See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/328232815

Automated specification of steel reinforcement to support the optimisation of

RC floors

Article  in  Automation in Construction · October 2018

DOI: 10.1016/j.autcon.2018.10.005

CITATION READS

1 104

4 authors:

Stathis Eleftheriadis Philippe Duffour

University College London University College London
19 PUBLICATIONS   78 CITATIONS    49 PUBLICATIONS   412 CITATIONS   

SEE PROFILE SEE PROFILE

Ben Stephenson D. Mumovic

Price & Myers University College London
2 PUBLICATIONS   10 CITATIONS    112 PUBLICATIONS   1,194 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

TOP-The Total Performance of Low Carbon Buildings in China and the UK View project

Innovative engineering approach for material, carbon and cost efficiency of steel buildings View project

All content following this page was uploaded by Stathis Eleftheriadis on 12 October 2018.

The user has requested enhancement of the downloaded file.

 Automation in Construction 96 (2018) 366–377

Contents lists available at ScienceDirect

Automation in Construction
journal homepage: www.elsevier.com/locate/autcon

Automated specification of steel reinforcement to support the optimisation T

of RC floors
⁎
S. Eleftheriadisa,b, , P. Duffourc, B. Stephensond, D. Mumovica
a
UCL Institute for Environmental Design and Engineering, University College London, UK
b
Department of Computer Science, University College London, UK
c
Department of Civil, Environmental & Geomatic Engineering, University College London, UK
d
Price & Myers LLP, 37 Alfred Place, WC1E 7DP London, UK

A R T I C LE I N FO A B S T R A C T

Keywords: A Building Information Modelling (BIM)-enabled computational approach was presented in this paper for the
Design automated specification of steel reinforcement to support the optimisation of reinforced concrete (RC) flat slabs.
Automation After importing slab geometries from BIM, the proposed procedure utilised internal forces output from Finite
Optimisation Element Model (FEM) to map required reinforcement in two stages. In the first stage, the reinforcement spe-
RC structures
cifications matched the spatial resolution of the FEM. In the second, the reinforcement was adjusted by imposing
Steel reinforcement
Floors
constructability functions to limit the number of arrangements in terms of zones and bar spacing. The aim of the
paper was to investigate the parametric capabilities of the proposed approach in the context of an optimisation
model for the generation of material-efficient structural designs. Numerical examples were presented to de-
monstrate the efficiency of the automated specification procedure. The material efficiency and the design
complexity of the developed reinforcement configurations were also assessed against a conventional solution
under realistic design conditions.

1. Introduction development and post-processing procedures, data sharing in multi-

disciplinary frameworks and algorithm selection among others ac-
Cast in situ reinforced concrete (RC) structures are prevalent in cording to a recent study that surveyed to what extend engineering
small and medium residential and office buildings. However, their de- companies in the UK use computational optimisation approaches [13].
tailed design and construction remain relatively low-tech and labour Despite practicing engineers' expertise and experience, conventional
intensive. The design of these structures involves specifying the place- structural design processes often result in sub-optimal solutions [14].
ment and size of reinforcing bars within the concrete matrix. This phase New design processes that increase the adoption of practical optimi-
of the design usually follows an iterative process, in which structural sation techniques by structural engineering practitioners are still ne-
engineers use a manual trial-and-error approach to find a sufficiently cessary [15]. Developments within Building Information Modelling
safe and economical structural solution [1,2]. (BIM) technologies and computational systems are expected to enhance
However, the trial-and-error method often demands considerable the integration of optimised structural designs with more efficient de-
amount of time and effort [3,4]. Most structural engineering problems sign [16] and construction [17–19] procedures. More specifically, Chi
could be formulated as optimisation problems that achieve optimum et al. [20] recognised five areas that will become more relevant with
structural performance whilst satisfying conflicting design constraints the development of BIM technologies in the context of structural en-
[5]. Reinforcement design is no exception and designing it this way gineering:
would achieve both automation and optimisation at the same time [4].
This paper proposes a numerical process to achieve this. • Adoption of structural optimisation during the early design stages
Structural optimisation techniques [6] have been applied in many • Parametric design specifications for enhanced sustainability per-
fields of engineering [7–12] however their uptake in engineering formance
practices is sometime met with some resistance. Common reasons for • Intuitive decision-making models supported by advanced visualisa-
this reluctance include issues such as training and attitude, modelling tion techniques

⁎
Corresponding author at: 14 Upper Woburn Place, London WC1H 0NN.
E-mail address: ucabele@ucl.ac.uk (S. Eleftheriadis).

https://doi.org/10.1016/j.autcon.2018.10.005
Received 27 June 2018; Received in revised form 1 October 2018; Accepted 3 October 2018
0926-5805/ © 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/BY/4.0/).
S. Eleftheriadis et al. Automation in Construction 96 (2018) 366–377

