Accepted Manuscript

Title: Computational Geometry in the Context of

BuildingInformation Modeling

Author: Daniel Ladenhauf Kurt Battisti René Berndt Eva

Eggeling Dieter W. Fellner Markus Gratzl-Michlmair Torsten
Ullrich

PII: S0378-7788(15)00180-2
DOI: http://dx.doi.org/doi:10.1016/j.enbuild.2015.02.056
Reference: ENB 5730

To appear in: ENB

Received date: 24-11-2014

Revised date: 13-1-2015
Accepted date: 25-2-2015

Please cite this article as: Daniel Ladenhauf, Kurt Battisti, René Berndt, Eva
Eggeling, Dieter W. Fellner, Markus Gratzl-Michlmair, Torsten Ullrich, Computational
Geometry in the Context of BuildingInformation Modeling, Energy & Buildings (2015),
http://dx.doi.org/10.1016/j.enbuild.2015.02.056

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
Figure(s)

t
ip
cr
us
an
M
d
te
p
ce
Ac

Page 1 of 31
Figure(s)

i
cr
us
an
M
ed
pt
ce
Ac

Page 2 of 31
Figure(s)

i
cr
us
an
M
ed
pt
ce
Ac

Page 3 of 31
Figure(s)

i
cr
us
an
M
ed
pt
ce
Ac

Page 4 of 31
Figure(s)

i
cr
us
an
M
ed
pt
ce
Ac

Page 5 of 31
Figure(s)

i
cr
us
an
M
ed
pt
ce
Ac

Page 6 of 31
Figure(s)

i
cr
us
an
M
ed
pt
ce
Ac

Page 7 of 31
Figure(s)

i
cr
us
an
M
ed
pt
ce
Ac

Page 8 of 31
Figure(s)

t
ip
cr
us
an
M
d
te
p
ce
Ac

Page 9 of 31
Highlights:

* Building information modeling (BIM) has gained attention in recent years.

* The BIM paradigm aims at a comprehensive digital representation of buildings.
* BIM contains implicitly the required data for energy performance calculations.
* Building energy analysis can be performed automatically based on BIM data.

t
ip
cr
us
an
M
d
p te
ce
Ac

Page 10 of 31
 Computational Geometry in the Context of
Building Information Modeling

t
Daniel Ladenhaufd , Kurt Battistia , René Berndtd , Eva Eggelingd ,

ip
Dieter W. Fellnerc,d , Markus Gratzl-Michlmairb , Torsten Ullrichd,∗
a A-NULL Bauphysik GmbH, Vienna, Austria

cr
b Ingenieurbüro Gratzl e. U., Taufkirchen, Austria
c Fraunhofer IGD & Technische Universität Darmstadt, Germany
d Fraunhofer Austria Research GmbH & Technische Universität Graz,

Inffeldgasse 16c, A-8010 Graz, Austria

us
an
Abstract

Building energy analysis has gained attention in recent years, as awareness for
energy efficiency is rising in order to reduce greenhouse gas emissions. At the
M
same time, the building information modeling paradigm is aiming to develop
comprehensive digital representations of building characteristics based on se-
mantic 3D models. Most of the data required for energy performance calcu-
d

lation can be found in such models; however, extracting the relevant data is
te

not a trivial problem. This article presents an algorithm to prepare input data
for energy analysis based on building information models. The crucial aspect
p

is geometric simplification according to semantic constraints: the building ele-

ment geometries are reduced to a set of surfaces representing the thermal shell
ce

as well as the internal boundaries. These boundary parts are then associated
with material layers and thermally relevant data. The presented approach,
Ac

previously discussed at the International Academic Conference on Places and

Technologies [1], significantly reduces the needed time for energy analysis.
Keywords: building information model, industry foundation classes, energy
efficiency, geometry simplification

∗ Correspondingauthor
Email address: torsten.ullrich@fraunhofer.at (Torsten Ullrich)
URL: www.fraunhofer.at (Torsten Ullrich)

