Feasibility of Developing a Physical Model Using Rapid Manufacturing Technologies by Referencing Remotely Sensed Data to Simulate Outdoor Environments

EPA/600/R-18/268 | October 2018
www.epa.gov/homeland-security-research
United States
Environmental Protectior
Agency
oEPA
The Feasibility of Developing a
Physical Model Using Rapid
Manufacturing Technologies by
Referencing Remotely Sensed Data
to Simulate Outdoor Environments
Office of Research and Development
Homeland Security Research Program

-------
This page left intentionally blank

-------
fj,EPA
United States	Office of Research and	EPA/600/R-18/268
Environmental Protection	Development	June 2018
Agency	Washington, D.C. 20460	www.epa.gov/nhsrc
The Feasibility of Developing a Physical
Model Using Rapid Manufacturing
Technologies by Referencing Remotely
Sensed Data to Simulate Outdoor
Environments
Office of Research and Development
National Homeland Security Research Center

-------
Acknowledgments
Contributions of the following individuals and organizations to this report are gratefully
acknowledged:
US Environmental Protection Agency (EPA) Project Team
Timothy Boe
Anne Mikelonis
Sang Don Lee
Worth Calfee
Katherine Ratliff
US EPA Technical Reviewers of Report
Leroy Mickelson
Michael Pirhalla
Joan Bursey
US EPA Quality Assurance
Eletha Brady Roberts
Ramona Sherman
Tetra Tech
Sujoy Roy
Thomas Loecherbach
Michael Uges
Howard Models
Edward Howard

-------
TABLE OF CONTENTS
Disclaimer	ii
List of Figures and Tables	iii
Acronyms and Abbreviations	v
Executive Summary	i
1	Introduction	1
2	Methodology	1
2.1	Model Inputs	2
2.2	Software Tools	8
2.3	Manufacturing Technologies	8
3	Quality Assurance/Quality Control (QA/QC)	9
4	Prototype Model	10
4.1	Prototype Inputs	10
4.2	Finished Prototype	13
4.3	Prototype Evaluation	15
5	Final Model	16
5.1	Final Model Inputs	16
5.2	Data Enhancements	18
5.2.1	Vertical exaggeration	18
5.2.2	Vertical exaggeration of the ground only	18
5.2.3	Profiles	19
5.2.4	Lowering of Roads	22
5.3	Final Model Development	26
6	Observations and Recommendations	28
7	Bibliography	30
Appendix A	31
i

-------
DISCLAIMER
The U.S. Environmental Protection Agency (EPA) through its Office of Research and
Development funded and managed the research described herein under EP-C-15-004, PR-ORD-
16-01029 to Tetra Tech. It has been subjected to the Agency's review and has been approved for
publication. Note that approval does not signify that the contents necessarily reflect the views of
the Agency. Any mention of trade names, products, or services does not imply an endorsement
by the U.S. Government or EPA. The EPA does not endorse any commercial products, services,
or enterprises. The contractor role did not include establishing Agency policy.
Questions concerning this document, or its application should be addressed to:
Timothy Boe
U.S. Environmental Protection Agency
Office of Research and Development
National Homeland Security Research Center
109 T.W. Alexander Dr. (MD-E-343-06)
Research Triangle Park, NC 27711
Phone (919) 541-2617
li

-------
LIST OF FIGURES AND TABLES
Figure 1: National Map Download Viewer	3
Figure 2: National Map Elevation Products at different ground spacings	3
Figure 3: National Map Elevation source data (LiDAR point cloud) in LAS format	4
Figure 4: Point classes present in the downloaded point cloud	4
Figure 5: Sample area with a LiDAR point cloud, displaying points classified as ground (brown)
and buildings (orange) only	5
Figure 6: Sample area showing ground points only	5
Figure 7: Zoomed-in area of LiDAR point cloud, displaying coverage of points. Points are
classified as ground (brown) and buildings (orange) only. Note the limited resolution for
buildings at this scale	6
Figure 8: DEM created from the LiDAR point cloud, interpolating between the individual points
in the cloud	6
Figure 9: Status of 3DEP (3D Elevation Program) program supported by the US Geological
Survey (USGS). Grey and green areas have LiDAR data that meet 3DEP requirements [4]	7
Figure 10: Area of Interest (AOI) for first 3D model	11
Figure 11: 3D view of digital surface model. Note the fuzzy edges of the buildings	12
Figure 12: Downloaded Elevation Products, a regular grid of the ground surface	12
Figure 13: Downloaded vector data from Openstreetmap.org	13
Figure 14: Building footprints merged with terrain. Assigned building heights based on estimated
number of stories	13
Figure 15: Physical model prototype 1 under construction	14
Figure 16: Completed physical model prototype 1	14
Figure 17: Physical model prototype 2 from Howard Models	15
Figure 18: Blue polygon was the proposed AOI for the final 4 by 10 feet scale print	17
Figure 19: Digital Surface Model (DSM) for the final physical model. Data gaps in the water
have not been filled in	17
Figure 20: DSM created for the final physical model, showing more detail	18
Figure 21: 2x vertical exaggeration. Only non-building areas have been exaggerated	19
Figure 22: Elevation profile along yellow transect, showing two buildings and a road surface.. 20
Figure 23: Profile across a road, exaggerated	20
in

