Automation in Construction 27 (2012) 67–77

Contents lists available at SciVerse ScienceDirect

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

Early stage multi-level cost estimation for schematic BIM models

Franco K.T. Cheung a,⁎, Jonathan Rihan a, Joseph Tah a, David Duce b, Esra Kurul a
a
Department of Real Estate and Construction, Oxford Brookes University, UK
b
Department of Computing and Communication Technologies, Oxford Brookes University, UK

a r t i c l e i n f o a b s t r a c t

Article history: Previous studies reveal the need for a tool to evaluate various aspects of building design in early design stages.
Accepted 9 May 2012 This paper proposes a multi-attribute based tool to address this need. The tool allows users to evaluate the
Available online 9 June 2012 functionality, economics and performance of buildings concurrently with building design. To illustrate the
mechanics of the tool, the paper details the cost estimation module that enables quick and intuitive explora-
Keywords:
tion of early stage design in a popular 3D modelling environment. Measurements are automatically extracted
BIM
Early stage design
from 3D models and profile driven estimates are revised in real-time. The data model uses a flexible unit rate
Cost model system that can easily be extended to other estimate dimensions such as carbon estimates and waste estima-
NRM tion. The approach illustrated in this paper is applicable to 3D modelling environments that support free form
3D CAD geometry for massing purposes other than the one chosen for this study.
© 2012 Elsevier B.V. All rights reserved.

1. Introduction that the industry adoption rate varies geographically. The SmartMarket
Report [25,26] suggests that there is a strong growth of adoption rate
Building Information Modelling (BIM) as an emerging process and from 28% in 2007 to 48% in 2009 in the US AEC community. A similar
technology has attracted many initiatives across the globe recently. survey by buildingSMART Australasia also reveals high adoption rates
BIM is a digital representation of the geometric and non-geometric amongst various stakeholders in Australia but the result was considered
data of a facility [27]. In contrast to the conventional design and con- to be heavily biased and the actual adoption rates were recommended
struction approach in which individual professionals predominantly to be a lot lower by the experts [1]. In a different survey by the National
work in their own specialised arena, the BIM approach to building de- Building Specification [28] looking at construction professionals in the
sign and construction emphasises on knowledge sharing in the form UK, predominantly architects from small and medium-sized enterprises
of a single data repository that develops throughout the life cycle of a (SMEs), revealed that only 13% of the respondents are currently using
building. The ability to store different types of information from various BIM. Major potential barriers for adoption have been identified in the
stakeholders at different building stages makes building models rich in literature which includes the lack of standards; presence of skill gaps;
information. BIM is considered to benefit the architecture, engineering hesitation to embrace technological changes within organisations; pres-
and construction (AEC) industry in terms of improving production effi- ence of externalities; ambiguity in data ownership and legal risks; and
ciency in the design and construction processes through shared infor- high investment cost [4,3,17,1]. Although it is evident that the adoption
mation and collaborative working amongst project participants, and rate is increasing, many argue that only large-scale projects implement
delivering buildings with better performance by taking into account BIM and the use by SMEs remains limited [1,2]. The situation is alarming
various attributes including whole life cost and environmental issues as as the majority of the AEC industry is made up of SMEs. For instance, 97%
design develops [9]. Actual benefits in measurable terms are evident of the construction industry in the EU have less than 20 employees and
from 32 case studies [15]. The applications of BIM are well documented 93% have less than 10 [34].
[9,3] and are expanding further to contract management such as e- BIM tools are widely available with comprehensive tools primarily
procurement [16] and facility management such as the facilitation of de- designed for architectural design such as Revit and ArchiCAD, and spe-
signer–user communication applications [36]. cialised tools such as Tekla and Structuralworks for structural design
BIM was first proposed during the late eighties to early nineties [8]. and fabrication, DesignEst Pro and Vico for cost estimating, and Design-
Its impact on the AEC industry is immense [38] although the adoption builder and VE Pro for energy assessment. Interoperability is the key un-
has been considerably slower than expected [4]. Recent surveys reveal derpinning the concept of BIM that allows various stakeholders to work
together and software to exchange information seamlessly. BIM tools
developed by individual software companies have been criticized as ad-
⁎ Corresponding author.
E-mail addresses: [email protected] (F.K.T. Cheung), [email protected]
dressing interoperability mainly amongst their own products rather
(J. Rihan), [email protected] (J. Tah), [email protected] (D. Duce), than other vendors' applications [18]. Although there are evolving stan-
[email protected] (E. Kurul). dards such as the Industry Foundation Classes (IFC) to ensure effective

0926-5805/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.autcon.2012.05.008
68 F.K.T. Cheung et al. / Automation in Construction 27 (2012) 67–77

