Journal of Building Engineering 44 (2021) 102952

Contents lists available at ScienceDirect

Journal of Building Engineering

journal homepage: www.elsevier.com/locate/jobe

Hospital layout design renovation as a Quadratic Assignment Problem with

geodesic distances
Cemre Cubukcuoglu a, b, *, Pirouz Nourian a, M. Fatih Tasgetiren c, I. Sevil Sariyildiz a, Shervin Azadi a
a
Department of Architectural Engineering and Technology, Chair of Design Informatics, Faculty of Architecture and the Built Environment, Delft University of Technology,
Delft, the Netherlands
b
Department of Interior Architecture and Environmental Design, Faculty of Architecture, Yasar University, Izmir, Turkey
c
Department of International Logistics Management, Faculty of Business, Yasar University, Izmir, Turkey

A R T I C L E I N F O A B S T R A C T

Keywords: Hospital facilities are known as functionally complex buildings. There are usually configurational problems that
Computational design lead to inefficient transportation processes for patients, medical staff, and/or logistics of materials. The Quadratic
Architectural space planning Assignment Problem (QAP) is a well-known problem in the field of Operations Research from the category of the
Hospital layout
facility’s location/allocation problems. However, it has rarely been utilized in architectural design practice. This
Quadratic assignment problem
Layout optimization tool
paper presents a formulation of such logistics issues as a QAP for space planning processes aimed at renovation of
existing hospitals, a heuristic QAP solver developed in a CAD environment, and its implementation as a
computational design tool designed to be used by architects. The tool is implemented in C# for Grasshopper
(GH), a plugin of Rhinoceros CAD software. This tool minimizes the internal transportation processes between
interrelated facilities where each facility is assigned to a location in an existing building. In our model, the
problem of assignment is relaxed in that a single facility may be allowed to be allocated within multiple voxel
locations, thus alleviating the complexity of the unequal area assignment problem. The QAP formulation takes
into account both the flows between facilities and distances between locations. The distance matrix is obtained
from the spatial network of the building by using graph traversal techniques. The developed tool also calculates
spatial geodesic distances (walkable, easiest, and/or shortest paths for pedestrians) inside the building. The QAP
is solved by a heuristic optimization algorithm, called Iterated Local Search. Using one exemplary real test case,
we demonstrate the potential of this method in the context of hospital layout design/re-design tasks in 3D.
Finally, we discuss the results and possible further developments concerning a generic computational space
planning framework.

1. Introduction 1.1. Problem definition

The term Space Planning refers to the processes aimed at arranging a Hospitals consist of a wide range of functional units, each serving for
spatial configuration having logistics-related objectives, ergonomics, different activities such as clinical, nursing, administration, service
and intended user-experience of a building [1]. The spatial configura­ (food, laundry, etc.), research, and teaching. There are also numerous
tion of a building is an abstract representation of the particular ways in types of user groups and materials, moving or being transported be­
which the spaces inside a building are related to one another. Mathe­ tween those various functions within the hospital. The flowrates of these
matically, spatial configurations are represented as graphs, which are movements in between the functional spaces can be modelled as a
typically labelled but may or may not be assumed as directed and graph/matrix indexed by the indices of the functional units, we can
weighted. Conventionally, if the graph in question is weighted and consider this as a graph describing the functional requirements. On the
directed, it is called a network. other hand, in an existing spatial configuration, how the spaces are inter-
related can be modelled as a dense graph/matrix encoding the distances

* Corresponding author. Department of Architectural Engineering and Technology, Chair of Design Informatics, Faculty of Architecture and the Built Environment,
Delft University of Technology, Delft, the Netherlands.
E-mail addresses: [email protected], [email protected] (C. Cubukcuoglu), [email protected] (P. Nourian), [email protected].
tr (M.F. Tasgetiren), [email protected] (I.S. Sariyildiz), [email protected] (S. Azadi).
Available online 13 July 2021
https://doi.org/10.1016/j.jobe.2021.102952
2352-7102/© 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
Received(http://creativecommons.org/licenses/by/4.0/).
17 February 2021; Received in revised form 14 June 2021; Accepted 3 July 2021
C. Cubukcuoglu et al.

between spaces of interest, computed with respect to the [temporal] local search) for equal area layout.
Journal of Building Engineering 44 (2021) 102952
length of geodesics or optimal paths; we may refer to this graph as the Moreover, Huntley and Brown (1991) [14] combined a genetic al­
spatial configuration graph. Therefore, one of the challenges of space gorithm (GA) and simulated annealing (SA) for an equal-area layout. Yip
planning can be seen as matching these two graphs such that the dis­ and Pao (1994) [15] proposed a hybrid technique that combined GA and
tance between pairs is small when the flow rate between them is large SA for equal-area layout design. Bland and Dawson (1994) [16]
and vice versa. addressed QAP for large-scale layout using a hybrid heuristic algorithm
According to the literature, around 67% of health and care em­ that combined SA and TS. Chiang and Chiang (1998) [17] proposed TS
ployees are not able to perform their jobs efficiently due to the unsuit­ and SA for facility layout. Kochar et al. (1998) [18] proposed a
able layout of the working spaces [2]. Especially nurses sometimes meta-heuristic using a genetic algorithm, called HOPE, for unequal area
spend more time walking than the activities related to patient care in a single and multi-row layouts. Chiang (2001) [19] addressed a modified
day [3] because of the spatial connectivity problems of the interrelated version of QAP with binary variables by proposing a TS algorithm for
spaces. In one study, it has been observed that 28.9% of nursing staff interdepartmental layout. Solimanpur et al. (2004) [20] used ant colony
time wasted on walking [4]. Previous studies also show that layout type optimization for inter-cell layout. Nourelfalth et al. (2007) [21] used
of nursing unit has an impact on the walking time of nursing staff, e.g. metaheuristic (Ant Colony Optimization with EGD local search) for
Ref. [5] established that nursing staff in the radial unit walked 4.7 steps equal-area layout. Jaramillo and Kendall (2010) [22] proposed a TS
per minute while the other staff working in rectangular unit walked 7.9 heuristic using different construction algorithms for machine layout.
steps per minute, which is significantly more. The poor placement of the Moslemipour and Lee (2012) [23] developed a SA approach for dynamic
clinics combined with the increasingly overwhelming volume of traffic layout. Pourvaziri and Pierreval (2017) [6] presented SA algorithm for
between them was causing delays and heavy congestion in hospitals [6]. dynamic facility layout design based on a QAP formulation. For an
All these examples show that spatial configuration has a great impact on extensive review of solution techniques for the QAP, refer to recent
the efficient functioning of hospital buildings in terms of walking dis­ survey paper by Singh and Sharma (2010) [24].
tances and transportation processes. Therefore, providing efficient
transportation processes by minimizing the walking distances between 1.2.2. QAP in hospital layout planning
interrelated spaces should be the major concern in hospital layout A relatively limited body of research has been published with respect
planning. to layout planning in hospitals.
The use of Quadratic Assignment Problem (QAP) is highly desirable Elshafei (1977) [25] firstly proposed QAP for locating the clinics
to deal with layout planning problems in hospitals. It is a well-known within a hospital department using an improvement heuristic in order to
problem in the field of Operations Research. Most of the facility layout optimize traveling distances of patients and delay in patient flows.
problems (FLP) are formulated as QAP to minimize the transportation Murtagh et al. (1982) [26] used QAP formulation for assigning 19 clinics
cost. The requirements for design in the production industry are parallel to predefined locations in order to minimize transport costs by devel­
to those in hospital design. Therefore, facility-planning methodologies, oping a new heuristic. Butler et al. (1992) [27] formulated a QAP for bed
which are widely used in industrial engineering, are needed in dealing allocation in a general-purpose hospital to minimize the distance be­
with hospital layouts. tween services taken by nurses using a constructive heuristic (CRAFT).
Hahn et al. (2001) [28] proposed QAP for assigning the facilities into
locations in order to minimize travel distance taken by all pairs in a
1.2. Related literature hospital comparing the results of several heuristics and metaheuristics
solution methods. Yeh (2006) [29] focused on adjacency of objects, the
QAP is developed by Koopmans and Beckmann (1957) [7]. It is one distance between objects, availability of space for object location, po­
of the most difficult computational problems in the NP-hard class. Hence sitions of objects in relation to others for a case study of a hospital with
solving them optimally in a reasonable time is a very challenging task. 28 facilities. Facility layout design formulated as QAP and solved by
Although exact problem-solving might be intractable, there exist heu­ simulated annealing with an annealed neural network. Chraibi et al.
ristics algorithms capable of solving the QAP even in large sizes with (2015) [30] minimized total traveling cost and rearrangement cost in a
nearly optimal solutions in a reasonable time. dynamic facility layout problem of the Operating Theatre (OT) depart­
ment of a hospital. Recently, Helber et al. (2016) [31] have proposed a
1.2.1. QAP in general hierarchical modelling approach. The first stage is formulated as QAP
Several solution techniques for the QAP have been suggested in the for assigning elements to locations using a fix-optimize heuristic by
literature. As early works of QAP for FLPs, Kaku et al. (1988) [8] pro­ considering transportation processes, locating some units on specific
posed a heuristic approach (exchange-improvement routine) for locations and ensuring the direct adjacency of two specific units. The
multi-story layout design. Kaku (1992) [9] proposed a procedure that second stage is detailed positioning within a location considering space
combined a constructive heuristic and exchange improvement for loop requirements in a large hospital facility. Zuo et al. (2019) [32] proposed
conveyor and linear-track layout cases. Rosenblatt (1992) [10] devel­ a QAP formulation for an emergency department of a hospital using a
oped a hybrid method that combined branch and bound framework with multi-objective tabu search algorithm by focusing on a real-case study.
heuristics for equal-sized departmental layout. A novel study by Li and As a result, most works focus on individual departments’ 2D layout
Smith (1995) [11] proposed a sample test-pairwise exchange heuristic optimization in the hospital such as OT planning and nursing units.
procedure (STEP) for dynamic facility layout. Urban (1998) [12] pre­ Considering the whole set of hospital departments in a 3D layout opti­
sented two heuristics (multi-greedy algorithm and GRASP) for dynamic mization is scarce in the literature. There is only one study in Ref. [31]
facility layout design. Ulutas and Sarac (2006) [12] addressed QAP with that focuses on whole hospital departments with a practical case study.
relocation cost thus handling a dynamic facility layout problem by Most of the application papers considered rectilinear distances (a.k.a.
developing a heuristic algorithm, called modified sub-gradient (MSG). Manhattan distances), not geodesic distances. To the best of our
Ramkumar et al. (2009) [13] proposed a new heuristic (iterated fast