-------
Figure 24: Profile across a row of buildings	
Figure 25: Profile of street detail along yellow transect,
values	
	21
Note the relatively small range in y-axis
	21
Figure 26: Profile along yellow transect. Note the point density inside the profile box around the
yellow line	22
Figure 27: Point cloud with point classes Ground, Building, Water and Road. A polygon layer of
the roads was used to classify road points	23
Figure 28: Detail of point cloud with road network and building polygons	23
Figure 29: Vertical profile of classified point cloud: ground (gray), road (red) and building
(green)	24
Figure 30: Vertical profile depicting points of class 'road' have been lowered to provide a 0.2
depth when scaled at 1:800	24
Figure 31: Resulting elevation model with roads unchanged	25
Figure 32: Resulting elevation model with roads lowered	25
Figure 33: For the 4' by 10' print, the print company used the Tetra Tech-supplied surface model,
which combined the LiDAR point cloud with building polygons and road polygons. Roads were
lowered	26
Figure 34: Milling machine used for developing final physical model by Howard Models	27
Figure 35: Finished segment of the model. Note the varying heights of buildings	27
Figure 36: Finished segment of the model demonstrating lowering of the roadways	28
Table 1. Vendor Evaluation Matrix	16
IV

-------
ACRONYMS AND ABBREVIATIONS
2D
Two-Dimensional
3D
Three-Dimensional
3DEP
3D Elevation Program
AOI
Area of Interest
CNC
Computer Numerical Controlled
DEM
Digital Elevation Model
DSM
Digital Surface Model
FGDC
Federal Geographic Data Committee
GIS
Geographic Information System
LiDAR
Light Detection and Ranging
SWMM
Storm Water Management Model
V

-------
EXECUTIVE SUMMARY
This report was prepared to evaluate the feasibility of using rapid manufacturing technologies
(i.e., computer numerical controlled (CNC) milling, printing) to develop three-dimensional (3D)
physical models as surrogates for outdoor field experiments. These models have the potential to
serve as test beds when working with contaminants that may otherwise pollute the ambient
environment or to evaluate specific environmental phenomena. Furthermore, physical models
serve as useful planning or situational tools for emergency responders and decision makers.
Two 3D physical models manufactured by separate technologies (CNC milling and 3D printing)
were evaluated for use as a potential stormwater test beds. Models were evaluated according to
their completeness, imperfections, manufacturing method, materials, and compatibility. CNC
milling was found to produce a more reliable product that was germane to conducting flow
experiments. Steps necessary for enhancing and preparing data for manufacturing were also
documented. This work serves as a resource for selecting optimal printing materials, hardware,
and procedures when developing3D physical models using rapid manufacturing technologies.
The study concludes with observations and recommendations for conducting future research.
1

-------
1 INTRODUCTION
Physical models provide a unique test bed for conducting experiments under controlled
laboratory conditions. Specifically, physical models may be used to work with contaminants that
cannot be released into the ambient environment and serve as useful communication and
planning tools. Historically, these models have required extensive resources to build and
generally lacked realism. With the increased availability of rapid manufacturing technologies
such as 3D printers and CNC milling machines, the display of topography, terrain, and
infrastructure through physical 3D models has become more widespread [1],
The underlying data needed to develop a 3D physical model consist primarily of remotely sensed
terrain elevation data, but may also include road networks, building footprints, water bodies, and
other geographical data [2], These types of data are often collected for specific projects at a
relatively high cost. However, a wealth of high-quality data is publicly available for download
over a large part of the United States. These datasets are commonly funded by a variety of
federal programs, such as the 3D Elevation Program (3DEP) (supported by the US Geological
Survey (USGS)). These geographical data can be used to create digital models, which serve as
inputs for manufacturing platforms to create physical 3D replicas.
When using high-quality geospatial data, accurate representations of surface properties can be
made to better mimic real-world conditions (when compared to traditional physical models).
This potentially improves model fits. Furthermore, environmental conditions can be altered (i.e.,
lessened or intensified) by modifying the dimensions of specific geographical features (i.e.,
ground surfaces, roadways, buildings). These enhancements enable a more effective approach to
evaluating specific environmental phenomena.
Consequently, there is a need to investigate the feasibility of using rapid manufacturing
technologies, and the resulting models for conducting controlled scientific experiments. The
objectives of this study were fivefold: (1) determine the feasibility of developing a physical
model using rapid manufacturing technologies by referencing remotely sensed data (i.e., LiDAR
and spatial extracts) with the purpose simulating outdoor environments; (2) understand whether
the scale at which the model operates is representative of the environmental conditions necessary
to conduct stormwater experiments, and the extent of data manipulation necessary to represent
such conditions; (3) compare and contrast rapid manufacturing technologies; (4) document the
steps required to build a physical model; and (5) document any gaps and recommendations for
continuing this research.
2 METHODOLOGY
The development of physical models consists of three primary components: (1) model inputs
depicting a given environment, (2) software tools for modifying the model inputs, and (3)
hardware and methods for manufacturing the model [2], The following sections describe these
processes in detail as well as important considerations one must contemplate when building a
physical model.
1

