You are on page 1of 8

Available online at www.sciencedirect.

com

ScienceDirect
Available online at www.sciencedirect.com

ScienceDirect
Procedia Manufacturing 00 (2019) 000–000
www.elsevier.com/locate/procedia
Procedia Manufacturing 44 (2020) 132–139

1st International Conference on Optimization-Driven Architectural Design (OPTARCH 2019)

Computer-aided approach to public buildings floor plan generation.


Magnetizing Floor Plan Generator.
Gavrilov Egora*, Schneider Svenb, Dennemark Martinc, Koenig Reinhardd
a
Bauhaus-University Weimar, 1st Novopodmoskovniy str. 4-46, Moscow 125171, Russian Federation
b
Bauhaus-University Weimar, Geschwister-Scholl-Strasse 8, 99423 Weimar, Germany
c
Bauhaus-University Weimar, Geschwister-Scholl-Strasse 8, 99423 Weimar, Germany
d
Bauhaus-University Weimar, Geschwister-Scholl-Strasse 8, 99423 Weimar, Germany

Abstract

For architects as well as developers and urban planners working on the floor plans or estimating the shape and
dimensions of large buildings is always a challenge. This task requires some knowledge and what is more important
– even with understanding of the process it is a very time-consuming task. One should take into consideration the
arrangement of all rooms as well as adjacencies and connections of main spaces. Presented project can be
considered as an exploration of various ways of generating floor plans for public buildings, which was followed by
creating a new algorithm for solving that task. Public buildings were chosen as a main target of conducted research
because of their complex and non-standardized structure. The aim was to try different previously described
approaches, invent completely new techniques and methods, choose the best of them and incorporate them into our
own generator. The creation of working generator prototype was an important target of the project and should not be
considered as its final stage, since this trial revealed some positive and negative sides of chosen approach and
therefore serves as an intermediate stage. Floor plans analysis was conducted in order to understand the structural
patterns of communication spaces in built public houses. One of positive outcomes of this investigation is that public
buildings do not share a lot of features in common, therefore this field of exploration is rather unobstructed and
provides a lot of possibilities for experiments.

* Corresponding author. Tel.: +7-903-6772744; +49-1522-3293010.


E-mail address: hellguz@gmail.com, egor.gavrilov@uni-weimar.de

2351-9789 © 2019 The Authors. Published by Elsevier B.V.


This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under responsibility of the scientific committee of the 1st International Conference on Optimization-Driven Architectural Design

2351-9789 © 2020 The Authors. Published by Elsevier B.V.


This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under responsibility of the scientific committee of the 1st International Conference on Optimization-Driven Architectural Design
10.1016/j.promfg.2020.02.214
2 Gavrilov E., Schneider S., Dennemark M., Koenig R./ Procedia Manufacturing 00 (2019) 000–000

Gavrilov Egor et al. / Procedia Manufacturing 44 (2020) 132–139 133

©
© 2019
2020 The Authors. Published
The Authors. Published by
by Elsevier
Elsevier B.V.
B.V.
This
This is
is an
an open
open access
access article
article under
under the
the CC
CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review
Peer-review under
under responsibility
responsibility ofof the scientific committee
committee of thethe 1st
1st International
International Conference
Conference on
on Optimization-Driven
Optimization-Driven
Architectural
ArchitecturalDesign
Design

Keywords: MagnetizingFPG; Floor plan generator; Floor plan generation; Evolutionary strategy; Magnetizing Floor Plan Generator

1. Introduction

Generally, for every different shape or position of a building’s footprint a whole new room structure should be
created. In this stage of building development, the final room program of the building is usually not completely
defined yet and can be subject to change. This uncertainty in the shape and room program of the building means that
a lot of draft and experimental solutions can be produced and later discussed and compared by the interdisciplinary
work team. However, as stated previously, the process of making viable solutions is rather time-consuming. At the
same time, this developmental stage does not install many strict constraints on buildings’ inner structure and overall
appearance, so it opens possibilities for computer-aided strategies of floor plan generation.
Specific features in the floor plans of public buildings include relatively complex connectivities which consist of
linked corridors, halls, foyers and vertical communication cores. Special requirements are usually applied to these to
ensure that the navigation inside the building is not too difficult for people. These connection structures (which
usually have different configurations in different building cases) should be considered at all stages of development,
since they provide the general layout of all main inner spaces. Taking this into account, one can argue that a specific
approach is required for floor plan making in public buildings – be it a human-generated plan or a computer-
generated one.
The objective of presented research was to understand the current development state of computer-based
approaches for floor plan generation and more importantly, to propose a new approach based on desired qualities:
flexibility, ease of use, maximal diversity of results and speed of generation. The resulting generator meets all these
qualities and shows the potential of searching for new methods in this field of researches.

