www.itcon.

org - Journal of Information Technology in Construction - ISSN 1874-4753

A PARAMETRIC APPROACH TOWARDS SEMI-AUTOMATED 3D

AS-BUILT MODELING
SUBMITTED: March 2023
REVISED: August 2023
PUBLISHED: December 2023
EDITOR: Robert Amor
DOI: 10.36680/j.itcon.2023.041

Yu-Chen Lee, Ph.D Student, corresponding author

Dept. of Civil, Architectural and Environmental Engineering, The University of Texas at Austin
ORCID: 0000-0001-7479-9968
email: [email protected]

Jong Won Ma, Ph.D., Assistant Professor

Dept. of Building, Civil and Environmental Engineering, Concordia University
ORCID: 0000-0001-5289-9183
email: [email protected]

Fernanda Leite, Ph.D., P.E., F.ASCE, Professor and John A. Focht Centennial Teaching Fellow in Civil
Engineering, Construction Engineering and Project Management Program
Dept. of Civil, Architectural and Environmental Engineering, The University of Texas at Austin
ORCID: 0000-0002-7789-4474
email: [email protected]

SUMMARY:Building Information Modeling (BIM) has been developed in response to the growing complexity of
construction projects. BIM implementation is beneficial throughout the entire building life cycle, and thus has been
widely adopted in new projects. However, BIM implementation in existing buildings is impeded by the lack of as-
built models. Conventionally, three-dimensional (3D) as-built BIMs are generated by experienced modelers, which
is time-consuming and error-prone. To cater to the need, Scan-to-BIM is a solution for automation in as-built BIM
generation. In the context of automated Scan-to-BIM, the parametric modeling process is worth investigating as
it has the ability to not only reconstruct 3D objects from a variety of categories from point clouds but also offer
flexibility to update objects simply by changing the values of the correlative parameters. Hence, this study proposes
a semi-automated framework for assisting 3D as-built modeling through a parametric modeling approach. The
presented methodology starts with wall boundary parameter extraction through 3D to 2D projection of the wall
segments and line detection techniques, followed by retrieving geometric parameters of all other non-wall elements
via CloudCompare. The extracted parameters are structured into a Microsoft® Excel® file and fed into Autodesk®
Dynamo for 3D BIM creation using a series of designed logic. To substantiate the viability, the proposed framework
is employed in two datasets containing structural, architectural, furniture, mechanical, and plumbing objects
(further categorized into structural, hosted and non-hosted elements). The Intersection over Union (IoU) of
structural elements was 96.35%, while the root-mean-square error (RMSE) of hosted and non-hosted elements
was 3.634 and 2.607 mm, respectively. This study established a universal methodology for semi-automated 3D as-
built modeling that can guide future research.

KEYWORDS: Scan-to-BIM, Parametric modeling, As-built BIM, Autodesk® Dynamo.

REFERENCE: Yu-Chen Lee, Jong Won Ma, Fernanda Leite (2023). A parametric approach towards semi-
automated 3D as-built modeling. Journal of Information Technology in Construction (ITcon), Vol. 28, pg. 806-
825, DOI: 10.36680/j.itcon.2023.041

COPYRIGHT: © 2023 The author(s). This is an open access article distributed under the terms of the Creative
Commons Attribution 4.0 International (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted
use, distribution, and reproduction in any medium, provided the original work is properly cited.

ITcon Vol. 28 (2023), Lee et. al., pg. 806

1. INTRODUCTION
The construction industry accounts for a significant portion of the U.S. economy (U.S. Bureau of Labor Statistics,
2021). According to the U.S. Bureau of Labor Statistics, the construction sector contributed 5.9 percent of all
nonfarm employment and 4.3 percent of Gross domestic product (GDP) in the U.S. in 2020. In spite of its crucial
economic role, the construction industry has a poor reputation for low productivity (Davila Delgado et al., 2019).
Given the complexity and uncertainty of projects, the industry has been plagued by construction waste, schedule
delays, quality assurance, increasing costs, and information transfer between multiple disciplines (Ullah et al.,
2019). For the sake of tackling obstacles in project management, Building Information Modeling (BIM) is one of
the most important advancements and has been widely adopted in the architecture, engineering, and construction
(AEC) industry in recent decades (Heaton et al., 2019). Adopting BIM in construction projects can assist decision-
makers in simulating and evaluating potential influences in a virtual environment (Gruber et al., 2018), facilitating
problem-solving with constrained labor and project costs. For instance, BIM can support clash detection, cost
estimation, design visualization, and effective cooperation to improve project performance (Leite, 2019; Volk et
al., 2014). BIM can be further integrated with other technologies, such as geographic information systems,
extended reality, and laser scanning, for a broader range of applications (Alizadehsalehi et al., 2020; Bansal, 2021;
Bosché et al., 2015).
In addition to implementing BIM in new construction projects, BIM adoption in existing buildings can better
facilitate operation and maintenance by reflecting changes made during the construction phase. As-built BIMs can
be used in non-invasive analysis to evaluate structural deformations in historic buildings preservation (Moyano et
al., 2022). Although a significant development in BIM applications has been made, the issue of automation in as-
built modeling is far from being settled. 2D paper-based drawings might not be well-preserved for existing
structures or heritage buildings, not to mention 3D BIMs. Therefore, creating as-built BIMs is the very first step
of BIM implementation in existing building structures. The traditional method of using 2D drawings for as-built
BIM creation has the possibility of causing errors by not capturing and updating variations during construction.
Scan-to-BIM can transform scanned point clouds with precise information into 3D BIMs by integrating 3D laser-
scanning and BIM. Some of the issues faced with Scan-to-BIM include dealing with occluded or hidden objects
in the scanned data, which can pose difficulties in accurately modeling these areas. However, despite these
challenges, the Scan-to-BIM technique offers advantages such as high efficiency and less human intervention
outweigh its limitations. It is noteworthy that currently developed software and tools for transforming point clouds
into as-built BIMs are still missing the capability to accurately model a variety of objects in real-world
environments without the necessity of manual intervention (Jung et al., 2018). Accordingly, the importance of
investigating an automatic Scan-to-BIM solution for efficient and rigorous 3D as-built BIM generation is self-
evident.
In light of automating Scan-to-BIM, object-based parametric modeling can create as-built BIMs with fully defined
parametric objects. Parametric modeling refers to automatically generating digital models for representing
elements through predefined parameters. Parametric modeling employs an objected-oriented approach in which
every object in the built environment is associated with its inherent attributes that should be identified and defined.
Parametric modeling is agile and flexible, where objects can be updated interactively by modifying the
corresponding value of predefined parameters (Barazzetti, 2016; Ma et al., 2022). Nevertheless, detailed logics or
algorithms for parametric modeling are not thoroughly investigated for modeling real-world objects. Accordingly,
this research seeks to establish a general framework leveraging Autodesk® Dynamo, a commercially available
BIM authoring software, Autodesk® Revit®. The aim is to facilitate the creation of as-built BIMs for indoor
objects utilizing a parametric approach that can be applied across a variety of contexts. The authors enrich the
Scan-to-BIM knowledge base both intellectually and practically by: 1) pinpointing essential parameters for
modeling prevalent indoor objects in the built environment, 2) presenting a well-structured workflow in
Autodesk® Dynamo for diverse object types, which could serve as a guideline for future research, 3) examining
the feasibility of the suggested methodology by implementing the proposed framework to model two distinct types
of real-world buildings, and 4) reducing the need for human intervention in as-built BIM generation.