• Numerical applications in realistic engineering examples support the reinforcement analysis for the slab component a set of
• Amplified collaboration and communication between design teams customised algorithms was constructed in juxtaposition with the FE
engine. If the relevant code constraints and limit states are satisfied
The current study investigates how Chi et al.'s [20] insights could be after the FE analysis, material schedules for the concrete and the steel
implemented in the context of RC building structures with flat slabs, reinforcement as well as the reinforcement zones in the slab are spe-
which are a very commonly used floor system with reinforced concrete cified. The same reinforcement specification procedure is repeated
structures. multiple times for the design optimisation of the slab using the NSGA-II
In the optimisation of flat slabs [21], the focus is on the careful until a set of optimised solutions is obtained and visualised in the Pareto
selection of slab thicknesses as the concrete in the slab constitutes the front. Thorough review of the optimisation principles as well as results
largest proportion of the floor material [22]. However, slab thickness from numerical examples can be found in [24].
optimisation is often limited by constructability constraints, which
dictate a small finite set of slab thickness options. On the other hand, 2.2. Automated reinforcement analysis
the optimisation of reinforcement could be achieved in various ways by
engineering practitioners, and could lead to significant material sav- The proposed automated design specification process comprises
ings. In a recent flat slab optimisation study, the reinforcement ac- three main stages:
counts for approximately 25% of the total material and construction
costs in the floor [21]. Similar figures have also been reported in other 1) Processing data from BIM so that the floor system can be analysed
studies conducted by Sahab et al. [23] and Eleftheriadis et al. [22,24] with a finite element (FE) model;
with flat slab systems. 2) Generating refined reinforcement maps that match the mesh size of
Previous studies have focused on the reinforcement optimisation of the finite element model;
different structural frames mainly without the implementation of BIM 3) Simplification and smoothing of the refined mesh to ensure it is
technologies [25–27]. However, in a recent study by Mangal & Cheng practical to build.
[4] the reinforcement (longitudinal and shear) of RC frames (beams and
columns) was optimised using a BIM-based approach. Little attention In this paper, Autodesk Robot Structural Analysis (RSA) 2016 was
was given on the automation of reinforcement specifications to support used for the FE calculations and Autodesk Revit for BIM as these are
the optimisation of RC flat slabs. currently very commonly used in the industry [16]. The entire com-
In this study an automated reinforcement specification process is putational process is driven automatically through code developed in
proposed to effectively support the optimisation of RC flat slabs, which C# using the Application Programme Interface (API) of RSA and Revit.
is a structural system that is used extensively in the UK and many other The project requirements define the material properties, load cases and
countries. Section 2 describes the computational processes used in this support coordinates which are directly transferred from BIM to the FE
study to automate the specification of steel reinforcement. Numerical model via the API. The slab-column connections are modelled in the FE
applications and the validation of the proposed computational model model assuming pinned supports. The limit state checks are specified
are presented in Section 3. The paper concludes with discussion in based on national or international codes. In this study, all structural
Section 4 and conclusions in Section 5. limit states were checked according to Eurocode 2 (EC2) [28,29].
Once the structural model is established in RSA, the calculations
2. Methods and models leading to the estimation of the slab reinforcement begin. Firstly, de-
tailed required reinforcement maps are calculated at the resolution of
2.1. Optimisation framework and context the finite element mesh size. These refined maps are then smoothed out
and simplified into practical reinforcement bar specifications. At the
The BIM-based approach that was initially proposed by Eleftheriadis end of this process, the reinforcement schedule as well as the detailed
et al. [22] for the optimisation of RC flat slabs and columns using single reinforcement weight of the slab are obtained. Fig. 2 shows the general
objective functions for cost and embodied carbon, was extended to si- computational workflow including the necessary processes for the cal-
multaneously evaluate multiple objectives deploying a bespoke NSGA-II culation of the required and the specified reinforcement.
algorithm with a FEM engine [24]. In this paper, special attention is
given to the computational modules and processes of the optimisation 2.3. Required reinforcement calculations
approach described in [24] that are responsible for the automated re-
inforcement specification of RC floor structures. The specification of the Using the FEM output, the required reinforcement is calculated as
reinforcement is an important component of the optimisation analysis an area of steel per unit length for top and bottom reinforcement in
enabling the computation of structural material quantities necessary for both directions at each node of the finite element mesh. An example of
the cost and embodied carbon calculations. Typically, the specification a map showing this required reinforcement is presented in Fig. 3 for a
of reinforcement in RC floors is completed by structural engineers generic slab component.
during the early design stages using aggregate reinforcement rates (the The map in Fig. 3 highlights the areas where no reinforcement is
quantities are given kg/m3). The main aim of the proposed reinforce- required as well as the areas where reinforcement is needed based on
ment specification procedure was to ensure that detailed yet practical the code restrictions. As the reinforcement follows the FE mesh, this
reinforcement topologies (layouts) and schedules (quantities) are uti- map represents the smallest amount of steel that must be provided for a
lised not only for optimisation purposes but also for the refinement of given FE mesh size. Punching shear reinforcement has not been speci-
construction drawings and design information at detailed design stages fied at this stage of the project and thus, it is not included in the scope
by the structural engineers through dedicated BIM interoperability of this study.
[24]. Because the optimisation of a structure involves multiple itera- Coons' method [30] is used to generate the finite mesh in the slab,
tions until a set of optimised solutions is adequately obtained (in some and the Wood & Armer method [31] is used in the calculation of the
cases thousands of iterations might be required depending on the moment for the required reinforcement in the slab. The finite element
complexity of the building case [24]), the detailed analysis of the slab mesh size can be adjusted by the user based on the project require-
reinforcement should be automated in an efficient and robust way. The ments. Herein it was initialised using the value the structural engineers
general workflow of the optimisation is shown in Fig. 1. used (0.5 m) in the tested building scenario under examination. This
The optimisation procedure begins by querying building and geo- allowed a direct comparison and assessment of the conventional designs
metric information directly from a BIM model into the FEM engine. To with the computer-generated scenarios presented in Section 3. Details