2
C. Cubukcuoglu et al.

knowledge, none of the works considered the combination of the QAP • Calculating spatial geodesic distances and entering the distance
Journal of Building Engineering 44 (2021) 102952
with the graph-theoretical aspect of computing spatial geodesic dis­ matrix as an input of the tool
tances (walkable, easiest, and/or shortest paths for pedestrians) inside • Proposing a heuristic algorithm for solving the QAP
the building in real-case scenarios. • Implementation of the model and heuristic optimization algorithm
• Running the solver, reflecting on the results, and comparing them to
1.3. Utilization of QAP in computational design the existing state of an actual hospital

QAP is first developed for facility layout planning but it is also useful 1.5. Gaps in the literature
for spatial layout planning in architecture. There are different facility
layout programs like CRAFT [33], COFAD [34], CORELAP [35], and The detailed literature review is given in section 1.2. We have
BLOCPLAN [36]. However, these tools have rarely been utilized in identified and summarised the following gaps in the literature con­
architectural design practice. Considering spatial layout planning as a cerning the layout optimization of existing hospitals:
fundamental aspect of architectural design that affects the functional
performance of hospital buildings, we argue that a systematic approach • There is no complete methodology for a 3D layout problem with a
for ensuring the effectiveness and efficiency of a layout schema should QAP formulation, most papers are focused on 2D layout and focused
be an integral part of any architectural design process, particularly in the on specific departments such as the operation theatre and the nursing
case of critically complex building such as hospitals. The idea of bridging units. However, we focus on the entirety of the hospital design
the gap between both the parametric CAD platforms and the layout problem.
design is not entirely new. There is a limited number of tools available in • There is no study on QAP that focuses on geodesic distances (espe­
Parametric CAD platforms to facilitate the space planning processes, cially in 3D). Most studies focus on rectilinear distances on a flat 2D
namely DeCoding Spaces, SpiderWeb, and Syntactic [37–39]. To the plan.
best of our knowledge, none of the tools proposed the QAP approach in • The combination of QAP with the graph theoretical aspects (way­
layout planning and none of them utilize the shortest paths for pedes­ finding) in real-world test-cases is unique.
trians. These three toolkits provide methods for formulating space • Modularization/Discretization of space layout problem both in terms
planning problems and analysing spatial configurations within the of modularization of the departments and the walkable space in
framework of Space Syntax theories. However, other than heuristic between the departments in 3D for solving a QAP.
force-directed graph drawing solvers, none of these tool suites proffer an • There is no implementation of QAP problem-solving in computa­
explicit formulation of the space planning problem as an optimization tional design methodologies.
problem. Our proposed methodology formulates the problem of layout • The Iterative Local Search algorithm is not entirely new, but it has
optimization as a Quadratic Assignment Problem, proposes a measure of not been implemented in the case of space layout of hospitals.
quality as the Logistic Cost Function, as to which a benchmark can be
created for the improvements on the inner walking/transportation costs 2. Problem formulation
within a complex layout. Finally, the proposed solver reduces these costs
to a minimum and finds a new spatial configuration. As explained in the literature review the Quadratic Assignment
Problem is well known in the field of Facilities Layout Planning but
1.4. Contributions relatively unknown in architectural layout design. Interestingly, when
the focus of the optimization task is on improving the configuration of an
The main goal of this paper is to formulate and solve a 3D space existing building, the Quadratic Assignment Problem arises naturally
planning problem methodology in the form of a Quadratic Assignment when we consider the efficiency of the spatial configuration of the
Problem (QAP), in the context of a re-design/renovation task, by building as a whole. QAP is a combinatorial optimization problem that
considering the effect of geodesic distances through a network of cir­ aims at allocating a set of facilities to a set of locations such that the total
culation spaces in 3D. The QAP is a well-known problem in the field of transportation cost is minimized. Total transportation cost is a function
Operations Research from the category of the facility’s location/allo­ of the flows between facilities and the distances between locations. The
cation problems. The methodology is first and foremost developed for QAP has a discrete representation of areas since the facility departments
hospital space planning; however, it can also be used in design and can only be assigned to predefined network locations. Essentially, the
optimization for other types of complex buildings, especially in order to problem of assignment here is an unequal area layout problem because
study reuse scenarios. The methodology is implemented and tested each department has a different required amount of surface area Table 1
partly in C# language [40] for McNeel’s Grasshopper3D [41] and partly (Appendix-A). Our model makes this discretization regular by topolog­
in VEX language [42] for SideFX′ Houdini [43]. The following steps are ically abstracting the whole 3D walkable floor space of the building as a
presented in this paper: voxelated domain [44,45]. Voxels or (volumetric pixels/picture cells)
are 3D regular units of space for partitioning a 3D volume into a Car­
• Formulating a layout problem in architectural design as a QAP based tesian grid of cells. The design domain, in this case, is voxelated for two
on geodesic distances main reasons: 1) to modularize the units of space so that the problem of
• Introducing a practical CAD workflow for applying QAP solvers to 3D layout can be formulated as an assignment problem, and 2) to use the
spatial layout problems explicit topological relations between adjacent voxels for constructing a
• Discretizing and modularizing the design space as a way of struc­ network model of space to compute the geodesic distances for the
turing the geodesic computation problem in 3D as well as relaxing computation of the logistic cost function (the objective function). In
and simplifying the unequal area QAP problem accordance with this tessellation, at a higher resolution corresponding to
• Estimating flows and entering the flow matrix as an input of the tool the structural grid of the building, we break down the departments in

