Factors Affecting Dimensional Precision of Consumer 3D Printing

International Journal of
Aviation, Aeronautics, and
Aerospace
Volume 2 | Issue 4
Article 2
9-21-2015
Factors Affecting Dimensional Precision of
Consumer 3D Printing
David D. Hernandez
Embry-Riddle Aeronautical University, hernad17@erau.edu
Follow this and additional works at: http://commons.erau.edu/ijaaa
Part of the Applied Mathematics Commons, Design of Experiments and Sample Surveys
Commons, Engineering Commons, Entrepreneurial and Small Business Operations Commons, and
the Technology and Innovation Commons
Scholarly Commons Citation
Hernandez, D. D. (2015). Factors Affecting Dimensional Precision of Consumer 3D Printing. International Journal of Aviation,
Aeronautics, and Aerospace, 2(4). https://doi.org/10.15394/ijaaa.2015.1085
This Article is brought to you for free and open access by the Journals at Scholarly Commons. It has been accepted for inclusion in International Journal
of Aviation, Aeronautics, and Aerospace by an authorized administrator of Scholarly Commons. For more information, please contact
commons@erau.edu.
Hernandez: Dimensional Precision 3D Printing
3D printing has been gaining more widespread usage, with falling prices and operational
simplicity bringing the tool out of the realm of corporations and into the hands of individuals.
Indeed, the techniques comprising today’s rapid prototyping – creating full-scale models that
reproduce the size, shape, and functionality of conventionally manufactured items – have made it
possible for individuals to create new products in shorter timeframes than whole corporations
could just a few short years ago. Roland DGA Corporation (2011) cites two major shifts in how
products are developed – an economic shift caused by rising costs associated with outsourced
manufacturing and an increase in entrepreneurship, respectively – which are pushing towards a
business model where conceptualization and productization are co-located. Three-dimensional
Computer-Aided Design (CAD) data and 3D scanning technology have both been made available
and refined through open-source communities, in addition to the availability of their for-profit
counterparts. The 3D printer forms the final component in a chain which turns ideas and
intellectual property into tangible product.
Rapid prototyping expedites the typical manufacturing process through the use of both
subtractive and additive technologies, as opposed to wholesale creation of customized tooling –
the traditional approach (CustomPartNet, 2009). A subtractive technology, such as CNC milling,
uses digital data to transform raw material by removing material in a predetermined fashion. By
skipping the step of creating typical manufacturing tooling, the same rapid processes, techniques,
and tools can be used to manufacture a wide-range of devices more quickly.
The most ubiquitous and economical consumer 3D printing devices make use of additive
technology – fused deposition modeling (FDM). FDM builds up a physical model layer-by-layer,
fusing higher layers of material to the layers beneath them to create new objects (Akande, 2015).
Though the march towards increasingly capable consumer printing has been steady, it is important
Published by Scholarly Commons, 2015
1
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
to note that economical 3D printing devices have not yet achieved a level of simplicity and
reliability comparable to that of the typical consumer devices that have achieved mass adoption.
In order to provide a quantitative analysis of this reliability, the study described in this paper
focused on dimensional precision of a consumer-grade, FDM printer. A full factorial design of
experiments (DOE) analysis was conducted, resulting in an Analysis of Variance (ANOVA)
design that shed light on the various factors that affect the use of FDM, in terms of dimensional
precision. The goal was to evaluate the limitations of the technology, to rule out factors that do not
contribute in a statistically significant fashion to print precision, and to provide a practical,
quantitative guide for optimizing results of consumer grade 3D printing for application as an
engineering tool.
Finite Deposition Modeling – An Emerging Technology
FDM raw material may consist of a variety of substances – often thermoplastics or
thermoplastics infused with other materials. The most common materials used for FDM are
Acrylonitrile butadiene styrene (ABS) and Polylactide (PLA), with their characteristics of
becoming a liquid substance with predictable flow properties in response to heat, while forming a
reliable solid once cooled (Liing Shian Colorant Manufacturer Co., Ltd., 2013). This process of
heating and cooling plastic, with some well-modeled aspects, is still susceptible to random
variation, with unpredictable results depending on the shape being printed. Differences in material
properties across manufacturers and even across different material lots from the same
manufacturer can result in very different printing results, requiring user intervention to refine
several printer parameters until usable prints are achieved (Boots Industries, n.d.). These include
extrusion rate, nozzle temperature, bed temperature and the properties of the design, itself. Several
papers in the public literature (e.g., Bakar, Alkahari, & Boejang, 2010; Luzanin, Movrin, &
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
2
Hernandez: Dimensional Precision 3D Printing
Plancak, 2013; Udroiu & Mihail, 2009) have attempted to quantify the effects of various usercontrollable factors on print quality. In at least one case (Luzanin, 2013), the investigators were
required to change their experimental plan when the printer was found to be incapable of printing
adequate test articles.
The focus of this paper is on the use of consumer-grade 3D printing to create engineering
prototypes of, tooling for, or finalized instances of mechanical devices. Unlike aesthetic uses of
FDM, a focus on accuracy - ability to meet precise physical dimensions, consistent shapes, and
predictable surface finish - is important in the case of engineered mechanical devices. 3D printing,
because of its additive nature, provides a capability to create unique components that cannot be
replicated via subtractive techniques. Consumer grade printing provides advantages in both
expense and turnaround time that represent a significant change in how certain engineering
challenges may be addressed. The measurement of fluid flow, for example, necessitates very
precise control of dimensioned parts with specific characteristics (The American Society of
Mechanical Engineers, 2004), which works counter to the concept of physical experimentation.