367
S. Eleftheriadis et al. Automation in Construction 96 (2018) 366–377

Fig. 1. Suggested workflow for the optimisation of RC floors using the approach described in [24]. The automated specification algorithms are part of the FEM engine
and utilised for the computation of reinforcement quantities, which are subsequently used in the objective functions of the NSGA-II algorithm.

on the mesh sensitivity and the computational performance of the al- An additional function was incorporated in the proposed computa-
gorithm can be found in [24]. tional process to reduce the peak bending moments obtained from the
These required reinforcement maps are used to compute the FEM that typically appear over column supports. This function de-
minimum and maximum values of required reinforcement for each creases the risk of reinforcement overestimation in the slab using the
element (hereafter called cell) in the FE model. For either top or bottom column strip approach based on Annex I EC2 [29]. A section is taken
and for a given direction, the minimum value from all the FE cells is across the bending moment diagram (i.e. in the y direction for moments
used to estimate the basic reinforcement mesh across the slab whilst the in the x direction) at the face of the column.
difference between the maximum and the minimum values is used to
calculate the additional reinforcement in the zones where it is necessary
(typically above the columns or mid-span areas).

Fig. 2. Automated reinforcement computations flowchart. The computations are organised in three main components: 1) Computation of required reinforcement
areas using geometric, building and loading data, 2) Computation of basic reinforcement mesh and additional reinforcement bars using various algorithmic functions,
3) Computation of reinforcement schedules and material lists.

368
S. Eleftheriadis et al. Automation in Construction 96 (2018) 366–377

directions top and bottom. The number shown on each cell corresponds
to the weight factors obtained from the reinforcement database and
presented in the previous section. For example, the value 3.6 kg/m2
corresponds to ø12 at 250 mm or the value of weight factor 2.5 kg/m2
corresponds to ø10 at 250 mm. The algorithm assigns a value of 0
where no additional reinforcement is needed. The total weight of the
reinforcement in the slab is computed by multiplying the obtained
weight factors with the cell areas and finally by adding all the cell
weights together. The same maps are also generated for the basic re-
inforcement mesh but with only one reinforcement type applied in all
the cells.

2.5. Constructability constraints

Two constructability constraint functions are implemented to en-

hance the practical applicability of the computational solutions: 1)
spacing function, and 2) zoning function.

Fig. 3. Fine reinforcement map in a slab component showing the gradient of 2.5.1. Spacing function
required reinforcement as computed by FEM. This is a common way to visualise The spacing function ensures that the additional reinforcement bars
the areas in a slab that require more reinforcement (dark red zones). For any are placed between the bars of the basic mesh by preserving equal
given slab, four maps like that are typically computed (2 maps in each direction distances. This functionality significantly reduces the fitting and in-
of reinforcement, for top and bottom meshes). (For interpretation of the re-
stallation time of reinforcement bars on site. In the spacing function,
ferences to colour in this figure legend, the reader is referred to the web version
the bar spacing specified for the computations of the basic mesh is
of this article.)
maintained and re-used in the additional reinforcement calculations.
The algorithm checks the reinforcement database for the combination
2.4. Steel reinforcement specification options with the defined spacing and then identifies the smallest safe
reinforcement option.
Once the required reinforcement maps are calculated the re- Fig. 6 shows the implementation of the spacing requirement in a
inforcement bar specification is computed using a database of re- notional slab panel. Maintaining the same spacing (dadd = dbasic) allows
inforcement options. The database is encoded using a text file which the additional bars (in red lines) to be placed at the same interval as the
includes information about the diameters of the bars, the spacing be- basic mesh bars which are shown with black lines, ensuring consistent
tween bars, the reinforcement areas and the steel weight. The ultimate spacing throughout the slab.
output from this process will be the specified reinforcement which will
almost always be higher than the required reinforcement produced by 2.5.2. Zoning function
the first stage. The zoning function simplifies the one-to-one mapping for the ad-
The assignment process for the specified reinforcement is executed ditional reinforcement by altering the selection of bars on adjacent FE
using the RSA API and the FEM results from the required reinforcement cells to reduce short length bars that often complicate reinforcement
calculations. A custom algorithm queries the text file and identifies the fixing on site.
option that matches more closely to the minimum required reinforce- Fig. 7 illustrates how the zoning algorithm is implemented on an
ment values. The aim of this recursive procedure is to minimise the example slab component based on the horizontal direction of re-
absolute difference between the required reinforcement area values at inforcement. A representative scenario is shown for the highlighted
every finite element node and the reinforcement area value of the data cells in row 10. All the cells in columns 9–16 are analysed to proceed
points in the text file with the reinforcement options. with the cell adjustments. For example, in the entire column 9 no re-
This process is repeated for all the finite element nodes in the slab. inforcement is necessary, and as a result, this cell is not altered. For
The same specification process is carried out for the assignment of the cells 10–16, the maximum value is obtained (6.32 which corresponds to
reinforcement in the basic steel mesh as well as the additional re- ø16 at 250 mm of reinforcement) and the neighbouring cells are as-
inforcement mesh. The detailed functionalities associated with the sessed against that value. If the value in the remaining cells are different
computation of the basic and the additional reinforcement (spacing and from the maximum value, then they are updated accordingly. The same
zoning) are presented in the subsequent sections. process is applied in the X and Y directions, top and bottom re-
An example of the assignment algorithm is shown in Fig. 4. In a inforcement.
floor plate with required reinforcement of 322 mm2/m, the algorithm
finds that ø10 at 225 mm (349 mm2/m) would be the most suitable 3. Numerical examples
reinforcement option as it minimises the difference between the re-
quired reinforcement and the available options in the database. To To test the automated specification method the case study from [22]
ensure that the specified reinforcement is safe the computed difference was used. Six optimised reinforcement designs are generated by the
needs to be a positive number (> 0). computational process following six design scenarios involving various
Once an appropriate reinforcement option is found the algorithm constructability constraints. These were analysed and compared to a
stores the corresponding bar spacing as it is used in the spacing function conventional design produced by practicing engineers (without auto-
of the additional reinforcement (See Section 2.5). Finally, the total mation or any formal optimisation). The computational method utilises
weight of the basic mesh is calculated multiplying the mass factors the same geometric properties and the same project constraints for the
integrated within the database given in kg/m2 and the entire slab area structural system as the conventional design.
in m2. A similar reinforcement assignment process is implemented for
the estimation of the additional reinforcement. 3.1. Computational scenarios
Fig. 5 shows the one-to-one weight mapping of additional re-
inforcement for a simplified slab component in all reinforcement Different simulation cases were analysed to evaluate the behaviour