-------
2.1 Model Inputs
Physical models are typically developed to mimic an environmental setting or area of interest
(AOI). An AOI was selected for development of the model, following an evaluation of multiple
metropolitan areas across the United States. The final selection of Cambridge, Massachusetts,
was based on the availability of a calibrated numerical stormwater model that the Cambridge city
government was willing to share. A subset of the City of Cambridge model was provided for
analysis, spanning an area with a reasonable amount of topographic variation including areas that
have historically exhibited flooding. The selected area contained commercial and residential
buildings and included the city's water supply. Access to an existing numerical model allows
joint exploration of stormwater flow and contaminant transport scenarios using both the physical
and computer model.
Physical models are either built by hand or manufactured by industrial-scale machinery.
Although some models do not require spatial accuracy (as commonly seen in architectural
models), those used to support scientific experiments or decision making-often require
substantial accuracy [2, 3], When developing highly-accurate models, geospatial elevation
products are commonly used, which consist of a raster (i.e., graphics) or grid points representing
elevation (Digital Elevation Models, DEMs) of a given terrain surface. These products are
typically derived from remotely sensed data (e.g., LiDAR), or ground-based surveys and are
presented as x-, y-, z-coordinates [4], Since remote sensing technologies such as LiDAR capture
the distance light or radio waves travel, ground features can be categorized (i.e., classified)
according to their height from the ground surface. Using this technique, terrain, vegetation, and
buildings can easily be distinguished from imagery.
A common source for elevation data is the National Map platform of the USGS
(https://viewer.nationalmap.gov/basic/. accessed August 15, 2018) as shown in Figure 1 [4], The
National Map provides two categories of elevation data: 1) Elevation Products and 2) Elevation
Source Data. For the purposes of this study, Elevation Products (i.e., DEMs) and Elevation
Source Data (i.e., LiDAR point cloud data) were used as shown in Figure 2 and Figure 3.
2

-------
^v| Cambridge Massachusetts
Advanced Search Potions
Data
C Boundaries - National Boundary Dalaset
~ Elevation Products (3DEP)
6E Elevation Source Data (3DEP)
Product Search Filter
[_]AU Subcategories
~	DEM Source (OPR)
snowAvattabtiity
~	Ifcar Digital Surface Model (DSM)
snow Availability
~	llsai Oitliotuctiliod Radai Imago (ORI)
Hydrography (NHDPIus MR, NMD, WBD)
Imagery - NAIP Plus (1 meter to 1 foot)
0 Use Map	OBexiPomt Ocunent Extent O Coordinates • Located Poml Polygon
<( CD Map indices	1 Degree 15 Minute 7 5 Minute All
Figure 1: National Map Download Viewer.
0 Elevation Products (3DEP)
¦
Product Search Filter
I I All Subcategories
O 1 arc-second DEM
Show Availability
~	l meter DEM
Show Availability
01/3 arc-second DEM
Show Availability
O 1/9 arc-second DEM
Show Availability
O 2 arc-second DEM - Alaska
Show Availability
~	5 meter DEM (Alaska only)
Show Availability
~	Contours (1:24,000-scale)
Show Preview
Figure 2: National Map Elevation Products at different ground spacings.
Description
3

-------
0 Elevation Source Data (3DEP)
Product Search Filter
~ All Subcategories
~	DEM Source (OPR)
Show Availability
~	Ifsar Digital Surface Model (DSM)
Show Availability
I I Ifsar Orthorectified Radar Image (ORI)
Show Availability
@ Lidar Point Cloud (LPC)
Show Availability
Data Extent
Varies
File Format
LAS/LAZ
Figure 3: National Map Elevation source data (LiDAR point cloud) in LAS format.
The LiDAR data provided by the National Map platform were pre-classified for ground, bridges,
rail, and water. Buildings had not been classified, but points reflected from buildings were
available, together with other points (i.e., noise), as a category titled 'unclassified' as seen in
Figure 4. One of the advantages of working with LiDAR is that unclassifi ed points can be
reclassified using specialized geographic information system (GIS) software. These unclassified
points can be binned (i.e., reclassified) based on some established criteria (typically height or
point spacing). For this project, Global Mapper was used to process the LiDAR points, classify
buildings, and create a digital elevation model. Representative maps are shown in Figure 5
through 8.
Figure 9 shows the 3DEP (3D Elevation Program) status as of 2017. Areas in gray or green have
existing or planned LiDAR data that meet the specifications necessary to create a physical
model. Thus, cities that fall within these areas can be used to develop physical models using the
same types of processes that have been employed in this study.
histogran of clas
sification of
po ints:
2635703
unclassif ied
<1>
1588922
ground <2)

578
noise (7>

155269
water <9)

7433
rail <10>

1571061
bridge deck
C17>
741503
Reserved for
flSPRS Definition <18)
Figure 4: Point classes present in the downloaded point cloud.
Description
4

-------
1000 ft
1500 ft
2000 ft
N
0
Figure 5: Sample area with a LiDAR point cloud, displaying points classified as ground
(brown) and buildings (orange) only.
N
<*>
tX U"	">s r
i—i—i—h
H	1	1	1
Figure 6: Sample area showing ground points only.
5

-------
Figure 7: Zoomed-in area of LiDAR point cloud, displaying coverage of points. Points are
classified as ground (brown) and buildings (orange) only. Note the limited resolution for
buildings at this scale.
'/*	V v" V -	f
1000 ft
2000 ft
Figure 8: DEM created from the LiDAR point cloud, interpolating between the individual
points in the cloud.
6