The approach of using changeable, disposable components in order to iterate towards an optimal
combination of test parameters, has previously been impractical. 3D printing could, among other
uses, provide a way to fabricate customized fluid flow test components that rival their significantly
more expensive and less-readily-manufactured (and, in turn, less practically customizable)
metallic counterparts. In order to meet the requirements for such applications, however, consumer
3D printing must first attain a level of dimensional consistency and precision that is, as yet, atypical
of FDM technology.
An example of the kinds of dimensional errors prevalent with consumer 3D printers is the
problem of creating simple round pegs and mating holes. A fastener, intended to fit within a
Published by Scholarly Commons, 2015
3
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
cylinder is often found to not mate correctly, and attempts to introduce simple linear correction
factors into slicer software have not been successful (Tmorris9, 2014). The positive-space peg
often exceeds the dimensions specified, and the negative-space hole typically falls short of the
guiding dimension. The two will not mate. Indeed, the use of non-3D-printed fasteners such as
screws and bolts is often complicated by this failure to achieve an accurately dimensioned hole
size. It should be understood that conventional finishing/machining techniques, such as drilling
and sanding, are not always appropriate for a hollow-core, fused-layer part. Holes added after the
fact can sometimes breach the component outer envelope, leaving the less-rigid inner honeycomb
– a standard weight/cost savings feature – exposed. Also, sanding fused layers of plastic does not
yield the same results with the same ease as can be expected with sanding a monolithic material.
Lack of predictability with FDM prints poses a significant challenge.
The study described in this paper investigated the various factors that affect the use of PLA
material with FDM, in terms of dimensional precision, to provide a quantitative guide for
optimizing results of consumer grade 3D printing for use as an engineering tool. Moreover, many
of the factors affecting print quality listed in the public literature do not appear to be based in a
mechanical understanding of how the printer operates, but rather, inconclusive anecdotal evidence.
I hypothesize that, once all confounding variables are accounted for, some of the oft-repeated
causes of dimensional variation – print speed, shell thickness, and print size – will be shown not
to be statistically significant contributors.
Method
3D printers have different make-up and functionality from one to the next. Consumer 3D
printers, however, have some similarities in how they are constructed (France, 2013). They
generally rely on some form of stepper-motor/belt drive system to move the print head in the x-y
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
4
Hernandez: Dimensional Precision 3D Printing
axis, as material is expelled, forming the particular shape desired. Another stepper-motor, with
some form of direct drive gear or threaded rod serves to raise and lower the print bed, creating the
desired layering of material. Another motor with a tensioning/drive system is used to move raw
material – in the form of a plastic filament – through the hot-end, which heats the material to a
liquid state, before it is expelled through a precisely machined nozzle. In such a system, there are
both theoretical device limitations on precision – motor step size, nozzle width, hot-end
temperature sensor accuracy – as well as practical expectations for variation. Loose belts,
inaccurate filament sizing, and nozzle clogging can all contribute to variation. It has even been
reported that something as seemingly innocuous as the size and surface characteristics of a clear
plastic Bowden tube in which the raw material travels before being liquified or the physical
location of the material spool relative to the printer can introduce filament friction and flexing,
resulting in print inaccuracies (Gr5, 2014, June 5; & Illuminarti, 2014, June 5). Inherent
technological limits and the associated tradeoffs are what I sought to investigate.
There are many intangible factors, however, which influence how a final printed-object
dimension deviates from the original CAD definition. The software used to define the object to be
printed, for instance, can contribute error (Creastudiostore, 2014). Slicer software, which is used
to convert the desired object into a layer-by-layer instruction set defining the toolpath which the
printer follows, can generate different results from the same initial specifications. Similarly,
default printer settings can fail to take into account the unique situations under which the print is
being executed. I attempted to research all of these influences and to remove those which
represented controllable variation in order to obtain a quantitative estimate of the dimensional
precision of FDM technology.
Published by Scholarly Commons, 2015
5
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
Equipment and the Printing Toolchain
The printer used in this study was an Ultimaker 2 (Ultimaker B.V., 2015a), with PLA
filament (nominal diameter, 2.85 mm). The printer is capable of manufacturing objects with a build
volume of 223 x 223 x 205 mm, using a layer resolution of as high as 0.06 mm. It is capable of
laying down material as fast as 300 mm/s, and it makes use of a heated print bed. Many features
of the printer – from the precise width of the filament to the tension under which that filament is
pushed through the system to the temperature of the hot-end/nozzle to the precise zero height of
the center, left, and right sides of the bed – are user-adjustable. This makes for a significant number
of variables to consider in optimizing output, before considering the slicer software specifically
designed to work with the printer, named Cura. The Cura interface allows users to preview the
simulated results of a print, layer-by-layer, as well as providing over 50 user-definable parameters
that control the print. Many of these are not categorical parameters, but require specific decimal
values. In the absence of fully understanding how many of these parameters affect the print, the
new user is advised simply to accept the default settings (Ultimaker B.V., 2015b). I discovered
through analysis, however, that some of these settings can directly introduce dimensional
inaccuracy, depending on the nature of the print. These settings are not unique to this particular
software/printer combination.
The CAD software used to develop the test articles for this study was CATIA V5R21.
CATIA is a feature-based, fully associative CAD application (Dassault Systemes, 2013). This
means that all models are generated with an inherent framework for subsequent applications in
engineering analysis or machining. The method of creating the model is stored within the model
itself. As model features are updated by the user, the associative nature of the application
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
6
Hernandez: Dimensional Precision 3D Printing
correspondingly updates the steps for creating the model, so that this information is always
represented and accurate.