369
S. Eleftheriadis et al. Automation in Construction 96 (2018) 366–377

Fig. 4. The assignment process utilises required reinforcement data obtained from FEM and a detailed reinforcement database which includes various reinforcement
combinations including data for bar diameters, spacing between bars, reinforcement area and weight factors. The algorithm uses two main operations before
computing the provided reinforcement. The algorithm searches for the option in the database that closely matches (min difference) the minimum required re-
inforcement, whilst ensuring that the difference is a positive number to avoid the selection of a suboptimum option.

of the computational constraints. Table 1 summarises the six different In scenarios 1 and 2, the spacing and zoning functions are switched
simulation scenarios that were tested in this study to assess how well off, which means that the computation of the reinforcement would be
the computational model performs against conventional practice. All the most optimum in terms of weight as no constructability constraints
six scenarios are evaluated by computing the total reinforcement are utilised. These scenarios would be better suited for automated
weight in the slab. The classification of the scenarios is based on two construction and fabrication processes such as reinforcement mats,
reinforcement databases (D1, D2) and spacing along with zoning which allow variable spacing and bars to be used. Scenarios 3 and 4
functions. utilise the zoning function, which allows the additional reinforcement
Database D1 consists of 24 reinforcement options using discrete sets bars to be organised based on constructability rules but not the spacing
of inputs for the bar sizes and the reinforcement spacing. In this data- constraint. Finally, scenarios 5 and 6 fully utilise both the spacing and
base (D1), the bar sizes vary from ø10mm to ø32mm and the spacing zoning functions. The results are expected to be similar to the results
between bars varies from 175 mm to 250 mm with 25 mm increments. obtained from the conventional design process.
The intention behind this classification is to help the algorithm identify
reinforcement combinations that are closer to what practicing en- 3.2. Conventional design
gineers would use in real projects.
However, these constraints can be superseded and more or different Fig. 9 shows the general floor layout and the BIM model of the
reinforcement combinations could be implemented if desired. This structural system. The structure comprises 275 mm flat slabs with a
would yield more efficient results as the distance between the actual 250 mm thick central core. The column grid consists of variable
reinforcement option and the required reinforcement could be further 3 m × 3 m bay configurations on both directions. In the X-direction, the
optimised. To investigate how much these common spacing variables grid includes 6.5 m, 5 m and 4.5 m spans, whereas in the Y-direction,
affect the calculation results a more detailed databased (D2) of 138 the spans are 5 m, 4 m and 6 m. The columns are designed by the
reinforcement options is also developed. D2 includes the same bar project engineers, and they are 400 × 400 mm. The combined dead
diameters as D1, but with more spacing options which varying from load of the structure (DL), superimposed load (SDL) on the slab is
50 mm to 250 mm with 10 mm increments. 2.5 kN/m2 and imposed load (IL) is 7.5 kN/m2. Concrete strength of
Fig. 8 shows the number of reinforcement options for both data- C32/40 and C50 were chosen for the slab and columns respectively,
bases with their corresponding areas of reinforcement. It can be ob- whereas steel grade of B500 was used for the reinforcement bars.
served that five times more reinforcement options are available in D2 The plans in Fig. 10 show the top and bottom required reinforce-
when compared to D1's maximum capacity (4596 mm2/m). ment maps in both directions for the slab of the tested building. Fig. 10

370
S. Eleftheriadis et al. Automation in Construction 96 (2018) 366–377

Fig. 5. One-to-one mapping of additional reinforcement with weight factors given in kg/m2 (cell sizes 0.5mx0.5 m) for all four reinforcement directions. The figure
highlights that the proposed approach can achieve a high level of customisation when it comes to the selection of additional reinforcement to closely match the
required reinforcement maps described in Fig. 3. For each FE cell additional reinforcement bars are computed following the assignment process defined in Fig. 4.

Fig. 6. Spacing function to adjust the additional reinforcement positions in the slab. This function ensures that dadd is equal to dbasic, which helps improve con-
structability and reduce errors on site.

371
S. Eleftheriadis et al. Automation in Construction 96 (2018) 366–377

Fig. 7. Zoning algorithm for the computation of additional reinforcement using a logic implemented by engineering practitioners to simplify the reinforcement
topologies.