interoperability, errors have been found in exchanges between tools Google 3D warehouse to share and download SketchUp models. Its
due to the lack of uniformity between the object schemas adopted by interoperability potential has not been fully exploited. This paper
individual BIM tool and the IFC objects and properties [31,30,19,32]. summarises part of the results from a funded research project on the de-
One approach to reducing these errors is through the introduction of velopment of innovative BIM decision making tools under the SketchUp
best practices for using each of the tools. However, care needs to be environment. This paper first introduces the multi-attribute knowledge-
taken to ensure the resulting best practices are not counter-intuitive based tool, the Low Impact Design Explorer (or LIDX), that assists designers
as rigid restrictions imposed to approaches to design problems or to evaluate early stage architectural models and provide feedback si-
designer's preferred workflows could render a BIM tool unpopular multaneously during the design process. The decision tool that contains
[40]. The dilemma is that in order to give informative feedback on de- a number of modules assessing various aspects of building design is cur-
sign, e.g. on its cost and performance, BIM tools need some level of rently under development. The focus of the paper is to explain the com-
formalised input but design is an intuitive and unstructured process in pleted cost estimating module in LIDX as building cost is well perceived
particular at early design stages. Apart from that, while a lot of commer- as one primary concern in development and designers are constantly
cial BIM tools are very powerful and contain a variety of add-on modules under pressure in designing to a cost limit. The novelty of the tool lies
to help in handling very specialised tasks, these tools and their add-ons in its ability to extract the highest level of geometric data from schematic
can be expensive to acquire. Many SMEs struggle just to survive through drawings to give real time cost estimate. The tool is able to update the
the current recession and cannot afford to invest on expensive BIM tools. estimate seamlessly as the designer defines the building envelop and in-
An alternative approach to deliver BIM is required to meet the needs of terior layout.
the industry comprising mainly of SMEs.
Cavieres et al. [6] argue that the current BIM tools are less able to 2. The LIDX approach
handle concurrency and integration at early design stages. The authors
urge the need for a re-distribution of efforts towards early knowledge Fig. 2 shows the flowchart of a conventional design management
integration and the provision of computational support during early approach focusing on design to work within a budget in conceptual
decision-making based on the arguments developed from three differ- design stage. Often, a cost planner is appointed to estimate building
ent studies; 1) knowledge on design economics and life cycle cost cost based on sketch drawings produced by the designer. Similarly,
needs to be fed in the conceptual design stages to maximise the use of it may involve other professionals to give advice on the functionality
knowledge [11] (as cited in [6]), 2) the potential impact for computa- and performance of the design. As design is an evolving process, the
tional tools to support decision is higher in the conceptual design time-lags between various assessment feedbacks, such as cost esti-
stage than the later stages [41] (as cited in [6]), and 3) the ability of de- mate and environmental performance assessment, and design develop-
cisions to affect costs and performance is higher and the cost of design ment make the conventional approach iterative, time-consuming and
changes is lower in the conceptual design stage than the later stages ineffective.
[29] (as cited in [6]). In response to the need for tools at early design The lack of a holistic assessment makes it difficult to achieve an
stages, Cavieres and others developed knowledge-based parametric optimal design solution considering that the design task is multi-
tools for concrete masonry walls. This study follows the arguments criteria based with increasing need to take environmental performance
supporting the need for early design stage tools but focuses on even ear- into account at early design stages. There are tools in the market to as-
lier and potentially more crucial design tasks; massing exercise and lay- sess the cost and performance of conceptual models. For instance,
out design. DProfiler by Beck Technology gives building cost estimates with ref-
In computer applications, it is rather common for users to employ erence to schematic building models. It can import 3D models from
one tool to deal with a type of task and another tool for a different type various design tools and has 3D modelling function. Simultaneous
of task even though the two tools may have overlapping functions to estimates can be generated by buildings modelled in DProfiler. Con-
handle both tasks. In the NBS BIM survey [28], a majority of the respon- trast to DProfiler which focuses estimating function in conceptual
dents were found to use Autodesk AutoCAD as their primary software to design stage, Vico Cost Planner by Vico Software allow users to plan
produce CAD drawings while more than half of the respondents used and estimate building cost as design evolves. Vico's cost estimates can
Google SketchUp as their secondary software to produce 3D drawings. be updated automatically with reference to imported building models.
An informal investigation by the authors following the result reveals Both DProfiler and Vico Cost Planner are primarily evaluation tools.
that a lot of architectural practices use Google SketchUp to study various They provide solutions that depend on import and export of models
potential schemes via 3D SketchUp models during the conceptual to facilitate interoperability. Some design tools also have functions to
design phase and develop detailed drawings mainly for the sake of evaluate cost and performance. For example, Conceptual Design Envi-
production information using other drafting tools such as AutoCAD. ronment in Revit allows users to obtain quantities from mass models,
The latter are often considered as more functional but less flexible create formulae and generate building estimates in real time. It does
to use. not however contain built-in cost databases. Also, it requires knowledge
SketchUp, on the other hand, is generally considered as an intuitive, of cost models to build them properly in the tool. In practice, it is not un-
simple and easy-to-use 3D design tool. It has a free version which has common for designers to use very coarse methods to carry out the neces-
been widely used by designers across different disciplines. Designers sary assessment or to use judgement by referring to their rules of thumb
across various disciplines including architecture, urban design, interior to evaluate their design for small sized projects due to the lack of budget
design and product design make use of SketchUp to carry out prelimi- for seeking advice from other consultants. There is a clear need for a
nary modelling exercise. Due to its popularity, relevant plug-ins have designer-focused system that can give simultaneous design assessment
been developed by the common BIM tools to import SketchUp files to on various aspects in the conceptual design stage.
their environment to develop BIM models. Although BIM models can This paper proposes the use of a knowledge based tool, Low Im-
be created from imported SketchUp model files, the authors argue pact Design Explorer (LIDX), for design assessment under the popular
that the workflow may create redundancy in definitions and more im- Google SketchUp environment. LIDX is a plug-in to SketchUp which
portant, may lose the opportunity to optimize the building design allows users to define building models and performs concurrent
since as described, earlier design decisions significantly influence the multi-attribute assessment (see Fig. 3). The evolving nature of design
performance and life cycle cost of buildings. Also, design purely based is inherent in the ethos of LIDX in that the system assists designers to
on intuition and experiences is likely to be problematic [12]. build SketchUp BIM models to suit the different needs in the conceptual
SketchUp is primarily a visualisation tool. It can be linked to Google design stage. For instance, a designer may need to first establish the
Earth and Google Maps to obtain the geographical information, and to mass layout plan by carrying out a massing exercise, then configure
 F.K.T. Cheung et al. / Automation in Construction 27 (2012) 67–77 69

Fig. 1. Schematic design workflow using LIDX and the SketchUp environment.

the shape and height of individual buildings within the plan and finally various purposes and assist designers to define BIM models for
divide the space within the buildings. evaluations. LIDX is currently under development. This paper illus-
The merit of LIDX lies in its ability to automatically extract geo- trates the concept by presenting its completed cost estimating
metric data from SketchUp drawings produced for the sake of the module.
70 F.K.T. Cheung et al. / Automation in Construction 27 (2012) 67–77

Fig. 2. Conventional design management.

