in high-rise modular building:
An expert-augmented cascade graph learning
and optimisation approach
This project aims to develop a computer-aided, designer-oriented generative DfX methodology for HRMB. It integrates graph learning, advanced optimisation techniques, and expert design knowledge in a cascading fashion to suit the complex, hierarchical, and reiterative nature of HRMB design.
Prof. Wilson Lu
01 Apr. 2023
31 Mar. 2026
High-rise modular building / Design for excellence / Generative design / Machine learning / Design knowledge management
About the Research
High-rise modular building (HRMB) is highly advocated to address the housing crisis in high- density cities around the world. In line with this advocate is the principle of design for excellence (DfX) that is vigorously explored to unlock the full potential of modular building. DfX encompasses ‘excellence’ criteria such as functionality, ease of manufacture and assembly, logistics, sustainability, and cost, which require multidisciplinary domain knowledge that is beyond the capability of any single designer. Computer-aided generative design seems to provide a promising strategy to handle the multifaceted knowledge requirement. However, scant attention has been paid to exploring how it can assist designers to achieve DfX, particularly in the complex context of HRMB.
This project aims to develop a computer-aided, designer-oriented generative DfX methodology for HRMB. It integrates graph learning, advanced optimisation techniques, and expert design knowledge in a cascading fashion to suit the complex, hierarchical, and reiterative nature of HRMB design. We employ graph learning in a top-down manner to progressively generate a rough building (floor plan), then flat design, and finally detailed module design. It then leverages advanced heuristic algorithms to optimise the generated design options from the bottom up, i.e., from module to flat and ultimately to building (floor plan). This approach mimics real-life design practice involving reiteration until the best or optimal solution is found. Since the reiterative processes are computationally onerous, smart algorithms will be devised to help designers explore the solution space effectively. These processes will be augmented by design knowledge and include human experts in the loop. We will pilot the research in Hong Kong’s HRMBs, a rich context for considering DfX in relation to factors including user groups, available construction technologies, manufacturing capacity, logistics, and site conditions.
The deliverables of this project are: (a) an ontology of HRMB design knowledge; (b) design knowledge analytical tools; (c) cascade graph learning models for generative design; and (d) DfX optimisation algorithms, which will be encapsulated into (e) a prototype system for calibration and application. The research will deepen our understanding of HRMB design by considering a wide range of excellence criteria, and may open up a new design paradigm through which humans and machines collaborate to deliver design value. The research can assist designers to explore potential design solutions more effectively. It can help unlock the full potential of HRMB in alleviating contemporary issues including housing shortages, widespread poverty, and productivity decline.