3
C. Cubukcuoglu et al.

terms of their surface area requirements into ‘multiple facilities, each of combinatorial optimization problem. The flow information refers to any
Journal of Building Engineering 44 (2021) 102952
which has exactly the area of one pixel’. Such a regular tessellation quantitative relationship score between any pair of items, typically the
brings about two main advantages: on the one hand computation of estimated volume of transporting materials or probability of transition
geodesic distance becomes straightforward on the network generated of medical staff in between facilities. We have considered the latter
from the voxels, and on the other hand the problem of assignment is probabilistic interpretation in our formulation. This score can also be
relaxed in that a single facility may be allowed to be allocated within interpreted from given adjacency requirements of a hospital, which are
multiple voxel locations, thus alleviating the complexity of the unequal typically recommended for ensuring effective logistics according to
area assignment problem. Each facility can be assigned to exactly one medical procedures, providing privacy or community, security, safety,
location and no location is assigned to more than one facility. Therefore, hygiene, congruence of noise levels, etcetera. The so-called RELChart
the number of facilities should be the same as the number of locations. table, e.g. the one used in this paper (Appendix B) is a matrix whose
The basic QAP model can be formulated as: entries indicate the relative importance of closeness between two de­
partments. Considering the central importance of this matrix in the
minimize ∑ ∑
n n
Di,j Tπi ,πj formulation of the QAP problem or even only in assessing the quality of a
π ∈ Pn i=1 j=1 particular assignment, it is important to compose this table with
objective information. However, in practice, such tables are often
Where, π is a vector of integers denoting a permutation of facilities at a composed following discussions of the board of directors of a hospital.
moment in time, Tπi ,πj denotes the [transportation] flow between a Nevertheless, for a building that does not exist yet, figuring out such
permuted facility πi = k, and another permuted facility πj = l and Di,j is importance ratings is a daunting task. For an existing building, however,
the distance between location i and j. Pn denotes the set of all permu­ these relative importance ratings can be replaced with the measured or
tations π(t) : N→N, where the superscript denotes the iteration time. estimated probabilities of transition between pairs of departments
Note that the cardinality of Pn will be n!, and so a brute-force search objectively. Given the technical difficulty of measuring such probabili­
becomes intractable as soon as the problem gets large, i.e. an order of ties in practice, estimating the probabilities according to the foreseen
complexity of O (n!); which in the case of our example would be about procedures is an alternative that has been shown to be feasible using a
searching within 64! = 1.2688693e + 89 possible permutations. Note Discrete Event Simulation (q.v. [46,47]).
that the objective function is measuring the expected travelled distance The number of spatial units must remain the same throughout the
for a typical building user; this is because the transition probabilities (e. QAP solving, and so, we firstly split the departments into spatial units
g. those given as percentage values in Table 2, Appendix-B) are considering the size of the designated modules and their area re­
dimensionless/unitless and that the travelled distance between every quirements. Then we divide the predicted flow rates mentioned in the
two nodes is multiplied with the probability of that transition. RELChart equally between the dividend units. We added this description
to the methodology section. The suggestion to set high intra-closeness
3. Methodology ratings for encouraging closeness between the split parts is very
logical and it is already implemented by setting high flow rates between
We have formulated the problem as a matter of reducing the logistic the divided units (100% closeness). The numbers written in our REL
cost function by choosing the right permutation of facilities within a set chart are percentages [0,100] which are interpreted at the end as
of existing locations. Due to the physical nature of the costs (distances) numbers in the range of [0,1] as transition probabilities. This makes it
and the dimensionless (unitless) meaning of the flow rates, the physical possible to have a physical interpretation of the objective cost function
unit of the objective function is the same as the distances. This means as an expected travelled time as explained in the section Application.
that the distances in between the departments in a 3D space must be However, if we raise these numbers to values higher than 100 (or 1) this
computed. In order to do so, we propose the following methodology: would disrupt the probabilistic interpretation and make it hard to
explain and justify the results.
1. Estimate the flow-rates as the transition probability between the
departments for a pedestrian (the flow-rates, in this case, are 3.2. Spatial geodesic distance
considered as given, see Appendix B);
2. Discretise the walkable space in between the departments and model In order to compute walking distances within a building, spatial
the topology of the connectivity between the discrete spaces as a geodesic (optimal paths on a network) are used in this method. In the
graph/network; mathematical field of graph theory, the geodesic distance between two
3. Compute the geodesic distance between the department locations in vertices in a graph is the total sum of the costs attributed to the edges
the discretized space and attribute the distances to the pairs of de­ connecting them through an optimal path. For constructing such paths
partments using the Floyd-Warshall algorithm; and computing geodesic distances, firstly, the set of all spaces (locations,
4. Heuristically improve (minimize) the objective cost function by try- corridors, and stairs) are discretized in a surface geometric model [45].
fitting various permutations of facilities over the locations using the The spatial network of the indoor walkable space of the building is then
Iterative Local Search algorithm. extracted from this mesh, a multi-source graph-traversal search is run,
and the distance matrix is obtained. The graph is constructed based on
6-neighbourhoods of voxels (i.e. voxels connected to their top, bottom,
3.1. Flows left, right, back and front neighbours) [48]. Then multiple A* searches
are run within the constructed graph to find the geodesic distance from
As stated above, the objective of the QAP is a function of flows be­ every location to every other location. Note that the location points (as
tween facilities and distances between locations as constant parameters marked in red and shown in Fig. 1) are exactly 64 voxels corresponding
and the only variable parameter is the permutation, hence it is called a to the larger sets of voxels whose areas are equal to 100 m2. These large

4
C. Cubukcuoglu et al.

Journal of Building Engineering 44 (2021) 102952

Fig. 1. The walkable space as a mesh (top) and its discretized voxel model (bottom); blue voxels: circulation areas & red voxels: location areas. (For interpretation of
the references to colour in this figure legend, the reader is referred to the web version of this article.)

areas are not included in the graph generated from the voxels because of an ILS algorithm in Ref. [53] has been tested on QAP instances arising
the model only needs to have the distance from their access points to from real-life problems as well as on several benchmark instances from
other access points. the QAPLIB [54]. Inspired by Refs. [53,55], we utilize the ILS algorithm
for the space planning tool presented in this paper. Details of the algo­
rithm that is considered in this paper are given in sub-sections below.
3.3. Heuristic problem solving

3.3.1. Solution encoding

As stated before, QAP is an NP-hard problem. Therefore, heuristic
The encoding scheme in our algorithm corresponds to a sequence of
optimization algorithms are seen as remedies for tackling this complex
integers that represents facilities in a feasible solution (permutation).
problem in large instances. In this tool, we selected Iterated Local Search
algorithm (ILS) [49–52] for problem solving. Recently, the performance

5
C. Cubukcuoglu et al.

3.3.2. Initial solution fourth position as a forward insertion

Journal toEngineering
of Building generate44 a new
(2021)solution
102952 π
(1)
=
In the Iterated Local Search (ILS) algorithm, the initial solution π(0) = [5, 4, 1, 2, 3]T whereas in the backward insertion, we remove π3 = 2
(0)

