Synthetic 3D Building Energy Model (BEM) Dataset Generation for Human + AI Synergies in Early-Phase High Performance Building Design

Abstract

Buildings account for nearly 40% of the global process and industrial greenhouse gas emissions, and within the building sector. And within the building sector, nearly 72% of the CO2 emissions are due operational carbon demand in the building sector. To meet the goals set out in the ParisAccord, we need to halve carbon emissions by the end of this decade: an aggressive and ambitious timeline. It has become an immediate and urgent need to design and construct buildings that are minimally intrusive in their impact on the environment and operate efficiently.Improvement of the building's energy performance must begin at the early design stage as the potential for sustainable intervention in the project's early phases is higher. Advances in computational methods for simulating a building's performance facilitate design decisionsthrough impact assessment to a great extent.Improvement of the building's energy performance must begin at the early design stage as the potential for sustainable intervention in the project's early phases is higher. The opportunity to improve building performance as the design progresses, is constantly reducing, while the cost of optimization is constantly increasing. One of the major design variables impacting the performance at the early phase is the building morphology/form and its associated variables. Sensible optioneering of the building form and its variables at the early stages of design - especially the conceptual design stage - can help improve the performance of the design early on in the design process at minimum cost. However, for effective early-phase design optimisation, there is a need to develop tools/methods that allow instantaneous evaluation of a large design sample space without much “lag” between the design and performance evaluation workflows. Also, in regular early phase sustainable design workflows, multiple stakeholders - like the designer, the performance analyst, etc. - areinvolved, which leads to a cognitive divide in the design process. For instance, while the architect designs the initial conceptual massing with a functional design intuition, theperformance analysts suggest design interventions with a performance oriented design intuition. This cognitive divide leads to a dipartite design cognition in early phase design. The underlying premise of this research is to support the notion of blurring the dichotomy between conflictingdesign intuitions through exploring Human + Artificial Intelligence (AI) synergies and their underlying foundational technological requirements with respect to early phase design, to enable intuitive high performance design scenarios with a centralized design cognition where there are no other stakeholders other than the designer and the design environment. With the paradigm shift of other industries towards Machine Learning (ML) and AI, recentresearch advances in ML and the increasing availability of Big-data in the AEC industry have bolstered multivariate problems like building performance evaluation to a great extent. Design optioneering with instantaneous performance feedback through a ML method called Surrogate Modelling is an upcoming and promising methodology, which delivers feedback based on knowledge through available data, rather than simulation. Also, paradigms in Concurrent HumanMachine Interaction (HMI) have explored avenues of augmenting the architectural design process through data-driven approaches. The type of data (input-output pairs) forming the basis for these paradigms depend on the target problem at hand. For early phase performance optimization - especially energy use optimization - Building EnergyModel (BEM) forms the primary data. The US Department of Energy defines BEM as a versatile, multipurpose tool that is used in new building and retrofit design, code compliance,green certification, qualification for tax credits and utility incentives, and even real-time building control. BEM is also used in large-scale analyses to develop building energy-efficiency codes and inform policy decisions. A comprehensive BEM dataset that is large and accessible can support the growing interest in ML and HMI research in the high performance building design field. However, common techniques to acquire large BEM datasets like manual 3D energy modeling and simulation, 3D scanning, etc. are very tedious and time consuming. Also, existing datasets are usually incomplete, inconsistent and very difficult to access, owing to data privacy and open-access issues. And lastly, these datasets are fixed, static and cant be easily reproduced for different use case scenarios, nor scalable to adapt to needs. This research aims to tackle this problem by introducing a novel framework to generate custom problem-specific synthetic BEM dataset that generates a user defined amount of context specific3D early phase building geometries and their associated BEM models, that is suitable for ML and HMI research. Synthetic datasets generated with this framework offer flexibility and customization in the generation process, making these datasets modular, reproducible andscalable. The framework uses the concepts of Generative Design, Geometry Manipulation, Simulation and Computational Automation to build the dataset. The dataset is qualified through the concepts of Preservationism and Sustainability, which are discussed in further sections of thisarticle. The research identifies the importance and multifacetedness of the impact of building geometry on its performance, and demonstrates the application of generated synthetic BEM datasets by developing an enactive, conversational design environment that allows the designerto make real-time sustainable design decisions based on instantaneous machine feedback, for an intuitive, centralized design intuition.