A data-driven framework to support performance-based early design of buildings: data and methodology

Abstract

Achieving resilient and sustainable infrastructure urges developing computational tools to explicitly consider performance objectives in all design and construction stages. The majority of critical decisions are made at earlier stages of the design. Early design can be substantially improved by incorporating quantitative methods to evaluate the consequences of these decisions. However, implementing quantitative methods poses several challenges, including imprecision of design variables and time- and effort-intensiveness of such assessments. This paper presents a modular framework to select suitable candidate structural systems, characterize their design parameters range, and communicate their expected hazard and environmental performance during their life cycles. The framework leverages a machine-learning-assisted workflow that performs mapping between crude design- and topology-related parameters and global hazard and environmental performance indicators. Next, a sequence of surrogate models with varying fidelity aids in performing the convergence-divergence cycle of early design. Lastly, a deep learning architecture with a customized loss function maps the result of simpler static analysis to the detailed description of seismic performance, linking early design to the next design stages. A case study is presented to illustrate the application of the framework to evaluate the embodied carbon and seismic-related repair cost of an inventory of 720 multi-story concrete frames with varying topologies in Charleston, South Carolina.