ARINet (2020)

Artificial intelligence and data-driven modeling are becoming more prominent in the building, and construction sectors. Physics-based models usually require significant computational power and a considerable amount of time to simulate output. Therefore, data-driven models for predicting the physical properties of buildings are becoming increasingly popular. The objective of this research is to introduce Artificial Neural Networks(ANNs)methods as a means of representing the physical properties of buildings. Achieving this goal will illustrate the future capacity of integrated neural networks in building performance simulations. The Annual Radiation Intensity Neural Network (ARINet) demonstrates the feasibility of using a 3D convolutional neural network to predict the surface radiation received by building façades. The structure of ARINet is composed of3Dconvolution, fully connected, and 3Ddeconvolution layers. In this research, it was trained on 1,692 datasets and validated by424 datasets generated by a physical simulator. ARINet showed errors in0.2% of the validation sets.

CoolVox (2021)

Data-driven models have become increasingly prominent in the building, architecture, and construction industries. One area ideally suited to exploit this powerful new technology is building performance simulation. Physics-based models have traditionally been used to estimate the energy flow, air movement, and heat balance of buildings. However, physics-based models require many assumptions, significant computational power, and a considerable amount of time to output predictions. Artificial neural networks (ANNs) with prefabricated or simulated data are likely to be a more feasible option for the environmental analysis conducted by designers during the early design phase. Because ANNs require fewer inputs and short computation times and offer superior performance and potential for data augmentation, they have received increased attention for predicting surface solar radiation on buildings. Furthermore, ANNs can provide innovative and quick design solutions, enabling designers to receive instantaneous feedback on the effects of a proposed change to a building’s design. This research introduces deep learning methods as a means of simulating the annual radiation intensities and levels of exposure of buildings without the need for physics-based engines. We propose the CoolVox model to demonstrate the feasibility of using 3D convolutional neural networks to predict the surface radiation on buildings’ façades. The CoolVox model accurately predicted the radiation intensities of building façades under different boundary conditions and performed better than AriNet (with average mean square errors of 0.01 and 0.036, respectively) in predicting the radiation intensity both with (validation error = 0.0165) and without (validation error = 0.0066) the presence of boundary buildings.

AIRVox (2022)

working paper

Computational Fluid Dynamics (CFD) has been used to solve the airflow and heat transfer related issues in buildings for the past few decades. Modeling CFD in building scale requires a tedious modeling workflow that also requires specific skills. The use of artificial neural networks (ANNs) in predicting airflow in buildings enables designers to check the potential opening positions or mass arrangements with instant design changes. The goal of this paper is to solve physics-based equations for the mean velocity and pressure fields with data-driven models. Data-driven turbulence models have gained attention with the development of the tensor basis neural networks. However, the application of 3DCNNs with simulated data for the building is new. The potential applicability of the data-driven airflow model in architectural design and consulting and the use of advanced modeling techniques are going to be discussed in this paper. The performance and computational time are compared with classical turbulence models as well as neural network models that do not preserve the physical properties of the velocity fields. Three different neural networks architecture was tested to get the lowest Mean Squared Error (MSE) on the given datasets, and the hyperparameters were tuned to deliver the prediction of the velocity and the pressure fields in given geometry configurations. The autoencoder with skip-connection structure of 3DCNNs performed best on the validation sets yielding the MSE at 0.0058.

DDES Dissertation (2022)

Building performance simulation (BPS) software using building information modeling (BIM) technology has been rigorously researched for integration into the building modeling workflow. Integrating a BPS workflow into BIM modeling will facilitate an optimized sustainable analysis and consulting. Efforts have been made to develop a green building XML (gbXML) format and other definitions, such as an information delivery manual (IDM). However, they have been mostly limited to energy model applications, such as EnergyPlus, and other types of BPS models for airflow, water, and other technologies. Recently, machine learning (ML) and deep learning (DL) have become more prominent in the building, architecture, and construction industries. One area ideally suited to exploit this powerful new technology is BPS for sustainable building design. As DL can handle large computational volumes in a short time, data-driven physical property-prediction models for buildings are becoming increasingly popular for their simplicity and high efficiency. For environmental analysis, artificial neural networks (ANNs) with prefabricated or simulated data are a more feasible option for designers than the physics-based method during the early design phase, and their use is expected to grow rapidly.

For computer vision, 3D data are an asset for ANN training because they provide rich information about geometry and related environment. Depending on the 3D data representation considered, different challenges may emerge when using trained ANN models. Hence, an interoperability framework is required to convert the building and environment-related geometries and information into relevant 3D matrices for ANN model training and utilization. However, to date, there has been no research on this topic in the BPS field; thus, this thesis proposes a new data interoperability framework for ANN models, with 3D buildings serving as inputs. A trial investigation was conducted on the developed framework using several ANN modeling studies on radiation and airflow simulation. These have been delineated in a comprehensive process map that includes the BPS requirement for ANN modeling, related subprocesses (i.e., building geometry and environmental levels), specific rules and methods, and the processing of input and output data. To accomplish this, data exchangers for the ANN models, a geometry representation tool (GRT), and a BIM specification tool (BST) were developed as computational tools. The comprehensive framework has been validated via case studies, demonstrating its applicability for different computer-aided design tools (Rhinoceros and Revit) and ANN models (Solar radiation and airflow) and the application potential of integrated ANNs in BPS and early-stage modeling.