AI for BIM-based sustainable building design: Integrating knowledge discovery and semantic data modelling for evidence-based design decision support

Sustainable building design requires an interplay between multidisciplinary input and fulfilment of various criteria, which need to align into one high-performing whole: the building. Building Information Modelling has already brought a profound change in Architecture, Engineering and Construction by enabling efficient collaborative workflows. Combined with the power of statistical and symbolic Artificial Intelligence approaches (e.g. machine learning, semantic query techniques, inference machines, etc.) and the richness of data, these technologies can foster accurate prediction of design outcomes and help uncover valuable hidden knowledge in the performance of the existing built environment. Such knowledge has the potential to create an information cycle that can redefine building design and serve as valuable evidence for design decision support. However, despite the technological advancements, the gap between designed and measured building performance remains. Design decision-making still, to a large extent, relies on rules of thumb and previous experiences, and not on sound evidence. Consequently, design practice is neither sufficiently data-driven nor evidence-based. Performance mismatches also occur due to inaccurate predictions and assumptions, lack of data integration and sharing across domains, poor modelling and collaboration, etc. Research has investigated possible solutions to eliminate these causes, but few attempts have been made to sever the problem at its core- the lack of feedback loop from building operation to design. In response to the latter, this research effort attempts to unlock the potential of Artificial Intelligence approaches to establish the missing feedback loop and enhance human decision-making capabilities. Therefore, this thesis aims to demonstrate how knowledge discovery, representation and retrieval techniques can be integrated to create the missing link between building operation and design and inform sustainable BIM-based design decision-making in an evidence-based, context-aware and user-centred way. To achieve the research objective, the thesis presents an in-depth analysis of the diverse building data sources and types and outlines how the data can be analysed to discover valuable knowledge. Based on the results of that analysis and an extensive literature review, a framework for performance-oriented design decision support relying on BIM, data mining and semantic data modelling is proposed. Furthermore, motif discovery and association rule mining are performed on operational building data from two use case buildings to uncover performance insights. The discovered knowledge is then represented in an ecosystem of (semantic) data to create a knowledge base enriched with building performance patterns. A significant challenge, namely the interpretation of the discovered knowledge, is approached using linked data and crowdsourcing techniques, which results in contextualised networks of building data and knowledge annotated by human domain experts. Finally, the thesis demonstrates how the created knowledge ecosystem can reach the building design professionals through evidence-based recommendations based on semantic relatedness between concepts and determined by the users’ profile and context. As such, the presented future-proof holistic technological approach enables a robust user-centred mechanism that allows knowledge discovery, representation, contextualisation and reuse and achieves the targeted, evidence-based decision support in BIM-based sustainable design processes.