Preprint submitted to Energy and Buildings January 12, 2015

Page 11 of 31
 1. Motivation

The building sector, which accounts for over 40% of all energy use in the

t
European Union, is the single biggest European energy-consuming sector. The

ip
reduction of needed energy as well as the increase of energy efficiency has be-
5 come a societal responsibility. Buildings constructed and designed with a high

cr
degree of energy efficiency will reduce energy consumption. As a direct impact
this will decrease the wastage of conventional fossil fuels resulting in a reduc-

us
tion of emissions of certain atmospheric pollutants and the damage caused by
the greenhouse phenomenon. Economically optimized buildings are cheaper in
terms of running costs than conventional ones with regards to the evolution of

an
10

prices of raw materials and energy.

The assessment of the energy performance of a building takes place in the
M
middle of the planning process, during the preparation of the mandatory energy
performance certificate. Major decisions for energy efficiency, such as geometry,
15 orientation, opening planning or construction, however, is done much earlier –
d

during project development, pre-design and design. At this early stage, however,
most of the details about quality of components, heating, ventilation and air
te

conditioning (HVAC) systems, etc., are still missing, which makes the full energy
efficiency rating impossible. However, the costs of the preliminary estimate of
p

20 energy efficiency stay the same since the main factor is the detection of areas of
ce

the thermal shell of the building.

Assuming that the design is most likely to change in the further planning
process, this estimation is to be executed repeatedly. By computer-assisted
Ac

creation of the building model (based on the budget versions of the ongoing
25 planning process), this effort could be significantly reduced through automation,
making the integral energy efficiency planning much more realistic.
Based on previous work [1], we present a solution to improve and simplify
the thermal analysis for complex buildings: For the use of simple forecasting
calculations complex planning processes have to be reduced to simple structures
30 (e.g. geometry and system configurations). The majority of needed data is

Page 12 of 31
 already included in an Industry Foundation Classes (IFC) model. The semi-
automatic creation of such simple models would allow planners a much easier
detection and control of the energetic impact of their plans and measures.

t
ip
2. Related Work

cr
35 The approach and its algorithms presented in this article include techniques
from various fields of application: data management in the context of Building

us
Information Modeling (BIM), energy performance analysis in the context of
Building Energy Modeling (BEM), and computational geometry with a focus
on semantic modeling.

an
40 J. Mitchell points out, that thermal performance of buildings has a long
history of application and has become a central focus today as a means to
M
improving building environments yet using less resources [2]. This has put a
new demand for multiple performance optimizations that embraces the basics
of structural stability, health, egress, fire protection etc. with environmental
d

45 qualities of noise control, daylighting, and indoor comfort.

The enhanced product assembly and quality as well as the application of
te

advanced manufacturing techniques have added further dimensions to the mod-

eling and analysis task. E. Franconi introduces the concept of a framework
p

for building energy modeling methods and processes [3]. The effort involves
ce

50 examining modeling tasks across modeling applications, creating a structure for

organizing them, and specifying the detailed development of shared components.
The digital representation of the design process is reflected by the Industry
Ac

Foundation Classes, an open data model for building and construction industry
data developed by buildingSMART, formerly International Alliance for Inter-
55 operability [4]. Intended to facilitate interoperability in the building industry,
it has become an ISO standard [5] and a commonly used format for building
information models.
The question how to close the gap between BIM and energy performance
simulation has been a subject of research in recent years. Hitchcock and Wong