2. BACKGROUND RESEARCH

2.1 Scan-to-BIM
Scan-to-BIM is a process that generates as-built BIMs from scanned point clouds, typically comprised of three
stages: scanning, registration, and modeling (Esfahani et al., 2021). This approach is widely applied to reconstruct

ITcon Vol. 28 (2023), Lee et. al., pg. 807

as-built information for managing or monitoring existing buildings and heritage structures, particularly when 2D
drawings or 3D models are absent (Allegra et al., 2020; Brumana et al., 2020; Pepe et al., 2020). With the
advancement of laser scanning technologies, the scanned point clouds with enriched geometric information enable
accurate and rapid data collection (Chen et al., 2018). However, traditional Scan-to-BIM practices often involve
significant manual efforts, which are labor-intensive, subjective, and prone to error, requiring experiential
professionals to perform the tasks (Tang et al., 2010).
In recent years, several research efforts have been placed on the automation of Scan-to-BIM. For instance, Macher
et al. (2021) presented a semi-automated approach for detecting windows from point clouds of building façades.
The proposed method combined geometric and radiometric information, utilizing intensity histogram analysis to
extract peaks representing different objects. In another study, Chen et al. (2018) developed a template-based
regional proposal mechanism leveraging the voxel grid method to identify major building components, including
beams, columns, roofs, stairs, and window frames. While the method conveniently omits the segmentation process,
enabling it to work on unstructured point clouds, it does necessitate user intervention and has a relatively low
precision rate. Son et al. (2015) provided an in-depth evaluation of three existing Scan-to-BIM software in terms
of automation level and performance in detecting pipes, including Trimble RealWorks®, Leica Cyclone, and
ClearEdge3D EdgeWise3D. The authors concluded that all software examined in the study leverages semi-
automated approaches, given their high reliance on user inputs for tasks such as selecting point clouds to be
modeled or defining necessary parameters.
Further contributing to reducing human input in Scan-to-BIM, Bosché et al. (2015) proposed a method that
specifically targets pipes, conduits, and ducts. The circular cross-section detection approach was incorporated with
the Scan-vs-BIM technique to identify and match objects from the as-built point clouds with the designed 3D BIM
of the project. Then, the as-built model was generated by the best-fitting centerline and radius from cross-sections
to create straight cylinders objects. While the proposed method is promising for automating the modeling process,
it carries a limitation of requiring readily available as-planned BIMs.
Though previous research shows significant progress in Scan-to-BIM automation, a certain extent of manual
intervention may not be avoided, especially in creating as-built models based on extracted geometric information.
Consequently, exploring automation in the modeling stage is a necessary step toward achieving a fully automated
Scan-to-BIM process.

2.2 Parametric modeling approach for Scan-to-BIM

Data collection and object recognition in Scan-to-BIM have attracted considerable attention recently. However,
studies focusing on the modeling stage remain limited, as summarized in Table 1. The workflow presented by
Andriasyan et al. (2020) utilizing a combination of Rhinoceros and Grasshopper-ArchiCAD illustrated the
potential of parametric modeling in reconstructing complex heritage buildings from point clouds. Similarly, a
recent study by Barazzetti (2016) applied an innovative semi-automated method. The proposed approach employed
Non-Uniform Rational B-Splines (NURBS) surfaces and curves to generate complex and irregular components,
further emphasizing the potential of parametric modeling in this sphere.
Yang et al. (2020) put forth a semi-automated methodology for generating steel structural components from point
cloud data through CloudCompare, MATLAB, and Autodesk® Dynamo. In this study, CloudCompare was used
for point cloud segmentation and several algorithms were deployed in MATLAB for extracting geometric
information. The proposed workflow incorporated Autodesk® Dynamo for as-built model generation,
demonstrating its capability to efficiently create parametric models. Nevertheless, the research was limited to four
common structural components, indicating the need for further development to enhance its applicability. In a
separate study, Wang et al. (2021) introduced a method to reconstruct mechanical, electrical, and plumbing (MEP)
components, which involves a 2D to 3D analysis for object detection and geometric information extraction. This
method underscored the potential of using Dynamo for parametric modeling. However, the framework had a
limitation as it required all MEP elements to be included in the model library, limiting its effectiveness with unseen
items.
Numerous studies have highlighted the potential of Dynamo for creating parametric BIMs. However, several
common challenges persist in the literature. For example, the lack of detail in explaining the procedure of designing
the Dynamo workflow hinders future research, strengthening the need for detailed guidelines to facilitate further
development of parametric modeling leveraging Dynamo. In addition, most previous research endeavors have
focused on individual areas of interest, such as structural or MEP systems, illustrating the need for a more

ITcon Vol. 28 (2023), Lee et. al., pg. 808

universally applicable method. Inspired by preceding studies, this research aims to address the aforementioned
issues to further enhance the modeling stage in Scan-to-BIM. Thus, the authors outline a comprehensive
framework for 3D as-built modeling through a semi-automated parametric approach leveraging Autodesk®
Dynamo and investigate its viability using two real-world datasets consisting of structural, architectural, furniture,
and MEP objects.