With 3D printing, it is currently standard practice to use a Standard Tessellation Language,
(.stl) file to store model information. As explained by Burns (1993/2015), “An StL
(‘StereoLithography’) file is a triangular representation of a 3-dimensional surface geometry. The
surface is tessellated or broken down logically into a series of small triangles (facets). Each facet
is described by a perpendicular direction and three points representing the vertices (corners) of the
triangle” (para. 3). Because the standardized format of an .stl file contains basic information about
the envelope of the model and its shape, but not the way it is built, there is a conversion process
that takes place through the printing toolchain. In the case of the Ultimaker 2, the CATIA model,
with precision as accurate as the user desires – the default being six decimal places in whichever
units the user prefers - is first converted to an .stl file which can have variable levels of precision.
This .stl file is then imported into CURA, where it is manipulated into a toolpath specific to the
Ultimaker printer – literally cartesian x, y, z coordinates for the print head to move, along with
extrusion commands which tell the printer how much filament to unspool into the hot-end. These
instructions are formatted in a standardized g-code file which is used by machine tools around the
world. The opportunity for errors and inaccuracy arises each time a conversion takes place.
With so many different parameters affecting prints, it is impractical to conduct a
quantitative study without limiting the scope of our analysis. After surveying several previous
studies on 3D printing quality (Akande, 2015; Bakar, et al., 2010; Luzanin, et al., 2013; Udroiu &
Mihail, 2009), I compiled a list of the key factors which those authors had concluded had the most
significant effect on the quality and dimensional accuracy of prints. I added to that list parameters
that my knowledge of the construction of the printer led me to believe could also affect the print.
Published by Scholarly Commons, 2015
7
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
The factors initially considered, in no particular order, are listed here: printing multiple model
envelopes on the same bed at one time; stl file tessellation precision; desired dimensions
incompatible with layer height or wall thickness settings; printing speed (which can be subdivided
into several different types of movement: print head travel speed between extrusions, bottom layer
speed, infill speed, top layer speed, outer shell speed, inner shell speed); extrusion rate adjustments
(or printer material thickness settings) and maximum speed of deposition; differences between
slicers’ toolpath algorithms; rate of change of acceleration (jerk) required by the print, which is a
function of shape; layer thickness; print fill density; overall size of the print; cooling fan interaction
with the print; and shell thickness.
Ruling Out Print Error and Inherent Toolchain Inaccuracy
Through trial and error, in setting up the experimental study, several direct contributors to
dimensional inaccuracy were identified and effectively removed from influencing the final data
set. This also allowed for a reduction in the number of factors considered as part of the quantitative
study. Each of those potential influences which were mitigated are addressed in this section.
The bottom layer of a 3D print, which directly contacts the print bed and serves as the base
for the layers above is an important contributor to dimensional inaccuracy that can be resolved. If
the bottom layer does not appropriately adhere to the print bed, the cooling process and material
shrinkage in layers above it will tend to pull at the smallest features, with the least surface area in
contact with the bed, causing sections of the print to warp. Sharp corners at the bottom of the print
pose a particular problem. I found that a layer of kapton tape or blue painter’s tape on the print
bed, together with the heated bed of the Ultimaker 2 and chamfering the sharp corners of areas in
contact with the bed made for a print that did not warp and was, moreover, so well attached as to
make removal difficult. Typically, the surface in contact with the print bed is the most uniform,
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
8
Hernandez: Dimensional Precision 3D Printing
not exhibiting the ridges associated with z-axis movement and forming a glazed surface due to
contact with the heated bed. I did find, however, that there was significant deformation and
inaccuracy in the first few layers of my prints, irrespective of the type of material used or the bed
temperature selected. There was an apparent melting of the material on the bed, such that it always
exceeded the outer desired envelope by approximately 0.5 mm, as illustrated in Figure 1.
. . . . .
. . . . .
. . . object
. .
Printed
. . . . .
. . . . .
Edge fillet
. . . . .
. . . . .
. . . object
. .
Printed
. . . . .
. . . . . .
Dimensional
innaccuracy
Printer bed
Printer bed
EXPECTED SHAPE
OBSERVED SHAPE
Figure 1. Graphical illustration of printer bottom layer inaccuracy.
This observation led to an analysis of the g-code file itself, deconstructing the x, y, z
positional commands and the extrusion requests. I found that Cura’s “advanced” settings allow for
definition of an initial layer thickness and extrusion percentage. The print nozzle is raised from the
surface of the bed by an amount greater than the standard layer height and uses greater-than-typical
material flow, allowing for non-uniformity in the bed and also to ensure proper adhesion. By
putting a simple circle and square shape through the slicer and analyzing the resultant g-code, it
was clear that with user changes to these parameters, the cartesian coordinates are changed to
compensate. An increased line width in the initial layer results in greater extrusion and in moving
the positional data for the print head to compensate for the increased thickness of the initial lines,
irrespective of the intended shape, which is cut-off at that initial layer height. It became obvious
Published by Scholarly Commons, 2015
9
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
that what seemed like a problem with material cooling and deformation had little to do with the
material. The printer setting for filament material thickness, which is used to ensure a consistent
extrusion rate, was actually being overridden by this additional factor in the Cura software, as was
the intended filleted shape.
Even with optimal leveling of the print bed – a critical component of quality prints –
minor variations in the bed and resulting differences in nozzle height from the bed for the initial
layer will affect the width of the material flowing from the nozzle. Raise the nozzle higher, and
the printed line becomes thinner. Press the nozzle closer to the surface, and the width of lines
expands. It was found by trial and error that the extrusion settings needed for the initial layer could
differ from subsequent layers by as much as a factor of two, in creating the same printed line width.