Page 13 of 31
60 analyze different efforts in this field, concluding that automatic data exchange
between BIMs and energy simulation is still an “elusive goal” [6].
The Lawrence Berkeley National Laboratory (Berkeley Lab) realizes the ap-

t
ip
proach most similar to the presented one: They are developing a methodology to
semi-automatically perform energy simulations based on original data from an

cr
65 IFC-based BIM. The methodology involves model checking (to ensure that the
BIM meets certain requirements), and, as its key step, automated data trans-

us
formation from the BIM to a format suitable for energy simulation [7]. For the
latter step, they developed an algorithm that splits up space boundaries (the
surfaces where spaces meet building elements) based on what is on the “other

an
70 side” of the building element, so that each resulting space boundary has con-
stant one-dimensional heat flow across its area. These split space boundaries
and building elements as vertices, as well as the physical connections between
M
them as edges, constitute paths of heat transmission in a building graph [8].
The main difference to our approach has its roots in the Austrian calculation
75 methodology for energy performance certificates. It is is defined in the OIB-
d

Richtlinie Energieeinsparung und Wärmeschutz and the ÖNORM B 8110. The

te

goal is to keep the effort for certificate issuance to a justifiable amount, yet pro-
vide “good enough” results, e.g. to enable meaningful comparisons of different
p

buildings on the market [9]. The input data necessary for such a calculation
80 includes:
ce

• building geometry data, namely the building elements’ area quantities,

gross area, and gross volume;
Ac

• the building elements’ thermal transmittance (U-values), or definition of

their material layers (material, thickness);

85 • heating-degree-days (in German: Heizgradtage) per year (determine the

period in which the space heating is used);

• monthly climate data (average temperature, solar radiation for horizontal

areas and all cardinal directions).

Page 14 of 31
 Another approach to facilitate exchange of IFC data between different tools
90 is the implementation of Model View Definitions (MVDs). This approach re-
sponds to the problem that one usually does not know what to expect from an

t
ip
IFC model: IFC, by its very nature, is rich, but flexible and redundant, because
it has to address various needs from architects, suppliers, engineers, and others.

cr
MVDs define subsets of the IFC schema for certain exchange tasks, outlining
95 what data are expected in a specific use case [10]. Official MVDs are being devel-

us
oped by buildingSMART, with the Coordination View being the first and most
widely implemented view [11]. For the Coordination View, there even exists a
“Space Boundary Add-On View” to support the usecase of energy analysis [12].

an
The specification of MVDs is done manually, and Venugopal et al. propose a
100 more formalized approach [13]. In any case, specifications are of course only the
first step, and it depends on BIM software vendors to implement them.
M
Meanwhile, the transformation from BIM to a format suitable for energy
analysis is based on computational geometry algorithms to simplify geometry
with respect to semantic constraints. These constraints are a set of rules formu-
d

105 lated by civil engineering experts with respect to thermal performance guidelines
te

and laws. Needed building information is extracted from the IFC data as much
as possible.
p

Applying computational geometry algorithms to BIM requires the incor-

poration of numerical robustness aspects. For an introduction to the robust-
ce

110 ness problems that arise with geometric computation we refer to the work of
Hoffmann [14]. A more recent discussion of the topic is the work of Kettner
et al. [15]. They discuss the problems of using floating point arithmetic for al-
Ac

gorithms that assume real arithmetic and show that simple algorithms may fail
under such conditions.
115 In the context of architecture and building information management, plane
based representations are of special interest. More specifically, Bernstein et
al. [16] and Krispel et al. [17] applied a plane based representation together
with exact floating point predicates [18] to calculate Boolean operations on solids
organized in binary space partitioning trees. Furthermore, the method has been

Page 15 of 31
120 applied for efficient robust polygonal boundary evaluation of swept Minkowski
sums [19].
Our algorithms use floating-point arithmetic as well. In contrast to the

t
ip
methods mentioned above, we employ additional information stored in IFC data
to cope with numerical issues: in most cases the result of a geometric operation

cr
125 (intersection, union, etc.) is (directly or indirectly) available. Therefore, it is
possible to use a snapping-based technique.

