Generating City-Scale 3D Finite Element Models from GIS Features and Heuristic-based Framing Models using Feed-Forward Artificial Neural Networks

Abstract

City simulation-based studies require numerous structural models of sufficient accuracy. In this study, a tool that automates the generation of three-dimensional (3D) finite element (FE) models given geographic information system (GIS) features is developed based on heuristic-based framing models. The developed tool utilizes feed-forward artificial neural networks (ANN), enhanced by topology-based partitioning and parallel computing by Message Passing Interface (MPI). Different sets of topologies are trained and tested using heuristic-based framing models to predict column locations and floor elevations given GIS features. Optimal topologies are then integrated into the frame model generating tool. With GIS features on building footprint and height as inputs, the developed tool performs several tasks: (1) pre-processing the given GIS features, (2) establishing the column locations, (3) generating the beam framing in the two-dimensional (2D) plane, (4) predicting the floor elevations, and (5) generating the 3D frame model. Using the developed tool, the 3D frame models of 8,753 and 7,400 target structures in two urban areas in Metro Manila, Philippines are created. With the generated 3D FE models, the application to city scenario seismic response simulation is demonstrated. Comparisons with a recent simulation-based study showed that the current ANN-based tool outputs more realistic framing configurations.