Project Details
Projekt
metis II - Artificial intelligence methods for auto-completion of designs based on semantic building information (BIM) for supporting architects in early design phases.
Subject Area
Structural Engineering, Building Informatics and Construction Operation
Term
from 2019 to 2024
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 419390235
The aim of the "metis II" project is to develop methods for auto-completion of building designs. Using artificial intelligence approaches, (partial) information is extracted from reference designs and proposed to the architect as additions to his own design. Thus, methods are developed to apply useful information from a semantic building model (BIM) in the design context. Methods are examined to suggest e.g. kitchens, corridors or bathrooms and e.g. their location to a given formalized spatial configuration of e.g. a living room and a bedroom. The methods to be developed must be able to recognize both the context of the building design and the user-specific context. For this purpose, methods of artificial intelligence (AI), especially case-based reasoning (CBR) and artificial neural networks (KNN), are applied, expanded and developed. The explainability of the system to be developed (XAI - Explainable AI) is a further focus of the project, since the automatically generated solution parts as well as the learned knowledge must be explained to the user - in addition to the usual search results.For the integration of (partial) information from digital semantic building information models (BIM) as semantic and topological design specifications (spatial arrangement), the steps of the CBR cycle are fully integrated into the design process and new methods for the domain-specific adaptation of CBR knowledge containers (case basis, similarity measure, vocabulary and adaptation knowledge) are developed. The current state of (CBR)-research results in acquisition and administrative deficits: A special challenge of CBR, especially of retrieve and retain steps, is the size and quality of the case base, since the largest and most high-quality database possible is required. However, the data must be acquired, processed and at the same time the quality of the cases must be ensured. In order to increase the quality of the database, deep-learning methods for targeted "forgetting" are examined in the project "metis II".The project "metis II" is based on the results of metis I ("metis - Knowledge-based search and query methods for the development of semantic information models (BIM) for use in early design phases" funded by the DFG from 2013-2017. In "metis I", approaches for drawing a retrieval for semantic building models (BIM) were investigated and methods were developed to process the information. The concept of semantic building fingerprints used for this has proven to be robust enough and the methodological approaches have been confirmed.
DFG Programme
Research Grants
Co-Investigators
Professor Dr. Klaus-Dieter Althoff; Professor Dr.-Ing. Frank Petzold