2. Previous methods overview

The topic of algorithms and computer-based approaches for large buildings plan generation is currently exposed
insufficiently in the field of scientific publications and thus needs to be discussed. Existing approaches for general
floor plan generation include constraint-based systems, agent-based systems, physically-based systems, evolutionary
algorithms etc. [1–5]. Some of these solutions provide rather powerful frameworks for generating small-scale plans,
such as plans of individual houses or chosen flats. All these solutions have their advantages and disadvantages;
however, only a few of them can generate relatively complex connection structures. Usually this means that these
algorithms cannot be used for offering variants of floor plans of public buildings due to reasons which were
uncovered previously.

3. Approach applied to find new working solutions

The question arises: What kind of algorithm could be applied to approach the set task? Research was conducted
to get a list of different techniques which could be in this context. The main idea was to invent as many
undiscovered and unpredictable approaches as possible and then to try mixing them in all different ways. This study
helped a lot to get several techniques which were not considered before. Naturally, most of them were not actually
usable due to different reasons, but this “playground stage” of the research was crucial for new insights to appear.
The factor of ‘surprise’ generated by computer and not hardcoded by developer, described in other papers [6], was
highly important in our work too. The most promising solutions were described step-by-step and then implemented,
Gavrilov E., Schneider S., Dennemark M., Koenig R / Procedia Manufacturing 00 (2019) 000–000 3

134 Gavrilov Egor et al. / Procedia Manufacturing 44 (2020) 132–139

tested and compared with each other. One of the approaches was translated into the generator which will be
introduced in this paper. It can be considered either as a separate project or as a base of a larger multi-level
generator, which can incorporate a number of different approaches for each of the task of floor plan creation [7].
Another important consideration would be the degree to which computer-based solutions could stay helpful for
solving real tasks. The most promising solution would be an algorithm which is capable of handling interactions and
changes from the user side on each of its steps, which can also complete the task by itself. However, the perspective
of making a tool which serves as one of the instruments in hands of an architect, developer or planner seems much
more realistic and reachable. This was taken into consideration as a general idea of the generator development.

4. MagnetizingFPG (Magnetizing Floorplan Generator)

One of the main outcomes of the research is the working floor plan generator. It is based on the previously done
work and thus incorporates several techniques which were discovered. Some of the ideas appeared in the very
beginning of the research, and some of them were discovered later. The generator is now in a development phase,
that is why the final description is subject to change, as well as the goals and outcomes of the research. The main
inputs of the generator include the room program with all areas, all required room connections, entrance point and
the boundary of the site. It works on a square grid, though the size of a cell can be altered to get more precise or
faster results.
The following statement was entered as a starting point: Each of the rooms in a building is somehow accessible
from any other room through a corridor. It means that the whole communication structure is interlinked and there is
one corridor structure, connected to all rooms. Of course, this statement cannot be said about all real-world
buildings. Two rooms can be connected directly, or through an inner corridor, or through another room, or they can
even share the same space and have no division (in this case the division is rather implicit) etc. Nevertheless, the
simplified structure was considered as more comprehensible representation for the first step of generation.
One can argue that this simplification can ruin the whole analysis that was briefly described in the Introduction.
However, if we take a closer look at the simplified structure that we operate with, we find a striking resemblance
with the common evacuation plan of the building (Fig. 1.). Since its main mission is to provide the fastest route
outside the building, it basically looks like a scheme in which every room is connected with the entrance. So, it
could be said that the first step of the described generation would be developing an evacuation plan, which can later
be converted into more intelligible communication network.

Fig. 1. Example of an evacuation plan in comparison to generated plan


4 Gavrilov E., Schneider S., Dennemark M., Koenig R./ Procedia Manufacturing 00 (2019) 000–000

Gavrilov Egor et al. / Procedia Manufacturing 44 (2020) 132–139 135

Fig. 2. Visual representation of one iteration. Room R1 should be adjacent to Kitchen, Laundry and WC

4.1. The base method:

1. A corridor is attached to each of the rooms. It can be placed along one, two or four sides of the room.
Corridors later form the whole communication network between rooms.
2. The first entrance room is placed near the entrance point.
3. All rooms which are adjacent to already placed room are sorted by the number of their total adjacencies.
The room R1 which has the most adjacencies (and therefore is the most ‘constricted’ room and the hardest
to place) is placed first. The whole structure of already placed corridors is analysed for this; the cells, which
meet further requirements can be considered as cells for placing a new room (Fig. 2):
a. A cell should be a part of a corridor structure
b. A cell should be not further away from all adjacent rooms of a room R1 than a given threshold
(values 2-4 work best)
c. There should be space around a cell, so the new room can be placed there
When a number of suitable cells are found, one of them is chosen randomly and the room R1 is placed near
this cell. It’s important that the corridor of the room R1 should be connected to the whole corridor structure.
4. If room R1 was placed successfully, then return to step 3. and place all other rooms one-by-one in the same
way.
5. If the next room cannot be placed, it means that either previous rooms were placed not compact enough or
there is no space left inside the boundary. To optimize the search of best placement solution we introduce
the iteration mechanism.

