Data centric simulation for performative feedback in early building design process

The research investigates the opportunities of a data-centric simulation process where Machine Learning (ML) is used as a simulation design driver to improve the efficacy of current early-phase design cycles. The study examines the possibilities of using data generated from several simulation iterations and creating a feedback process to improve the design process and reduce feedback dependency from more advanced design stages. The aim is to improve the decision-making in the initial design phase relative to standard solution searching methods, e.g., standard optimization algorithms. The study will examine if past simulations can be processed and utilized to become design vectors for future projects where performance evaluation is provided in real-time. This approach moves the simulation benefits from the validation stage to the solution exploration phase, allowing for the assessment of more options with informed outcomes. The proposed methodology intends to generate data through a finite element analysis to feed into a Machine Learning model that uses Deep Convolutional Neural Networks (DCNN). This model will then process the data to predict the material volume and consequently the Embodied Carbon (EC) of a structural element from a Topology Optimization (TO) process. The proposed procedure would encourage a comprehensive iterative cycle that is structurally and environmentally sensible. The research investigates the boundaries of machine learning in the AEC design process and the possibilities of enabling a multidisciplinary integration of simulations in the design process.