[π 1 , π2 , .., πn ]T is constructed randomly, which is a permutation of the from the current solution and insert into the second position as π(1) =
integers between 1 and n, where n is the number of facilities.
[5, 2, 4, 1, 3]T .
3.3.3. Perturbation scheme
3.3.4. Local search
In the ILS procedure, the initial solution is perturbed with swap and
After the “perturbation” of the current solution, we apply a “local
insertion neighbourhoods to escape from local minima. In this paper,
search” based on swap neighbourhood. In the swap local search, the
random swap and insertion neighbourhoods are employed. The swap
perturbed solution π(1) goes under a swap local search procedure. The
operator exchanges two facilities in a solution, whereas the insertion
iteration counter is fixed at 1 at the beginning, we select two facilities
operator removes a single facility from a solution and inserts it into a
randomly and simply swap them. If the new solution obtained after the
random position in the solution. As an example, to a swap operator,
swap neighbourhood is better than the current solution, it is replaced
suppose that we are given a current solution π(0) = [5, 4, 2, 1, 3]T . Two with the current solution and the iteration counter is again fixed at 1,
facilities are randomly selected and they are exchanged. As an example, otherwise, we keep the current solution as it is. And the iteration counter
we randomly choose the facility π 4 = 1 and π2 = 4 in order to swap
(0) (0)
is increased by 1. The swap local search is repeated until the iteration
them. Thus, we end up with a solution as π = [5, 1, 2, 4, 3]T . In addi­
(1) counter is reached at the number of facilities n. The pseudo-code of the
tion, as an example of an insertion operator, we apply forward or swap local search is given in Algorithm 1.
backward insertion with an equal probability. Suppose that we are given
Algorithm 1. Swap Local Search
a current solution π(0) = [5, 4, 2, 1, 3]T . Assume that we randomly choose
3 = 2. Then, we remove it from the solution and insert it into the
π (0)

6
C. Cubukcuoglu et al.

Journal of Building Engineering 44 (2021) 102952

dubbed f and we can add a condition to the Swap procedure and change
it to:
The general framework of the ILS algorithm is given in Algorithm 2.
if ( a ∕ ∈ f ) then Swap πa and πb
∈f ∧ b∕
Briefly, the initial solution is constructed randomly. Then, a swap local
search is applied to the initial solution. A loop-based on the termination
criterion is started. Repeatedly, perturbation and swap local searches are
applied to the current solution until a termination criterion is satisfied. If Algorithm 2. The General Framework of the Iterated Local Search
we need to implement a fixed-department constraint, we can define a set Algorithm

7
C. Cubukcuoglu et al.

Journal of Building Engineering 44 (2021) 102952

optimization as well as the amount of improvement in objective value.

The permutation results are also shown on the number sliders, e.g. if the
first slider has resulted as 2, then the second facility is placed to the first
4. Test & implementation location and so on. The QAP Solver component is shown in Fig. 2 when
working on a toy problem with three facilities to be assigned to three
The purpose of our implementation at this point was to test the al­ locations.
gorithms. More specifically, the purpose was to verify whether the al­ For computing the geodesic distances, we first extract a set of meshes
gorithms work as expected in terms of the correctness of the results and representing the connective spaces such as corridors, stairs, and ramps
to validate whether the results are improved. We have implemented the [if any]; then we voxelate these spaces using openVDB [56]; then
method presented above partly using the C# programming language and construct a network out of the voxels, and then calculate shortest paths
developed a space planning tool, called QAP Solver as an add-on for from all locations to each other location, i.e. the same set of locations
McNeel’s Grasshopper3D [41] software application and partly as a VEX will be used as both origins and as destinations. Technically, the loca­
add-on for SideFX′ Houdini [43] for computing network geodesics and tions are first mapped onto their closest points/voxels on the network.
the corresponding distance matrix. The output of the process will be the matrix of distances [Di,j ]n×n where n
QAP Solver component implements Algorithm 2 for solving a QAP is the number of locations. This output is directly used in the QAP solver.
instance based on the given input data (flow and distance matrices). The working principle of the QAP solver can be shown in the video
Inside the component, the first initial solution is generated randomly, available online (https://www.youtube.com/watch?v=Lv52qy1OjSw).
and then the optimization algorithm is run after Boolean toggle is set to
“True” mode; this will trigger the generation of new permutations, for 5. Application
each of which the objective function of the QAP is evaluated and re­
ported. In each generation, the component is capable of showing the In this section, we articulate an outlook for using the QAP tool in
change of decision variables (permutation) on a collection of number space planning and design of existing hospitals in a larger context.
sliders by realizing a slider update procedure inside the component. In Due to the nature of the QAP method, it can be used in case of the
this way, users of the tool can see how well the layout is being improved following scenarios in redesigning a hospital building, i.e. a building
over the generations in terms of the value of the fitness function and at with a set of facilities is to be moved to another or the same building
the same time see the generations in real-time, as the number sliders that with the same number of locations:
encode the permutation are used to pick and change the colours of
rooms in the 3D model. In addition, the tool allows users to set the 1. When we know the transportation flows [logistics of pedestrians or
maximum number of trial times (t max in Algorithm 2). The user is materials] between the facilities of a building;
expected to connect as many number sliders as the number of facilities 2. When we can estimate the flows between the facilities of a building,
(functional units). All sliders should be connected to the QAP Solver 8 e.g. by utilizing a simulation procedure such as Discrete Event
component. The output of the component presents the result of the Simulation [46].
 C. Cubukcuoglu et al.
9

Journal of Building Engineering 44 (2021) 102952

Fig. 2. QAP_Solver component in Grasshopper3D (left) and the geodesic/network distance computing in houdini (right).
C. Cubukcuoglu et al.

Journal of Building Engineering 44 (2021) 102952

Fig. 3. Definition of rectangular location boxes’ floor surfaces (consisting of 4 structural grid pixels based on building axes).