-------
' jr	Gulf of
, Mexico
Explanation
In-Progress and Existing Data that
Meet 3DEP Specification
^USGS
sdencB for a changing world
Puerto Rico and US Virgin Islands
U.S. Department of the Interior
U.S. Geological Survey
National Geospatial Program
G17PS00746 / G17AS00116 Attachment F
3D Elevation Program: FY17 Status of 3DEP Quality Data
| lidar
| ifsar (Alaska)
Other lidar data
No publicly available lidar data (or ifsar in
Alaska)	
Pacific
Ocean
Federated
States of
Micronesia
Atlantic
Ocean
Map showing
the a real extent
and quality level of
planned, in progress,
and existing publicly
available lidar (ifsar in
Alaska) data identified
by the U.S. Interagency
Elevation Inventory
(USIEI) that meet
3DEP base-level
specification as of
August2017.3DEP base-
level specification data are
defined as quality level 2 or better
lidar data (ifsar in Alaska) and 8
years old or newer. The inventory was
produced in partnership by the U.S.
Geological Survey and the National
Oceanic and Atmospheric Administration.
While every attempt has been made to
accurately inventory projects that are
publicly available, some errors and
omissions may occur.
as of August 2017
Northern Mariana
Islands
Salpan
J
Tim an
Guam
American Samoa
Oloasga
Figure 9: Status of 3DEP (3D Elevation Program) program supported by the US Geological Survey (USGS). Grey and green
areas have LiDAR data that meet 3DEP requirements [4].
7

-------
2.2 Software Tools
Software tools are critical for modifying and converting elevation datasets for use in rapid
manufacturing hardware. There is a wide range of software available for modifying LiDAR data.
These tools vary in cost (proprietary vs. open source) and level of sophistication. The
functionality of many of these tools overlap and can be used interchangeably. For example,
profiles across a LiDAR point cloud can be displayed in Terrascan, in LP360, and in the LiDAR
extension of GlobalMapper. Even though capabilities overlap, each tool has its own strength:
Terrascan is used predominantly for automatic classification of point cloud data, LP360 is often
preferred for interactive quality assurance/quality control QA/QC, and Global Mapper offers a
lower cost and easy to use alternative for the novice user. Sections 4 and 5 of this report
highlight how each set of software was used. The following software is capable of working with
LiDAR data1:
Terrascan by Terrasolid: this LiDAR-processing software is well suited to analyze
and classify LiDAR from very large LiDAR data sets
(http://www.terrasolid.com/prodiicts/teiTascanpaee.php. accessed July 23, 2018).
Terrascan runs on top of a Bentley Microstation (described below).
Bentley Microstation: This software platform for two- and three-dimensional design
and drafting is widely used in the architectural and engineering industries. This
software platform generates 2D/3D vector graphics for visualization
(https://www.bentley.com/en/prodiicts/brands/microstation, accessed July 23, 2018).
LP360: This production tool works with large LiDAR data sets and is well suited for
quality control by analyzing elevation profiles of the data. LP360 exists as a
standalone version and as an extension to ESRI ArcMap.
ArcMap by ESRI: This GIS software is used for management and viewing of
geospatial data.
GlobalMapper* version 19, with LiDAR module: This versatile tool combines
many of the tasks that can be achieved with the above software tools at a reasonable
cost. The software has multiple options for exporting data to various formats suited
for 3D printing. Global Mapper was used throughout most of this study.
Web browser and uGet: For a bulk download of larger data sets, The National Map
suggests the use of uGet.
Lasinfo*: This tool can be run in command line mode to summarize the properties of
numerous .las files.
2.3 Manufacturing Technologies
Rapid prototyping technologies offer great potential in the scientific and emergency management
fields. Physical models depicting real-world environments can quickly be developed using user-
defined or remotely sensed data. The resulting models can be used as scientific testbeds for
simulating outdoor landscapes or as planning tools.
1 [*] denotes the software was used in this study.
8

