Failure Evaluation of Steel Plate Shear Walls in Multi-Storey Steel Buildings Under Seismic Excitation Using Convolutional Neural Networks

Multi-storey steel buildings may be susceptible to structural damage under seismic loading. Shear plate walls are often integrated in the structural system of this type of buildings in order to restrict the lateral response. This article aims, therefore, to propose a methodology for the automatic evaluation of failure on the shear plate walls of multi-storey steel buildings using computer vision. Physics-based non-linear dynamic finite element models have been developed and solved for a range of geometries, shear plate wall thicknesses and seismic loading from past events. Images depicting failure on shear plate walls given as equivalent plastic strain contour plots are included in the output data of the parametric simulations. Then, Convolutional Neural Networks (CNNs) are introduced, predicting the failure distribution on shear plate walls. The input parameters are the geometric properties of the buildings and the seismic event intensity, and the output parameters is the equivalent plastic strain images. This scheme was tested on random buildings with satisfactory accuracy. The proposed methodology can be adopted and used within structural digital twin solutions.