Data-driven model for seismic assessment, design, and retrofit of structures using explainable artificial intelligence

Abstract

Retrofitting building designs is crucial given the global aging infrastructure and increased in frequency of natural hazards like earthquakes. While traditional data-driven models are widely used for predicting building conditions, there has been limited exploration of recent artificial intelligence (AI) techniques in structural design. This study introduces a novel explainable AI framework that utilizes data-driven models for assessing, designing, and retrofitting of structures. The framework highlights the key global features of the model and further investigates them locally to adjust the input design parameters. It suggests the necessary changes in these inputs to achieve the desired structural performance. To achieve this, the framework employs interpretability techniques such as feature importance, feature interactions, Shapley Additive exPlanations, local interpretable model-agnostic explanations, partial dependence plot (PDP), and individual conditional expectation to highlight the important features. Additionally, a novel counterfactual) technique is applied for the first time as a design tool in seismic assessment and retrofitting of structures. The effectiveness of this framework is validated on a real benchmark structure through nonlinear time history analysis and natural earthquakes. The results show that the proposed framework is highly effective, especially under design-level earthquake conditions in achieving the necessary change in stiffness and strength of structures to meet the required seismic design objectives across different earthquake scenarios. This framework holds promise for wider adoption and applications in various other structural and civil engineering domains.