Table 1 a and b are the maps for the bottom reinforcement. Fig. 10 c and d on
Simulation scenarios summary using different reinforcement databases and the other hand, show the top reinforcement. The areas highlighted in
algorithmic components. red represent the zones with higher reinforcement requirements com-
1 2 3 4 5 6 pared to the areas highlighted in orange or yellow. The required re-
inforcement maps from the FEM are used to estimate the actual re-
Reinforcement database D1 D2 D1 D2 D1 D2 inforcement in the slab. The large peaks observed above the columns
Spacing constraint ✓ ✓
shown in Fig. 10 c and d are distributed across a larger area of column
Peak reduction ✓ ✓ ✓ ✓ ✓ ✓
Zoning constraint ✓ ✓ ✓ ✓ strips by averaging the bending moments as explained in Section 2.3.
The reinforcement specified by the project engineers is summarised
in Table 2. The basic mesh for the top and the bottom reinforcement is
ø10 at 200 mm, whereas the additional top reinforcement is ø16 at
200 mm and the additional bottom reinforcement is ø10 at 200 mm.
The total weight of the reinforcement in the slab is 4502 kg excluding
any laps. The conventional design includes spacing of 200 mm, which is
commonly used in practice, and it contains only 2 different bar dia-
meters (Ø10, Ø16). The number of different bars is used in this study as
a measure of design complexity when assessing the optimised design
options.
Fig. 11 highlights the zoning strategy of the reinforcement in the
conventional design. A basic mesh is applied both to top and bottom
reinforcement, and additional bars are added where more reinforce-
ment is required.

3.3. Verifications checks

3.3.1. Weight assessment

A simple check was carried out to ensure the feasibility of the
Fig. 8. Reinforcement data points for D1 and D2 databases (Logarithmic scale). computational process outputs. This was carried out by comparing the
The horizontal axis shows the number of different reinforcement options in the
reinforcement weight and layout for the conventional design to that
two databases D1 and D2. The vertical axis shows the corresponding area of
produced by the automated process imposing constraints on bar dia-
reinforcement for every reinforcement option.
meters and spacing. A simplified reinforcement database with 6 avail-
able bar diameters (ø10, ø12, ø16, ø20, ø25, ø32) and fixed spacing

372
S. Eleftheriadis et al. Automation in Construction 96 (2018) 366–377

Fig. 9. Structural BIM model and slab layout for the tested building scenario.

between bars (200 mm) was imposed as these are the alternatives the reduced from 25 mm to 20 mm, which is a more realistic solution for
project engineers used in their actual design proposal. The reinforce- the given building and load case.
ment in the slab was calculated using the computational process. The
total weight of reinforcement from the computational method for this 3.4. Simulation results
simplified database is 4517 kg which is a very close match (≈100%) to
that of the conventional design. The reinforcement layouts produced by 3.4.1. Material use
the computational design were also assessed against the conventional Fig. 14 summarises the total weight of the reinforcement in the slab
design. Fig. 12 shows an example of that comparison for the bottom for the six scenarios as obtained from the computational process against
reinforcement in the X direction. the conventional design. Results show that the selection of reinforce-
The reinforcement layout of the computational design (left) matches ment database has a small impact on the weight for the same con-
exactly the one the engineers specified in the conventional design structability constraints. The database with more detailed reinforce-
(right). In both designs the basic mesh consists of ø10 bas at 200 mm ment options (D2) can yield more optimised designs by only 2% when
spacing but with a few differences in the additional bar diameters as compared to database D1 with more limited reinforcement options.
highlighted in Fig. 12. The conventional design uses only ø10 bars Fig. 15 shows an example of this application considering the re-
(lines in blue) for the additional reinforcement but in the computational quired reinforcement data from all the FE cells in the slab for the
design 83% of the additional bars is ø10 with the remaining bars being bottom reinforcement (X direction). Despite the more accurate mapping
ø12 (lines in green) and ø16 (lines in purple). of the required reinforcement with database D2, its total reinforcement
This difference is mainly attributed to the accurate representation of weight is marginally improved against D1 due to the small differences
the bending moments shown in Fig. 10 a. This is also a good example obtained between the two provided reinforcement options and the
that demonstrates how the engineering practitioners often rationalise small occurrence of those differences. The most material-efficient op-
the specification of reinforcement across the entire slab. Similar beha- tions emerged in the scenarios where the constructability functions
viour of the computational process was observed for the reinforcement were disabled (in scenarios 1 and 2).
specification on the rest of the reinforcement layouts (top and bottom). The results show significant material savings, which can be attrib-
Overall, it can be observed that the computational design can effec- uted to the algorithm's one-to-one mapping of the required reinforce-
tively create a very close approximation of realistic reinforcement ment. The scenario using the D1 database reduces the total weight of
schedules and layouts. reinforcement in the slab by 21% when compared to the conventional
design and by 23% when using the D2 database. Furthermore, in sce-
3.3.2. Peak reduction narios 3 and 4 the total reinforcement weight is higher than scenarios 1
Fig. 13 a shows the reinforcement options for top and bottom me- and 2 by approximately 12% due to the rationalisation of reinforcement
shes in both directions for the building without reducing the bending occurred by the zoning function. On the other hand, scenarios 5 and 6
moment peaks. If the peaks are not effectively treated, the bars on top closely match the weight estimations for the conventional design by 3%
reinforcement can reach up to 25 mm in diameter which is obviously a and 5% respectively. This suggests that the zoning and spacing algo-
conservative value and is typically to be avoided in this kind of struc- rithms are effective at generating practical reinforcement designs that
tures unless completely necessary. To eliminate these situations, the are close to those obtained through conventional practice.
peak reduction subroutine described in Section 2.3 was implemented to
reduce the peak moments on the column by evenly distributing them on 3.4.2. Design complexity
the column strip zones. The resulting conditions for the top reinforce- The design complexity of the tested scenarios is assessed using a
ment bars and the adjusted peak reinforcement requirements are shown measure of the total number of the different reinforcement types im-
in Fig. 13 b. Using this algorithmic approach, the largest bar diameters plemented in each design. Currently engineering practitioners try to
which were observed above the columns zones can be effectively minimise the number of different bar diameters and spacing types used