While trial and error with the Cura settings can alleviate this dimensional inaccuracy, creating a
raft for all the test articles – a solid piece of printed material on which the actual desired print rests,
raised off the bed – ensured that this particular source of dimensional inaccuracy did not play a
role in the study measurements.
In analyzing the coordinate information in the g-code provided to the printer, with the
simple circle and square shapes, I also ruled out variation between the CAD dimensions and the
printer toolpath. Exporting from CAD to an .stl file involves a process of tessellation – representing
the shape using triangles - which can have varying precision. I found that with CATIA, there is a
setting for “3D Accuracy” which can be set between 0.01 and 10.00. This value is hidden to all
but the most astute user, and it does not affect either the displayed CAD model or any of the
standard file formats. Even saving a .stl file will not reveal any variation due to this parameter. If,
however, we save an .stl file and then import the file back into CATIA, we can see a marked effect
from these parameter settings, as shown in Figure 2. The effect of this differing tessellation
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
10
Hernandez: Dimensional Precision 3D Printing
precision would be to create approximations of curves with differing levels of precision, resulting
in measured dimensional inaccuracy. Consider the case of an octagon representing a circle.
Depending on where a measurement of diameter is taken, the result will vary from that of a uniform
circle. Setting “3D Accuracy” to 0.01 ensured that tessellation inaccuracy did not play a role in the
study.
Early prints showed that an additional source of dimensional variation was rooted in the
physical dimensions of the printer and the discrete amounts of material that it is designed to
extrude. Creating an object with a z-axis dimension that is not a multiple of the selected layer
height can introduce dimensional error. Analysis of the g-code output from Cura demonstrates that
the z-axis position of the print head does not match the desired dimensions when they are not a
multiple of the selected layer height. The software does not attempt to extrude fractional layers in
order to maintain accuracy. Similarly, the nozzle width of 0.4 mm limits the dimension of our
horizontal shell thickness. These width and height constraints prevent the extrusion of excess PLA
material which can create dimensional errors – attempting, for instance, to print two lines in the
same physical space. The constraints create dimensional variation of their own, however, by not
printing precisely what the user requests. The user needs to understand the limitations of the
printer. The smallest layer height for deposition on the Ultimaker 2 is 0.06 mm, and all prints for
this study were created with this layering resolution. Because the z-axis mechanism of the printer
differs from the x and y axes, and because the z-resolution is smaller by nearly a factor of ten than
the 0.4 mm horizontal line thickness, and because the z-axis movement is relatively slow and
unidirectional (one step per printed layer) I focus exclusively on x and y dimensions in this study.
Optimization in the z dimension is left as an exercise for a future investigation.
Published by Scholarly Commons, 2015
11
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
Figure 2. Isometric (left) and top orthographic (right) views of half sphere exported to .stl file
with different tessellation settings (top to bottom image variation). Dimensional accuracy is
compromised by the different quantity of lines used to represent the shape, resulting in an
increasingly imperfect approximation of the curved surface.
Extrusion rate and its relationship to the speed of the print head – literally, the volume of
plastic material which the hot-end is able to deposit in order to keep up with the movements of the
print head – is listed in the public literature as a factor in print quality. Underextrusion – the failure
to deliver sufficient material to maintain printing – can be detrimental to quality and accuracy, but
it often prevents completion of prints at all. Figure 3 demonstrates a print with an infill setting of
20%, which was suffering from underextrusion.
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
12
Hernandez: Dimensional Precision 3D Printing
Figure 3. The differences between the image at left, exhibiting poor, frayed extrusion
characteristics and the image at right, with crisp infill, are related merely to an increase in infill
percentage. The right image, though printed at a denser infill rate, used a pattern which exhibited
less jerking motion, allowing for more consistent adhesion between layers, resulting in a cleaner
print.
Counterintuitive as it may seem, increasing from 20 to 25% infill – actually increasing the
amount of material the printer must deposit – can help alleviate extrusion problems. In the case of
Figure 3, the increase to 25% resulted in the printer using a different printing pattern. Rather than
attempting to make several ninety-degree turns to create a cross-hatched pattern, box-by-box, the
printer laid down straight lines across the full length of the print. By alternating the direction of
these layers of parallel lines, the printer created the same cross hatch pattern with minimal
extrusion problems. Extrusion issues stem from a combination of print head motion and inability
to deposit material quickly enough. The tell-tale frayed lines visible in the left image of Figure 3
are all but eradicated from the right image of the same figure.
It should be understood that extrusion rate is generally considered as a threshold factor.
Below this threshold – a certain volume of material per unit time – extrusion rate should not pose
an issue to quality (Illuminarti, 2014, Jan 17), as demonstrated in Figure 4. For most functional
prints, the motion and jerk exhibited by the print head is not within user control. It is best simply
to leave a wide margin for error, such that the maximum extrusion capability of the printer is not
Published by Scholarly Commons, 2015
13
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
exceeded. This margin is sometimes overlooked, because extrusion rate margin results in longer
duration prints.
Figure 4. Author’s print of Illuminarti extrusion test – front (left) and top (right) views –
demonstrating that as extrusion speed increases, the printer has difficulty maintaining the
outflow sufficient to fuse to the lower adjacent layer while maintaining circular motion. Past a
threshold of ~8 mm^3/second, the motion of the printer hot-end pulls the plastic away from the
intended circular path. Above ~5 mm^3/second light ca be seen shining between some poorly
deposited layers.
