Machine learning integrated portfolio-based strategic building asset management

The data for strategic asset management (SAM) contains multi-functional features and is usually built with a complex structure of attributes. Effective management and utilisation of the building data are essential for monitoring projects in a sustainable and whole lifecycle manner and avoiding potential risks for all the project stakeholders. A common challenge FM professionals face is exposure to an information-saturated and data-rich facility management environment. Yet, a limited amount of research has focused on quality and project portfolio-level management of the building data required for SAM services. Even fewer research studies have examined how building data migration, collection and management processes can benefit machine learning (ML) related technologies based on the existing unstructured project data. This thesis aims to explore how the facility management industry can improve portfolio-based strategic asset management/planning by applying ML algorithms in asset data collection and management processes. By applying an abductive research approach, this study first uses a case study approach to evaluate the current portfolio-based SAM practice and data quality. Then, the preliminary research outcome from more than ten different projects (21 for image classification and 12 for text classification) is gathered and further processed to evaluate the effectiveness of image and text classification applications over data collection and management processes of portfolio-based SAM. The preliminary findings from the case study confirm the challenges of current portfoliobased SAM, which include: (1) the lack of any guiding framework for SAM professionals to control their documentation flow; (2) data interoperability issues across different projects; (3) the lack of sufficient data entries for conducting some core SAM services, and; (4) the lack of utilising portfolio-based management intelligence to power the SAM decision making. It is found in the second phase study that deep-neural-based text and image classification algorithms are effective in supporting data interoperability and in remedying the lack of insufficient data attributes issue (e.g., the lack of 'manufacture' information) raised across the project portfolio. This is because the ML-enabled approach can provide an automated data collection solution, combined with the developed SAM information management frameworks and structures. This study contributes to the body of knowledge as it (1) both theoretically and empirically evaluates the data management in different aspects of effective portfolio-based SAM, (2) validates the applicability of applying deep-neural-based image and text classification algorithms for automated data migration, collection, and management processes and (3) develops a structure to bridge SAM documentation flow and ML applications.