373
S. Eleftheriadis et al. Automation in Construction 96 (2018) 366–377

Fig. 10. Maps showing the required areas of reinforcement for (a, b) bottom, (c, d) top reinforcement as obtained from the FEM model – Results displayed in mm2/m.

to ease fabrication on site. It is expected that a larger number of re- reinforcement types (13 in total).
inforcement types would result in more optimised designs due to a more A similar behavior in the model is observed in scenarios 3 and 4
accurate representation of the required reinforcement maps. Fig. 16 which implies the importance of the spacing function for the reduction
shows the total reinforcement weight for the six scenarios in ascending of the design complexity of the reinforcement design. On the other
weight order against their corresponding number of reinforcement hand, in scenarios 5 and 6, the constraints imposed by the zoning and
types. spacing functions seem not to affect the total number of reinforcement
Restricting the database of available reinforcement to commonly types used in the slab. Thus, the selection of the reinforcement database
used diameters and arrangements is crucial. Scenario 2, which is the becomes less relevant. Overall, it can be observed that more complex
most optimum configuration in terms of reinforcement weight uses 35 designs result in more material-efficient solutions. Therefore, under-
different reinforcement types. Instead, scenario 1 achieves slightly standing the design requirements and building conditions is necessary
worse performance by approximately 2% using significantly fewer when setting up the model.

Table 2
Conventional reinforcement quantities.
Basic mesh Weight factors (kg/m2) Weight (kg) Additional bars Weight factors (kg/m2) Weight (kg)

Bottom reinforcement Ø10 at 200 mm 3.085 1357 Ø10 at 200 mm 3.085 777
Top reinforcement Ø10 at 200 mm 3.085 1357 Ø16 at 200 mm 7.9 1011
2714 1788
Total 4502

374
S. Eleftheriadis et al. Automation in Construction 96 (2018) 366–377

Fig. 11. Top and bottom reinforcement layouts for the conventional design. Blue and red lines are used to show the reinforcement bars on X- and Y-directions
respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 12. Comparison of reinforcement layout (Bottom X) between the computational and the conventional approaches. The lines in blue colour correspond to ø10 at
200 mm reinforcement bars, in green colour to ø12 at 200 mm bars and in purple to ø16 at 200 mm bars. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)

4. Discussion and implementation 4.1. Design and construction synergies

There is an increasing demand by engineering practitioners to op- The proposed design optimisation and specification practices could
timise various elements of buildings structures, to improve efficiencies create new synergies between design and construction processes like
[24], reduce costs or enhance the environmental performance of their the one proposed by Bamtec Reinforcement [34] that allows engineers
designs [32,33]. To effectively improve the current practices in to more accurately design the reinforcement in concrete slabs with
common engineering projects most of the design specification proce- variable bars and spacing using rolling meshes. This technology can
dures will have to be automated. The paper presented a computational reduce the steel-fixing time of reinforcement, whilst offering steel
approach that automates the reinforcement specification in RC floors savings by accommodating various combination of bars.
supporting a BIM-based optimisation of RC structures. Additionally, in recent studies robotic technologies appear to im-
prove the fabrication processes in the construction industry [35] and

375
S. Eleftheriadis et al. Automation in Construction 96 (2018) 366–377

Fig. 14. Total reinforcement weight in kg for the six computed scenarios and
the conventional scenario.

Fig. 13. (a) Summary of additional reinforcement options without peak re-
ductions for top and bottom reinforcement in both directions, (b) Comparison
between the obtained diameters for top reinforcement in both direction with
(dotted lines) and without (solid lines) the peak reduction algorithm. In both
figures, the horizontal axis shows in how many FE cells each reinforcement type Fig. 15. Discrepancies between D1 (solid line) and D2 (dotted line) databases in
occurs. The vertical axis shows the corresponding weight of the different re- the mapping of required reinforcement for the bottom reinforcement in X di-
inforcement options in kg/m2. rection. Due to the larger number of spacing options in D2 database against D1,
a closer approximation of the required reinforcement can be observed.
particularly in concrete structures [36], whilst allowing the construc-
tion of more complex reinforcement at little or no extra cost. Using the the collaboration with architects and contractors as frequent changes in
proposed reinforcement specification as an integrated front-end to these the design could be accommodated in a cost-effective manner.
automated construction technologies has the potential to yield truly
optimised engineering design workflows that can manage the increased
5. Conclusions
complexity of building structures whilst enabling resource efficient and
cost effective solutions in a timely manner.
The study proposed an automated reinforcement specification ap-
proach that integrates code verification within a BIM-based optimisa-
4.2. Design collaboration tion procedure. The developed computational process used structural
BIM data to initiate the FEM analysis, and it incorporated several
The traditional approaches for steel reinforcement specification re- smoothing functions applying constructability constraints to obtain
quire manual calculations, which are often time-consuming and error- designs of acceptable complexity. The practical dimensions of the new
prone. The rapid assessment of the required reinforcement in the slab design model were evaluated in real building scenarios against solu-
could assist engineering practitioners in the early stages of the design tions generated by structural engineering practitioners. It was found
development when various slab configurations are analysed in a short that the constructability constraints significantly influence the results.
time. The time savings associated with the reinforcement analysis are The proposed automated method can effectively achieve detailed re-
expected to be considerable as manual data processing is not needed in inforcement designs. In the future, the proposed automated design
the proposed approach. In addition to the increase in productivity, the process could be integrated with new fabrication processes to achieve
parametric nature of the proposed specification significantly improves more efficient structural design systems overall.

