Efficient binary serialization of IFC models using HDF5

Conference Contribution

Krijnen, T.F. & Beetz, J. (2016). Efficient binary serialization of IFC models using HDF5. ICCCBE2016: 16th International Conference on Computing in Civil and Building Engineering, Osaka, July 6-8, 2016 Read more: Medialink/Full text



The Industry Foundation Classes (IFC) are a common file-based open standard to describe Building Information Models. An IFC file can describe a building model to a level of detail suitable for production use and unite information pertaining to all stakeholders involved in a construction project. IFC files can possibly constitute up to gigabytes of data. Processing the full extent of this data can be time consuming. Considering the multi-disciplinary nature of our industry it may also be unnecessary for the use case at hand. Therefore, the retrieval of relevant subsets, whether spatially, based on discipline, or others, is necessary to effectively consume such datasets in downstream applications.
However, prevalent encoding forms of IFC models are text-based. And even though, in terms of file size, the most prevalent encoding, called IFC-SPF, can be rather efficient, by nature, it does not facilitate random access seeking in the file and no ordering is imposed to the definition of elements in the file. Therefore, at worst, the entire file needs to be traversed in order to find instances of interest. Furthermore, text-based data is slow to parse in comparison to its binary equivalent.
This paper introduces a binary serialization for IFC models as an alternative to prevalent text-based formats. It is based on an existing open standard called HDF5. An implementation for the translation of conventional IFC instance models into HDF5 is provided under and open source license. HDF5 is a binary and hierarchical data format. The hierarchical nature allows random access to specific instances. Other benefits include transparent compression and mechanisms for linking and mounting external files. The compressed HDF5 format yields a significant reduction of file sizes as compared to IFC-SPF models. In three use cases is assessed that extracting data from the model, can occur in near-constant time in relation to the size of the model, contrary to linear time using IFC-SPF models.
The translation into HDF5 files follows an existing ISO standardized mapping from EXPRESS instance models, the parent standard of IFC. The self-documenting nature of HDF5 enables incorporating additional attributes that are not part of the schema. In order to improve visualisation one can cache calculated information such as triangulated geometry for complex CSG geometries that are computationally complex to compute. In addition, incorporating inverse attribute values as part of the instantiation allows to further optimize the generation of subgraphs.