For purposes of this study, it was possible to optimize the repetitive printing of test articles
such that jerk is not believed to have played a significant role. A single instance of apparent error
due to jerk, as shown in Figure 5, could be accounted for and ruled out by changing the way infill
was applied. Cura maintains settings for “infill overlap %” – how far over the innermost line of
the print shell or outer print envelope the infill is allowed to overlap. It should be noted that it may
be possible for future implementations of slicer software to account for jerk, in creating the printer
toolpath, and to limit movements such that they do not detrimentally affect print quality.
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
14
Hernandez: Dimensional Precision 3D Printing
Tight change in direction visibly
flings material rather than typical
deposition edge of print
Figure 5 – Evidence of the potential detrimental effect of too high print head jerk. Because of
how infill cross hatch pattern is positioned relative to the envelope of the print towards the
bottom of the image, it necessitates a significant change in acceleration forces, causing material
to build up on the edge.
From Table 2, the highest speed attained during extrusion maneuvers – actual deposition
of material – was 50 mm/sec. Since the printer nozzle diameter is 0.4 mm, the area of the nozzle
is Pi*(diameter/2)2 = 0.126 mm2. Multiplying area of the nozzle by the speed of the nozzle
movement yields a maximal extrusion rate of 6.28 mm3/sec, which is within the capabilities of the
Ultimaker 2, as demonstrated in Figure 4.
Finally, certain types of prints will necessitate lifting and moving the print head after
printing the current layer or feature. During these moves, it is possible that additional material is
released or dropped unintentionally in a region of the print. In order to prevent this, Cura allows
users to make use of a feature called retraction. From France 2013,
Retraction… will greatly improve the quality of your prints. By retracting the hot
filament with the extruder motor during travel moves, plastic oozing is prevented. The
length of filament to retract before moving to the next extrusion path will depend wildly
Published by Scholarly Commons, 2015
15
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
on the motor and gearing you have… Extra length on restart is the length of plastic you’d
like to extrude after traveling to a new path and prior to moving again… The only
application for this may be when your extruder has serious problems starting up again
after retraction… (pp 42-43).
In order to prevent the dimensional errors that could occur because of the retraction of
filament and the need to restart between layers – re-priming the nozzle – this feature was disabled
for the test articles, in favor of combing. From Ultimaker B.V. (2015c): “Combing is the act of
avoiding holes in the print for the head to travel over. If combing is ‘Off’ the printer head moves
straight from the start point to the end point and it will always retract.” Retraction and the
associated nozzle re-priming, then, was removed as a source of dimensional imprecision.
Design of Experiments
In order to evaluate the effect of the factors selected as well as any interaction effects
between these factors, it was decided that a DOE approach would be conducted with two levels
associated with each factor. In creating physical test articles, we are limited by the printer size, by
material expenditures, and by the time duration required to print the test articles. With the
Ultimaker’s performance specifications, some simple 5x5x5 inch prints can take over thirty hours
to print at the fastest settings. Also, occasional failed prints can necessitate a restart, wasting that
significant print time. Size and speed settings were selected such that the longest test article print
required ten hours.
Procedure
A full factorial DOE approach was selected with replication factor of sixteen – creating
more than one sample data point in each treatment group - in order to maximize the information
that could be extracted about interaction effects and to improve the accuracy in each selected
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
16
Hernandez: Dimensional Precision 3D Printing
treatment group. Using a full factorial design, it is a relatively simple matter to analyze main effects
and interactions effects. The balanced nature of a full factorial design is such that for any factor
setting, there are an equal number of data points for all settings of all other factors, such that the
effects of individual factors can be isolated by averaging over all other effects. The same holds for
interactions between factors.
Hypothesizing that there may be some inherent error in the way the printer prints different
shapes, positive versus negative space, or multiple components in a single print, it was appropriate
to create test articles that yielded samples of different types of physical configurations. While even
two samples from the smallest print exhibited reasonable physical separation and variation, it
should be noted that all samples are not strictly independent. Measuring in multiple places on the
same print opens up the possibility of some correlation between samples. At ten hours per print,
with N = 128, printing each sample independently was impractical. If this study reveals statistically
significant factors, a further decomposition of those particular factors with greater sample
collection can be performed as part of a more targeted future study. Each test article was designed
to yield sixteen measurement samples.
Test articles were printed in a random sequence to limit the effect of extraneous factors.
With ANOVA testing, it is customary to calculate the power of the study as a method of selecting
sample size. For a full factorial DOE, however, it is only possible to estimate power a priori
(Ruttimann & Wegener, 2015). With multiple measurements taken from different areas of each
test article in this study, there is added complexity in any power calculation. Since the practical
capabilities to print more test articles was a limiting factor, I considered the Mee, R.W. (2009)
recommendation of ten samples per treatment group in order to account for within run variation,
and I decided upon sixteen observations per treatment group.
Published by Scholarly Commons, 2015
17
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
Figure 6. Isometric view of large square-shaped test article with measurement point locations for
outer square and thickness of gap to adjacent rectangle
Figure 7. Isometric view of large square-shaped test article with measurement point locations for
inner square and thickness of adjacent rectangle
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
18
Hernandez: Dimensional Precision 3D Printing
Figure 8. Isometric view of large circular-shaped test article with measurement point locations
for outer circle and thickness of gap to adjacent arc
Figure 9. Isometric view of large circular-shaped test article with measurement point locations
for inner circle and thickness of adjacent arc
Since my approach is not, in point of fact, seeking to find a statistically significant effect
as an academic exercise, but rather to optimize our use of a practical technology, the concern that
this study might misclassify a factor which could have otherwise been viewed as statistically
Published by Scholarly Commons, 2015
19
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
significant (Type II error) is mitigated by calculation of p-values for each factor and the
interactions between factors. Despite my use of the standard alpha = 0.05, it is possible for us to
note factors that show a relatively low probability of occurring randomly and, as mentioned
previously, future studies can make use of larger numbers of samples targeted to the particular
factor in question to increase power. This approach allowed me to maintain sample sizes that are
practical.