us
3. Algorithm

The goal of the algorithm is to create simple building models for energy anal-

an
ysis based on complex building information models. Currently, this task is done
130 manually, i.e., by re-creating models from scratch. The ideal algorithm would
M
do the same as a human expert who is modeling the data for energy analysis
by looking at and using the data from the IFC model. Faced with a complex,
yet incomplete representation of a real-world building, he or she decides, which
d

data are relevant and which are not based on semantic understanding.
135 A problem with IFC, or BIM in general, is that the semantics are often not
te

integrated in the way it would be possible. One cannot rely on all necessary
data to be available as a lot of attributes and relationships are optional, and
p

MVDs are not yet implemented on a broad basis.

ce

In response to the data quality issues in many IFC models, the input re-
140 quirements are kept to a reasonable minimum and many data are determined
through the semantically grouped geometries. The input data required for the
Ac

algorithm are:

• building elements (walls, slabs, windows, doors);

• spaces (there should be “no air” in the building that is not inside a space);

145 • full 3D geometry associated with all building elements and spaces, repre-
sented as geometry type than can be triangulated;

Page 16 of 31
 • thermally relevant data (material layers, window properties, etc.) associ-
ated with the building elements.

t
In a preprocessing step, all geometries are triangulated, so that the algo-

ip
150 rithm does not have to deal with the various geometry representations or differ-
ent local coordinate systems. Furthermore, the triangulation eliminates curved

cr
surfaces, which are not supported by energy analysis tools and thus need to be
approximated by planar surfaces. Subsequently, the triangulated geometries are

us
simplified by unioning the triangles that belong to the same object and lie on
155 the same plane. After this geometry preprocessing step, all geometries, both in-
put (building elements and spaces) as well as output (boundary parts, described

an
below), are represented as polygon soups. Each polygon is defined by the list of
points in global 3D coordinates, the normal vector (in its normalized form), and
the distance of its host plane from the origin. It also holds a reference to the
M
160 building element, as it is important for the algorithm that each polygon “knows”
the building element it belongs to, i.e., the semantic information is maintained.
d
For the collection, a map with the normal vector as key and the polygons sorted
by the distance as values is used for easy access during boundary detection. As
te

slight imprecision within the normal vectors cannot be ruled out, a tolerance
165 for the keys is applied within the implementation of the map.
p

Significant parts of the geometric simplification are performed in 2D space.

ce

Polygons are converted from 3D to 2D by assuming their host plane as xy-plane.

That is, the x-axis is an orthonormal vector to the normal vector, and the y-axis
is the cross product of this vector and the normal vector.
Ac

170 A central object for the algorithm is a “boundary part”. In BIMs, there are
tangible building elements (walls, slabs, doors, windows, etc.), and spaces. Even
though relationships between instances of these two basic types can be modeled
in IFC, they remain separated from each other. A wall, for example, might
have several spaces on either side. For energy analysis, this must be modeled
175 in a different way: the spaces on both sides of a building element constitute
“parts”, i.e., building elements are split up if necessary. Consequently, each

Page 17 of 31
 boundary part is associated with exactly one building element and one or two
spaces (one for exterior boundary parts, two for interior ones). The geometry
is homogeneous in orientation, i.e., all polygons share the same normal vector.

t
ip
180 The geometric simplification algorithm finds the boundary parts based on
the building element and space geometries in three steps:

cr
1. detecting the boundary per space;
2. merging boundary parts between spaces;

us
3. filling gaps in the boundary.

Algorithm 1 Find Boundary for Space

an
1: polygons p have a normal vector n and a distance from origin d

2: let P be the set of all polygons of any building element

3: let Ps be the set of all space face polygons
M
4: for all pi ∈ Ps do
5: Pc ← {p ∈ P : n = ni ∧ d > di }
6: pb ← ∅ (an empty polygonal geometry)
d

7: for all pj ∈ Pc do
te

8: if pj ∩ pi 6= ∅ then
9: add pj ∩ pi \ pb to the collection of boundary polygons
pb ← pb ∪ pj ∩ pi
p

10:

