The building sector is one of the largest consumers of energy and contributors to greenhouse gas emissions in the European Union (EU), representing 35% of energy-related EU emissions in 2021. In response, the EU has set ambitious targets, aiming to become the world’s first climate-neutral continent by 2050.
This objective is at the heart of the European Green Deal and is legally binding under the European Climate Law. The EU seeks to significantly reduce emissions through increased energy efficiency and a higher rate of building renovations (source). Achieving these goals requires innovative approaches to urban planning and energy management.
In support of this effort, the CoE Hidalgo2 project is developing advanced high-performance computing (HPC) solutions to model and simulate complex urban environments through the Urban Building pilot. The latter aims to create detailed and scalable simulations of urban areas to optimize energy consumption and inform renovation strategies.
The Urban Building pilot application leverages cutting-edge CI/CD (Continuous Integration/ Continuous Deployment) methodologies to streamline the development and deployment of simulation tools across various computational platforms. These platforms include EuroHPC JU supercomputers, which provide the necessary computational power to handle the large-scale simulations required for accurate urban modeling.
By integrating detailed terrain modeling with building and vegetation data, the Urban Building (UB) application seeks to create a comprehensive and realistic representation of urban environments. This model will support various analyses, such as energy efficiency assessments and solar mask computation, contributing to the EU’s broader objectives of reducing carbon footprints. To this aim, the CEMOSIS team leading the UB pilot proposed several challenging internship projects to the Master students to enhance their skills in computational modelling, data analysis, and simulation techniques. These projects will enable students to work on real-world problems, contribute to the development of the UB application, and gain valuable experience in interdisciplinary collaboration and innovative urban planning solutions.
In the following, we present the most promoted internships for this year with a focus on their results, highlighting the key contributions made by the students and the impact of their work on the Urban Building (UB) application.
1. LOD1 building integration
This project focuses on the Level of Detail 1 (LOD1) geometric reconstruction of buildings. LOD1 is a fundamental level that represents buildings as simple block models, capturing the basic shape and height of their structures. Despite its simplicity, LOD1 is essential for various large-scale urban analyses and simulations.
The development of a robust geometric reconstruction tool for LOD1 buildings has several key benefits:
● Improved Accuracy: accurate 3D models are essential for reliable energy simulations and urban planning.
● Efficiency: automated tools reduce the time computation required to generate building models, allowing for large-scale analyses.
● Flexibility: the tool can be adapted and extended for more detailed levels of building reconstruction in the future.
By focusing on LOD1, this project lays the groundwork for more complex modelling efforts, ensuring that foundational models in UB are accurate and reliable. The objective of our work is to generate a surface mesh, representing the building’s external envelope, that outlines the outer surface of the building and ensures that all indoor features, such as internal and external walls, slabs and roof are faithfully captured.
The methodology for creating a building’s surface mesh involves several key steps. First, the building base is triangulated, extruded to form a 3D structure, and then the walls and floors are created and merged with the base. For the building roof, a straight skeleton extrusion is used to shape it, followed by height adjustments and wall extrusion, before merging the roof with the base. After individual buildings are modelled, they are merged into a single mesh. Once the surface mesh is complete, volume meshing can be performed. Lastly, quality checks are conducted on both the surface and volume meshes to ensure they meet the required standards.
The implementation of the mesh generation for a single building includes, first, the surface mesh generation, which provides a detailed exterior view of a building’s geometry. Next, the volume mesh is processed to illustrate the building’s internal structure and separations, showing spatial arrangements and material distribution. Markers label internal and external walls, floors, roofs, and other vertical faces for easier identification and analysis.
When applied to multiple buildings, surface and volume meshes show how different structures interact and compare in a common environment.
For the district scale, which includes over 700 buildings, both surface and volume meshes are generated to analyze large-scale urban planning. However, handling such large datasets presents computational challenges due to the sheer size of the output files.
2. LOD2 building generation
This project aims to develop a tool for the geometric reconstruction of buildings at Level of Detail 2 (LOD 2), based on IFC (Industry Foundation Class) standards. This involves generating volumetric meshes with constraints on conformality and watertightness, essential for simulations using Finite Element Methods (FEM). To achieve the latter objective, the Kinetic algorithms developed by the Inria Titane Team are employed. The process includes validating and possibly correcting the mesh from IFC files, automating mesh generation with Kinetic, performing quality tests, and ensuring correspondence between IFC elements and mesh elements.
To simulate physical phenomena like heat propagation, the meshes used must meet strict criteria, such as being free of self-intersections, watertight, and outwardly oriented. While a mesh may appear visually fine, internal issues can occur, particularly in the junctions between the building components, such as illustrated in the next figure, that can affect simulation accuracy. The Kinetic project aims to repair such errors using algorithms from CGAL automatically. The project seeks to achieve automatic mesh repair on different mesh types, such as STL or MSH, by studying potential issues specific to these formats.
The second issue is related to the orientation of the mesh. The mesh needed to be outwardly oriented, and each point required an associated normal vector. To test this, we applied the Kinetic algorithm to some basic elements like a cube. We used Gmsh to define an initial cube without considering the orientation of the faces. After that, we created an enhanced cube using CGAL functions to orient the mesh. One remarks that if the cube face is wrongly oriented, it will result in incorrect normals, causing problems in generating the other faces. Some faces might not even be generated. We fixed this issue by performing preprocessing on the mesh. In the following, we present the results.
Another issue we tackled is the multiple self-intersection in the mesh. In this case, the algorithm would end up with a segmentation fault error. Here is an example of a self-intersected mesh. Only the two cubes self-intersecting worked out; other use cases with several cubes result in a segmentation error.
We used the kinetic algorithm to handle all the building geometry issues mentioned above. The methodology for implementing the Kinetic theory involves several key steps. First, the mesh is pre-processed to align with the algorithm’s requirements, ensuring compatibility. Throughout this process, it’s crucial to maintain the correspondence between the input and output labelling of the mesh elements. The methodology also includes an exploration of existing reports on Kinetic theory and its applications. Once problematic elements are identified, they will be addressed using the Kinetic Surface Reconstruction KSR algorithm implemented in the CGAL library. The latter algorithm allows the repair of mesh with different topological issues, such as non-watertight mesh, self-intersected mesh or mesh with holes. It takes as input a point cloud with associated normals representing a 3D model decomposes the point cloud into different detected planar shapes and reconstructs the mesh. Furthermore, metrics are defined to assess mesh quality, checking for properties such as closure and orientation, and comparing the bounding boxes of the original and output meshes. To do so, we examined the bounding boxes by extracting point sets from both the input and output for specific markers. The bounding box for each point set is then calculated, and the relative distance between the two bounding boxes is determined to estimate any differences. Finally, the analysis counts how many bounding boxes exhibit a difference greater or less than 10–20%, providing insight into the accuracy and consistency of the mesh reconstruction process.
In the following, we present the results of the process described above on a use case building.
Author: Cemosis – UNISTRA
Editor: Kyriaki Daskaloudi