-------
Two prominent manufacturing technologies were evaluated as part of this study: (1) CNC; and
(2) 3D printing. CNC is described as a subtractive process (i.e., milling) where a block of
material, typically consisting of a polymer plastic or foam, is cut or shaved to form a finished
product [3], The cutting process is guided by a CAD program that defines the end-products'
dimensions and shape. CNC was first developed in the 1940s and has since been used to
manufacture industrial scale products [3], 3D printing is a relatively new technology that has
surged in popularity in recent years. In contrast to CNC, 3D printing is an additive process where
a product is printed in a 3D fashion using layers of plastic (typically thermoplastic polymer) [1],
Like CNC, 3D printed products are designed in a CAD-like program.
Though there are many similarities in the two manufacturing technologies featured in this study,
there are two stark differences: (1) given that 3D printing is a relatively new technology, there
are still anomalies in the design and printing process that may create imperfections in the
finished product. However, CNC, a well-established manufacturing process, is more likely to
create a reliable and imperfection-free product; and (2) 3D printing is a relatively inexpensive
manufacturing option. 3D printing requires very little training. Startup costs average between
$500-1,000. CNC typically requires large and expensive equipment to operate and may lack
practicality, except in large industrial settings.
3 QUALITY ASSURANCE/QUALITY CONTROL
The purpose of this study was to determine the feasibility of developing a physical model using
rapid manufacturing technologies. The work and conclusions presented as part of this study were
empirical and observational - no scientific experiments were performed. Secondary data were
used to inform the area of extent, scaling, and development of the physical model. These data
were collected as part of a larger EPA-led effort assessing the fate and transport of biological and
radiological contaminants through urban landscapes [5], Technical area leads evaluated the
quality of the data collected by this effort (i.e., secondary data), and based on their expert
opinion, determined if the data should be considered in the design or implementation of the
physical model. Collected literature was evaluated according to the "Literature Assessment
Factor Rating" as shown in Appendix A. All supporting documentation of the secondary data
considered worthy for inclusion were cited. However, no experimental confirmation of
secondary data (e.g., accuracy, precision, representativeness, completeness, and comparability)
was conducted as part of this study.
Geospatial products used in the manufacturing of the physical model abided by geospatial data
standards and conformed with the USGS National Geospatial Program standards. DEM data
generated as part of LiDAR met, at minimum, specifications defined in the USGS LiDAR Base
Specification (https://pubs.iisgs.gOv/tm/1 Ib4/. accessed August \). 3DEP products were
documented using the Federal Geographic Data Committee (FGDC) content standard for
geospatial metadata and can be located through the Data.gov Open Government Initiative. No
further efforts were made to confirm the spatial accuracy of the GIS inputs or the physical
models created as part of this study.
9

-------
4 PROTOTYPE MODEL
To better understand the scale and type of information necessary to build a physical model, a
small 16-inch square prototype was developed. Four companies were contracted to conduct this
work. One company was unable to deliver a prototype in a timely manner, and one was unable to
process the files provided. Two of the selected companies could develop prototypes.
The model builders were given flexibility to select the best methodology to develop the
prototype, with the understanding that the method would need to be scaled up to develop a model
with dimensions of 4 feet by 10 feet. The prototypes were evaluated using the following criteria:
• Data compatibility: the printing method is compatible with the digital elevation data;
• Successful print: the printing method successfully produced a viable product depicting
the AOI;
• Aggregate print: the print exists as a single unit and did not require extensive
modifications; and
• Imperfections: the model was free of blemishes and other anomalies that might otherwise
impact the quality of experimental work.
One of the two model builders used CNC milling to prepare the ground surface and developed
individual 3D prints of the buildings. The 3D approach required that the prints be manually
glued onto the terrain. The second model builder used an automated CNC router to build the
terrain and buildings as a single piece. This second model builder, Howard Models, of Toledo,
OH, was selected for additional work.
4.1 Prototype Inputs
Figure 10 displays the DSM inside the AOI and aerial imagery as a background. The term DSM
is frequently used to describe an elevation model that includes the ground and the visible surface,
such as buildings and possibly vegetation. The Global Mapper software allows direct loading of
these data directly into the viewer.
The process of building the digital model within Global Mapper was broken into two parts: (1)
terrain and (2) buildings. Elevation Source Data were downloaded, which contain all points, i.e.,
3D coordinates from ground surface, buildings, vegetation, etc. Points on the ground were
classified as 'ground' in the default dataset. No additional processing was performed on the
terrain (i.e., ground surface).
The remaining points were defined as 'unclassified', indicating that points representing buildings
did exist in the "noise." Further, there was no distinction between points on buildings and points
on trees. For this reason, a tool in Global Mapper was run to extract buildings from a point cloud,
which generates vector polygons representing the footprints of each building. For the purposes of
this study, a building is defined by one polygon, with a single elevation (i.e., a flat roof).
However, when the building polygons (flat roofs) were derived from the point cloud, it was
discovered that the point cloud contained individual points around the edges of the roofs but just
10

-------
outside the building outline. Therefore, even though the rooftops were flat and smooth, the
buildings looked fuzzy around the edges as shown in Figure 11.
To smooth the edges of buildings, ground surfaces and buildings were processed separately. The
Elevation Products (i.e., bare earth) were used to represent the ground surfaces. Unlike the
Elevation Source Data, the Elevation Products do not contain building elevations, only ground
surfaces as seen in Figure 12. For buildings, OpenStreetMaps (www.openstreetmap.org)
infrastructure data were used (Figure 13). The OpenStreetMaps building footprints are 2D
polygons with distinct edges. Flowever, the building footprints are void of elevation information.
Google Street View was used to visually estimate the number of floors for each building within
the AOI. Building heights were estimated by assuming a fixed height per story. The
OpenStreetMaps building footprints were extruded vertically using Global Mapper according to
their established height value. The final model showing the buildings merged with the terrain is
shown in Figure 14. The final digital model was converted to a .stl file (stereolithography CAD
software) and uploaded to the milling machine.
Figure 10: Area of Interest (AOI) for first 3D model.
11

-------
Figure 11: 3D view of digital surface model. Note the fuzzy edges of the buildings.
S_19TCG225S5CLTlFf_20?
LUSgS.tgP.OPg.m.Mt.CMiSP.fantf,^ 19.A2J0»S. l9ro52iS935.riFf^9i
Terrain
Model
Boundaries
Figure 12: Downloaded Elevation Products, a regular grid of the ground surface.
12

-------
Figure 13: Downloaded vector data from Openstreetmap.org.
Figure 14: Building footprints merged with terrain. Assigned building heights based on
estimated number of stories.
4.2 Finished Prototype
Figure 15 shows the first prototype under construction with individual printed buildings that
were manually glued to the terrain model. Figure 16 shows the completed prototype. The
resulting model from the second model builder, Howard Models, is shown in Figure 17.
13

-------
Figure 15: Physical model prototype 1 under construction.
Figure 16: Completed physical model prototype 1.

