{"id":5298,"date":"2026-04-01T16:01:59","date_gmt":"2026-04-01T14:01:59","guid":{"rendered":"https:\/\/blog.innovation4e.de\/?p=5298"},"modified":"2026-04-01T16:02:01","modified_gmt":"2026-04-01T14:02:01","slug":"digital-buildings-part-4-knowledge-graphs-and-ai-for-automated-inventory-assessment","status":"publish","type":"post","link":"https:\/\/blog.innovation4e.de\/en\/2026\/04\/01\/digital-buildings-part-4-knowledge-graphs-and-ai-for-automated-inventory-assessment\/","title":{"rendered":"Digital Buildings \u2013 Part 4:\u00a0Knowledge Graphs and AI for Automated Inventory Assessment"},"content":{"rendered":"\n

Technologies such as the IoT, artificial intelligence (AI), and big data can help buildings operate more efficiently and sustainably. Fraunhofer ISE is conducting research across various working groups on the application of AI and digitalization in the construction industry. What opportunities do these developments offer for a green transformation of the building sector? How can digital processes help address key challenges in the construction industry, such as the shortage of skilled workers and cost pressures?<\/span><\/i> <\/span><\/p>\n\n\n\n

Our interview with ISE digital researcher Simon G\u00f6lzh\u00e4user takes the \u201cDigital Buildings\u201d series on our Innovation4E blog\u2014which began as a three-part series\u2014into its next phase.<\/span><\/i> <\/span><\/p>\n\n\n\n

You work in the \u201cCognitive Buildings<\/a>\u201d research group at Fraunhofer ISE. What are your\u00a0main areas\u00a0of focus?<\/h2>\n\n\n\n

I conduct research on methodologies for the efficient and automated digitization of building data, with a focus on building\u00a0services, as well as their\u00a0subsequent\u00a0digital structuring and modeling. Especially in existing buildings, data on building\u00a0services \u2013 like heating,\u00a0ventilation\u00a0and air conditioning systems (HVACs) –\u00a0is in most cases incomplete and scattered across highly heterogeneous sources and formats, which often makes the\u00a0initial\u00a0assessment at the start of renovation projects or operational\u00a0efficiency\u00a0optimizations\u00a0a laborious task. In addition, I focus on time series forecasting as well as data-driven system\u00a0identification\u00a0and control of building\u00a0energy\u00a0systems using neural networks. Digital system models can be helpful here, as they\u00a0contain\u00a0detailed knowledge about a system\u2019s functions and topology.<\/p>\n\n\n\n

Digital technologies offer\u00a0great potential\u00a0for the energy transition in the building sector. Which approaches excite you?<\/h2>\n\n\n\n

A major challenge in digitizing and structuring building data lies in how the collected information is digitally represented. The use of knowledge graphs as a data model, in conjunction with Semantic Web ontologies, offers\u00a0great potential\u00a0here. In a knowledge graph, entities\u00a0appear as the\u00a0nodes\u00a0of the graph, while\u00a0the\u00a0edges\u00a0represent\u00a0the relationships between the entities. If the model includes, for example, a pipe connected to a pump, then the pipe and the pump each form a node in the graph.\u00a0The relationship \u201cis connected to\u201d is in turn modeled by a directed edge from the \u201cpipe\u201d node to the \u201cpump\u201d node.\u00a0In a dictionary-like manner, ontologies then\u00a0provide clearly defined terms for the available entities and their relationships within a specific domain,\u00a0e.g.\u00a0for building\u00a0energy\u00a0systems.\u00a0<\/p>\n\n\n\n

What are the advantages of modeling building and facility information using knowledge graphs?\u00a0<\/h2>\n\n\n\n

Digital information can not only be captured in a knowledge graph, but also semantically linked to one another in a machine-readable format. This results in a functional digital model of the facility within the context of building\u00a0energy\u00a0systems. This can be used for a wide variety of applications:\u00a0For example, for automated verification of the system\u2019s topology and the design of its components, or as a basis for control and fault detection\u00a0algorithms\u00a0in the context of monitoring and operational optimization. A functional system model in the form of a knowledge graph can also be used to create a digital twin of the system or the building. In conjunction with\u00a0appropriate ontologies, this yields a uniform and standardized digital representation of the captured information that is reusable, expandable, and traceable.\u00a0For existing buildings in particular, knowledge graphs thus offer the possibility of digitizing building data in a lean and flexible manner.\u00a0<\/p>\n\n\n\n