Table 1: Overview of Previous Studies on Parametric Modeling Approaches.

References Focused Method/Tool Automation Accuracy Limitations

Objects Level
• Lower 3D meshing accuracy
Andriasyan et Heritage Rhinoceros and Semi- Standard deviation: in volumetric components.
al., (2020) building Grasshopper- automated 68.28 pixels
• The algorithm can only
components ArchiCAD
represent the overall shape,
volume, and deformations of
the building components.
Barazzetti Complex and NURBS surfaces Semi- Error: ±2 to ±4 mm • Requires creating custom
(2016) irregular and curves automated Revit family objects for
components modeling.
Yang et al. Steel structural CloudCompare, Semi- RMSE: 3.8 to 11.8 • Limited to four common
(2020) components MATLAB, and automated mm structural components.
Autodesk®
Dynamo
Retrieval rate: • Requires all MEP elements to
Wang et al. MEP 2D to 3D analysis Automated 91.3% be included in the model
(2021) components and Autodesk® library.
Dynamo RMSE: 2.45 mm
• Limited to MEP components
that are only aligned with the
orientation of the building.

3. RESEARCH METHODOLOGY

3.1 Overall Methodology

Building upon on our previous study (Ma et al., 2022), this research leverages parametric modeling to generate
3D as-built models from scanned point clouds for a variety of indoor objects in the built environment. This study
assumes that all objects have been properly separated from raw point clouds, as point cloud segmentation is beyond
the scope of this research. Moreover, the proposed method is assuming the pre-segmented point clouds are nearly
flawless and accurate, as errors or inconsistencies in the segmentation could influence the accuracy of the extracted
geometric information. The authors acknowledge that this assumption is ideal and is not always feasible in practice.
Therefore, the proposed framework can still function to handle less-than-perfect conditions, with the understanding
that modeling accuracy might be compromised.
As shown in Figure 1, the proposed research methodology consists of two stages: pre-processing and parametric
modeling.

1. Pre-processing stage
The authors projected wall segments to 2D images and conducted line detection to obtain boundary
information. Python scripts and measurement tools in CloudCompare are utilized to extract parametric
information of all other non-wall objects. The parameter sets for each object category are then archived
in a Microsoft® Excel® file.

2. Parametric modeling stage

This stage first starts with importing the Microsoft® Excel® file into Autodesk® Dynamo, which is
subsequently followed by the creation of wall, floor, and roof elements to outline the skeleton of the
structure. Then, hosted elements such as doors and windows can be generated by specifying their host
walls. Hosted elements are elements that require attachment to a host wall. Simultaneously, non-hosted

ITcon Vol. 28 (2023), Lee et. al., pg. 809

 elements including tables, chairs, and other types of furniture are created. These non-hosted elements are
characterized as movable and rotatable objects that can be placed anywhere without needing attachment
to host walls.

Figure 1: Workflow of the proposed framework.

3.2 Dataset
This research utilizes two benchmark datasets for case studies. The first case is the Stanford 3D Indoor Scene
Dataset (S3DIS), which comprises point clouds of 271 rooms in 6 indoor areas at Stanford University (Armeni et
al., 2016). Each point in the S3DIS is annotated and objects are categorized into 13 groups, such as column, wall,
floor, bookcase, table, and so on. The S3DIS stores segmented point clouds as .txt files, which are named after the
object type they represent (e.g. bookcase_1.txt or window_1.txt). For the first case, area five from the S3DIS was
selected for applying the proposed framework as it encompasses the largest number of rooms (68 in total) and
diverse environments.
The Medical Center Mechanical Room dataset is chosen for the second implementation case. The dataset consists
of 3D BIMs featuring MEP system elements like pipes and elbows. However, this dataset does not include pre-
segmented point clouds. As this research focuses on enhancing the modeling stage of Scan-to-BIM, the
segmentation process falls outside the scope of this study. Instead, the authors employ the dataset’s as-designed
BIMs to extract necessary information using Autodesk® Dynamo under the assumption that the geometric
information accurately represents the actual objects, mimicking the conditions of well-segmented point clouds.
This methodological choice allows us to focus on evaluating the potential and constraints of the proposed
parametric modeling approach in a controlled setting. Except for these noted differences, the second case followed
the same workflow to generate as-built models as the first case.

3.3 Geometric Information Requirement

When it comes to as-built modeling, the modeling parameters differ according to the type of object. For example,
when modeling wall elements, necessary parameters include the coordinates of the start and end points that are
representing wall centerlines (Figure 2), along with the wall’s starting level and height. The generation of floors
and roofs requires coordinates of multiple points to form enclosed curves as well as the start and end levels for
each curve. In the case of wall-dependent objects, such as doors, windows, bookcases, and boards, modeling
requires the start and end coordinates of their host walls. The calculation of distances from the centers of these

ITcon Vol. 28 (2023), Lee et. al., pg. 810

wall-based objects to the start point coordinates of their respective host walls is also necessary. In addition, the
specific family types and dimensions of those objects are needed for accurate sizing. For other objects that are not
attached to walls, the parameters include the center's location, family types, and object dimensions. Table 2
summarizes the information necessary for each object type, providing a detailed overview.

Figure 2: Illustration of wall centerline.

Table 2: Required information for different object types.