-------
Figure 17: Physical model prototype 2 from Howard Models.
4.3 Prototype Evaluation
The process of developing the prototypes was instructive in helping define an established
methodology for building the final model. Upon evaluating the prototype, the following
decisions were made:
• Howard Models was chosen to build the final model because the overall approach using a
single CNC machine appeared simpler than construction that involved a manual step of
gluing the buildings as required by the 3D printing process.
• The 3D printed model appeared to show ripples on its surface, likely due to an artifact of
the 3D printing process. This artifact may be an anomaly of this printing instance (i.e.,
inadequate printing material) or due to subpar printing hardware. It was decided that this
anomaly may interfere with experimental work and was not representative of the real-
world environment.
• Using the same vertical and horizontal scale for the model would result in the road
surfaces being visible but would show very little channelization for experiments with
fluids. Thus, a different methodology would be needed to represent the roads, possibly by
performing vertical exaggeration of the elevation. Based on the findings of the literature
review and the Cambridge stormwater model, the velocities seen in the hydraulic model
would likely not match the velocities of the physical model. Therefore, a vertical
exaggeration would be necessary to represent the channelization of flow along roads [5,
6]-
• Although higher resolution data are collected for urban area studies, the lower-resolution
free public LiDAR data were adequate for developing the types of physical models
envisioned in this work. However, higher-resolution data (especially at larger scales) may
15

-------
reduce the need for manipulating data to exaggerate features to enhance a given
phenomenon (in the case of this study, flow channelization).
Table 1, below, shows a succinct comparison of the vendors and manufacturing processes
evaluated as part of this study.
Table 1. Vendor Evaluation Matrix
Metrics
Vendors (V)
Howard Models
V2
V3
V4
Data Compatibility (Y/N)
Yes
Yes
No1
N/A
Successful Print (Y/N)
Yes
Yes
No2
No3
Manufacturing Type
CNC milling
3D printing
N/A
N/A
Aggregate Print (Y/N)
Yes
No4
N/A
N/A
Imperfections (Y/N)
No
Yes5
N/A
N/A
Print Material
High density
polyurethane foam
Polylactic acid
N/A
N/A
1 Vendor was unable to convert .las files into a usable file format.
2 Vendor was unable to work with the provided topography data (i.e., LiDAR).
3 Vendor became unresponsive during the model building process.
4 Vendor's building process required buildings be manually glued to the ground surface.
5 The surface of the model contained ripples; glued buildings created crevices that may collect water.
5 FINAL MODEL
Howard models (i.e., CNC milling) was selected to develop a final model measuring 4 feet by 10
feet. The methodology developed as part of the prototype was slightly modified to smooth the
appearance of buildings and other vertical features. Furthermore, a process for normalizing the
road depth (i.e., to enhance flow channelization) was developed.
5.1 Final Model Inputs
The final model covered a much larger area, requiring a greater extent of elevation data. Figure
18 shows the extended AOI. Like the prototype, LiDAR point cloud data (.las format) was used
to simulate ground surfaces. Building footprints were derived using OpenStreetMap and height
values were manually assigned by referencing Google Street View. Street vector polygons were
also added to the model. These layers were used to smooth and create a constant width and depth
for roadways. Figure 19 and Figure 20 show the surface model that was provided to the model
building company. Data were exported from Global Mapper both in .img and in .stl format.
16

-------
V ^/^cCambridge..H:griancl:
' N e i g h b o r h oo d- N i n e _
resh Pond
,°Strawberry Mill.
Figure 18: Blue polygon was the proposed AOI for the final 4 by 10 feet scale print.
Figure 19: Digital Surface Model (DSM) for the final physical model. Data gaps in the
water have not been filled in.
17

-------
Figure 20: DSM created for the final physical model, showing more detail.
5.2 Data Enhancements
Because the LiDAR data lacked clarity to depict curb heights, and a model of this scale would
likely prevent flow channeling, modifications to the street and ground features were necessary.
5.2.1 Vertical exaggeration
Vertical exaggeration is a common tool in displaying or analyzing map data. The reason can be
explained with mapping scale. Using the Cambridge model as an example, the pilot model scale
was 1:1,200, and for the larger model 1:800. An area of 1,200 ft. by 1,200 ft. would be
represented in the model by 1 ft. by 1 ft. If the same scale is used for the vertical, a variation of
the terrain of 1 ft., would be equivalent to 0.0008 ft. or 0.01 inches, which is insufficient to
permit channelization water in an experimental set up [6, 7
5.2.2 Vertical exaggeration of the ground only
If vertical exaggeration was applied uniformly to the model, includi ng buildings, tall buildings
would be unrealistically large. Therefore, points were classified into specific ground and building
bins, which allowed for vertical exaggeration of select ground points (or specific bins). Using
this approach, ground surfaces were selectively extruded by 1 ft., leaving the buildings to scale
[5], GIS software such as GlobalMapper, Tarascan, and ArcMap can be used for this task. The
completed exaggeration is displayed using Global Mapper in Figure 21. Although feasible, the
aesthetic result of the ground surface exaggeration was not positive, and this approach was not
pursued in the physical model.
18