11: end if
ce

12: end for

13: end for
Ac

185 In the first step, boundary parts are detected for each space as if this space
was the only one. That is, it is assumed that all boundary parts of a space
are exterior. Thus, at this point, each boundary part has only one space, and
geometrically it corresponds to the part of the outer face of its building element
vis-à-vis the space face. Each space’s boundary geometries are detected as
190 described in Algorithm 1. The polygon map makes finding the right polygons
to consider easy: starting from a space face, only those polygons with the same

Page 18 of 31
 normal vector and a greater distance to the origin have to be checked, starting
with the polygon closest to the space face. The actual geometric operations
are performed in 2D space: intersections of building element polygons with the

t
ip
195 space face polygon become part of the boundary. By comparing each checked
polygon to the boundary geometry that has been detected so far, it is ensured

cr
that polygons “further out” the building geometry are not wrongly detected.
Additionally, an upper bound (e.g., a distance of 1 m to the space face) is used

us
to avoid unnecessary checks. The resulting boundary geometries are grouped
200 per building element and stored in boundary parts. The result of this step
performed for all spaces of an exemplary building story is shown schematically

an
in Figure 1.
Windows and doors form a special case. Their geometries typically include
many details that are not relevant for energy modeling, e.g., detailed framing,
M
205 window handles, door knobs, etc. On the other hand, quite relevant data for
windows are not included in IFC: there is no information about what part of
the geometry is glass and what part is framing. This would be useful for energy
d

analysis as it allows calculating the length of the glass edge bond. Because this
te

information cannot be modeled in IFC, the framing geometry is not relevant for
210 the algorithm’s output. Thus, window and door geometries are replaced by their
p

oriented bounding boxes beforehand. Moreover, after the boundary detection,

it is ensured that window/door boundary parts lie on the same plane as the
ce

respective wall boundary parts.

During the first step, all boundary parts have been detected. However,
the neighbor relationships between spaces, whether next to each other or on
Ac

215

top of each other, have not been considered yet. Thus, in the following step,
those pairs of boundary parts that lie between two spaces are detected and
merged into one boundary part that is associated to both spaces. To qualify
as a “merging candidate” pair, two boundary parts have to belong to the same
220 building element, bound two different spaces, and have opposite normal vectors.
If these conditions are met, the two geometries are intersected in 2D space. An
empty intersection means that the pair does not constitute an interior boundary

Page 19 of 31
 t
ip
cr
us
an
M
d
p te
ce
Ac

Figure 1: The intermediate results of each of the three processing steps by example: each
space is treated as if it was the only one; hence, all boundary parts are exterior and belong
to only one space (top). The original boundary parts are merged and split up depending on
the neighboring spaces (middle). Afterwards, gaps in the corners are filled (bottom).

10

Page 20 of 31
 part, thus no action is required. If there is an intersection, a new boundary
part, associated to both spaces, is created. Its geometry corresponds to the
(2D) intersection and is placed in the center of the two original geometries.

t
225

ip
The original boundary parts are not discarded immediately, because they
may be merged into more than one interior boundary part. For example, con-

cr
sider a segment of a wall with one space on one side, and two spaces on the
other side. In such a case, three original boundary parts are merged into two

us
230 interior boundary parts, while the original part belonging to the single space on
one side is split up and merged into both of those.
After all interior boundary parts have been merged, the number of parts

an
is down to the final number, and every part is at its final position. The last
step to be performed on the geometry is the elimination of gaps between parts
235 which occur depending on the simplification scheme. Since the spaces repre-
M
sent net volumes (i.e., the space geometries do not include building elements),
gaps at the corners are unavoidable as the boundary parts are placed on the
outside of building elements. Adjacent non-horizontal boundary parts can be
d

determined by checking the IFC relationship IfcRelConnectsElements for their

te

240 building elements. If this relationship is not set in the model, adjacency can
be heuristically determined by checking whether the smallest distance between
p

any pair of points of two boundary parts is within a threshold. For each pair of
adjacent boundary parts, gaps are filled by intersecting their host planes.
ce