3. Cost module in LIDX a built-in automation facility in LIDX that simulates the process of
costing-a-design in practice as design develops. To enhance the appli-
In the conceptual design stage in which LIDX is pitched, informa- cability of the tool, automated cost estimates are reported in the same
tion is deemed to be coarse and limited. LIDX extracts as much geo- fashion as the convention of the industry. The tool makes use of the
metric data as possibly available to generate estimates. The tool Royal Institution of Chartered Surveyor (RICS) new rules of measure-
does not dictate to designers to use a particular approach to define ment (NRM) [39], widely used by building cost planners in the UK, as
them in the SketchUp environment or to make premature decisions the basis for the representation of building cost. Benefiting from fa-
on building specifications. It makes use of a number of cost models miliarity, cost data required for approximate estimating are available
depending on the level of data available in a SketchUp model. Instead from price books such as Spon's [22], Franklin & Andrew [14] and BCIS
of relying on an individual's preference and knowledge of cost [35].
models, LIDX chooses the cost models according to the type of infor- The cost database in LIDX comprises 14 entries, namely: 1) ele-
mation available; i.e. more sophisticated methods are employed for ment type (element, sub-element or component), 2) element name
cost estimation as the level of information detail develops. By doing (e.g. frame), 3) element description (e.g. Timber frame incorporating
this, it eliminates the subjectivity of individuals in choosing the ap- Structural Insulated Panels), 4) sub-element/element unit (e.g. m2), 5)
propriate cost models. Building cost estimates are generated through minimum element unit rate, 6) maximum element unit rate, 7) density

Fig. 3. Multi-dimensional design management.

 F.K.T. Cheung et al. / Automation in Construction 27 (2012) 67–77 71

of sub-element, 8) reference of density (e.g. GIA), 9) default assumption

(yes or no), 10) sundries allowance (low end), 11) sundries allowance
(high end), 12) data source reference (e.g. in-house), 13) adjustment re-
mark, and 14) type of rate (e.g. basic or extra over).
If data descriptions from more than one source are very similar
but their corresponding unit rates are different, the average rate is
used for estimating. Other methods such as Monte Carlo simulation
based on assigned probability density functions of unit rates can be
applied to LIDX to give probabilistic results but it was decided that
this would not be implemented in the early stage of development of
the tool for the sake of simplicity.
Although many experimental cost models have been proposed in the
last 40 years such as statistical and network-based models [5,21,24,
13,33,37,7,23,10,20], few are able to challenge the existing cost estimating
approach and their application in practice remains very limited. Practi-
tioners still prefer cost models that are simple to understand and easy to
use. All cost models used in practice are basically quantity-based which Fig. 4. Site definition functionality over a map imported from Google Maps within SketchUp.
can be expressed by the following equation: The site boundary is shown partially drawn in red.

ϕðCÞ ¼ ∑ q·r ð1Þ

ðq;r Þ∈C These inferences are determined based on rules suggested by ex-
perienced architects that were consulted during the development of
the tool. Further empirical studies can be done to ensure that the in-
where C is an arbitrary sized set of quantity and rate pairs (q,r) that
ference accurately reflects what designers want. LIDX does however
make up the costs for the model, q represents unit quantities of the cur-
let the user change the inferred building element specifications as
rent cost item and r represents unit rate of the current cost item.
per the user's discretion subsequently if they choose to develop its
When text appears in the superscript of quantity or rate such as
design further (see process 3 in Fig. 1).
q [GIA] or r[GIA], it denotes the measurement class that the quantity or
Next the user masses out a simple shape of the building he wishes
rate is with respect to (in this example gross internal area, or GIA),
to cost using the usual SketchUp shape tools. In the background LIDX
and where a number appears in the superscript of a the unit rate it rep-
monitors the shape as it changes and updates measurements automati-
resents a position in the NRM hierarchy, for example r〈1〉 with super-
cally. This is an important consideration as the tool integrates tightly
script 〈1〉 denotes that the quantity and rates q 〈1〉 is from the NRM‘1.
with existing massing workflow approaches, and does not require a dif-
Substructure’ group element.
ferent set of tools for the user to learn to mass their building. Fig. 5 shows
Three quantity-based cost models are described in NRM for early
a building mass created with the tool. While the user defines shape of
stage estimating purposes; i.e. the floor area model, functional unit
the building, the estimated cost (i.e. L1 cost estimate) is updated by
model and elemental model. These models are all quantity-based models
the tool after each change to the building size and shape.
comprising various types of quantities and unit rates.
3.1.3. Defining space with horizontal elements
3.1. Workflow in LIDX
Once the building mass is created and the user is satisfied with the
shape, he may start defining the vertical space of the building. The user
In LIDX, the cost model is selected (L1–L4) according to the level
can decide the number of floors and the storey height, and LIDX auto-
of information available from the schematic design. The different
matically generates each floor for the mass model in SketchUp. Fig. 6a
levels of information were identified through a workflow study with
shows the building mass model with floors generated by LIDX based
the architects and cost planners participating in the project. The iden-
on Fig. 5's model. In parallel to the automatic generation of building
tified workflow is shown in Fig. 1, and comprises five processes:
floors, LIDX assumes default specifications for building elements to esti-
1. Defining site boundary, mate building cost (i.e. L2 cost estimate). The user may change the spec-
2. Massing, ifications of building elements in the cost library to customize the model.
3. Defining space with horizontal elements,
4. Defining space with vertical elements and
5. Defining specific components for individual elements.

