"From B-Rep to BEM: Robust Framework for Generating Knowledge-Graph BIMs and Building Energy Models"

"    This article introduces a large mixture B-rep model datasets to evaluate and refine the developed module. Comprehensive building models including 100 parametric models (dataset 1), 570 sketch design models (dataset 2), and 15 complex real-building models (dataset 3) was curated, for the validation on the topology accuracy, analytical model accuracy, robustness, the and the efficiency in different applications.        A series of parametric models (dataset 1) are created by an auto-mass-generation module Evomass \\cite{Wang2020} for the sake of derivable space topology and accurate *.idf modelling. The grasshopper module Evomass can create randomized multi-storey models with several shape parameters. A post-process script has been developed and applied to randomly create inner walls and windows to finish the thermal zoning on the parametric models following the grid axis of the building. The space volume would be maintained during the generation, which can be used to restore the space topology as well as build the energy model by Honeybee-grasshopper. Therefore, the dataset 1 is a completed auto-generated dataset with *.obj, ground truth *.owl, and ground truth *.idf for each case.        The sketch models (dataset 2) are sourced from architecture students and were specifically intended to emulate the characteristics of models produced during the early design stages. Accordingly, no restrictions are imposed regarding model cleansing, modeling techniques, or the specific design software utilized (provided an .obj file export was possible). To further challenge the module's robustness, several redundant or complex elements were intentionally introduced into the model testing sets. These included items such as shading devices, individual stair steps, and decorative fa\u00e7ade components. The sole modeling constraint mandate that walls be represented as single surfaces rather than as two faces defining a thickness. Consequently, this dataset can be considered representative of the diverse range of model inputs encountered in practical scenarios.         The real-building models (dataset 3) are collected from important green building design cases with relatively more complex spaces and geometries. The buildings were selected according to their complexity and shapes: Oblique and Curve shape contains non-orthogonal walls and faces creating small deviation and creak; Complex floor arrangement includes high space cross multiple storeys; High-rise Oblique shape have numeric faces and spaces requires high effort on calculation; and the Complex decorated building includes numeric redundant shading and decorating elements disturbing the space and face identification. All buildings are carefully modelled with their shape, storey, and fa\u00e7ade. Especially, considering the applications in performance analysis, the connectivity between rooms with inner walls, doors and windows are restored. This dataset is included to validate the robustness of the framework\u2019s application on real construction and design workflow.    In which, dataset 1 and dataset 3 is open-access in this material. Due to privacy concerns of student modelers, Dataset 2 will not be made publicly available."