element model of a building is subjected to a progressively increasing lateral load that mimics the inertial forces of an earthquake, and the response is tracked until key structural members reach their deformation limits. The procedure produces a capacity curve, typically simplified into a bilinear acceleration-displacement relationship characterised by an equivalent vibration period, a yield acceleration and an available ductility."

"The team's answer is a hybrid deep neural network that fuses two architectural paradigms: Long Short-Term Memory networks, which model sequential dependencies, and Convolutional Neural Networks, which specialise in spatial pattern recognition. The seismic behaviour of a frame building depends simultaneously on how forces propagate vertically through the storeys and on how stiffness and strength are distributed across the floor plan, and neither architecture alone captures both."

"Each building is described by three complementary sources of information: nine global parameters characterising the structure as a whole, eight local parameters for each reinforced concrete column, and a mask that encodes the building's layout on a 50 by 50 grid spanning a 50 metre square."

"The fifty LSTM outputs are arranged as columns of a two-dimensional matrix and combined with the layout mask to form a two-channel image, processed by a convolutional network. The output is concatenated with the nine global building parameters and passed through fully connected layers that produce twenty regression targets."