Object Type Family Type Required information

 Location coordinates of the start point and end points representing wall
centerlines
Walls  Starting level and height
 Wall type
Structural elements

 Location coordinates of points forming closed curves

Floors and ceilings  Floor or ceiling type

 Location coordinates of host walls

Hosted Doors, windows,
 Length from the center of hosted objects to the start point of its host wall
elements bookcases, and whiteboards
 Family type and dimensions
Non-hosted  Location coordinates of the object's center
Tables and chairs  Family type and dimensions
elements

Mechanical  Location coordinates of the object's center

Mechanical equipment  Family type and dimensions
elements
 Location coordinates of the start and end points representing pipe
centerlines
Pipes and flex ducts
 Level
Plumbing elements  Family type and diameter
 Location coordinates of the object's center
Pipe fittings  Family type and diameter

3.4 Parametric Information Extraction – Python scripts, CloudCompare, and

Microsoft® Excel®
In the pre-processing phase, the authors utilized scripts to project 3D point clouds onto 2D images, then used the
Hough transform (Hough, 1962) to extract lines representing walls, as illustrated in Figure 3. The Hough transform
is a method for extracting shape characteristics from an image via a voting procedure. Fig. 3b shows the outcome
of the line detection algorithm, where green lines represent detected lines. The wall centerline information is
derived by calculating the median line between two lines and their intersections. The coordinates of start and end
points for each centerline segment are then recorded to a Microsoft® Excel® spreadsheet on a sheet named "Wall."

ITcon Vol. 28 (2023), Lee et. al., pg. 811

Figure 3: Examples of (a) projected point cloud image and (b) line detection image.

Upon acquiring the wall boundary information, the authors developed Python to automatically extract geometric
information for all other non-wall objects within the indoor space. These scripts search through all folders in the
parent folder, looking for an "Annotations" folder where point cloud files are stored. The proposed methodology
presumes that object physical sizes can be approximated by bounding boxes, which enables an efficient
representation and subsequent analysis of data (as illustrated in Figure 4). For hosted elements, the center location,
family type, and dimensions (length, width, and height) of the bounding box were determined. The scripts
identified the nearest wall for these objects by computing distances to previously outlined walls. For non-hosted
objects, the algorithm determined the center location, as wall information was not required. The information was
recorded in the spreadsheet containing the wall centerline data, with separate sheets created for different object
types. Sheets in the output file use object type as sheet name, such as “Door,” “Bookcase,” and “Window.” Each
sheet contains object-specific information, as mentioned in Table 2. Furthermore, the processing time for each
room’s point clouds was also documented, as shown in Figure 5. Table 3 shows the number of objects processed
and the time taken for different room types using the script. The total time spent processing all rooms in area five
of the S3DIS is 1426.8 seconds for a total of 2338 objects, equating to approximately 0.6 seconds per object. All
computations were performed on a system equipped with an Intel® Core™ i7-8700 CPU, 64 GB RAM, operating
on a 64-bit system.

Figure 4: Object represented by a bounding box with dimensions approximating its size.

Table 3: Summary of total processing time and number of objects by room type.
Room Type Total Time (seconds) Number of Objects
Conference Room 65.68 131
Hallway 789.39 292
Lobby 2.11 45
Office 526.34 1708
Pantry 0.05 19
Storage 38.84 95
WC 4.39 48
Total 1426.8 2338

ITcon Vol. 28 (2023), Lee et. al., pg. 812

Figure 5: Example of the script in action, documenting the time spent for processing each room.

CloudCompare serves as the tool to gather information relating to Revit family types. The authors utilized standard
Revit family objects, which are universally accessible to users. This strategy ensures that the proposed approach
can be reproduced by users who may not have access to custom objects. Using built-in family items could also
streamline the process, as custom objects may require additional information or specific import procedures.
However, it should be noted that built-in family types are not exhaustive and may not be a perfect fit for every
real-world object. Therefore, the authors have to perform their judgment to select the most appropriate match. In
this study, point cloud files are imported into CloudCompare, and researchers determine the default family type
most similar in shape to the point clouds to represent each object. Also, the authors utilize measurement tools in
CloudCompare to gather rotation angle information. The manually gathered information using CloudCompare
consists of identifying the family type that best matches the shape of the objects and the rotation angle information.
The chosen family type and orientation details are then appended to the corresponding spreadsheet, equipping it
with crucial information for the 3D modeling process.

Figure 6: An example of the output of geometric information extraction.

The methodology of the study is designed specifically for efficient implementation in Revit using the Dynamo
plugin. While other BIM tools, such as Grasshopper for Rhino, can perform similar tasks, their compatibility and

ITcon Vol. 28 (2023), Lee et. al., pg. 813

functionality may differ based on project requirements. Thus, the optimization is currently focused on using Revit
and Dynamo. In sum, this research extracts essential geometric information using a combination of Python scripts
and CloudCompare, compiling it into a Microsoft® Excel® spreadsheet containing several worksheets for
different element types, as demonstrated in Figure 6.

3.5 Parametric Modeling – Autodesk® Dynamo

The framework proposed in this research selects Autodesk® Dynamo for the parametric modeling process.
Dynamo is an add-on for Autodesk® Revit® and a visual programming language that provides high flexibility for
users to design and execute sophisticated workflows. Utilizing customized nodes along with built-in nodes, users
can design algorithms or automate tasks to assist in 3D BIM generation simply by connecting and sequencing
nodes. Dynamo is especially appropriate for modeling objects with complex design features, as it can help reduce
human error in geometric modeling (Kontoudaki and Georgopoulos, 2022). Furthermore, once a generic workflow
is established within Dynamo, it can be efficiently applied to similar structures. Given the aforementioned
advantages, Dynamo is selected as a facilitator for efficient parametric modeling.

4. PARAMETRIC MODELING PROCESS – CASE 1

As mentioned previously, area five of S3DIS consists of indoor rooms with standard office objects (including
walls, floors, ceilings, doors, windows, bookcases, and boards). Figure 7 illustrates the configuration of the rooms
in area five of S3DIS. The parametric modeling process of this dataset begins with importing data from the
Microsoft® Excel® file, followed by creating structural elements that define the room outlines. Subsequently,
hosted and non-hosted objects can be created simultaneously. For each object type, the design logic and the
required information are different. Therefore, the following section dives deeper into the parametric modeling
process for the three primary object categories: (1) structural elements, (2) hosted elements, and (3) non-hosted
elements.
The designed logic contains built-in nodes and nodes from 'Clockwork for Dynamo' (Dieckmann, 2013) and
'Springs for Dynamo' (Venkov, 2015) packages, which are open access resources published by Dynamo users. To
facilitate the parametric modeling process and serve the needs, the authors designed a customized node'
GetPointsFromExcel(xyz)' that allows users to obtain x, y, and z coordinates of points from a Microsoft® Excel®
file by specifying the file path, name of the worksheet, and the index of columns.