We illustrate the application of the proposed QAP method can be necessarily stay at a single location. This is obviously a limitation but at
used in computational space planning in the context of an exemplary the same time, it might be beneficial in light of a higher level of satis­
case study hospital (corresponding to the first scenario described above). faction with the logistics requirements. Based on the structural system of
All hospital departments are considered in the layout optimization the building, we define the locations as rectangular spaces, in line with
problem; however, it would be also possible to perform a similar pro­ the structural axes. This ensures that during the renovation the struc­
cedure on the inner spatial layout of only one department like Operation tural system does not have to be modified. Afterward, we tessellate each
Theatre layout or Intensive Care Unit layout, i. e QAP at a higher level of rectilinear floor space into four quadrangular faces. Each quad face re­
detail. The chosen hospital is a state hospital with a capacity of 250-beds fers to a square-like space surrounded by vertical columns axes and
for the in-patient wards. It is a 9-story building including basement and horizontal beam axes of the existing building (as shown in Fig. 3). A list
ground floors. In this model, we excluded the locations placed on the of spaces with the number of needed rectangular boxes is given in
basement and the 7th level of the building in the existing situation; Appendix-A. The flow matrix is given in Appendix-B. For calculating the
because, the 7th level has only a terrace area, which is not suitable for distance matrix, the spatial network of the building is given in Fig. 4.
placing any facility; and that the spaces located in the basement have The RELChart table in Appendix-B is a matrix whose entries indicate
some specific features and their locations cannot be changed. These the relative importance of closeness between two departments.
spaces that are excluded from the model can be listed as a mortuary, a Considering the central importance of this matrix in the formulation of
worship-space, a bunker, parking lots, and storage. Furthermore, some the QAP problem or even only in assessing the quality of a particular
of the functional units may require some specific locations due to a assignment, it is important to compose this table with objective infor­
specific feature in hospitals, e.g. the emergency department and main mation. However, in practice, such tables are often composed following
entrance should be at the ground level of the hospital. The locations of discussions of the board of directors of a hospital. Nevertheless, for a
these facilities are excluded in the model by making their locations building that does not exist yet, figuring out such importance ratings is a
constant in proper locations. Based on this, there are 34 facilities daunting task. For an existing building, however, these relative impor­
considered in this case model for renovating the hospital layout. tance ratings can be replaced with the measured or estimated proba­
Utilizing the QAP method is limited, theoretically, to equal-area bilities of transition between pairs of departments objectively. Given the
layout problems. However, each functional unit differs with respect to technical difficulty of measuring such probabilities in practice, esti­
its space requirements in this case. Although this is arguably an inherent mating the probabilities according to the foreseen procedures is an
limitation of the methodology, we can relax the requirements such that alternative that has been shown to be feasible using a Discrete Event
this is no longer a limitation. Since each location is represented with Simulation in Ref. [47]. The particular table added in the appendices of
[modular] discrete spaces (boxes in 3D), the number of needed boxes for this paper, however, is a RELChart produced by collating expert in­
each facility can also differ. As a new approach to adapt QAP to unequal terviews, site visits, design guidelines/standards, and recommendations
departments, we can repeat each facility according to the area from the scientific literature.
requirement in the flow matrix and distribute/divide the flows accord­ Regarding the computational results, the proposed heuristic algo­
ingly. For instance, assuming that each box has a capacity of 100 m2 and rithm for the QAP is tested on an Intel Core-i7 computer, with 2 GB of
that the cardiology department needs 300 m2 then we can define this RAM. Maximum trial time is taken as 50 000 iterations with a seed
department 3 times in the flow matrix and then divide the row and the number 5. Permutations of the existing and proposed layout are given
column corresponding to this space in the flow matrix by 3 and use the detailly in Appendix-C. Based on this table, existing fitness is 483751000
results in the new rows and columns. By this repetition, the number of expected travelled steps (roughly equal to 60 cm, i.e. the small voxel-size
facilities becomes the same as the number of locations, which is defined in the model). After the optimization by QAP Solver, the new fitness is
as 64 in the model; while this also entails that a single facility may not 411578400. The improvement in the objective value is 72172600 steps

10
C. Cubukcuoglu et al.

Journal of Building Engineering 44 (2021) 102952

Fig. 4. Extracting “spatial network” of the building (top), continuous version of paths (middle), a discrete version of paths (bottom), (red dots are location points).
(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

(adjusted to 721726.00 after dividing by 100 for converting flow per­ facilities by 469121.90/5000 = 93.82438 person hours for a typical day.
centages to probability fractions), which equals about 469121.90 m of Note that as we explained in the definition of the unit of the objective
travelled distance. To put this result in a more concrete context, let us function, this is the ‘expected travelled distance’ (or the time spent on
assume that an average person can walk 5 km per hour; then this number walking) for a typical building user on a typical day of operation. This
means that we have reduced the time spent for walking in between the means that the improvement can be attributed to the building as a whole

11
C. Cubukcuoglu et al.

Journal of Building Engineering 44 (2021) 102952

Fig. 5. Existing layout.

12
C. Cubukcuoglu et al.

Journal of Building Engineering 44 (2021) 102952

Fig. 6. Proposed layout.

rather than an individual person. The existing layout and the new layout layout, interrelated spaces with inpatient have moved to various loca­
result for the case study hospital model are visualized in Figs. 5 and 6. tions. In the proposed layout, operation rooms are located closer to the
Regarding the design results, the placement of the inpatient areas is patient wards and there exists ease of access between these facilities
still located at the top levels of the building as expected due to the with a vertical short connection. Critical patient flow processes and the
daylight requirements of patient wards. However, in the proposed shared equipment entail better access between intensive care units and

13
C. Cubukcuoglu et al.

operating rooms. In the new layout, intensive care units and operation infeasible, the seed ofJournal
the heuristic solver can be changed to find
of Building Engineering 44 (2021) 102952
rooms have a better connection since Intensive Care Units are the another configuration. The single configuration found as an example
neighbour of the Operating Theatre at the upper levels of the building. In in this paper seems to be feasible. However, in practice, more ex­
addition, delivery rooms have to get closer to Operation Theatre, which periments are needed to list layout alternatives and choose the one
is an advantage for shared staff and facilities. The kitchen and dining with the least transformation costs and/or the best suitability with
hall were located at the last level of the building whereas they are respect to other architectural criteria.
located at lower levels in the proposed layout. This provides quicker • The proposed methodology bases the reconfiguration problems on a
food transportation from the kitchen to the medical spaces e.g. dining & completely discretized and modularized design space, and the pro­
kitchen area has horizontal access to outpatient and vertical access to posed algorithm is reasonably fast, it would be theoretically feasible
inpatient. In the new layout, all departments related to maternity like to dissect the spatial units into smaller units and generalize the
Paediatrics Intensive Care Unit, Paediatrics Outpatient, Delivery Rooms method to broaden the application areas and the versatility of the
and Obstetrics departments are located close to each other. method for incorporating more diverse validity constraints.
Laboratories and diagnostic units are located adjacent to each other
on the ground floor and become closer to the outpatient departments as Integrating QAP into computational design workflows can, to say the
expected. In addition, inpatient departments were placed in the areas least, provides awareness of the logistics performance of the building in
that have one single corridor in the old layout. Whereas, these de­ terms of the expected walking time for personnel, and in that sense, it
partments are mostly located in areas with a radial layout structure. can even be used as an informative tool for conceptual design of new
Surgical outpatients like Obstetrics, Orthopedy, and Urology de­ buildings as well as re-designing existing buildings.
partments are located closer to the operating rooms in the new layout.
Transportation processes between interrelated spaces became more 7. Limitations & future work
efficient with the proposed approach.
The method presented in this paper is only suitable for reconfigu­
6. Conclusions ration of existing buildings, especially because it requires computing the
distance between available locations for computing the main objective
This paper introduces a new computational space planning meth­ function. Even though most hospital buildings have a regular structural
odology based on the well known Quadratic Assignment Problem, pre­ grid, it must be noted that our proposed way of dissecting departments
sents a heuristic solver for it and presents the test results on a hospital re- into modular areal units is only feasible on such highly modular and
design case study. One of the novelties of the presented methodology is regularly structured buildings. It must be noted that in our problem
that it utilizes the spatial network of the existing building for computing formulation we consider all facilities to be accommodatable in all
geodesic distances, which are then used as inputs of the QAP model. available locations, while in reality there might be facilities that can
Results show that objective value is reasonably minimized, and design only be accommodated in certain locations due to particular technical
results seem more logistically efficient. We have estimated an aspect of requirements. While this constraint is handled by excluding a list of fixed
the operational cost of the building as expected travel time of em­ facilities, we have disregarded the exchangeability of other de­
ployees/users of the building using our objective cost function. It must partments. Our methodology does not take the contiguity constraints
be noted that this is a matter of ex-ante assessment and not a mea­ into account explicitly as hard constraints, e.g. in cases where we split a
surement, as measuring the actual travel time would require tracking the facility into 2 or 3 facilities to fit it into our modularized spaces. We
personnel inside the building and fall out of the scope of this paper. The cannot enforce the new units to stay contiguous/adjacent to each other
time saving achieved with our methodology based on Operations during the optimization process; however, by adding extra closeness
Research has achieved an estimated reduction of around 90 person- ratings in between the split parts, we relax such constraints and add
hours for a typical operational day of the hospital. Due to its aggre­ them to the objective function effectively. Moreover, this limitation can
gate nature, such a reduction should be of interest for the management also be considered in another way: that the obtained results, which may
of the hospital as it implies not only a reduction of costs but also not strictly entail the initially conceived contiguity, can be used to
implicitly an increased comfort for the employees, users, and thus a reflect on the programme of requirements and consider revising it, e.g.
higher-quality service. The contributions of the paper can be recapitu­ considering two Cardiology departments if a significant expected travel-
lated as below: time saving can be made by splitting it into two departments. This can be
observed from the objective function. As future work, limitations of the
• The obtained results, i.e. the new configuration and its correspond­ proposed workflow to the existing buildings can be addressed by
ing logistic cost function, reveal a major difference made by reas­ modifying the problem-formulation, for instance by a Mixed-Integer
signing the departments to alternative locations, hence validating Programming formulation of the hospital layout problem. More
the major contribution of the paper on improving existing layouts. advanced optimization algorithms for solving the QAP can be proposed
The reduction of the logistic cost function in this case corresponds to e.g. populated version of the proposed algorithm. The geodesic distances
a total reduction of around 93 person-hours of expected travel time computed on the voxelated corridors are currently more accurate than
between departments of the hospital for a typical day. Euclidean distance but still quite simplistic in that they do not consider
• We have considered the physical constraints pertaining to the size of the cost of waiting times for the elevators, nor do they differentiate
the departments and assumed that the departments can be accom­ between going downstairs and upstairs with distances on the same level.
modated into modular/rectilinear spatial units (colored in the pic­ In general, instead of measuring distance in meters, it would be more
tures). The newly found assignment can be visually inspected from general to measure distance in travel time/effort, to also account for
the point of view of an architect/manager and it seems to be a path complexity for the visitors, for instance by using the Easiest Paths
feasible/logical assignment in terms of other constraints that are not weighting [57] Corridor Allocation Problem (CAP) [58] can be added,
taken into account in this formulation. If the configuration is deemed which has the same fitness structure with the QAP but with an extra