-------
Figure 21: 2x vertical exaggeration. Only non-building areas have been exaggerated.
5.2.3 Profiles
Profiles were plotted perpendicular to roads, including roads and buildings. The purpose of these
profiles was to prepare a recommendation for vertical exaggeration for roadways based on the
variation and details that exist in the data. The profiles show the limitations of the data set.
Profiles were visualized in Global Mapper (Figure 22-26). The same tasks can be done in other
3D point cloud processing applications such as LP360 or in Terrascan.
19

------- t Profile/Line of Sight File Path Setup Display Options Calculate Click Polygon to Select and Edit Lidar Points

-------
Polygon to Select and l
: Global Setting from Toolbar
From Pos 324071.029.4694680 665
To Pos: 324007.324.4694579.516
Figure 24: Profile across a row of buildings.
File Path Setup Display Options Calculate
Use Global Setting fron
From Pos; 323977.582,4694569.297
To Pos: 323980.367,4694580.536
Figure 25: Profile of street detail along yellow transect. Note the relatively small range in
y-axis values.
21

-------
Path Setup Display Options Calculate
i Select and Edit I

Global Setting from Toolbar
ToPos: 323980.367, 4694580.5361
Figure 26: Profile along yellow transect. Note the point density inside the profile box
around the yellow line.
5.2.4 Lowering of Roads
Following construction of the prototypes and examination of the road profiles, un-exaggerated
road surfaces would be insufficient to permit channelized flow in future stormwater experiments
with the model (Figures 27-31). To promote flow in the final model, a minimum height
difference was necessary for roadways. Unlike buildings, road points are at the same elevation as
ground points in the 3D point cloud returns. To classify road points, a more accurate dataset may
be used or superimposed on the points to reclassify them as road points. A GIS "Impervious
Surface Layer—Road Category" from the City of Cambridge was used to re-classify the LiDAR
points as "point class roads." Ftaving points assigned to four separate classes (i.e., ground,
building, road, water and unclassified) allowed for a simplified approach to modifying the
elevation of certain classes - in this case, strictly roadways. This approach produced a flow
model that encouraged surface flows along roadways (Figures 32-33).
22

-------
Figure 27: Point cloud with point classes Ground, Building, Water and Road. A polygon
layer of the roads was used to classify road points.
It®
Si##
Figure 28: Detail of point cloud with road network and building polygons.
23

-------
Figure 29: Vertical profile of classified point cloud: ground (gray), road (red) and building
(green).
Figure 30: Vertical profile depicting points of class 'road' have been lowered to provide a
0.2 depth when scaled at 1:800.
24

-------
Figure 31: Resulting elevation model with roads unchanged,
Figure 32: Resulting elevation model with roads lowered.
25

-------
Figure 33: For the 4' by 10' print, the print company used the Tetra Tech-supplied surface
model, which combined the LiDAR point cloud with building polygons and road polygons.
Roads were lowered.
5.3 Final Model Development
The final physical model (4 feet by 10 feet) was built using the enhanced surface model (i.e.,
lowered roads). The surface model was converted to .img and .stl formats, the file format used by
the CNC printer. The supplied terrain model was milled from high density polyurethane foam
using the Thermwood ES 919-3 Axis CNC Router (http://www.thermwood.com/index.html.
26

-------
accessed August 15, 2018). For the final model, 15 pounds of polyurethane foam was used.
Figure 34 shows the final model built using the CNC milling process.
The final model took 48 hours to mill and weighed approximately 200 lbs. The height of the
model measured 6 in. from the base to the highest point. The physical model was sliced into two
4 by 5 inch segments for easier mobility.
Figure 34: Milling machine used for developing final physical model by Howard Models.


Figure 35: Finished segment of the model. Note the varying heights of buildings.
27

-------

Figure 36: Finished segment of the model demonstrating lowering of the roadways.
6 OBSERVATIONS AND RECOMMENDATIONS
This study demonstrates the feasibility of developing a physical model using rapid
manufacturing technologies by referencing remotely sensed data (i.e., LIDAR and spatial
extracts) with the purpose simulating outdoor environments. Such a model serves as unique test-
bed for homeland security research efforts, particularity when working with contaminants that
may otherwise pollute the ambient environment or as environmental surrogates for evaluating
fate and transport of contaminants. Furthermore, physical models serve as useful planning or
situational tools for emergency responders and decision makers. These models can be
manufactured quickly following a di saster or prior to an exercise for planning at the strategic,
tactical, or operational level.
In addition to the findings above, several other observations were made during the preparation
and implementation of the model:
This approach shows potential in the field of atmospheric and fate and transport
modeling where scaled models may be used to simulate real-world conditions and
outcomes when outdoor experiments risk damaging or contaminating the
environment or considerable resources are required to execute a well-controlled field
study. For instance, surface chemical interactions and fluvial experiments may be
conducted in controlled laboratory conditions. This approach may serve as an
alternative to typical outdoor stormwater experiments and to commonly used
28