4.2. The iteration mechanism:

1. Execute the base method 3-5 times, save obtained branches of variants in a list
2. Sort the results according to the evaluation function (generally number of the rooms placed or the total
area of the rooms placed provide the most comprehensive evaluation results)
3. For 2-3 best branches do the following:
Gavrilov E., Schneider S., Dennemark M., Koenig R / Procedia Manufacturing 00 (2019) 000–000 5

136 Gavrilov Egor et al. / Procedia Manufacturing 44 (2020) 132–139

a. remove last 1-5 placed rooms


b. continue the method from this point, try to place as much rooms as possible
c. compare results with the parent-variant
d. if resulting solutions are better, add them to the main list
4. Now we have more variants in our list, we can sort them again and keep only 3-5 best of them
5. One iteration is finished, the next iteration means returning to step 2 and doing all further steps again

The process of iterating serves as a quasi-evolutionary strategy, the importance of which in the field of floor plan
generation was described by other authors [8].
After the basic generation is finished some minor enhancements are made to the final solution, including
removing dead ends from the corridor system, adjusting the working grid to the given site boundary, adding defined
halls/foyers to the corridor structure and some other slight changes.
In its current state the generator can fit all or almost all the rooms into the boundary in couple of seconds or
couple of minutes, if the boundary is relatively large. A lot of tests were conducted, and results are quite impressive
already, considering the possible improvements which would be added later.

5. Structure representation in Grasshopper workspace

Ease of use was considered as a crucial feature since the beginning of a project. Therefore, a simple solution for
managing a room program was created for the Grasshopper environment (Fig. 3). It consists of a schematic room
program and a few number sliders and other inputs, which are helping to control the basic values for generation.
This enables user to set basic parameters, such as a room’s name, room’s area, room’s dimensions, connections
between rooms, entrance location, type of each space (room/hall). Grasshopper and Rhinoceros were chosen as the
working environment, because they are quite commonly used for tasks of optimization or generation in architecture
and provide the needed flexibility.
The resulting solution provides two main components: HouseInstance and RoomInstance. The HouseInstance
component acts as a container for all the information about one single room structure and connects directly to the
MagnetizingFPG (Magnetizing Floorplan Generator) component. HouseInstance holds such information as the
house’s boundary and entrance point and referring topological structure of rooms.
This topological structure is presented by RoomInstance components, which contain information about each
single room – name, area, id, dimensions (optional), type of space (room/hall). These RoomInstance components
can be connected to each other with colorful wires, which color is changed based on the color of house, to which
rooms are assigned.

Fig. 3. Grasshopper environment with RoomInstance and HouseInstance components


6 Gavrilov E., Schneider S., Dennemark M., Koenig R./ Procedia Manufacturing 00 (2019) 000–000
Gavrilov Egor et al. / Procedia Manufacturing 44 (2020) 132–139 137

Fig. 4. Generation preview in Rhinoceros 6 environment

6. Further development options and strategies

Though the MagnetizingFPG algorithm already reached some important milestones, we absolutely agree, that a
lot of practical questions remain unanswered. The first and the most relevant is how the algorithm operates with
different floors. The basic idea is to ‘freeze’ some rooms like staircases, restrooms and kitchens in the same place on
each of the floors and develop the remaining structure around these spaces. The only apparent problem is to make
sure that the corridor structure will be interlinked, since it is crucial for the whole algorithm. One of the possible
solutions would be to connect parts of the structure together by creating new corridors in case it will be disconnected
and then check the remaining adjacencies.
The second question is how we can make the resulting floorplan more compact; namely, how we can fit it into
the given boundary, so that no unintentional corners or edges appear. This task was partially solved by using the
method of spring-based systems, fully described in other papers [9,10].
The other important aspect is optimizing the base method of placing rooms. Since the complexity of each
iteration increases with each improvement, new optimizations should be introduced to speed up the whole process.