376
S. Eleftheriadis et al. Automation in Construction 96 (2018) 366–377

particle swarm optimization with time varying acceleration coefficients, World

Appl. Sci. J. 13 (12) (2011) 2484–2494 (ISSN 1818-4952).
[15] M. Kripka, G. Medeiros, A. Lemonge, Use of optimization for automatic grouping of
beam cross-section dimensions in reinforced concrete building structures, Eng.
Struct. 99 (September 2015) 311–318, https://doi.org/10.1016/j.engstruct.2015.
05.001.
[16] S. Eleftheriadis, D. Mumovic, P. Greening, Life cycle energy efficiency in building
structures: a review of current developments and future outlooks based on BIM
capabilities, Renew. Sust. Energ. Rev. 67 (January 2017) 811–825, https://doi.org/
10.1016/j.rser.2016.09.028.
[17] S. Song, J. Yang, N. Kim, Development of a BIM-based structural framework opti-
mization and simulation system for building construction, Comput. Ind. 63 (9)
(December 2012) 895–912, https://doi.org/10.1016/j.compind.2012.08.013.
[18] J. Tulke, J. Hanff, 4D construction sequence planning – new process and data
model, Proceedings of the CIB W078 24th International Conference on Information
Technology in Construction, Maribor, Slovenia, 2007 [Online]. Available http://
www.irbnet.de/daten/iconda/CIB21474.pdf , Accessed date: 1 October 2018.
[19] J. Tulke, M. Nour, K. Beucke, A dynamic framework for construction scheduling
based on BIM using IFC, Proceedings of the 17th IABSE Congress, Chicago, USA,
2008, https://doi.org/10.2749/222137908796292119.
[20] H.-L. Chi, X. Wang, Y. Jiao, BIM-enabled structural design: impacts and future
developments in structural modelling, analysis and optimisation processes, Arch.
Comput. Meth. Eng. 22 (1) (January 2015) 135–151, https://doi.org/10.1007/
s11831-014-9127-7.
[21] M. Aldwaik, H. Adeli, Cost optimization of reinforced concrete flat slabs of arbitrary
configuration in irregular highrise building structures, Struct. Multidiscip. Optim.
54 (1) (July 2016) 151–164, https://doi.org/10.1007/s00158-016-1483-5.
Fig. 16. Relationship between total reinforcement weight and design com- [22] S. Eleftheriadis, P. Duffour, P. Greening, J. James, D. Mumovic, Multilevel com-
plexity. putational model for cost and carbon optimisation of reinforced concrete floor
systems, 34th International Symposium on Automation and Robotics in
Construction (ISARC 2017), Taipei, Taiwan, 2017 9781510844735.
Acknowledgements [23] M. Sahab, A. Ashour, V. Toropov, Cost optimisation of reinforced concrete flat slab
buildings, Eng. Struct. 27 (3) (February 2005) 313–322, https://doi.org/10.1016/j.
engstruct.2004.10.002.
This research has been made possible through funding provided by [24] S. Eleftheriadis, P. Duffour, P. Greening, J. James, B. Stephenson, D. Mumovic,
the Engineering and Physical Sciences Research Council (EPSRC) and Investigating relationships between cost and CO2 emissions in reinforced concrete
from Price & Myers LLP via the UCL EngD Centre (Grant number: EP/ structures using a BIM-based design optimisation approach, Energ. Buildings 166
(May 2018) 330–346, https://doi.org/10.1016/j.enbuild.2018.01.059.
G037159/1) in Virtual Environments, Interaction and Visualisation and
[25] B. Saini, V. Sehgala, M. Gambhirb, Genetically optimized artificial neural network
this is gratefully acknowledged here. based optimum design of singly and doubly reinforced concrete beams, Asian J.
Civil Eng. 7 (6) (2006) 603–619 [Online]. Available https://ajce.bhrc.ac.ir/Portals/
References 25/PropertyAgent/2905/Files/6268/603.pdf , Accessed date: 1 October 2018.
[26] M. Hadi, Neural networks applications in concrete structures, Comput. Struct. 81
(6) (March 2003) 373–381, https://doi.org/10.1016/S0045-7949(02)00451-0.
[1] C. Coello, A. Christiansen, F. Santos Hernandez, A simple genetic algorithm for the [27] F. Ahmadkhanlou, H. Adeli, Optimum cost design of reinforced concrete slabs using
design of reinforced concrete beams, Eng. Comput. 13 (4) (December 1997) neural dynamics model, Eng. Appl. Artif. Intell. 18 (1) (February 2005) 65–72,
185–196, https://doi.org/10.1007/BF01200046. https://doi.org/10.1016/j.engappai.2004.08.025.
[2] R. Rao, A material selection model using graph theory and matrix approach, Mater. [28] A. Bond, O. Brooker, A. Fraser, A. Harris, T. Harrison, A. Jones, R. Moss,
Sci. Eng. A 431 (1–2) (September 2006) 248–255, https://doi.org/10.1016/j.msea. R. Narayanan, R. Webster, How to design concrete structures to eurocode 2 - the
2006.06.006. compendium, MPA The Concrete Centre, London, UK, (2011) [Online]. Available