3.1.1. Defining site boundary

The workflow begins with the user defining the site boundary
using the site boundary function in the tool (see Fig. 4). The user
may select the location of their site using the existing Google Maps
functionality within SketchUp to geo-locate the site and import the
selected map into SketchUp prior to the definition of a site boundary.
Once the user has defined the site border over the imported map LIDX
can calculate the site area.

3.1.2. Massing
Before the user can generate cost estimates, they first need to de-
fine the building profile of their intended building model, such as the
use type, structure type, and ground condition etc., to allow the sys- Fig. 5. Building massing stage. The user changes the building shape using standard
tem to infer the appropriate building material type, foundation mate- Google SketchUp shape tools. Estimates are updated immediately after each change
rial type and external work type to use in the calculation. in the building shape.
72 F.K.T. Cheung et al. / Automation in Construction 27 (2012) 67–77

Fig. 6. Building floor plan. Using the floor editing tools, the building floors can be edited quickly and easily to try out different floor layouts. Once floors have been created (a), the
internal spaces can be defined by adding walls (b), and openings such as doors and windows can be added (c) by simply right-clicking on the wall and selecting the appropriate
opening type.

3.1.4. Defining space with vertical elements This means that when the whole building is selected, high level building
Once the floor geometry has been created for the building, the wide estimates and options are displayed (see Fig. 7), and when editing
user can define zones by using the standard SketchUp line tool to de- at floor level, room functional profile or wall options are displayed.
fine wall partitions on the floor geometry to create the desired floor When the user wishes to apply a specific rate to a building component,
layout. LIDX monitors these changes in the background and for each such as specify a particular type of internal wall construction, they sim-
line drawn it attaches additional properties to the line geometry object ply need to select the item and uses the editor GUI to browse for the de-
associated with the wall representation. Fig. 6b shows a zone created on sired unit rate. This specified rate is then used to produce an estimate
the 1st floor of the building. The user can create openings such as doors that reflects the user's specifications (i.e. L4 cost estimate).
and windows by simply right-clicking on the desired wall and selecting
the appropriate opening type. The openings can be moved around the
4. Cost models in LIDX
floor, scaled, rotated, or can moved to fit with other walls. Fig. 6c
shows a more developed floor plan with two door openings added,
There are four levels of data associated to the workflow described
and illustrates how openings affect the 3D visualisation of the walls
above. LIDX generates estimates seamlessly as the design evolves. The
and zones. Functional type can then be assigned to an individual build-
estimates are calculated according to the quantity based cost model
ing zone by using the right-click menu as shown in Fig. 7. LIDX has a de-
as described in Eq. (1). Four levels of data are used according to the
fault functional profile containing specifications of components for
workflow to determine the type of quantities that are used in esti-
building elements for each functional type. Cost estimate is then gener-
mating, and are described in the following sections.
ated according to the model with space defined (i.e. L3 cost estimate).

3.1.5. Defining specific components for individual elements 4.1. Level 1 estimate
Using the context sensitive editor window, the user can further de-
fine the composition of specific components for individual elements. As The L1 model gives ballpark estimates. In LIDX, it is applied to esti-
the user adds more detail to the building model the LIDX GUI changes mate the cost of a development when only mass models are available
depending on the context of the building component being selected. in the design; i.e. at the stage when buildings are represented by mass

Fig. 7. A functional profile may be applied to any zone on the floor model. This is done by right-clicking on the selected face and selecting the desired functional type (a), once done
the profile is applied to the zone with the associated colour (b). Profiles are managed using an editor window.
 F.K.T. Cheung et al. / Automation in Construction 27 (2012) 67–77 73

blocks and designers use those mass blocks to appraise the potential allows LIDX to give responsive estimates by capturing the cost implica-
and restrictions of a development as part of the feasibility study or the tions of specification changes.
study to establish design requirements. The quantity-based model as At the level 2 estimate, the C〈2〉 quantities and rates become:
described by Eq. (1) can be applied to the mass model estimate where
q represents the various types of areas and r represents the cost per 〈2〉
L2 jC ¼ C〈2:1〉∪C〈2:2〉∪C〈2:3〉∪C〈2:4〉∪C〈2:5〉∪C〈2:6〉∪C〈2:7〉∪C〈2:8〉 ð7Þ
unit area. The L1 cost model is composed of 2 parts: (1) building work
elements represented by ground floor area (GA) for substructure and
 Which can be written more compactly in a shorter form as:
GIA for superstructure, and (2) external work representing by site area.  〈2〉 〈2:1 2:8〉
where the L2 jC notation explicitly defines the cost
L2 C ¼C
These two cost quantities for the L1 estimate are combined to give: level referred to for a given set C of quantity and rates, and can be omit-
ted when cost level is clear from the context. Similarly, using the same
CL1 ¼ CE ∪CB ð2Þ
shorthand notation C〈3〉 at L2 becomes: L2 C〈3〉 ¼ C〈3:1−3:3〉:


where CE defines the external work and CB defines the building work el- Tables 4 and 5 show the breakdown of the respective element
ements for each building type. CL1 is used to calculate the final L1 cost groups and the associated measurements used.
estimate as follows: Using these quantities, the L2 cost model is defined as follows:

