Data-driven Reduced-Order Models for Multidisciplinary Design Optimization Process

Multidisciplinary Design Optimization (MDO) is a model-based simulation and optimization process that integrates multiple disciplines with conflicting objectives and design constraints to allow a more affordable design. In the Architecture, Engineering and Construction (AEC) sector this method still in the research and testing phase compared to the automotive and aerospace industries. However, the ability of MDO to extend the number of solutions examined through automation requires significant computational resources. In this context, the following paper explores the advantages of reducing simulation times using the AI-based reduced-order models (ROM). This data-driven method combines Artificial Intelligence and system modelling techniques to reduce computational complexity as Digital Twin (“As Designed”) and it can be used to speed up system design and optimization analyses. This paper presents a test application that explores how AI-based ROM can support the MDO process, which has already been applied to an AEC retrofit project. The case study is a classroom of an existing building where fluid dynamics, thermal and comfort performances have been optimized to support decisions in the conceptual design phase. Although the simulations were successful, a high computational complexity emerged, making it difficult to extend the simulations to the entire building and to more disciplines. The digital experiment carried out in this paper is about speeding up the process and making simulations easier compared to the legacy approach based on high computational simulations. The digital experiment carried out in this paper is about physics phenomena in buildings, which are only a part of the architecture performance and quality. This is an early example of demonstrating how AI-based ROMs can accelerate MDO simulations to make it scalable up the entire AEC design process in the future.