[3] A. Adedeji, Searching for the optimal design variables of RC flat slab by the reactive http://greganagno.com/download/Reinforced%20Concrete/How%20to%20design
taboo, Trends Appl. Sci. Res. 6 (4) (2011) 375–385, https://doi.org/10.3923/tasr. %20concrete%20structures%20using%20Eurocode%202.pdf , Accessed date: 1
2011.375.385. October 2018.
[4] M. Mangal, J. Cheng, Automated optimization of steel reinforcement in RC building [29] O. Brooker, How to design reinforced concrete flat slabs using Finite Element
frames using building information modeling and hybrid genetic algorithm, Autom. Analysis, MPA The Concrete Centre, London, UK, (2006) [Online]. Available
Constr. 90 (June 2018) 39–57, https://doi.org/10.1016/j.autcon.2018.01.013. https://www.concretecentre.com/Publications-Software/Publications/How-to-
[5] W. Bennage, A. Dhingra, Single and multiobjective structural optimization in dis- design-reinforced-concrete-flat-slabs-using.aspx , Accessed date: 1 October 2018.
crete-continuous variables using simulated annealing, Int. J. Numer. Methods Eng. [30] S. Coons, Surfaces for Computer-Aided Design of Space Forms, Massachusetts
38 (August) (1995) 2753–2773, https://doi.org/10.1002/nme.1620381606. Institute of Technology, Cambridge, MA, USA, 1967 [Online]. Available http://
[6] M. Cohn, A. Divonitzer, Application of structural optimization, J. Struct. Eng. 120 www.dtic.mil/dtic/tr/fulltext/u2/663504.pdf , Accessed date: 1 October 2018.
(2) (1994) 617–650, https://doi.org/10.1061/(ASCE)0733-9445(1994)120:2(617). [31] R. Wood, G. Armer, A. Hillerborg, The theory of the strip method for design of slabs,
[7] M. Jahjouh, M. Arafa, M. Alqedra, Artificial Bee Colony (ABC) algorithm in the Proc. Inst. Civ. Eng. 41 (2) (1968) 285–311, https://doi.org/10.1680/iicep.1968.
design optimization of RC continuous beams, J. Struct. Multidiscip. Optim. 47 (6) 7755.
(June 2013) 963–979, https://doi.org/10.1007/s00158-013-0884-y. [32] S. Eleftheriadis, P. Duffour, D. Mumovic, BIM-embedded life cycle carbon assess-
[8] W. Jenkins, Towards structural optimization via the genetic algorithm, Comput. ment of RC buildings using optimised structural design alternatives, Energ.
Struct. 40 (5) (1991) 1321–1327, https://doi.org/10.1016/0045-7949(91)90402-8. Buildings 173 (August 2018) 587–600, https://doi.org/10.1016/j.enbuild.2018.05.
[9] G. Zavala, A. Nebro, F. Luna, C. Coello Coello, A survey of multi-objective meta- 042.
heurtistics applied to structural optimization, Struct. Multidiscip. Optim. 49 (4) [33] S. Eleftheriadis, Y. Schwartz, R. Raslan, P. Duffour, D. Mumovic, Integrated building
(April 2014) 537–558, https://doi.org/10.1007/s00158-013-0996-4. life cycle carbon and cost analysis embedding multiple optimisation levels, Building
[10] M. Bensdoe, O. Sigmund, Topology Optimisation: Theory, Methods and Simulation and Optimization 2018, Cambridge, UK, 2018 [Online]. Available
Applications, Springer, 2002, https://doi.org/10.1007/978-3-662-05086-6. http://discovery.ucl.ac.uk/10056381/ , Accessed date: 1 October 2018.
[11] F. Glover, Heuristic for integer programming using surrogate constraints, Decis. Sci. [34] Bamtec, Bamtec system, [Online]. Available http://www.bamtec.com/Bamtec_en/
8 (1) (January 1977) 156–166, https://doi.org/10.1111/j.1540-5915.1977. Bamtec_System.htm, (2017) , Accessed date: 1 October 2018.
tb01074.x. [35] T. Sandy, M. Giftthaler, K. Dorfler, M. Kohler, J. Bunchli, Autonomous repositioning
[12] M. Pincus, Monte Carlo method for the approximate solution of certain types of and localization of an in situ fabricator, IEEE International Conference on Robotics
constrained optimization problems, Oper. Res. 18 (6) (November–December 1970) and Automation (ICRA), Stockholm, Sweden, 2016, https://doi.org/10.1109/ICRA.
967–1235, https://doi.org/10.1287/opre.18.6.1225. 2016.7487449.
[13] A. Tiwari, P. Hoyos, W. Hutabarat, C. Turner, N. Ince, X.-P. Gan, N. Prajapat, Survey [36] T. Wangler, E. Lloret, L. Reiter, N. Hack, F. Gramazio, M. Kohler, M. Bernhard,
on the use of computational optimisation in UK engineering companies, CIRP J. B. Dillenburger, J. Buchli, N. Roussel, R. Flatt, Digital concrete: opportunities and
Manuf. Sci. Technol. 9 (January 2015) 57–68, https://doi.org/10.1016/j.cirpj. challenges, RILEM Tech. Lett. 1 (2016) 67–75, https://doi.org/10.21809/
2015.01.003. rilemtechlett.2016.16.
[14] B. Ahmadi-Nedushan, H. Varaee, Minimum cost design of concrete slabs using

377

View publication stats