A Framework for generating building models for graph-based neural networks

This Thesis engages in the investigation of a graph-based approach for the creation of realistic building models from an Artificial Intelligence. In the scope of this study a methodology was developed that allows the automated creation of building models in Industry Foundation Classes (IFC) format. Therefore, a thorough investigation of the IFC schema is presented. The creation of the files is done with a .NET application that is written in the programming language C#. The software library XBIM Essentials offers fundamental functions, which are designed specifically for the development of IFC schema conform software solutions to manipulate building models. Additionally, a software solution was developed that reads data from IFC models and imports them into a graph database. Finally, graph-based neural networks, like generative adversarial networks (GraphGANs) and graph recurrent neural networks (GraphRNN) are evaluated for learning the IFC building models structure and relationships as well as generating realistic models.     «

This Thesis engages in the investigation of a graph-based approach for the creation of realistic building models from an Artificial Intelligence. In the scope of this study a methodology was developed that allows the automated creation of building models in Industry Foundation Classes (IFC) format. Therefore, a thorough investigation of the IFC schema is presented. The creation of the files is done with a .NET application that is written in the programming language C#. The software library XBIM...     »