-------
numerical stormwater models (such as the EPA Storm Water Management Model
(SWMM) family) [5],
Both manufacturing methods (i.e., CNC and 3D printing) were found suitable for
achieving the objectives set forth by this study. However, the CNC milling method
produced a more reliable product that contained fewer imperfections when compared
to the 3D printed method. The anomalies seen on the 3D printed model may be a
result of inadequate printing materials or hardware. As the 3D printing technology
progresses and the printing process becomes more standardized, these anomalies will
likely become less apparent. Given its ability to develop prototypes requiring little
training, resources, and laboratory space, the authors recommend 3D printing be
reconsidered for future experiments.
When using geospatial data to build physical models, it is imperative that the
resolution and scale of those data be accurate enough to represent the phenomena
being investigated (i.e., low-resolution data may simplify or minimize geometries that
would otherwise have a tangible effect in the real-world). When high-resolution data
are unavailable, supplemental data may be used to correct poorly represented features
(as seen in this study). Coordination among all parties (i.e., scientists, GIS specialists,
and manufacturers) is critical to ensure the use of compatible datasets, spatial
projections, and minimization of data processing.
An empirical review of representative materials for use in a scaled physical model
was conducted in parallel to this study [5], However, laboratory-scale experiments are
necessary to determine the likeness of these surfaces (or potential surrogates) on
urban materials based on their interaction with surrogates of biological and
radiological contamination. Furthermore, additional testing is needed to determine the
compatibility of model building and coating materials when sterilized for potential
reuse.
To support future stormwater experiments, the potential fluid flow velocity of the
roadways was evaluated. These velocities were extracted from the literature and
evaluated using a stormwater model developed by the City of Cambridge. The
velocities seen in the stormwater model would likely not match those of the physical
model (barring any enhancements to the surface data). Because of this finding,
vertical features would need to be modified to simulate real-world surface flows (as
reported in section 5.2). Using this approach, a vertical exaggeration depth of 0.8-1
inches was assumed [5, 6], Experimental testing is necessary to confirm these
findings.
29

-------
7 BIBLIOGRAPHY
1.	Rayna, T., and Ludmila, S. (2016). From rapid prototyping to home fabrication: How 3D
printing is changing business model innovation. Technological forecasting and social change.
Journal of Technological Forecasting and Social Change. 102 (214-224).
2.	Roy, S B., Ungs, M.J., Boe, T., Mikelonis, A., Lee, S.D., Calfee, M., and Ratliff, K. (2018).
Use of small scale physical models for conducting transport and decontamination
experiments, presentation at U.S. Environmental Protection Agency International
Decontamination Research and Development Conference, May 2018, Durham, NC.
3.	Everything you need to know about CNC machines: Creative mechanisms blog.
https://www.creativemechanisms.com/blog/everything-you-need-to-know-about-cnc-
machimes (accessed August 6, 2018).
4.	National Geospatial Program Standards and Specifications.
https://nationalmap.gov/standards/ (accessed May 10, 2018).
5.	Ungs, M.J., Roy, S.B., Boe, T., Mikelonis, A., Lee, S.D., Calfee, M., and Ratliff, K. (2017)
Theoretical development of scale modeling of stormwater transport of biological agents and
radionuclides in urban landscape, draft Technical Memorandum prepared for US
Environmental Protection Agency.
6.	Peakall, J., and Warburton, J. (2017). Surface tension in small hydraulic river models - The
significance of the Weber number. Journal of Hydrology (New Zealand). 35(2) 199-212.
30

-------
APPENDIX A
Literature Assessment Factor Rating
Rate each factor from 0 (not applicable) to 5 (strongly applicable) and total for the overall rating.
Title of Article:
NHSRC Reviewer: Rating (1-5)
Focus
The work not only addresses the area of inquiry under consideration but also
contributes to its understanding; it is germane to the issue at hand.

Verity
The data are consistent with accepted knowledge in the field or, if not, the new
or varying data are explained within the work. The data fit within the context
of the literature and are intellectually honest and authentic.

Integrity
The data are structurally sound and present a cohesive story. The design or
research rationale is logical and appropriate.

Rigor
The work is important, meaningful, and non-trivial relative to the field. It
exhibits sufficient depth of intellect rather than superficial or simplistic
reasoning.

Utility
The work is useful and professionally relevant. It makes a contribution to the
field in terms of the practitioners= understanding or decision-making on the
topic.

Clarity
The work is written clearly, not dependent on jargon. The writing style is
appropriate to the nature of the study.

Soundness
The extent to which the scientific and technical procedures, measures,
methods, or models employed to generate the information is well documented
and reasonable for, and consistent with, the intended application.

Uncertainty and
Variability
The extent to which the variability and uncertainty (quantitative and
qualitative) in the information or in the procedures, measures, methods, or
models are evaluated and characterized.

Evaluation and
Review
The extent of independent verification, validation, and peer review of the
information or of the procedures, measures, methods, or models.

Total:
Overall Rating:
35—45 High quality article
25—34 Moderately high quality article
15—24 Lower quality article but with some useful information (please explain below)
<15 Unacceptable/Do not use
31

-------
vvEPA
United States
Environmental Protection
Agency
PRESORTED STANDARD
POSTAGE & FEES PAID
EPA
PERMIT NO. G-35
Office of Research and Development (8101R)
Washington, DC 20460
Official Business
Penalty for Private Use
$300

-------