Room categories

Office (61.8%)
Hallway (22.1%)
Storage (5.9%)
Restroom (2.9%)
Conference room (4.4%)
Lobby (1.5%)
Pantry (1.5%)

Figure 7: Configuration of rooms in area five of S3DIS.

4.1 Structural elements

Structural elements encompass walls, floors, and ceilings. For walls, the process involves forming centerlines by
retrieving start and end point data. Floors and ceilings employ a similar strategy, using point data to establish the
outlines and then translating these outlines vertically to represent the ceiling. Figure 8 shows an example of the
designed logic for wall objects.

ITcon Vol. 28 (2023), Lee et. al., pg. 814

Figure 8: The designed logic for wall objects.

4.2 Hosted elements

Hosted elements include items that are attached or related to structural elements, such as windows, doors, or
shelves. The geometric information, such as host walls' start and end points, is crucial in this process. Figure 9
provides an example of the required geometric information of a hosted object, where the first six columns indicate
the location coordinates of the start and end points of host walls. The term “chord length” represents the distance
between the object’s center and the starting point of its respective host wall. With the extracted information, the
designed logic for hosted elements locates the center of the objects on host wall baselines and places the objects
at these specified locations. It is important to note that object dimensions are automatically adjusted post-placement
based on object dimensions.
StartPoint.X StartPoint.Y StartPoint.Z EndPoint.X EndPoint.Y EndPoint.Z ChordLength FamilyType Bookcase_DistoffFloor Bookcase_Width Bookcase_Depth Bookcase_Height
-12.773 17.249 0 -7.748 17.249 0 0.5 WallBased_Shelving_8 0 1 0.35 2.8
-12.773 17.249 0 -7.748 17.249 0 1.5 WallBased_Shelving_4 1.4 1 0.35 1.4
-12.773 17.249 0 -7.748 17.249 0 2.5 WallBased_Shelving_4 1.4 1 0.35 1.4
-12.773 17.249 0 -7.748 17.249 0 3.5 WallBased_Shelving_8 0 1 0.35 2.8
-12.773 17.249 0 -7.748 17.249 0 4.5 WallBased_Shelving_8 0 1 0.35 2.8
-7.748 20.404 0 -12.773 20.404 0 2.725 WallBased_Shelving_4 1.4 0.9 0.35 1.4
-7.748 20.404 0 -12.773 20.404 0 3.625 WallBased_Shelving_4 1.4 0.9 0.35 1.4
-7.748 20.404 0 -12.773 20.404 0 4.525 WallBased_Shelving_4 1.4 0.9 0.35 1.4
-12.773 17.249 0 -7.748 17.249 0 1.5 WallBased_Cabinet1 0 1 0.45 0.75
-12.773 17.249 0 -7.748 17.249 0 2.5 WallBased_Cabinet1 0 1 0.45 0.75

Figure 9: An example of geometric information of hosted objects.

4.3 Non-hosted elements

Non-hosted elements, such as furniture, provide more flexibility as they are movable and rotatable. Their
placement requires only the location coordinates of the object's center, along with dimensions and rotation angles
if necessary.

4.4 Code Blocks

Instead of using substantial nodes to create the designed scripts, code blocks can be deployed to minimize the
number of nodes so as to reduce computation. A code block in Dynamo is a node containing numbers, strings,

ITcon Vol. 28 (2023), Lee et. al., pg. 815

formulas, and scripts to perform more sophisticated tasks that allow users to customize and interact with all the
functionalities in Dynamo (Jezyk et al., 2019). Moreover, customized functions are stored in a code block and can
be recalled anywhere within Dynamo. For example, Figure 10 illustrates a code block being used to locate and
select host wall elements.

Figure 10: Example of code block application.

4.5 Results
Figure 11 presents the established logic for non-hosted elements. Figure 12 provides an example of the as-built
models generated by the proposed framework. The results highlight the potential of creating 3D as-built BIMs
from well-segmented segmented point clouds with structural, hosted, and non-hosted elements given the necessary
information (as displayed in Figure 13). In addition, this generic modeling process is versatile and applicable to
various element types.

Figure 11: The designed logic for non-hosted elements.

ITcon Vol. 28 (2023), Lee et. al., pg. 816

Figure 12: Examples of the as-built BIMs generated by the proposed methodology.

Figure 13: A comparison of point clouds and 3D BIM.

Figure 14 compares the modeling time between the manual modeling approach and the proposed method. The
manual modeling process is time-consuming in that it generally takes up to four hours to construct a 3D BIM for
a room. On the other hand, the parametric approach is more efficient and less labor-intensive. The time needed for
creating as-built BIMs leveraging the parametric approach for a room decreased over time. The initial stage of
designing a logical workflow takes a few hours to construct, test, and modify for one room. After accomplishing
the generic logic, the average time spent for a single room is significantly reduced. Take an office room as an
example; the average modeling time decreased to 35 minutes after establishing the designed logic. In addition, the
dimensions and rotation angles of objects are modified simultaneously.

ITcon Vol. 28 (2023), Lee et. al., pg. 817

 Manual modeling process

Parametric modeling process (initial stage)

Parametric modeling process (final stage)

0 1 2 3 4 5
Time (hours)

Extract necessary information Object creation Object modification

Figure 14: Modeling time comparison chart.

Table 4: RMSE for various parameters of hosted and non-hosted objects.

