Application of AI to Advance Structural Performance and Resiliency Quantification

Abstract

This research is focused on the development of a multi-level seismic impact framework for regional seismic response and damage assessment. The work presented here uses a detailed database of buildings in Wellington. First, structural and site characteristics that were necessary for development of response frameworks were added to the database including the estimated building periods, design accelerations and site periods. Next, a one-DOF framework that can be used to roughly identify structures where the seismic demands may have exceeded the design accelerations was presented and demonstrated using a small earthquake in Wellington. Then, machine learning driven computational methods to cluster buildings into typologically similar groups and select representative indicator structures were evaluated. Two prominent unsupervised machine learning clustering approaches are utilized to cluster the mixed categorical and numerical building database: namely k-prototype and k-mean. A novel autoencoder deep learning neural network is also designed and trained to convert the mixed data into a low-dimensional subspace called latent space and feed this into k-mean algorithm. The autoencoder method is demonstrated to be more effective at clustering buildings into useful typological clusters for seismic response analysis. Then, the concept of using indicator buildings for regional response modelling is introduced. Indicator buildings within each cluster were selected and modelled, with supplementary models generated by modifying the stiffness of the base indicator building models account for building flexibilities across each cluster (for a total four models per cluster). A case-study was undertaken using the 2016 Kaikōura earthquake, and the response of the models are utilized to estimate the seismic response of all buildings in the database. Results from the case-study suggest that the indicator building approach can be effective in estimating building drifts and accelerations across a range of buildings. Ongoing work is focused on developing methodologies to correct for building flexibility across each building cluster, developing more detailed structural models and instrumenting indicator buildings, and applying additional machine learning techniques for damage and response estimation.