In principle, the described algorithm works for any orientation of bound-

245 aries, whether vertical, sloping, or horizontal. However, the filling algorithm
would have trouble with filling gaps between horizontal boundary parts and
Ac

certain compositions of vertical boundary parts, for example those of an round

wall, or a continuous boundary part in one storey and multiple boundary parts
directly above it, with horizontal boundary parts at different elevation (because
250 of different slab thickness) in between. Moreover, staircase openings in slabs
are to be ignored in the energy model (the horizontal boundary part continues
through the opening). For these reasons, horizontal boundaries are treated as
a special case: the presented algorithm steps are performed story by story, and

11

Page 21 of 31
 t
ip
cr
us
Figure 2: Profile view of IFC space geometries in an office building. Building stories can easily
be identified based on the spaces z-coordinates.

an
the horizontal boundaries are added afterwards. Hence, the information about
255 which spaces belong to which story is needed first. This can be retrieved from
the spatial structure relationships in IFC. If this information is not available,
M
our algorithm is able to optionally detect the stories based on the elevations of
the space geometries. As shown in Figure 2, the stories can be detected by gaps
d
between the space geometries throughout the building, which can be identified
260 by the spaces’ minimal and maximal z-coordinates. These gaps also determine
te

the exact elevation of the horizontal boundary parts beforehand: they lie on the
centerlines between stories.
p
ce

4. Examples

A prototypic implementation of the algorithm has been tested with five

reference models: a simple bungalow; a simple two-story home; a two-story
Ac

265

twin home (see Figure 3); a complex office building with six stories: a basement
garage, large shop spaces in the first floor, many small office spaces in the upper
floors, and a roof terrace; a high-rise office building with 21 stories (see Figure 4).
All results have been compared to corresponding, manually simplified mod-
270 els; i.e. an expert on energy calculations prepared the reference solution for each
test case and documented the simplification process. The result, illustrated in
Figure 5, is the simplified geometry in a CAD model and a step-by-step descrip-

12

Page 22 of 31
 t
ip
cr
us
an
M
d
p te
ce
Ac

Figure 3: The presented algorithm has been tested using five test cases. The first test
case (top) and the second one (middle) are simplistic data sets, which are used to test the
algorithm’s main features. The third test case (bottom) is a realistic data set of a two-story
twin home.

13

Page 23 of 31
 t
ip
cr
us
an
M
d
p te
ce
Ac

Figure 4: While the first three test cases are residential houses (see Figure 3), the fourth and
the fifth data set represent office buildings: an office building with six stories (top) and an
office building with 21 stories (bottom).

14

Page 24 of 31
 t
ip
cr
us
an
M
Figure 5: Each use case consists of a BIM model, a simplified CAD model, and a description
of the simplification process. For example, the BIM model of the office building illustrated in
Figure 4 (top) has been simplified manually by an expert. The resulting CAD model (top) is
d

used to validate the presented algorithm, while the description (bottom; in German) guided
te

the algorithm design phase.

tion, which explains the simplification steps, provides explanatory information

p

and mentions typical traps.

ce

275 The step-by-step description guided the development process of the algo-
rithm and the resulting simplified CAD model has been used to validate the
results. Apart from some problem cases discussed below, the implementation
Ac

works robustly, and the results comply with the requirements (see Figure 6).

5. Discussion

280 Although the prototype shows that the algorithm works fine for most models,
even if they are incomplete to some degree, a few unresolved issues remain:

1. Incorrect Spatial Structure and Multi-Story Spaces

Many buildings have spaces that span over multiple building stories, e.g.,

15

Page 25 of 31
 t
ip
cr
us
an
M
d
p te
ce
Ac

Figure 6: Comparison of input (left) and output (right) geometries for a high-rise office
building with 21 stories. The boundary parts lie on the outer edges of exterior building
elements, and on the centerline of interior ones. Windows are shifted to align with their host
elements. A cross-section is cut near the top to make the inside visible.