Table 1
Relevant Specifications for Mitutoyo CD-6" CS Digital Caliper
Resolution
Accuracy
Repeatability
0.01 mm
+/- 0.02 mm
0.01 mm
Dimensional data was obtained by direct measurement with an m-type vernier digital
caliper – model Mitutoyo CD-6” CS with key specifications as listed in Table 1. In order to
minimize measurement error in the N = 128 data points, a consistent procedure was used with only
one operator capturing all data points, consistent with the best practices for accurate, repeatable
measurement as outlined in Juran (2010).
Table 2 lists the various levels of factor settings used in my DOE analysis.
Since the dimensions the printer is expected to reproduce vary in scale, and with any known
fixed bias removed from influencing data, it was appropriate to work with dimensional deviation
as a percentage of the expected original dimension rather than simply a fixed dimension. The
measured dimension on the test article, the fixed deviation from expected original, and the
percentage error (positive for enlargement or negative for reduction in sizes) were captured for
each of the N = 128 samples. From this point forward, however, this paper will address the
percentage dimensional error values, unless otherwise noted.
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
20
Hernandez: Dimensional Precision 3D Printing
Table 2
Low and High Settings for Test Article Printing
Speed
Cura Descriptor
Low (0)
Print Speed
12.5
Travel Speed
Bottom layer speed
25
5
High (1)
50 (400% speed of
low test article)
100
20
Units
Infill speed
10
40
Top/bottom speed
Outer shell speed
Inner shell speed
3.75
7.5
7.5
15
30
30
mm
mm/s
mm/s
mm/s
mm/s, Cura
calculates the ratio
of this to “Print
Speed” to set infill
speeds (Illuminarti,
2013)
mm/s
mm/s
mm/s
Size
N/A
40 x 20
130 x 65 (325%
size of low test
article in each
dimension)
Shell
Shell Thickness
0.4
1.6
mm
Additional
Relevant
Settings
Layer Height
0.06
0.06
mm
0.12
0.12
mm
25
All
0.5
100
15
25
All
0.5
100
15
%
N/A
mm
%
%
Top/Bottom
Thickness
Fill Density
Enable Combing
Fan full on at height
Fan speed
Infill overlap
Published by Scholarly Commons, 2015
21
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
Figure 10 is a graphical representation of the factor level settings for each of the eight
experimental treatment groups.
{101}
Factor Settings:
{Speed Size Shell}
{111}
5
7
{100}
{110}
4
6
{011}
Speed
{001}
1
3
{000}
{010}
Size
2 of speed,
Figure 10. Graphical representation of a two-level, full 0factorial design with factors
size, and shell thickness. Numbers in circles are base ten representations of the binary number
formed from the concatenated factor level selections; arrows show the direction of increase of
the factors.
Results
Square Test Articles
The square test articles were printed and corresponding data collected. Using the full
factorial data, it is possible to construct plots of main effects, provided in Figure 11. The main
effects demonstrate the trends in the mean percentage dimensional deviation data as the factors of
print speed, shell thickness, and test article size were varied. These trend lines represent the effects
of the factors in isolation.
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
22
Hernandez: Dimensional Precision 3D Printing
Figure 11. Main effects plots for square test article.
With the results of the design of experiments revealed, an ANOVA analysis can further
elucidate which respective main and interaction effects are statistically significant. This analysis
begins with testing the necessary assumptions and conditions that make the analysis valid. GraceMartin (2012, May) provided guidance on the appropriate testing of assumptions for experiments
similar to my full factorial DOE with eight treatment groups. For the square test articles, a
Kolmogorov-Smirnov test was applied to the percentage dimensional error data. Across all N data
points for the square test articles, the mean percentage error was -0.387% (SD = 0.559). Calculating
Published by Scholarly Commons, 2015
23
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
the z scores within each of the eight treatment groups – z = (X – Mi) / si , where the subscript i
denotes a statistic for the treatment group i – the eight resultant z vectors were each subjected to
the Kolmogorov-Smirnov test. The hypothesis is that the vector, z, comes from a standard normal
distribution at the 5% significance level. The test validated my hypothesis across all treatment
groups.
Figure 12. Interaction effects plots for square test article.
Because the data set is relatively large, by the central limit theorem, the data is sufficiently
normal to apply further statistical analysis with the assumption of normality, but the empirical
testing above was performed for completeness. Figure 13 is a plots of the empirical cumulative
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
24
Hernandez: Dimensional Precision 3D Printing
distribution function obtained from the N = 128 data points, using the overall mean and standard
deviation of the samples to normalize and provide a visual comparison to the standard normal
curve.
1
0.9
Relative Cumulative Frequency
0.8
0.7
0.6
0.5
0.4
0.3
0.2
Empirical CDF
Standard Normal CDF
0.1
0
-4
-3
-2
-1
z
0
1
2
Figure 13. Plot of the overall empirical cumulative distribution function (CDF) of z-values for
all square test article percentage deviation sample data (all treatment groups combined) from the
square test article versus a standard normal CDF for visual comparison of normality.