L1 ¼ ϕðCL1 Þ·p^ ð3Þ L2 ¼ ϕðCL2 Þ·p^ ð8Þ

p^ ¼ 1 þ p〈10〉 þ p〈11〉 þ p〈14〉 ð4Þ where the set of L2 building quantity and rates is defined as:

where L1 is the total estimated L1 cost, ϕ(.) is the costing function defined CL2 ¼L2 jCE ∪L2 jCB ð9Þ
in Eq. (1), p^ is the percentage allowance factor combining p〈10〉, p〈11〉 and
p〈14〉 and are the allowances for preliminaries, overheads and risk, respec- and using the compact notation defined above, the building L2 quantity
tively and correspond to the NRM group elements 10, 11 and 14. and rates are defined as:
The cost quantities CE and CB are defined as follows:
nB
n

½SAŠ
o L2 jCB ¼ ∪ 〈1−9〉:
L2 jCi 10
CE ¼ q ; r〈6〉 ð5Þ i¼1

nB
CB ¼ ∪ Ci 〈1〉
C〈2〉 C〈3〉 C〈4〉 C〈5〉 C〈6〉 C〈7〉 C〈8〉 C〈9〉
∪ i ∪ i ∪ i ∪ i ∪ i ∪ i ∪ i ∪ i ð6Þ 4.3. Level 3 estimate
i¼1
 
where quantity q[SA] with superscript [SA] represents the site area mea- In the level 3 estimate, the L3 C〈2:1−2:8〉 and L3 C〈3:1−3:3〉 quantities and
surement, r 〈6〉 is the estimated cost per unit of site external area works, rates are calculated using more detailed geometric information from
and nB is the total number of building types. A given set of quantities the model, and are used to provide a more accurate estimate of the
and rates expressed as C〈n〉 denotes that it is generated using quantities NRM superstructure, and fittings and furnishings group elements. The
and rates appropriate for the NRM hierarchy position n, so for example quantities are extracted using the measurements shown in Tables 2
C〈1〉 with superscript 〈1〉 denotes that the quantity and rates C〈1〉 is from and 3 respectively.
the NRM‘1. Substructure’ group element, C〈2〉 is from the NRM‘2. Sub- The cost assumptions generated in the L2 model for GIA unit cost rates
structure’ group element, and so on. are transferred to the L3 cost model, but they now use more accurate
Since geometric data of a SketchUp mass model is extractable, they measurements (i.e. they use the actual geometric data in the 3D model)
are used to calculate the various areas according to the rules as de- and the cost estimate becomes more accurate. For each unit rate GIA as-
scribed in NRM. The measurements and rates extracted for the GA sumption at L2, a corresponding L3 unit rate is found for the respective
(ground floor area) measurement in the NRM substructure group ele- NRM measurement (Tables 2 and 3) of the same element at L3.
ment are the same measurement extracted for L2 and L3 and is deter- The L3 cost model is defined as follows:
mined as shown in Table 1.
L3 ¼ ϕðCL3 Þ·p^ ð11Þ
4.2. Level 2 estimate
where the set of L3 building quantity and rates is defined as:
A default breakdown of the cost per unit of GIA by building ele-
ments in the NRM fashion is incorporated with reference to the in- CL3 ¼L3 jCE ∪L3 jCB ð12Þ
ferred specifications. As the user changes the inferred specifications,
LIDX uses L2 cost models, largely based on the element quantities ex- with the L3 quantity and rates defined as:
tractable from SketchUp, to perform element cost estimation. Individual
nB
element costs estimated from L2 cost models are used to replace the de-
L3 jCB ¼ ∪ 〈1−9〉:
L3 jCi ð13Þ
fault element cost from the L1 estimate. This substitutional approach i¼1

Table 1
Table illustrating how the NRM Substructure element group measurements for L1, L2 and L3 estimates is calculated within the tool. Measured geometry is shown in red.

1. Substructure GA (m2) Substructure is represented by the area of the ground

floor measured to the internal face of the external perimeter walls.
74 F.K.T. Cheung et al. / Automation in Construction 27 (2012) 67–77

Table 2
Table illustrating how the NRM Superstructure element measurements for L3 estimates are calculated within the tool. Measured geometry is shown in red.

2.1. Frame GIA (m2) Floor perimeter is found a horizontal intersection at each floor height.
Area calculation excludes external wall depth (shown in yellow).

2.2. Upper floors TUFA (m2) Area is calculated as for 2.1, but excludes ground floor.

2.3. Roof RA (m2) Calculated as the area of the roof as viewed from plan.

2.4. Stairs and ramps SR (nr) The number of stair units within the building, cost is by stair type.

2.5. External walls EW (m2) Calculated using the perimeter of the external wall for
each floor multiplied by the floor to floor height.

2.6 Windows and external doors WED (m2) For windows cost is calculated by finding the total glazing and
external door area in m2.

2.7. Internal walls and partitions IW (m2) Calculated using the centre line of the internal walls and partitions
multiplied by the floor height.

2.8. Internal doors ID (nr) Calculated by counting the number of doors nr for each door type.

4.4. Level 4 estimate work for lighting, and contain single unit costs (cost based is per item
added) such as a bath or shower unit.
As the user provides more details to the building model, they may When applied to an RFA quantity, each of these functional profiles
begin to define specific components, such as particular wall types and generates a set of quantity and rate pairs that can be incorporated in
constructions. This data is applied to individual geometry elements to the cost model, and is defined as follows: given a functional profile
and can be tailored to the level of detail required by the designer. F and an area quantity q [RFA] for a room space in a building, then let
Where this information is available it replaces the more general L3 es- F q½RFAŠ denote the application of the functional profile F to the
timate used for that component. area q [RFA], and is defined as:
The quantity measurements and rates of the NRM categories 4 and
5 for the L4 estimate are determined using a functional profile model. The
 n
 o n
 o
 ½RFAŠ ½RFAŠ ½RFAŠ  ½RFAŠ ½UŠ  ½UŠ
F q ¼ q ;r r ∈F ∪ 1; r r ∈F : ð14Þ
functional profile model uses the room floor area (RFA) measurement of
each room space within a building (e.g. bedroom, or living room) to cal-
culate fittings, furnishings and services associated with that use profile. The result is a set of quantity and rate pairs scaled appropriately
For instance a bathroom may need to include area based works (unit by q [RFA] and is compatible with the costing function ϕ(.) in the cost
cost based on RFA of the space) for things like plumbing, and electrical model.
 F.K.T. Cheung et al. / Automation in Construction 27 (2012) 67–77 75

Table 3
Table illustrating how the NRM finishes element measurements for L3 estimates are calculated within the tool. Measured geometry is shown in red.