7. Overview

The outcomes of current development phase have already proved the applicability of the chosen approach,
although a lot of work is still required for stable and more controlled results. The research and the generator
development led to the discovery of some previously unconsidered though important questions. For example: how
strong is the connection between the topological building floor plan and its geometrical plan [11]? Is the set of
connection structure, room areas and boundary enough for generating relatively viable plan? How one can overcome
the restrictions set by square grid? Could an algorithm generate a better communication structure than human can?
All of these questions are highly relevant to the research and thus need to be answered, or at least investigated.
Magnetizing Floor Plan Generator serves as the first milestone for this research, since it accomplishes all the
goals that were set on the beginning of this project and produces unusual though quite rational results. It also serves
Gavrilov E., Schneider S., Dennemark M., Koenig R / Procedia Manufacturing 00 (2019) 000–000 7

138 Gavrilov Egor et al. / Procedia Manufacturing 44 (2020) 132–139

as a tool for exploring the possibilities of packing large amount of rooms into complex boundary and for studying
the relation between user inputs and a resulting plan.
Aside from reaching the goals of creating a generator that inherits the stated qualities – flexibility, ease of use,
maximal diversity of results and speed of generation – other outcomes of the research can be considered as helpful
achievements for further works.
The system of room structure representation in Grasshopper environment proved to be intuitive and quick to use.
It is extremely helpful for storing and managing a great amount of values while working with a floor plan generation
for a large building. We also received feedback from some users who already started using this generator in their
projects in Grasshopper and were able to tweak the named structure of room program representation in Grasshopper
considering their needs and requests.
The study of corridor and hall structures in public buildings, which is not fully elaborated in this paper, presented
some interesting results worth considering in the next part of the research. For example, such structures could look
very thoughtful and easy to comprehend. In contrast, some look rather like a very complex system of corridors and
dead ends. It often depends on building type and size, but in some cases such different systems are dictated by
building form and a complex room adjacencies structure. The interesting part is how the presented generator also
builds corresponding, similar structures based on the given input. A detailed study of these relations between inputs
and resulting plan will be also conducted as part of the research.

Fig. 5. Generated variants, which were created on base of plans of iconic houses. The original structure of these houses was translated to
topological structure, which was used as an input to the MagnetizingFPG. Here are presented some of the most interesting and diverse results of
generation without any other additional changes.
8 Gavrilov E., Schneider S., Dennemark M., Koenig R./ Procedia Manufacturing 00 (2019) 000–000

Gavrilov Egor et al. / Procedia Manufacturing 44 (2020) 132–139 139

References

[1] Schneider S, Fischer J-R, Koenig R. Rethinking Automated Layout Design: Developing a Creative Evolutionary Design Method for the
Layout Problems in Architecture and Urban Design. Des. Comput. Cogn. ’10, 2011, p. 367–86. doi:10.1007/978-94-007-0510-4_20.
[2] Koenig R, Schneider S, Knecht K. KREMLAS: Entwicklung einer kreativen evolutionären Entwurfsmethode für Layoutprobleme in
Architektur und Städtebau. 2012.
[3] Nourian P, Rezvani S, Sariyildiz S. Designing with Space Syntax: A configurative approach to architectural layout, proposing a
computational methodology 2013;1:357–65.
[4] Donath D, Böhme LFG. Constraint-Based Design in Participatory Housing Planning. Int J Archit Comput 2008;6:97–117.
doi:10.1260/147807708784640081.
[5] Schaffranek R. Parallel planning An experimental study in spectral graph matching. SSS 2015 - 10th Int Sp Syntax Symp 2015:1–14.
[6] Elezkurtaj T, Franck G. Algorithmic support of creative architectural design. Organization 2002:1–16.
doi:http://www.iemar.tuwien.ac.at/publications/.
[7] Fricker P, Hovestadt L, Braach M, Dillenburger B, Dohmen P, Rüdenauer K, et al. Organised Complexity Application of Statistical Design
in Large-Scale Building Projects Chair for Computer Aided Architectural Design CAAD. Predict Futur 25th ECAADe Conf Proc
2007;25:695–701.
[8] Dipl. Evolutionary Algorithms in Urban Planning Tomor ELEZKURTAJ & Georg FRANCK. 2001.
[9] Koenig R, Schneider S. Hierarchical structuring of layout problems in an interactive evolutionary layout system. Artif Intell Eng Des Anal
Manuf AIEDAM 2012;26:129–42. doi:10.1017/S0890060412000030.
[10] Arvin SA, House DH. Modeling architectural design objectives in physically based space planning. Autom Constr 2002;11:213–25.
doi:10.1016/S0926-5805(00)00099-6.
[11] Medjdoub B, Yannou B. Separating topology and geometry in space planning. CAD Comput Aided Des 2000;32:39–61.
doi:10.1016/S0010-4485(99)00084-6.

You might also like