The ANOVA analysis (results in Table 3) revealed that none of the main or interaction
effects observed were statistically significant. The p-value closest to the selected alpha = 0.05
corresponded to the size factor, with an approximately 10% probability that the observed
differences between treatment groups were due to random chance.
Published by Scholarly Commons, 2015
25
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
Table 3
Three-Way Analysis of Variance of Factors Affecting Dimensional Percentage Deviation in
Square Test Articles
Source
df
SS
MS
F
p
speed
size
shell
speed * size
speed * shell
1
1
1
1
1
0.00002
0.00015
0.00002
0.00005
0.00005
0.00002
0.00015
0.00002
0.00005
0.00005
0.30
2.64
0.38
0.85
0.82
0.584
0.107
0.541
0.359
0.367
size * shell
Error
Total
1
121
127
0.00001 0.00001
0.00694 0.00006
0.00724
0.11
0.743
Circular Test Articles
The circular test articles were printed and corresponding data collected. Using the full factorial
data, it is possible to construct plots of main effects, provided in Figure 14. Across all N data points
for the circular test articles, the mean percentage error was -0.573% (SD = 0.607).
Figure 14. Main effects plots for circular test article.
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
26
Hernandez: Dimensional Precision 3D Printing
Figure 15. Interaction effects plots for circular test article.
For the circular test articles, a Kolmogorov-Smirnov test was applied to the percentage
dimensional error data. The eight treatment groups were each subjected to the KolmogorovSmirnov test at the 5% significance level. The test validated the hypothesis of a normal distribution
across all treatment groups. Figure 16 contains a plots of the empirical cumulative distribution
function obtained from the N = 128 data points, using the overall mean and standard deviation of
the samples to normalize and provide a visual comparison to the standard normal curve.
Published by Scholarly Commons, 2015
27
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
1
Empirical CDF
Standard Normal CDF
0.9
Relative Cumulative Frequency
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-6
-5
-4
-3
-2
z
-1
0
1
2
Figure 16. Plot of the overall empirical cumulative distribution function (CDF) of z-values for
all circular test article percentage deviation sample data (all treatment groups combined) from
the square test article versus a standard normal CDF for visual comparison of normality.
The ANOVA analysis (results in Table 4) revealed that only one of the main or interaction
effects observed was statistically significant. The p-value for the size factor, showed an
approximately 0.9% probability that the observed differences between treatment groups were due
to random chance, well below the chosen 5% significance level.
Table 4
Three-Way Analysis of Variance of Factors Affecting Dimensional Percentage Deviation in
Circular Test Articles
Source
df
speed
size
1
1
shell
speed * size
speed * shell
size * shell
Error
Total
1
1
1
1
121
127
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
SS
MS
F
p
0.00029 0.00029
0.00056 0.00056
3.59
6.97
0.061
0.009
0.00008
0.00016
0.00001
0.00003
0.00968
0.01081
0.96
2.04
0.13
0.41
0.330
0.156
0.719
0.524
0.00008
0.00016
0.00001
0.00003
0.00008
28
Hernandez: Dimensional Precision 3D Printing
Because multiple types of data were collected for each test article, it was deemed
potentially instructive to chart the data, categorized by type, considering the differently sized test
articles, the different axes from which the dimensional data was obtained, and whether the
dimension was from an inner, outer, gap, or thickness location, as defined in Figures 6, 7, 8, and
9. The summary chart of averages is shown in Figure 17.
Figure 17. Mean percentage deviation for specific categories of dimension types.
It is apparent from the average percentage dimensional deviation data that the thicknesstype data exhibits a greater error than other dimensions. A relatively small sample set can be
expected to result in studies with low power, where it may be possible to erroneously fail to
detect a statistically significant difference. Although an ANOVA analysis is unreliable with
Published by Scholarly Commons, 2015
29
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
small numbers of data points in each treatment group – in this case only four data points per
group – it was deemed potentially instructive to perform a design of experiments ANOVA
analysis on the square articles’ thickness data, alone, in order to search for patterns that might
lead to future studies performed with greater samples obtained for these specific, targeted data
types. The results appear in Table 5.
Table 5
Three-Way Analysis of Variance of Factors Affecting Percentage Deviation in the "Thickness"
Dimension of Square Test Articles
Source
speed
size
shell
speed * size
speed * shell
size * shell
Error
Total
df
1
1
1
1
1
1
25
31
SS
0.00004
0.00009
0.00000
0.00001
0.00006
0.00001
0.00095
0.00115
MS
0.00004
0.00009
0.00000
0.00001
0.00006
0.00001
0.00004
F
0.94
2.28
0.00
0.16
1.68
0.16
p
0.341
0.144
0.958
0.688
0.206
0.688
For the most significant interaction effect – print speed interacting with shell thickness – the
confidence interval was calculated and the plot appears in Figure 18.
Discussion
The mean of results for all factors with the square test articles were centered around a
dimensional deviation of approximately -0.387%, a value which is comparable to many typical
techniques for manufacturing mechanical components. By way of example, using subtractive
manufacturing techniques such as milling and drilling to achieve a hole similar in size to the 32.5
mm inner dimension of the test articles used in this study, Sandvik AB (2011) lists the achievable
precision grade as IT9/10, with 33 mm as the outer limit for these particular grades. IT10 would
corresponds with a dimensional tolerance of +/- .219% (Coban Engineering, 2015).
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
30
Hernandez: Dimensional Precision 3D Printing
Figure 18. 95% confidence interval plots for square test article thickness data, demonstrating a
statistically insignificant but notable difference when printing at high speed with thickest shell
settings. Circles represent estimated population marginal mean values. Abscissa numeric values
represent fractional dimensional deviation.