3.1. Wall finishes WF (m2) Calculated using the surface area of the
walls m2 for each wall finish type.

3.2. Floor finishes FF (m2) Calculated using the surface area of the
floor m2 for each floor finish type.

3.3. Ceiling finishes CF (m2) Calculated using the surface area of the
ceiling m2 for each ceiling finish type.

Eq. (14) simply states that the functional profile for a given RFA The NRM group elements 4 and 5 at level 4 are calculated together
quantity measurement q [RFA] for an internal building space, is defined using the functional profiles that have been assigned to each usable
as the set of all RFA based functional rates (i.e. area based rates) com- building space in the building:
bined with the unit rate costs for any units associated with that pro-
nS
file (such as fittings or furniture included in the profile). The profile F 〈4−5〉 ½RFAŠ
¼ ∪ F i jqi
L4 jC ð19Þ
itself is just a collection of RFA rates and unit rates, i¼1

n o where nS is the total number of usable building spaces within the

½RFAŠ ½RFAŠ ½UŠ ½UŠ
F ¼ r 1 ; …; r nRFA ; r 1 ; …; r nU ð15Þ building.

4.5. Multi-level cost estimation

where nRFA and nU are the number of area and unit rates respectively.
The L4 cost model is defined as follows:
One advanced feature of LIDX is its ability to update cost estimates
simultaneously as the design develops. The tool can assess the implica-
L4 ¼ ϕðCL4 Þ·p^ ð16Þ tions of changes in building mass and building type, and reflect them in
the unit quantities and the cost per unit quantities respectively. Thus, L1
where the set of L4 building quantity and rates is defined as: estimates produced from different mass models permit users to analyse
the cost of alternative options spontaneously taking into account build-
CL4 ¼L4 jCE ∪L4 jCB ð17Þ ing type and size.
This substitutional approach is automatically applied throughout
with the L4 quantity and rates defined as: the design process as the user specifies more design details. The propor-
tion of the total cost made up by the respective levels can be shown to
nB
〈1−3〉 the user so that they cannot only see the current cost estimate, but to
L4 jCB ¼ ∪ L4 jCi ∪L4 jCi〈4−5〉∪L4 jCi〈6−9〉: ð18Þ
i¼1 what level of cost estimation each part of the building represents.

Table 4
Building specific L2 cost models adopted according to the NRM taxonomy in LIDX.

Ref Group element Elements Quantity Unit rate Cost model

Cost/m2 GA for substructure

 〈1〉
1 Substructure – Ground floor area (GA) L2 C

2 Superstructure 2.1 Frame Gross internal area (GIA) Cost/m2 GIA for each sub-element  〈2〉
L2 C
2.2 Upper floors
2.3 Roof
2.4 Stairs and ramps
2.5 External walls
2.6 Windows and external doors
2.7 Internal walls and partitions
2.8 Internal doors
Cost/m2 GIA for each sub-element
 〈3〉
3 Finishes 3.1 Wall finishes Gross internal area (GIA) L2
C
3.2 Floor finishes
3.3 Ceiling finishes
Cost/m2 GIA for fittings and furnishings
 〈4〉
4 Fittings and furnishings – Gross internal area (GIA) L2 C

5 Services – Gross internal area (GIA) Cost/m2 GIA for services L2 C
 〈5〉
76 F.K.T. Cheung et al. / Automation in Construction 27 (2012) 67–77

Table 5
External works and contractor allowances for L2 cost models adopted according to the NRM taxonomy in LIDX. Note: Reference items 12 (project/design team fees) and 13 (other
development/project costs) in the NRM are not adopted in LIDX.

Ref Group element Quantity Unit rate Cost model

6 Complete buildings and building units – Lump sum allowance for complete buildings and building units: default value is 0 r〈6〉
7 Work to existing buildings – Lump sum allowance for work to existing building (default value is 0) r〈7〉
8 External works External work area (EA) Cost/m2 EA for external works L2 jC
〈8〉

9 Facilitating works – Lump sum allowance for facilitating works (default value is 0) r〈9〉 
10 Main contractor's preliminaries – Percentage allowance (p〈10〉) for preliminaries; default value is 10% p10 ⋅L2 C〈1−9〉
11 Main contractor's overheads and profit – Percentage allowance (p〈11〉) for overheads and profit; default value is 5% p11 ⋅L2 C〈1−9〉
14 Risk – Percentage allowance (p〈14〉) for risk; default value is 10% p14 ⋅L2 C〈1−9〉

