Generative DfX in high-rise modular building: An expert-augmented cascade graph learning and optimisation approach

Professor Lu, Weisheng Wilson   (Project Coordinator (PC))

Dr Xu Jinying   (Co-principal investigator)

Dr Chen Junjie   (Co-principal investigator)

Professor Crolla Kristof Henri Chris Ghislain Jozef   (Co-principal investigator)

Professor Webster Christopher John   (Co-principal investigator)

Professor Xue Fan   (Co-principal investigator)

Dr Chu Yiu Leung   (Collaborator)

Dr. LAOVISUTTHICHAI Vikrom   (Collaborator)

Mr. LOU Jinfeng   (Collaborator)

Dai Jianguo   (Co-Investigator)

Gao Shang   (Co-Investigator)

Tan Tan   (Co-Investigator)

Dr Zhu Jingxiang   (Co-Investigator)

Professor Shen Geoffrey Qiping   (Co-Investigator)

36

2023-04-01

5310000

Generative DfX in high-rise modular building: An expert-augmented cascade graph learning and optimisation approach

1) High-rise modular building 2) Design for excellence 3) Generative design 4) Machine learning 5) Design knowledge management

Building and ConstructionDesign

Engineering (E)

C7080-22G

Collaborative Research Fund (CRF) - Group Research Project 2022/2023

2022

On-going

1. The proposed research aims to develop a generative design for excellence (DfX) methodology by focusing on the complexity of high-rise modular building (HRMB) and harnessing the power of expert knowledge and machine learning. It is positioned in the global trend of modular building renaissance to tackle the chronic issues such as housing shortages, widespread poverty, labour and building ‘double aging’, lacklustre productivity, poor occupational health and safety, and the logistics and supply chain crisis. It embraces the philosophies of design for manufacture and assembly (DfMA) and DfX made possible through technological advancements such as Industry 4.0, building information modelling (BIM), virtual design and construction (VDC), machine learning, optimisation techniques, and unprecedented computational power. Many of the opportunities instigated by the new design philosophies and the computing technologies have been much discussed but few of them have been adequately explored to solve challenging problems such as HRMB design. 2. The prevailing methodology in generative design is to generate design options first and then optimise them up to performance criteria. We adopt this methodology but with innovative adaptions. Our research integrates graph learning, advanced optimisation techniques, and design knowledge management into a self-contained cascade framework, whereby the three techniques are synergized to accommodate the hierarchical, reiterative, and complex nature of HRMB design. There are three interconnected lines of this research. The first line employs graph learning in a top-down manner to progressively generate designs from building (floor plan), to flat, and to module design options. The second line leverages advanced heuristic algorithms to optimise the generated design options towards excellence from the bottom up, i.e., from module, to flat, and finally to building (floor plan). Both the top-down generation and bottom-up optimisation will be augmented by the third line of research, which harnesses design knowledge and has human experts in the loop. 3. The project has five objectives/work packages (WPs): WP1: To build data infrastructure for generative DfX; WP2: To develop cascade graph learning models for generative HRMB design; WP3: To develop a multi-stage optimisation methodology for generative DfX; WP4: To augment the generative DfX methodology with expert design knowledge; and WP5: To encapsulate the methodology in a prototype platform and further refine it by conducting case studies in various DfX scenarios. 4. Five tangible deliverables can be expected:(a) An ontology of HRMB design knowledge (e.g., DfX grammars, genotypes, codes, rules, best practices, and performance metrics); (b) Design knowledge analytical tools (e.g., space syntax analytics, justified graph representation); (c) Cascade graph learning generative design models; (d) Multi-stage DfX optimisation algorithms; and (e) A prototype system comprising backend servers, plug-ins for mainstream design software, and Web-based BIM user interface to encapsulate the methodology and support its calibration and application. 5. The novelty of the research lies in the following aspects:(1) Applying generative design and DfX principles to challenging HRMB design problems; (2) Articulating the nexus of design option generation and optimisation in a two-way hierarchy formed by HRMB design from floor plan, to flat, to module; (3) Developing cascade graph learning models for generative design; (4) Using advanced heuristic algorithms for multi-stage DfX optimisation; (5)Harnessing existing expert knowledge in generative DfX; and (6) Re-centring the human in generative DfX to explore delivery of design value through human and machine collaboration. 6. The significance of this proposed research is both theoretical and practical. On the theoretical side, the research will add to our understanding of HRMB design by considering generative design and DfX. It exploits contemporary computing power in a field that is highly creative and predominantly conducted by human design professionals. It may open up a new design paradigm through which humans and machines work collaboratively to deliver design value. On the practical side, the research can assist designers to explore the potential design solution space more efficiently. It can help to unlock the full potential of HRMB in alleviating pressing societal issues as mentioned at the outset of this section.