14
C. Cubukcuoglu et al.

Journal of Building Engineering 44 (2021) 102952

decision on locating the facilities on either sides of a corridor. Finally, solver, Writing – review & editing, Supervision. M. Fatih Tasgetiren:
the proposed layout optimization tool can be potentially extended to Conceptualization, Methodology, Software, QAP solver, Writing – re­
solve a Multi-Objective QAP [59,60] in further versions. view & editing, Supervision. I. Sevil Sariyildiz: Conceptualization,
Methodology, Supervision. Shervin Azadi: Software, Geodesic
CRediT authorship contribution statement Distance.

Cemre Cubukcuoglu: Conceptualization, Methodology, Software, Declaration of competing interest

QAP solver, Writing – original draft. Pirouz Nourian: Conceptualiza­
tion, Methodology, Software, Geodesic Distance, Slider Update in QAP The authors declare that there is no conflict of interest in this paper.

Appendix

Appendix-A
Table 1
A List of Spaces with a needed number of boxes (modular units of roughly 100 m2)

Facility Name Number of Boxes Needed

NEUROLOGY 1
OBSTETRICS 1
INTERNAL MEDICINE 1
CARDIOLOGY 1
PEDIATRICS 1
GASTROENTEROLOGY 2
DIAGNOSTIC UNITS 2
LABORATORIES 2
ADMINISTRATION-1 1
ORTHOPEDY 1
PHYSIOTHERAPY 2
HEMODIALYSIS 2
INFECTION DISEASES 2
SURGERY ICU 2
GENERAL ICU 1
PEDIATRY ICU 1
ICU WAITING AREA 1
ENT 1
UROLOGY 1
EYE DISEASES 2
BRAIN&GENERAL SURGERY 1
DERMATOLOGY 1
INPATIENT-1 2
DELIVERY ROOMS 3
OPERATION THEATRE 5
ADMINISTRATION-2 4
CONFERENCE HALL 3
LIBRARY 1
INPATIENT-2 4
INPATIENT-3 4
INPATIENT-4 4
KITCHEN 2
DINING HALL 2
Total number of locations (modular units) 64

15
C. Cubukcuoglu et al.
Appendix-B Journal of Building Engineering 44 (2021) 102952
The table below is a RELChart produced by collating expert interviews, site visits, design guidelines/standards, and recommendations from the
scientific literature.
16
C. Cubukcuoglu et al.

Journal of Building Engineering 44 (2021) 102952

Appendix-C
Table 3
The Existing Assignment of the Case Study Hospital

The Existing Assignment The Proposed Assignment

Location Index
Facility Index Facility Name Facility Index Facility Name

0 44 CONFERENCE HALL-1 56 INPATIENT-4-1

1 35 OPERATION THEATRE-1 19 SURGERY ICU-1
2 22 PEDIATRY ICU 29 DERMATOLOGY
3 7 DIAGNOSTIC UNITS-1 30 INPATIENT-1-1
4 8 DIAGNOSTIC UNITS-2 61 KITCHEN-2
5 3 CARDIOLOGY 15 HEMODIALYSIS-1
6 4 PEDIATRICS 25 UROLOGY
7 2 INTERNAL MEDICINE 50 INPATIENT-2-3
8 1 OBSTETRICS 48 INPATIENT-2-1
9 0 NEUROLOGY 55 INPATIENT-3-4
10 5 GASTROENTOLOGY-1 49 INPATIENT-2-2
11 9 LABORATORIES-1 32 DELIVERY ROOMS-1
12 6 GASTROENTEROLOGY-2 2 INTERNAL MEDICINE
13 10 LABORATORIES-2 54 INPATIENT-3-3
14 15 HEMODIALYSIS-1 60 KITCHEN-1
15 16 HEMODIALYSIS-2 47 LIBRARY
16 11 ADMINISTRATION-1 13 PHYSIOTHERAPHY-1
17 12 ORTHOPEDY 63 DINING HALL-2
18 13 PHYSIOTHERAPHY-1 57 INPATIENT-4-2
19 14 PHYSIOTHERAPY-2 6 GASTROENTEROLOGY-2
20 19 SURGERY ICU-1 4 PEDIATRICS
21 20 SURGERY ICU-2 17 INFECTION DISEASES-1
22 21 GENERAL ICU 3 CARDIOLOGY
23 17 INFECTION DISEASES-1 31 INPATIENT-1-2
24 18 INFECTION DISEASES-2 44 CONFERENCE HALL-1
25 23 ICU WAITING AREA 52 INPATIENT-3-1
26 30 INPATIENT-1-1 23 ICU WAITING AREA
27 31 INPATIENT-1-2 59 INPATIENT-4-4
28 26 EYE DISEASES-1 1 OBSTETRICS
29 28 BRAIN&GENERAL SURGERY 0 NEUROLOGY
30 27 EYE DISEASES-2 41 ADMINISTRATION-2-2
31 25 UROLOGY 9 LABORATORIES-1
32 24 ENT 26 EYE DISEASES-1
33 36 OPERATION THEATRE-2 45 CONFERENCE HALL-2
34 37 OPERATION THEATRE-3 7 DIAGNOSTIC UNITS-1
35 38 OPERATION THEATRE-4 12 ORTHOPEDY
36 39 OPERATION THEATRE-5 46 CONFERENCE HALL-3
37 32 DELIVERY ROOMS-1 11 ADMINISTRATION-1
38 29 DERMATOLOGY 40 ADMINISTRATION-2-1
39 33 DELIVERY ROOMS-2 24 ENT
40 34 DELIVERY ROOMS-3 62 DINING HALL-1
41 48 INPATIENT-2-1 28 BRAIN& GENERAL SURGERY
42 49 INPATIENT-2-2 51 INPATIENT-2-4
43 45 CONFERENCE HALL-2 35 OPERATION THEATRE-1
44 50 INPATIENT-2-3 10 LABORATORIES-2
45 47 LIBRARY 14 PHYSIOTHERAPY-2
46 51 INPATIENT-2-4 39 OPERATION THEATRE-5
47 46 CONFERENCE HALL-3 16 HEMODIALYSIS-2
48 52 INPATIENT-3-1 37 OPERATION THEATRE-3
49 53 INPATIENT-3-2 22 PEDIATRY ICU
50 54 INPATIENT-3-3 34 DELIVERY ROOMS-3
51 55 INPATIENT-3-4 36 OPERATION THEATRE-2
52 56 INPATIENT-4-1 38 OPERATION THEATRE-4
53 57 INPATIENT-4-2 20 SURGERY ICU-2
54 58 INPATIENT-4-3 43 ADMINISTRATION-2-4
55 59 INPATIENT-4-4 53 INPATIENT-3-2
56 60 KITCHEN-1 8 DIAGNOSTIC UNITS-2
57 61 KITCHEN-2 21 GENERAL ICU
58 62 DINING HALL-1 33 DELIVERY ROOMS-2
59 63 DINING HALL-2 5 GASTROENTEROLOGY-1
60 40 ADMINISTRATION-2-1 42 ADMINISTRATION-2-3
61 41 ADMINISTRATION-2-2 27 EYE DISEASES-2
62 42 ADMINISTRATION-2-3 58 INPATIENT-4-3
63 43 ADMINISTRATION-2-4 18 INFECTION DISEASES-2