Object Type Family Type Parameter RMSE (mm)
Hosted Windows Width 4.28
Height 1.94
Center Location 3.72
Doors Width 0.55
Height 0.35
Center Location 3.81
Boards Width 2.16
Height 0.25
Center Location 1.54
Bookcase Width 3.32
Height 5.05
Depth 1.07
Center Location 5.47
Non-hosted Table Width 1.98
Height 2.73
Length 0.66
Center Location 4.93
Chair Width 2.63
Height 1.07
Depth 3.66
Center Location 2.28

The Intersection over Union (IoU) is a widely recognized metric in computer vision, designed to quantify the
degree of overlap between predicted data and ground truth, thereby providing a measure of accuracy. Within the
framework of Case Study 1, the authors employed an IoU analysis to compare geometric information, which was
autonomously extracted using Python scripts, with the ground truth data, as illustrated in Figure 15. The ground
truth data was derived from as-built BMIs that were manually modeled by an experienced Revit user. This data is
assumed to represent the ground truth. As a result, the overall IoU for structural elements was 96.35%. For hosted
and non-hosted elements, the researchers calculated the errors associated with each parameter (such as width,
height, and center location), and these findings are compiled in Table 4. The root-mean-square error (RMSE) for
hosted and non-hosted elements were 3.634 and 2.607 mm, respectively.

ITcon Vol. 28 (2023), Lee et. al., pg. 818

Figure 15: Illustration of IoU for structural elements in Case Study 1.

5. PARAMETRIC MODELING PROCESS – CASE 2

The Medical Center Mechanical Room dataset was chosen to verify the proposed framework's applicability to
varied element types. The dataset is a 3D model containing structural, mechanical, and plumbing elements. This
second case implementation primarily focuses on mechanical and plumbing system elements, elaborating on the
designed logic for their creation. ‘Clockwork for Dynamo’, ‘Springs for Dynamo’, and ‘MEPover’ (MEPover,
2016) packages are mainly utilized in the designed logic for the second case study.

5.1 Parametric Information extraction

Figure 16: The designed logic for information extraction of MEP elements.

ITcon Vol. 28 (2023), Lee et. al., pg. 819

In this second case, Autodesk® Dynamo is utilized for extracting the required information. The designed logic
selects all elements from a specified category (for instance, duct elements) and extracts their location coordinates
and other information. This information is organized into a list and store in a spreadsheet for future use. Figure 16
presents the designed logic for extracting necessary information of mechanical equipment elements.

5.2 Mechanical and Plumbing Elements

The creation of mechanical equipment elements follows the workflow for non-hosted objects, requiring the
location coordinates of the object’s center, family type, and dimensions. On the other hand, plumbing elements
can be divided into two categories: (1) ducts and pipes, and (2) fittings.
5.2.1 Ducts and pipes
Creating pipes and ducts requires the coordinates of the start and end points, representing their centerlines. The
designed logic generates duct and pipe elements after forming these centerlines, as depicted in Figure 17.

Figure 17: The designed logic for duct and pipe elements.

5.2.2 Fittings
After creating ducts and pipes, fittings that connect sections of ducts or pipes are generated manually. Since
existing Dynamo nodes are not capable of creating fittings with specified size and family type, the authors
developed a Dynamo script to serve the need. Dynamo Player is applied to execute the designed script more
efficiently since it allows users to operate the script in Revit without Dynamo. Figure 18 describes how Dynamo
Player works in Revit. Users can significantly expedite the modeling process by selecting pipes or ducts to be
connected and assigning family types with Dynamo Player.

ITcon Vol. 28 (2023), Lee et. al., pg. 820

Figure 18: Dynamo script execution with Dynamo Player.

5.3 Results
Figure 19 shows the model generated by the proposed framework. The designed logic can collect the necessary
information into a spreadsheet with well-organized worksheets presenting each object type. The outcome
highlights the potential of the proposed parametric approach to be versatile across different disciplines, including
mechanical and plumbing systems. However, it is important to note that further automation of this method is
currently limited due to the restricted availability of Dynamo packages for MEP.

Figure 19: Mechanical and plumbing elements generated by the proposed method.

In the presented case study, all computations were performed on a system equipped with an Intel® Core™ i7-8700
CPU, 64 GB RAM, operating on a 64-bit system. The process of extracting geometric information was completed
in 12 seconds, while the creation of 3D BIMs was accomplished in 37 seconds. The authors utilized Dynamo
Players to manually create connections for pipes and ducts, adding an additional 41 minutes to the total time.
Accordingly, the overall time spent in generating this model was slightly under 42 minutes.
The Industry Foundation Classes (IFC) is an object-oriented data model that is standardized and platform neutral,

ITcon Vol. 28 (2023), Lee et. al., pg. 821

thus can be used in any BIM software. In this study, we utilized the quantity of elements in each IFC type to obtain
the accuracy rate between the as-designed and the generated BIM, as presented in Table 5. The comparative
analysis was restricted to mechanical and plumbing elements, with an overall retrieval rate of 93.9%.
Table 5: Accuracy comparison for various IFC Types in Case Study 2.
IFC Type Ground Truth Generated BIM Retrieval Rate

IfcDistributionPort 1711 1612 94.2%

IfcDuctFitting 293 242 82.6%
IfcDuctSegment 556 555 99.8%
IfcPipeFitting 6 5 83.3%
IfcRelConnectsPorts 574 550 95.8%
IfcSystem 275 243 88.4%