A designer may also want to appraise buildings with different consequences of a design change as they make them, and has not previ-
specifications during or after a massing exercise. Throughout the various ously been accomplished within the SketchUp environment.
processes in the proposed workflow, the user may open the property
editor and see clearly what has been assumed in the current cost 5.1. Defining the building floor plan
model and make changes if necessary. The user may also export the
various cost models based on different assumptions in building mass 5.1.1. Implementation of cost estimation
shape and/or specification for analysis in a spreadsheet environment. The implementation of the cost estimation module draws upon
The specification for individual elements can be inferred according to several developed system components. These core subsystems are:
the building type selected. Alternatively, the designer can specify indi- geometry processing, BIM editing tool, database management, estima-
vidual elements by selecting other options in the tool or creating a tion, and user interface.
new description together with an assumption on the unit cost. To generate a cost estimate for a building, a set of rules are applied
to a building profile (shown in the dialogue in Fig. 8a) to generate a set
5. Cost estimation tool of building inferences. Appropriate unit rates are found by searching a
cost database for unit rates that match these default inferences. Each
The costing tool is implemented as a Google SketchUp plug-in unit rate is paired with the appropriate measurement quantity that
module, and as such the majority of the tool is written in the Ruby the geometry subsystem has automatically extracted from the model
scripting language. For graphic user interface components, the tool to create a set of nR quantity and rate pairs C ¼ fðqi ; r i Þgni¼1
R
, where nR
uses Adobe Flex for both simple information dialogues and more com- is the number of unit rates in the estimate. Each of the unit rates is asso-
plex tool windows. ciated with a specific position in the NRM hierarchy and is used to allo-
The motivation behind the choice of Flex as the GUI system was cate the quantity and rate pairs to the appropriate group in the cost
the consistent ease of use and simplicity offered by the Flex framework model. A cost estimate is then found by applying the costing function
and the built in Flex web service support, which meant that the number ϕ(.) to each group of quantity and rate pairs to generate a cost report
of external dependencies is kept at a minimum. which is displayed to the using NRM report structure (see Fig. 8b).
The tool transports structured data as JSON objects between the The user may easily and quickly explore different building profiles and
SketchUp Ruby environment, via JavaScript, to the adobe Flex GUI immediately see a new cost estimate, or provide new unit rates to refine
components. Since JSON is well supported by both the Ruby environ- the building cost estimate. The LIDX cost module makes the iterative pro-
ment and the Flex GUI system, it is a natural choice for passing data cesses of designing-to-a-cost and costing-a-design more efficient.
between the two systems.
As with existing commercial BIM tools, the proposed SketchUp 6. Conclusion
costing tool can determine building measurements directly from the
building geometry that the user of the tool is editing. This facilitates The model proposed in this paper provides an intuitive method to
a real-time iterative design model so that the user can see the incorporate cost (and multi-level) estimation into the early stage of

(a) (b)

Fig. 8. GUI showing the building context view (a), and the corresponding NRM report breakdown (b). Changing the profile yields a set of building material inferences, and changes
to the model are updated in real-time.
 F.K.T. Cheung et al. / Automation in Construction 27 (2012) 67–77 77

