This repo stores the dataset used in the paper "Learning to Simulate and Design for Structural Engineering". The paper is published to ICML 2020 and here are the linkes to the Paper and the Video. Here is the ICML page.
Each json contains a list of 100 graphs where each graph represents a building strucuture. The graph is stored as a dictionary.
Assume we have a building structure with N bars(columns and beams) and K stories. Then, the corresponding structural graph has N+1(Ground).
- end point location
Nx6: [x1, y1, z1, x2, y2, z2] (unit: ft) - cross-section
Nx9: One hot vector for cross-section. Please refer to 'cross_sections.json'. - if_beam
Nx1: 1 if True, 0 if False - if_boundary_beam
Nx1: 1 if True, 0 if False - if_roof_beam
Nx1: 1 if True, 0 if False - metal_deck_area_on_beam
Nx1: Each metal deck panel sits across only two beams on its longer edges. The areas (unit: ft^2) are multiplied by the per-area load in surface load cases. Please refer to the Appendix of the paper for details of the load case setups.
edges are stored with two lists: senders and receivers. The two lists of the same edge type should have the same length. Elements in the list are bar ID, the same ass index in the bar_nodes list.
Column-column edge
- column_column_senders: upper-story column ID
- column_column_receivers: lower-story column ID
Beam-column edge
- beam_column_senders: beam ID
- beam_column_receivers: column ID
Column-graph edge
- column_grounds_senders: First-story column ID
- column_grounds_receivers:
ground node ID = N
For a K-story building, this list contains K+1 elements, where each element is a list containing all bar IDs in that story and the last element is the ground node ID = N.
Corresponding to the bar_id_in_stories, drift_ratio is a list containing K+1 elements. Each elements contains four drift ratios [Ex_x, Ex_y, Ey_x, Ey_y] (x and y component of drift ratio under seismic Ex and Ey load cases). The last element for the story with a single ground node is [0,0,0,0].