Overall 3415 3207 93.9%

6. CONCLUSIONS
This study presents a 3D semi-automated as-built modeling framework by applying a parametric approach, which
is divided into pre-processing and parametric modeling stages. The proposed method extracts necessary modeling
information through the integration of scripts and CloudCompare. Most of the required information is extracted
automatically, while CloudCompare is utilized to manually determine family types and orientation angles. The
parametric modeling is performed by Dynamo, which allows users to create 3D models with high efficiency and
provide high flexibility with powerful built-in functions and can also execute Python scripts for more sophisticated
tasks. Also, the proposed approach leverages built-in family objects in Revit allows widespread access without
searching for objects from other sources. This study elaborates the design process for parametric modeling in
Dynamo with details, including the required parameters, the entire designed logic for some element types, to serve
as a guideline for future research in Scan-to-BIM. Building on our previous study (Ma et al., 2022), this semi-
automated method has been applied to two real-world datasets to validate the practicality of implementing the
designed workflow for modeling structural, architectural, furniture, mechanical, and plumbing elements. Based on
the results, it can be inferred that the proposed framework has promising potential in generating 3D BIMs and
significantly reduce modeling time, thus indicating its practicality and benefits for Scan-to-BIM applications.
Unlike the parametric modeling approach, the manual modeling process is time-consuming and error-prone.
Experienced modelers have to visually identify and generate objects by referencing imported point clouds.
Meanwhile, they have to create multiple new family types to fit the actual sizes of recognized objects. That is, they
may neglect some objects or create objects with incorrect sizes. Once a change has been made to the dimension of
an object, modelers have to revisit the created family object to reflect the change. On the other hand, the parametric
modeling approach with Dynamo has the potential to improve modeling efficiency by defining required parameters
and manipulating nodes and scripts, as it can reduce human effort in creating tons of family objects and save time
spent on updating the parameters of multiple objects individually. Consequently, the proposed framework has the
potential to create as-built models efficiently and facilitate Scan-to-BIM.
The proposed framework, while promising, does have a few areas of improvement. First of all, the proposed
method relies on accurate point cloud segmentation, which may not always be feasible or achievable. Secondly,
existing Dynamo packages for assisting MEP system modeling are too limited to perform specific tasks. Plus, the
proposed research method utilizes built-in family objects in Autodesk® Revit® with limited specifications. Lastly,
the devised workflow should be examined to validate its applicability and accuracy in other disciplines, such as
electrical elements. These limitations highlight areas for enhancement within our framework and underscore
potential future research in the Scan-to-BIM field.
To further advance the parametric modeling approach and overcome the aforementioned limitations, future studies
could focus on several key areas. One effective approach is to employ deep learning-based instance-level semantic
segmentation algorithms, which can not only classify points at a category level but also at an instance level,
enabling more accurate modeling of distinct objects within the same category. Another focus area would be the
development of customized nodes and packages for MEP system elements, which could expand the versatility of
Dynamo in this context. Furthermore, expanding the family object library within Revit to accommodate a broader

ITcon Vol. 28 (2023), Lee et. al., pg. 822

range of specifications can help reflect real-world objects more accurately. It is important to note that increasing
the number of object types might increase the recognition process and potentially compromise accuracy,
highlighting the need of maintaining a balance between versatility and accuracy in future development. Lastly,
investigating the adaptability of the developed workflow across other disciplines would verify its broad
applicability. These focus areas have the potential to considerably enhance the modeling process, paving the way
toward achieving a fully automated Scan-to-BIM process.
In conclusion, this study highlights the applicability and adaptability of the parametric modeling approach. The
proposed method encompasses a wide variety of indoor objects, aiming at reducing modeling time and human
intervention, which is a significant contribution to the field of Scan-to-BIM. The extraction of geometric
information and creation of 3D BIM elements, traditionally considered labor-intensive and error-prone, are mostly
automated in our semi-automated approach. In the proposed method, manual intervention is primarily required for
the determination of family types and orientation angles using CloudCompare and the selection of ducts for
connection. Although manual intervention is still necessary, this method substantially diminishes the likelihood of
human errors and reduces modeling time. The workflow represents one of the early attempts to model a variety of
indoor objects using parametric modeling at once, which adds to its strength and uniqueness. Overall, the findings
from this research support the use of parametric modeling as a promising technique for Scan-to-BIM, and future
research can build upon this work to improve and expand the methodology further.

ACKNOWLEDGMENTS
The authors would like to thank G4 Spatial Technologies for their support in providing the Medical Center
Mechanical Room dataset that facilitated the development of this research.

REFERENCES
Alizadehsalehi, S., Hadavi, A., Huang, J.C., 2020. From BIM to extended reality in AEC industry. Autom.
Constr. 116, 103254. https://doi.org/10.1016/j.autcon.2020.103254
Allegra, V., Di Paola, F., Lo Brutto, M., Vinci, C., 2020. Scan-To-BIM For The Management Of Heritage
Buildings: The Case Study Of The Castle Of Maredolce (Palermo, Italy), in: The International Archives
of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Presented at the XXIV
ISPRS Congress, Commission II (Volume XLIII-B2-2020) - 2020 edition, Copernicus GmbH, pp.
1355–1362. https://doi.org/10.5194/isprs-archives-XLIII-B2-2020-1355-2020
Andriasyan, M., Moyano, J., Nieto-Julián, J.E., Antón, D., 2020. From Point Cloud Data to Building Information
Modelling: An Automatic Parametric Workflow for Heritage. Remote Sens. 12, 1094.
https://doi.org/10.3390/rs12071094
Armeni, I., Sener, O., Zamir, A.R., Jiang, H., Brilakis, I., Fischer, M., Savarese, S., 2016. 3D Semantic Parsing
of Large-Scale Indoor Spaces, in: 2016 IEEE Conference on Computer Vision and Pattern Recognition
(CVPR). Presented at the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR),
pp. 1534–1543. https://doi.org/10.1109/CVPR.2016.170
Bansal, V.K., 2021. Integrated Framework of BIM and GIS Applications to Support Building Lifecycle: A Move
toward nD Modeling. J. Archit. Eng. 27, 05021009. https://doi.org/10.1061/(ASCE)AE.1943-
5568.0000490
Barazzetti, L., 2016. Parametric as-built model generation of complex shapes from point clouds. Adv. Eng.
Inform. 30, 298–311. https://doi.org/10.1016/j.aei.2016.03.005
Bosché, F., Ahmed, M., Turkan, Y., Haas, C.T., Haas, R., 2015. The value of integrating Scan-to-BIM and Scan-
vs-BIM techniques for construction monitoring using laser scanning and BIM: The case of cylindrical
MEP components. Autom. Constr., 30th ISARC Special Issue 49, 201–213.
https://doi.org/10.1016/j.autcon.2014.05.014
Brumana, R., Oreni, D., Barazzetti, L., Cuca, B., Previtali, M., Banfi, F., 2020. Survey and Scan to BIM Model
for the Knowledge of Built Heritage and the Management of Conservation Activities, in: Daniotti, B.,
Gianinetto, M., Della Torre, S. (Eds.), Digital Transformation of the Design, Construction and
Management Processes of the Built Environment, Research for Development. Springer International
Publishing, Cham, pp. 391–400. https://doi.org/10.1007/978-3-030-33570-0_35