17
C. Cubukcuoglu et al.

Appendix-D Journal of Building Engineering 44 (2021) 102952

Table 4
Nomenclature

Notation Explanation

QAP Quadratic Assignment Problem

C# C sharp programming language
GH Grasshopper
CAD Computer Aided Design
ILS Iterated Local Search
FLP Facility Layout Planning
STEP A sample test-pairwise exchange heuristic procedure
GRASP Greedy randomized adaptive search procedure
MSG Modified sub-gradient
GA Genetic algorithm
SA Simulated annealing
TS Tabu search
EGD Extended great deluge
OT Operating Theatre
2D Two dimensional
3D Three dimensional
OR Operations Research
QAPLIB A Quadratic Assignment Problem Library
CAP Corridor Allocation Problem
A* A-star
π A vector of integers denoting a permutation of facilities at a moment in time
Tπi,πj The [transportation] flow between a permuted facility πi = k, and another permuted facility πj = l
Pn The set of all permutations
Di,j The distance between location i and j
t_max Maximum number of trial times
n Number of facilities & locations
CRAFT Computerized Relative Allocation of Facilities Technique
COFAD COmputerized FAcilities Design
CORELAP Computerized Relationship Layout Planning

References [15] P.P.C. Yip, Y.-H. Pao, A guided evolutionary simulated annealing approach to the
quadratic assignment problem, IEEE Trans. Syst. Man. Cybern. 24 (1994)
1383–1387, https://doi.org/10.1109/21.310514.
[1] Space Planning Basics, fourth ed., Wiley, (n.d.), https://www.wiley.com/en-gb/
[16] J.A. Bland, G.P. Dawson, Large-scale layout of facilities using a heuristic hybrid
Space+Planning+Basics%2C+4th+Edition-p-9781118882009. (Accessed 13
algorithm, Appl. Math. Model. 18 (1994) 500–503, https://doi.org/10.1016/0307-
December 2019). accessed.
904X(94)90328-X.
[2] A. Chraibi, S. Kharraja, I.H. Osman, O. Elbeqqali, A mixed integer programming
[17] C. Wen-Chyuan, C. Chi, Intelligent local search strategies for solving facility layout
formulation for solving operating theatre layout problem: a multi-goal approach,
problems with the quadratic assignment problem formulation, Eur. J. Oper. Res.
in: Proc. - IESM Conf. Oct. 2013, Rabat, Morocco, 2013.
2217 (1998) 457–488, https://doi.org/10.1016/S0377-2217(97)00285-3.
[3] R. Ulrich, X. Quan, H. Systems, C. Architecture, A. Texas, The role of the physical
[18] J.S. Kochhar, B.T. Foster, S.S. Heragu, Hope, A genetic algorithm for the unequal
environment in the hospital of the 21 st Century : a once-in-a-lifetime opportunity,
area facility layout problem, Comput. Oper. Res. 25 (1998) 583–594, https://doi.
Environment 439 (2004) 69.
org/10.1016/S0305-0548(97)00100-7.
[4] L.D. Burgio, B.T. Engel, A. Hawkins, K. McCormick, A. Scheve, A descriptive
[19] W.-C. Chiang, Visual facility layout design system, Int. J. Prod. Res. 39 (2001)
analysis of nursing staff behaviors in a teaching nursing home: differences among
1811–1836, https://doi.org/10.1080/00207540110035192.
NAs, LPNs, and RNs, Gerontol. 30 (1990) 107–112.
[20] M. Solimanpur, P. Vrat, R. Shankar, Ant colony optimization algorithm to the inter-
[5] M.M. Shepley, K. Davies, Nursing unit configuration and its relationship to noise
cell layout problem in cellular manufacturing, Eur. J. Oper. (2004).
and nurse walking behavior: an AIDS/HIV unit case study, AIA Acad. J. 6 (2003)
[21] M. Nourelfath, N. Nahas, B. Montreuil, Coupling ant colony optimization and the
12–14.
extended great deluge algorithm for the discrete facility layout problem, Eng.
[6] H. Pourvaziri, H. Pierreval, Dynamic facility layout problem based on open
Optim. 39 (2007) 953–968, https://doi.org/10.1080/03052150701551461.
queuing network theory, Eur. J. Oper. Res. 259 (2017) 538–553, https://doi.org/
[22] J.R. Jaramillo, A.R. McKendall Jr., The generalised machine layout problem, Int. J.
10.1016/j.ejor.2016.11.011.
Prod. Res. 48 (2010) 4845–4859, https://doi.org/10.1080/00207540903117840.
[7] T.C. Koopmans, M. Beckmann, Assignment problems and the location of economic
[23] G. Moslemipour, T.S. Lee, Intelligent design of a dynamic machine layout in
activities, Econom. J. Econom. Soc. (1957) 53–76.
uncertain environment of flexible manufacturing systems, J. Intell. Manuf. 23
[8] B.K. Kaku, G.L. Thompson, I. Baybars, A heuristic method for the multi-story layout
(2012) 1849–1860, https://doi.org/10.1007/s10845-010-0499-8.
problem, Eur. J. Oper. Res. 37 (1988) 384–397, https://doi.org/10.1016/0377-
[24] S.P. Singh, R.R.K. Sharma, A review of different approaches to the facility layout
2217(88)90202-0.
problems, Int. J. Adv. Manuf. Technol. 30 (2006) 425–433, https://doi.org/
[9] B.K. Kaku, R. Rachamadugu, Layout design for flexible manufacturing systems,
10.1007/s00170-005-0087-9.
Eur. J. Oper. Res. 57 (1992) 224–230, https://doi.org/10.1016/0377-2217(92)
[25] A.N. Elshafei, Hospital layout as a quadratic assignment problem, Oper. Res. Q. 28
90044-A.
(1977) 167–179.
[10] M.J. Rosenblatt, B. Golany, A distance assignment approach to the facility layout
[26] B.A. Murtagh, T.R. Jefferson, V. Sornprasit, A heuristic procedure for solving the
problem, Eur. J. Oper. Res. 57 (1992) 253–270, https://doi.org/10.1016/0377-
quadratic assignment problem, Eur. J. Oper. Res. 9 (1982) 71–76, https://doi.org/
2217(92)90047-D.
10.1016/0377-2217(82)90013-3.
[11] W.J. Li, J.M. Smith, An algorithm for quadratic assignment problems, Eur. J. Oper.
[27] T.W. Butler, K.R. Karwan, J.R. Sweigart, G.R. Reeves, An integrative model-based
Res. 81 (1995) 205–216, https://doi.org/10.1016/0377-2217(93)E0162-Q.
approach to hospital layout, IIE Trans. 24 (1992) 144–152, https://doi.org/
[12] T.L. Urban, Solution procedures for the dynamic facility layout problem, Ann.
10.1080/07408179208964211.
Oper. Res. 76 (1998) 323–342.
[28] P.M. Hahn, J. Krarup, A hospital facility layout problem finally solved, J. Intell.
[13] A.S. Ramkumar, S.G. Ponnambalam, N. Jawahar, A new iterated fast local search
Manuf. 12 (2001) 487–496, https://doi.org/10.1023/A:1012252420779.
heuristic for solving QAP formulation in facility layout design, Robot. Comput.
[29] I.C. Yeh, Architectural layout optimization using annealed neural network, Autom.
Integrated Manuf. 25 (2009) 620–629, https://doi.org/10.1016/j.
ConStruct. 15 (2006) 531–539, https://doi.org/10.1016/j.autcon.2005.07.002.
rcim.2008.03.022.
[30] A. Chraibi, S. Kharraja, I.H. Osman, O. Elbeqqali, Multi-agent system for solving
[14] C.L. Huntley, D.E. Brown, A parallel heuristic for quadratic assignment problems,
Dynamic Operating Theater facility layout problem, IFAC-PapersOnLine. 28 (2015)
Comput. Oper. Res. 18 (1991) 275–289, https://doi.org/10.1016/0305-0548(91)
1146–1151, https://doi.org/10.1016/j.ifacol.2015.06.238.
90029-Q. 18
C. Cubukcuoglu et al.