16

Page 26 of 31
 elevator shafts. The correct way to model such spaces is to split them
285 up into one “sub-space” per story. IFC supports this with the IfcSpatial-
StructureElementCompositionType. When multi-story spaces are modeled

t
ip
as ordinary single spaces, the optional story detection, which groups spaces
into stories based on their geometries, fails. In that case, the stories have

cr
to be retrieved from the IFC spatial structure. Unfortunately, this data is
290 often incomplete and/or incorrect; see Figure 4 (bottom right).

us
2. Exterior Spaces
Another space-related problem case occurs in models, where exterior sec-
tions of the site (e.g., carports, lawn areas) are modeled as IfcSpace. This

an
would not be problematic, if these spaces were correctly marked as Ex-
295 ternal, which is seldom the case; see Figure3 (bottom; trees on the right
hand side).
M
3. Clashing Geometries
A surprisingly frequent problem in many models is geometry clashes, i.e.,
areas that are occupied by multiple building elements. This can lead
d

300 to non-deterministic behavior of the algorithm, as clashing polygons of

te

different building elements might be randomly detected as boundaries.

As many clashes occur in corner areas (e.g., clashes between walls and a
p

slab, or between two orthogonal walls), the gap filling approach actually
helps to avoid resulting issues. Other than that, there is not much that the
ce

305 algorithm can do, because it cannot decide which of two clashing polygons
to prefer.
Ac

6. Conclusion

The proposed algorithm serves an important purpose in making energy per-

formance calculation easier and faster at a time where the ideas of BIM are on
310 the rise. It does that by simplifying the geometry and extracting other relevant
data from models based on the open standard IFC. This brings dramatic time
savings compared to manually remodeling buildings just for the purpose of a

17

Page 27 of 31
 single energy performance simulation, a practice that is quite common today. It
can also become a preferable alternative to exports from proprietary CAD tools
once IFC-based shared BIMs are standard.

t
315

ip
The tests of the implemented prototype against the reference model set con-
firmed what had been suggested by the literature: data quality in BIM and IFC

cr
is a major issue, and one cannot expect perfect models. Such flaws exist for two
main reasons. The first, poor implementation of IFC export by BIM authoring

us
320 software. IFC’s wide and deep scope, is addressed by buildingSMART with
Model View Definition specifications and corresponding software certification.
The second reason is probably harder to overcome: BIMs are often created with-

an
out future use cases other than generation of plans and visualizations in mind,
which, strictly speaking, makes them “non-BIMs”.
325 While the prototype software already produces correct results for basic ref-
M
erence models, it will be further improved by testing it against a larger set
of real-world models. A choice of different simplification strategies would al-
low the resulting geometry to reflect additional norms other than the Austrian
d

ÖNORM. One focus of future work will be the development of a user interface
te

330 that presents intermediate results of the geometry simplification and allows the
user to check and modify where necessary for handling ambiguous aspects of
p

models.
ce

Acknowledgment

The authors gratefully acknowledge the generous support of the Austrian

Ac

335 Research Promotion Agency (FFG) for the research project GINGER (Graph-
ical Energy-Efficiency-Visualization in Architecture), grant number 840190 as
well as the support of the European Commission within the DURAARK project
founded by the program “ICT-2011-4.3-Digital Preservation”.

18

Page 28 of 31
 References

340 [1] D. Ladenhauf, R. Berndt, E. Eggeling, T. Ullrich, K. Battisti, M. Gratzl-

t
Michlmair, From Building Information Models to Simplified Geometries for

ip
Energy Performance Simulation, Proceeding of the International Academic
Conference on Places and Technologies 1 (2014) 669–676.

cr
[2] J. Mitchell, BIM & Building Simulation, Proceedings of Building Simula-

us
345 tion: Conference of International Building Performance Simulation Associ-
ation 12 (2011) K1–K7, [link].