design. This prototype system has been implemented as a module [9] C. Eastman, P. Teicholz, R. Sacks, K. Liston, BIM Handbook: A Guide to Building In-
formation Modeling for Owners, Managers, Architects, Engineers, Contractors,
within the free version of Google SketchUp, however the paradigm and Fabricators, John Wiley and Sons, NJ, 2008.
can be easily applied in other editing environments, and could be in- [10] M. Emsley, D. Lowe, A. Duff, A. Harding, A. Hickson, Data modelling and the appli-
tegrated into higher end systems for early stage estimation. cation of a neural network approach to the prediction of total construction costs,
Construction Management and Economics 20 (6) (2002) 465–472.
This study has also demonstrated that the SketchUp environment [11] W. Fabrycky, B. Blanchard, Life-cycle Cost and Economic Analysis, Prentice Hall,
can be augmented to include a simple early stage BIM system, with 1991.
the intention of addressing the misconception that it is impossible to [12] S. Fenves, H. Rivard, N. Gomez, Seed-config: a tool for conceptual structural
design in a collaborative building design environment, Artificial Intelligence in
use as a BIM tool, and potentially paving the way for more development Engineering 14 (3) (2000) 233–247.
of similar BIM tools in this area. [13] R. Flanagan, G. Norman, The relationship between construction price and height,
LIDX is composed of a number of modules. For instance, the Chartered Surveyor Building and Quantity Surveying Quarterly 5 (4) (1978)
68–71.
gbXML module from within LIDX is able to export building data di-
[14] Franklin and Andrews, UK Building Blackbook: The Cost and Carbon Guide Hutchins:
rectly into the other energy simulation software (e.g. DesignBuilder) 2011, Franklin and Andrews, 2010.
for energy assessment, to demonstrate interoperability. To do this the [15] J. Gao, M. Fischer, Framework and Case Studies Comparing Implementations and
internal LIDX representation of an early stage building designed using Impacts of 3D/4D Modeling across Projects, Center for Integrated Facility Engi-
neering Technical Report #TR172, March 2008, Stanford University.
the LIDX tool (which is modelled internally as a hierarchy of building [16] A. Grilo, R. Jardim-Goncalves, Challenging electronic procurement in the AEC sec-
objects and components) is interpreted for export into the gbXML for- tor: a BIM-based integrated perspective, Automation in Construction 20 (2)
mat. From this representation, the building's 3D geometry is automati- (2011) 107–114.
[17] N. Gu, K. London, Understanding and facilitating BIM adoption in the AEC indus-
cally created from the 2D schematic floor representation using a try, Automation in Construction 19 (8) (2010) 988–999.
constructive geometry inspired approach and building part associations [18] R. Jardim-Goncalves, A. Grilo, Building information modeling and interoperabili-
and properties. However, given that this paper focuses on presenting ty, Automation in Construction 19 (4) (2010) 388–397.
[19] Y. Jeong, C. Eastman, R. Sacks, I. Kaner, Benchmark tests for BIM data exchanges of
the early stage design cost module from within LIDX, the implementa- precast concrete, Automation in Construction 18 (4) (2009) 469–484.
tion details of the other modules are left for discussion in future work. [20] G. Kim, J. Yoon, S. An, H. Cho, K. Kang, Neural network model incorporating a ge-
More information about the modules in the LIDX system can be found netic algorithm in estimating construction costs, Building and Environment 39
(11) (2004) 1333–1340.
in the video demonstrations published on the project's YouTube chan- [21] V. Kouskoulas, E. Koehn, Predesign cost-estimation function for buildings, Journal
nel, http://www.youtube.com/user/tsbLIB/videos. of the Construction Division 100 (4) (1974) 589–604.
In future versions of the prototype, the early stage BIM data struc- [22] D. Langdon, Spon's Architects' and Builders' Price Book 2011, , 2010.
[23] H. Li, Neural networks for construction cost estimation, Building Research & In-
tures created using the tool could also be exported in a format that
formation 23 (5) (1995) 279–284.
can then be imported by other tools such as Revit using standards [24] R. McCaffer, Some examples of the use of regression analysis as an estimating
such as IFC. The exported model can then be developed further within tool, The Quantity Surveyor 8 (1975) 1–6.
the design tools, allowing the model to be carried through the complete [25] McGraw Hill Construction, Interoperability in the Construction Industry.
SmartMarket Report, Interoperability Issue, , 2007.
design process. [26] McGraw Hill Construction, Interoperability in the Construction Industry.
SmartMarket Report — The Business Value of BIM, 2009.
[27] National Institute of Building Sciences, United States National Building Information
Acknowledgements Modeling Standard Version 1 - Part 1: Overview, Principles, and Methodologies,
Final Report, December 2007.
[28] S. Hamil, What BIM is and How It is Being Used, CIBSE BIM: who benefits? conference
This research was supported by the Design and Decision Tools for Low presentation. 2nd December 2010.
Impact Buildings grant (Project No. 400151) from the UK Technology [29] The American Institute of Architects, Integrated Project Delivery: A Guide, Version 1,
Strategy Board (TSB) and Engineering and Physical Sciences Research 2007.
[30] J. Plume, J. Mitchell, Collaborative design using a shared IFC building model —
Council (EPSRC). The authors would also like to thank the participating learning from experience, Automation in Construction 16 (1) (2007) 28–36.
industrial partners of the research project, Best Foot Forward, Zedfactory [31] R. Sacks, C. Eastman, G. Lee, Parametric 3D modeling in building construction
Ltd., DesignBuilder Software Ltd. and advisor it sO works, for their contri- with examples from precast concrete, Automation in Construction 13 (3)
(2004) 291–312.
bution to the study. [32] R. Sacks, I. Kaner, C. Eastman, Y. Jeong, The Rosewood experiment — building in-
formation modeling and interoperability for architectural precast facades, Auto-
mation in Construction 19 (4) (2010) 419–432.
References [33] A. Saeed Karshenas, Predesign cost estimating method for multistory buildings,
Journal of Construction Engineering and Management 110 (1984) 79.
[1] Allen Consulting Group, Productivity in the buildings network: assessing the im- [34] F. Schultmann, N. Sunke, Study on the Future Opportunities and Challenges of
pacts of building information models, Report to the Built Environment Innovation EU-China Trade and Investment Relations — Study 9: Construction, European
and Industry Council, October 2010. Commission Trade, 2007.
[2] Y. Arayici, P. Coates, L. Koskela, M. Kagioglou, C. Usher, K. O'Reilly, Technology [35] Service, B. C. I., BCIS Wessex Comprehensive Building Price Book 2011–2012,
adoption in the BIM implementation for lean architectural practice, Automation Royal Institution of Chartered Surveyors (RICS), 2011.
in Construction (2010). [36] W. Shen, Q. Shen, Q. Sun, Building information modeling-based user activity simula-
[3] S. Azhar, M. Hein, B. Sketo, Building information modeling (BIM): benefits, risks and tion and evaluation method for improving designer–user communications, Automa-
challenges, Proceedings of the 44th ASC Annual Conference, Auburn, Alabama, tion in Construction 21 (1) (2012) 148–160.
2-5 Apr 2008, pp. 627–634. [37] R. Skitmore, B. Patchell, Developments in contract price forecasting and bidding
[4] P.G. Bernstein, J.H. Pittman, Barriers to the Adoption of Building Information techniques, Quantity Surveying Techniques: New Directions, BSP Professional
Modeling in the Building Industry, Autodesk Building Solutions, San Rafael, CA, Books, London, 1990, pp. 75–120.
2004. [38] Stanford University Centre for Integrated Facility Engineering, Specific Report
[5] J. Buchanan, Development of a cost model for the reinforced concrete frame of a Needed. Center for Integrated Engineering: CIFE Technical Reports, , 2007.
building, M. Sc. Project Report, Department of Civil Engineering, Loughborough [39] R. Tilyard, RICS New Rules of Measurement, Royal Institution of Chartered Sur-
University of Technology, 1969. veyors (RICS), 2009.
[6] A. Cavieres, R. Gentry, T. Al-Haddad, Knowledge-based parametric tools for con- [40] Ž. Turk, Phenomenologial foundations of conceptual product modelling in archi-
crete masonry walls: conceptual design and preliminary structural analysis, Au- tecture, engineering and construction, Artificial Intelligence in Engineering 15
tomation in Construction 20 (6) (2011) 716–728. (2) (2001) 83–92.
[7] K. Chau, Monte Carlo simulation of construction costs using subjective data, Con- [41] L. Wang, W. Shen, H. Xie, J. Neelamkavil, A. Pardasani, Collaborative conceptual
struction Management and Economics 13 (5) (1995) 369–383. design — state of the art and future trends, Computer-Aided Design 34 (13)
[8] C. Eastman, Modeling of buildings: evolution and concepts, Automation in Con- (2002) 981–996.
struction 1 (2) (1992) 99–109.