Abstract
Testing the preliminary design of a building is usually carried out to ensure that the requirements for its programme, space sizes, adjacencies, travel distances and access to daylight are satisfactorily met. Increasingly, this task is being implemented through automated plan layout generation tools. However, these tools struggle to create realistic and functional floor plans. In this paper, our aim is to generate plausible floor plans with fully functional circulation systems by combining Machine Learning (ML), evolutionary optimisation and parametric methods. We focused on generating buildings with functioning circulation, consisting of primary circulation with building entrances, horizontal and vertical connections and secondary circulation that connects all the rooms. We started by analysing school buildings, as a test case scenario, which showed that their zoning, circulation and topological relationships consistently follow design patterns. This enabled us to specify algorithmic rules for room placement. The floor plans also allowed us to generate the dataset needed to train a neural network and a decision tree. To achieve this, we traced the overall zones (functional and circulation spaces) of the ground floor layout and encoded it with its main features such as its area-to-perimeter ratio and the shape of its convex hull. We also varied the geometry of the layout without modifying its topology to add more training test cases for the neural network. Once the neural network and decision tree have been trained, our implementation proceeded in three main phases. First, the user inputs the approximate building shape using a simple closed curve. Selecting either the trained neural network or the decision tree, the user is presented with a matching building layout from a gallery which is geometrically transformed to closely fit the user’s sketch. Second, the spatial program is packed into the overall layout parametrically, using an algorithm based on the aforementioned set of rules. Finally, three fitness criteria (adjacencies, travel distances and daylight) are measured and optimised using an evolutionary solver. The result is a novel and optimised plausible floor plan with a functional circulation system. While the current workflow is designed to generate only the ground floor, we plan to expand it to generate multi-storey buildings. Our analysis has found that extracting design patterns, matching sketches using ML and optimising the result using an evolutionary solver is significantly more efficient than manual processes. The paper concludes with a discussion of the potential of integrating artificial intelligence in the early phases of design and the need to share standardised and richer building data beyond architectural projection drawings.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Chaillou, S. (2020). Archigan: Artificial intelligence x architecture. Architectural intelligence (pp. 117–127). Springer.
Chen, Q., et al. (2020). Intelligent home 3d: Automatic 3d-house design from linguistic descriptions only. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.
Ekici, B., et al. (2019). Performative computational architecture using swarm and evolutionary optimisation: A review. Building and Environment, 147, 356–371.
Green, O. (2020). An introspective approach to apartment design.
Hu, R., et al. (2020). Graph2Plan: Learning floorplan generation from layout graphs. arXiv preprint arXiv:2004.13204
Liu, Y., et al. (2020). Exploration of campus layout based on generative adversarial network. In The international conference on computational design and robotic fabrication. Springer.
Nauata, N., et al. (2020). House-GAN: Relational generative adversarial networks for graph-constrained house layout generation. arXiv preprint arXiv:2003.06988
Para, W., et al. (2020). Generative layout modeling using constraint graphs. arXiv preprint arXiv:2011.13417
Schwartz, Y., et al. (2021). A decision support tool for building design: An integrated generative design, optimisation and life cycle performance approach. International Journal of Architectural Computing, 1478077121999802.
Steve Baer, S. D. (2021). Hops component. https://developer.rhino3d.com/guides/grasshopper/hops-component/
Tian, R. (2020). Suggestive site planning with conditional GAN and urban GIS data. In The international conference on computational design and robotic fabrication. Springer.
W3Schools. (2021). Machine learning—Decision tree. [cited 2021]. https://www.w3schools.com/python/python_ml_decision_tree.asp
Yao, J., et al. (2021). Generative design method of building group-Based on generative adversarial network and genetic algorithm.
Zhao, C., et al. (2021). Two generative design methods of hospital operating department layouts based on healthcare systematic layout planning and generative adversarial network. Journal of Shanghai Jiaotong University (science), 26(1), 103–115.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2023 The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
About this paper
Cite this paper
Marcinkeviciute, D., Jabi, W. (2023). Plausible Layout Generation Using Machine Learning, Evolutionary Optimisation and Parametric Methods. In: Mora, P.L., Viana, D.L., Morais, F., Vieira Vaz, J. (eds) Formal Methods in Architecture. FMA 2022. Digital Innovations in Architecture, Engineering and Construction. Springer, Singapore. https://doi.org/10.1007/978-981-99-2217-8_14
Download citation
DOI: https://doi.org/10.1007/978-981-99-2217-8_14
Published:
Publisher Name: Springer, Singapore
Print ISBN: 978-981-99-2216-1
Online ISBN: 978-981-99-2217-8
eBook Packages: EngineeringEngineering (R0)