ITcon Vol. 28 (2023), Lee et. al., pg. 823

Chen, J., Cho, Y.K., Kim, K., 2018. Region Proposal Mechanism for Building Element Recognition for
Advanced Scan-to-BIM Process 221–231. https://doi.org/10.1061/9780784481264.022
Davila Delgado, J.M., Oyedele, L., Ajayi, A., Akanbi, L., Akinade, O., Bilal, M., Owolabi, H., 2019. Robotics
and automated systems in construction: Understanding industry-specific challenges for adoption. J.
Build. Eng. 26, 100868. https://doi.org/10.1016/j.jobe.2019.100868
Dieckmann, A., 2013. Clockwork For Dynamo Package [WWW Document]. URL
https://github.com/andydandy74/ClockworkForDynamo (accessed 10.18.21).
Esfahani, M.E., Rausch, C., Sharif, M.M., Chen, Q., Haas, C., Adey, B.T., 2021. Quantitative investigation on
the accuracy and precision of Scan-to-BIM under different modelling scenarios. Autom. Constr. 126,
103686. https://doi.org/10.1016/j.autcon.2021.103686
Gruber, C., Weiner, T., Zuchtriegel, R., 2018. BIM for tunnelling for a company – Approaches and strategies.
Geomech. Tunn. 11, 366–373. https://doi.org/10.1002/geot.201800017
Heaton, J., Parlikad, A.K., Schooling, J., 2019. Design and development of BIM models to support operations
and maintenance. Comput. Ind. 111, 172–186. https://doi.org/10.1016/j.compind.2019.08.001
Hough, P.V.C., 1962. Method and means for recognizing complex patterns. US3069654A.
Jezyk, M., Mode Lab, John Pierson of Parallax Team, 2019. What’s a Code Block | The Dynamo Primer [WWW
Document]. URL https://primer.dynamobim.org/07_Code-Block/7-1_what-is-a-code-block.html
(accessed 4.11.22).
Jung, J., Stachniss, C., Ju, S., Heo, J., 2018. Automated 3D volumetric reconstruction of multiple-room building
interiors for as-built BIM. Adv. Eng. Inform. 38, 811–825. https://doi.org/10.1016/j.aei.2018.10.007
Kontoudaki, A., Georgopoulos, A., 2022. HBIM Library Development For A Doric Temple Column. Presented
at the The International Archives of the Photogrammetry Remote Sensing and Spatial Information
Sciences, Nice, France.
Leite, F., 2019. BIM for Design Coordination: A Virtual Design and Construction Guide for Designers, General
Contractors, and MEP Subcontractors, 1st ed. Wiley.
Ma, J.W., Lee, Y.-C., Leite, F., 2022. A Practical Application Using Parametric Modeling for As-Built BIM
Generation from Point Clouds 830–838. https://doi.org/10.1061/9780784483961.087
Macher, H., Roy, L., Landes, T., 2021. Automation of windows detection from geometric and radiometric
information of point clouds in a scan-to-BIM process. Int. Arch. Photogramm. Remote Sens. Spat. Inf.
Sci. XLIII-B2-2021, 193–200. https://doi.org/10.5194/isprs-archives-XLIII-B2-2021-193-2021
MEPover, 2016. MEPover Package [WWW Document]. URL
https://sites.google.com/site/bimstallatie/dynamo/mepover-package (accessed 10.18.21).
Moyano, J., Gil-Arizón, I., Nieto-Julián, J.E., Marín-García, D., 2022. Analysis and management of structural
deformations through parametric models and HBIM workflow in architectural heritage. J. Build. Eng.
45, 103274. https://doi.org/10.1016/j.jobe.2021.103274
Pepe, M., Costantino, D., Restuccia Garofalo, A., 2020. An Efficient Pipeline to Obtain 3D Model for HBIM and
Structural Analysis Purposes from 3D Point Clouds. Appl. Sci. 10, 1235.
https://doi.org/10.3390/app10041235
Son, H., Kim, C., Turkan, Y., 2015. Scan-to-BIM - An Overview of the Current State of the Art and a Look
Ahead. ISARC Proc. 1–8.
Tang, P., Huber, D., Akinci, B., Lipman, R., Lytle, A., 2010. Automatic reconstruction of as-built building
information models from laser-scanned point clouds: A review of related techniques. Autom. Constr.
19, 829–843. https://doi.org/10.1016/j.autcon.2010.06.007
Ullah, K., Lill, I., Witt, E., 2019. An Overview of BIM Adoption in the Construction Industry: Benefits and
Barriers, in: Lill, I., Witt, E. (Eds.), 10th Nordic Conference on Construction Economics and
Organization, Emerald Reach Proceedings Series. Emerald Publishing Limited, pp. 297–303.
https://doi.org/10.1108/S2516-285320190000002052

ITcon Vol. 28 (2023), Lee et. al., pg. 824

U.S. Bureau of Labor Statistics, 2021. Labor Productivity for Selected Construction Industries [WWW
Document]. URL https://www.bls.gov/lpc/construction.htm#footnote1 (accessed 3.3.22).
Venkov, D., 2015. Springs for Dynamo Package [WWW Document]. URL
https://github.com/dimven/SpringNodes (accessed 10.18.21).
Volk, R., Stengel, J., Schultmann, F., 2014. Building Information Modeling (BIM) for existing buildings —
Literature review and future needs. Autom. Constr. 38, 109–127.
https://doi.org/10.1016/j.autcon.2013.10.023
Wang, B., Yin, C., Luo, H., Cheng, J.C.P., Wang, Q., 2021. Fully automated generation of parametric BIM for
MEP scenes based on terrestrial laser scanning data. Autom. Constr. 125, 103615.
https://doi.org/10.1016/j.autcon.2021.103615
Yang, L., Cheng, J.C.P., Wang, Q., 2020. Semi-automated generation of parametric BIM for steel structures
based on terrestrial laser scanning data. Autom. Constr. 112, 103037.
https://doi.org/10.1016/j.autcon.2019.103037

ITcon Vol. 28 (2023), Lee et. al., pg. 825