[3] E. Franconi, Introducing a framework for advancing building energy mod-

an
elling methods & processes, Proceedings of Building Simulation – Confer-
ence of the International Building Performance Simulation Association 12
(2011) K8–K15.
M
350

[4] T. Liebich, Y. Adachi, J. Forester, J. Hyvarinen, K. Karstila, K. Reed,

S. Richter, J. Wix, Industry Foundation Classes – IFC2x, Edition 3, Tech-
d

nical Corrigendum 1, International Alliance for Interoperability 3 (2007)

te

1–653.

355 [5] International Organization for Standardization (ISO) 16739:2013 (Industry

p

Foundation Classes (IFC) for data sharing in the construction and facility
management industries).
ce

[6] R. J. Hitchcock, J. Wong, Transforming IFC Architectural View BIMs for

Energy Simulation, Proceedings of Building Simulation – Conference of
Ac

360 the International Building Performance Simulation Association 12 (2011)

1089–1095.

[7] V. Bazjanac, Implementation of semi-automated energy performance sim-

ulation: building geometry, Proceedings of the International Council for
Research and Innovation in Building and Construction 78 (2009) 595–602.

19

Page 29 of 31
365 [8] C. M. Rose, V. Bazjanac, An algorithm to generate space boundaries for
building energy simulation, Engineering with Computers 12 (2013) 1–10,
[link].

t
ip
[9] U. Pont, B. Sommer, A. Mahdavi, Ein Vergleich der Ergebnisse von sta-
tionärer und instationärer Berechnung thermischer Energiekennzahlen an-

cr
370 hand bestehender Objekte in Wien, Building Performance Simulation in
a Changing Environment – Proceedings of the German-Austrian IBPSA

us
Conference 3 (2010) 515–522.

[10] C. Eastman, P. Teicholz, R. Sacks, K. Liston, BIM Handbook: A Guide

an
to Building Information Modeling for Owners, Managers, Designers, Engi-
375 neers and Contractors, Wiley John + Sons, 2011, [link].

[11] J. Beetz, L. v. Berlo, R. d. Laat, P. v. d. Helm, bimserver.org – An Open

M
Source IFC Model Server, Proceedings of the International Council for
Research and Innovation in Building and Construction 78 (2010) 1–8.
d

[12] M. Weise, T. Liebich, R. See, V. Bazjanac, T. Laine, IFC Implementation

te

380 Guide: Space Boundaries For Thermal Analysis, INPRO Building infor-
mation model 19 (2009) 1–68.
p

[13] M. Venugopal, C. Eastman, R. Sacks, J. Teizer, Improving the Robustness

of Model Exchanges Using Product Modeling Concepts for IFC Schema,
ce

Computing in Civil Engineering 75 (2011) 611–618, [link].

385 [14] C. M. Hoffmann, The Problems of Accuracy and Robustness in Geometric

Ac

Computation, Computer 22 (1989) 31–39.

[15] L. Kettner, K. Mehlhorn, S. Pion, S. Schirra, C. Yap, Classroom examples

of robustness problems in geometric computations, Computational Geom-
etry 40 (2008) 61–78, [link].

390 [16] G. Bernstein, D. Fussell, Fast, Exact, Linear Booleans, Eurographics Sym-
posium on Geometry Processing 28 (2009) 1269–1278.

20

Page 30 of 31
 [17] U. Krispel, T. Ullrich, D. W. Fellner, Fast and Exact Plane-Based Rep-
resentation for Polygonal Meshes, Proceeding of the International Confer-
ence on Computer Graphics, Visualization, Computer Vision and Image

t
ip
395 Processing 8 (2014) 189–196.

[18] J. R. Shewchuk, Robust Adaptive Floating-Point Geometric Predicates,

cr
Proceedings of the Annual Symposium on Computational Geometry 12
(1996) 141–150.

us
[19] M. Campen, L. Kobbelt, Polygonal Boundary Evaluation of Minkowski
400 Sums and Swept Volumes, Eurographics Symposium on Geometry Pro-

an
cessing 29 (2010) 1613–1622.
M
d
p te
ce
Ac

21

Page 31 of 31