Your latest innovation is an AI-based app for digitizing system diagrams. How did this come about?<\/h2>\n\n\n\n

System diagrams<\/span> (or P&ID diagrams) <\/span>are an essential part of the documentation for building <\/span>energy<\/span> <\/span>systems. They describe the layout of the installed systems as a schematic diagram and, in <\/span>most<\/span> cases, are available for existing buildings only as image files or PDFs<\/span>, <\/span>or even just in paper form. If a renovation or operational<\/span> efficiency<\/span> optimization is to be planned for a building, the system diagram must first be manually <\/span>drawn<\/span> <\/span>into the <\/span>appropriate software<\/span>, which is <\/span>a time-consuming <\/span>manual <\/span>process. This is exactly where we <\/span>started with<\/span> our \u201c<\/span>DiMASH<\/span><\/a>\u201d project.<\/span><\/span><\/p>\n\n\n\n

How exactly does the app work?<\/span><\/span><\/h2>\n\n\n\n

We have developed AI-based algorithms and tools that use advanced <\/span>C<\/span>omputer <\/span>V<\/span>ision and image processing techniques to scan a <\/span>system diagram<\/span> and convert the content into a digital model of the <\/span>system<\/span> <\/span>with minimal user interaction. The captured data is structured as a knowledge graph based on an ontology also developed within the project, which <\/span>consolidates<\/span> and extends the concepts of publicly available ontologies <\/span>established<\/span> in the field of application. The digital <\/span>system<\/span> <\/span>model can then be further edited in the app and, for example, <\/span>directly linked<\/span> to time-series data from a monitoring system.<\/span><\/span> <\/span><\/p>\n\n\n\n

\"\"<\/a>
From analog system diagrams to digital system models \u2013 the app from Fraunhofer ISE\u2019s \u201cDiMASH\u201d project makes it possible (visualization of the prototype). \u00a9 Scherr+Klimke AG<\/figcaption><\/figure>\n\n\n\n

You want to continue developing the app in collaboration with software manufacturers. What is the added value for future partners and end customers?<\/h2>\n\n\n\n

For end customers, such as <\/span>engineering design offices<\/span>, our <\/span>prototype<\/span> <\/span>and the AI-based digitization pipeline it <\/span>contains<\/span> <\/span>offer <\/span>significant time<\/span> savings during the <\/span>stocktaking <\/span>process, which can reduce overall <\/span>job<\/span> <\/span>processing time and increase throughput. Potential partners from the software industry, such as manufacturers of relevant CAD design software, can offer their customers unique functionality with significant added value by integrating the tools developed in \u201cDiMASH\u201d into their portfolio.<\/span><\/span> <\/span><\/p>\n\n\n\n

Where do you see further room for development?<\/span><\/span><\/h2>\n\n\n\n

In current and future projects, we aim to further explore the potential of <\/span>L<\/span>arge <\/span>L<\/span>anguage <\/span>M<\/span>odels (LLMs) for capturing, digitizing, and processing unstructured building data. In addition to system diagrams, there are many other diverse sources of information <\/span>regarding<\/span> building <\/span>energy<\/span> <\/span>systems and buildings in general. One example is data point lists. These are tables that describe all measured values and control variables of a system, such as temperature sensors or valve positions. These tables each follow specific naming conventions, for which, however, there are no universal standards<\/span> <\/span>established<\/span>. This means that the information <\/span>required<\/span> to set up system <\/span>monitoring<\/span> usually <\/span>must<\/span> be <\/span>tracked down<\/span> manually and assigned to the correct system components. In the \u201c<\/span>GraphEET<\/span><\/a>\u201d project, we are therefore investigating how monitoring processes<\/span> <\/span>can be automated<\/span> <\/span>based on data point lists<\/span> and related<\/span> documentation<\/span> using LLMs and knowledge graphs.<\/span><\/span><\/p>\n\n\n\n

Title photo: \u00a9 iStock.com \/ Henvry<\/p>\n","protected":false},"excerpt":{"rendered":"