[31] S. Helber, D. Böhme, F. Oucherif, S. Lagershausen, S. Kasper, A hierarchical facility [46] M.M. Gunal, A guide for building hospital simulation models, Heal. Syst. 1 (2012)
layout planning approach for large and complex hospitals, Flex. Serv. Manuf. J. 2 17–25. Journal of Building Engineering 44 (2021) 102952
(2014) 5–29, https://doi.org/10.1007/s10696-015-9214-6. [47] C. Cubukcuoglu, P. Nourian, I.S. Sariyildiz, M.F. Tasgetiren, A discrete event
[32] X. Zuo, B. Li, X. Huang, M. Zhou, C. Cheng, X. Zhao, Z. Liu, Optimizing hospital simulation procedure for validating programs of requirements: the case of hospital
emergency department layout via multiobjective tabu search, IEEE Trans. Autom. space planning, Software 12 (2020) 100539, https://doi.org/10.1016/j.
Sci. Eng. 16 (2019) 1137–1147. softx.2020.100539.
[33] G.C. Armour, E.S. Buffa, A heuristic algorithm and simulation approach to relative [48] S. Laine, A topological approach to voxelization, Comput. Graph. Forum 32 (2013)
location of facilities, Manag. Sci. 9 (1963) 294–309. 77–86, https://doi.org/10.1111/cgf.12153.
[34] J.A. Tompkins, R. Reed Jr., An applied model for the facilities design problem, Int. [49] A.S. Ramkumar, S.G. Ponnambalam, N. Jawahar, A new iterated fast local search
J. Prod. Res. 14 (1976) 583–595. heuristic for solving QAP formulation in facility layout design, Robot. Comput.
[35] R.C. Lee, CORELAP-computerized relationship layout planning, Jour. Ind. Engg. 8 Integrated Manuf. 25 (2009) 620–629.
(1967) 195–200. [50] M.S. Hussin, T. Stützle, Hierarchical iterated local search for the quadratic
[36] C.E. Donaghey, V.F. Pire, Facilities layout with BLOCPLAN, Univ. Houst. TX Work. assignment problem, in: Int. Work. Hybrid Metaheuristics, 2009, pp. 115–129.
Pap., 1991. [51] H.R. Lourenço, O.C. Martin, T. Stützle, Iterated local search: framework and
[37] A. Abdulmawla, M. Bielik, P. Buš, M.-C. Chang, M. Dennemark, E. Fuchkina, applications, in: Handb. Metaheuristics, Springer, 2019, pp. 129–168.
Y. Miao, K. Knecht, R. Koenig, S. Schneider, DeCodingSpaces toolbox for [52] T. Stützle, Iterated local search for the quadratic assignment problem, Eur. J. Oper.
grasshopper. Computational analysis and generation of street network, Plots and Res. 174 (2006) 1519–1539.
Buildings (2017). [53] M.F. Tasgetiren, Q.-K. Pan, Y. Ozturkoglu, O.K. Cotur, Variable block insertion
[38] R. Schaffranek, SpiderWeb. https://www.food4rhino.com/app/spiderweb, 2012. heuristic for the quadratic assignment problem, 2017 IEEE Congr. Evol. Comput.
[39] P. Nourian, S. Rezvani, S. Sariyildiz, A syntactic architectural design methodology, (2017) 1765–1770.
in: Meet. Ninth Int. Sp. Syntax Symp, 2013. Seoul, Korea. [54] R.E. Burkard, S.E. Karisch, F. Rendl, QAPLIB–a quadratic assignment problem
[40] Introduction to the C# Language and the, NET Framework | Microsoft Docs, (n.d.), library, J. Global Optim. 10 (1997) 391–403.
https://docs.microsoft.com/en-us/dotnet/csharp/getting-started/introduction-t [55] M.F. Tasgetiren, Q.-K. Pan, P.N. Suganthan, I.E. Dizbay, Metaheuristic algorithms
o-the-csharp-language-and-the-net-framework. (Accessed 22 January 2020). for the quadratic assignment problem, 2013 IEEE Symp. Comput. Intell. Prod.
accessed. Logist. Syst. (2013) 131–137.
[41] D. Rutten, Grasshopper3D, (2007). [56] OpenVDB (n.d.), https://www.openvdb.org/. (Accessed 22 January 2020).
[42] VEX language reference (n.d.), https://www.sidefx.com/docs/houdini/vex/lang. accessed.
html. (Accessed 22 January 2020). accessed. [57] P. Nourian, F. van der Hoeven, S. Rezvani, S. Sariylidiz, Easiest paths for walking
[43] Houdini - 3D modeling, animation, VFX, look development, lighting and rendering and cycling : combining syntactic and geographic analyses in studying walking and
| SideFX (n.d.), https://www.sidefx.com/. (Accessed 22 January 2020). accessed. cycling mobility, Proc. 10th Int. Sp. Syntax Symp. (2015) 1–15.
[44] P. Nourian, R. Gonçalves, S. Zlatanova, K.A. Ohori, A. Vu Vo, Voxelization [58] A.R.S. Amaral, The corridor allocation problem, Comput. Oper. Res. 39 (2012)
algorithms for geospatial applications: computational methods for voxelating 3325–3330, https://doi.org/10.1016/j.cor.2012.04.016.
spatial datasets of 3D city models containing 3D surface, curve and point data [59] C. Sanhueza, F. Jiménez, R. Berretta, P. Moscato, PasMoQAP: a parallel
models, Methods (Duluth) 3 (2016) 69–86, https://doi.org/10.1016/j. asynchronous memetic algorithm for solving the Multi-Objective Quadratic
mex.2016.01.001. Assignment Problem, in: Evol. Comput. (CEC), 2017 IEEE Congr, ., 2017,
[45] S. Azadi, P. Nourian, topoGenesis: an open-source python library providing pp. 1103–1110.
topological structures and functions for Generative Systems and Sciences. https://d [60] C. Cubukcuoglu, M.F. Tasgetiren, I.S. Sariyildiz, L. Gao, M. Kucukvar, A memetic
oi.org/10.5281/ZENODO.4006514, 2020. algorithm for the bi-objective quadratic assignment problem, Procedia Manuf 39
(2019) 1215–1222.

19