From the standpoint of creating components with repeatability and precision, addressing
the several factors described in the method section of this study – low tessellation precision,
improper slicer settings, bottom layer inaccuracy, extrusion/flow miscalibration, exceeding printer
extrusion rate capabilities, improper bed leveling, print warpage due to poor adhesion, selection of
dimensions which are not multiples of the print nozzle width or the selected layer height, infill
settings which result in excessive jerk at edges of the print, and retraction errors due to re-priming
the nozzle – appears to have removed a significant fraction of observed print dimensional
Published by Scholarly Commons, 2015
31
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
imprecision. None of the test article measurements exhibited an average error exceeding one
percent, except the thickness-type dimensions.
The DOE analysis for the square test article reveals some interesting trends with respect to
main effects (Figure 11). Thicker outer shells, smaller sizes, and slower speed all appear to
decrease dimensional precision. The differences are not, however, dramatic and, indeed, were
shown by the subsequent ANOVA analysis (Table 3) to be statistically insignificant. The observed
trends are still instructive, however, particularly when looking at the interaction effects shown in
Figure 12. Increasing speed to increase precision might appear counterintuitive, but the interaction
plots between speed and shell thickness and between size and speed, respectively, reveal that the
increased print speed provides greater precision when printing with thicker outer shells or smaller
objects. When the test article wall thickness was narrow or the object was relatively large, greater
precision was achieved with a slower print speed. It would appear that with greater material
deposited in close proximity – either to build up a shell or because the object is physically small –
the lingering of the print head may cause melting or expansion of previously deposited material.
It may also be possible that an inability to reliably repeat positioning upon subsequent passes
results in print imprecision, whereas a quicker traversal somehow minimizes the effect. This is a
consideration for a future study.
In addition, because a print head passing through an arc naturally deposits greater material
at the inside of the arc versus the outside, there is an inherent inaccuracy that comes with printing
arcs as opposed to straight lines. This inaccuracy becomes more pronounced with smaller objects,
as the excess material deposited represents a greater fraction of the object dimension. Slicer
software manufacturers and individual users have attempted to introduce arc compensation
(Alexrj, 2013) without success. Other sources cite material properties as being the source of error
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
32
Hernandez: Dimensional Precision 3D Printing
in hole dimensions. Hodgson (n.d.) writes, “Plastic shrinks when cooling. Different kinds of plastic
exhibit different shrinkage, which might also depend on temperature. Because of such shrinkage,
circular (or polygonal) holes laid by the extruder at the nominal diameter will end up smaller after
cooling” (para. 4). For these reasons, the ANOVA analysis results demonstrating a statistically
significant factor in size for the circular test articles can be interpreted not as an inherent inability
to print small circles or arcs but as a type of printing for which trial and error are required in order
to compensate, until a more encompassing model is developed. Again, this is an area for future
study.
Figure 17 shows that the greatest differences between circular and square test articles
occurs for smaller test articles. Beyond those differences, the two types of shapes do not
demonstrate an appreciable difference other than what can be expected due to random variation.
Because of the relatively precise values of dimensional deviation revealed in the study, small
contributors of inaccuracy may be considered to factor into the quantities observed. Measurement
error is a contributor worthy of note. In particular it was appreciably more difficult to measure the
small circular test article versus the small square. It was difficult, due to the nature of the test article
size and the adjacent gap/thickness component, to obtain consistent positional contact with the
small circular test article. As the items are plastic, and hence pliable, the reduced surface area for
the small circles to contact the caliper also translates into the pressure applied by the caliper
possibly flexing the smaller articles. This may well be a significant contributor to the appreciably
poorer average dimensional deviation exhibited in Figure 17 for small circular articles. Indeed, in
looking at the interaction effects, it appears that the smaller and thinner circular test articles (Figure
15) deviate quantitatively and trend-wise from their square counterparts (Figure 12), but the larger,
thicker articles do not.
Published by Scholarly Commons, 2015
33
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
As explained in a National Physical Laboratory guide on good measurement practices
(Flack, 2014),
The use of polymer materials for machine parts is extensive and they require accurate
dimensional measurement. These materials are often soft and the measuring force applied
by ordinary callipers and micrometers can deform them, resulting in inaccurate
measurements. Consequently, this has led to the development of constant force dial
calipers…that allow the measurement of materials that are easily deformed. (pp. 20-21)
Though such constant force calipers were not available for this study, Flack (2014) goes on to
explain that there is an additional source of measurement error when using calipers with circular
test articles:
Inside diameter measurements made with M-type vernier calliper involve measurement
errors that are inherent to the design of the jaws. These errors are more significant when
measuring small holes and result from the measuring face of the jaws being offset from
the centre line of the hole. It is therefore necessary to take these errors into consideration
and make necessary compensations or use another type of instrument if greater accuracy
is required. (pp. 22)
Future study planning should take into consideration the use of different types of
instruments in order to achieve consistent measurement reliability across test articles – from drill
gauges to 3D scanning.
As the overall dimensional precision of the test articles was better than anticipated, and the
observations of interest represented less than one-half of one percent dimensional variation, it was
appropriate to consider additional factors and reevaluate the potential causes of variation with
much greater resolution appropriate to this new context, to pave the way for future studies.
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
34
Hernandez: Dimensional Precision 3D Printing
Focusing on the largest errors in Figure 17, thickness-type dimensions seemed to pose a particular
difficulty for the printer. The ANOVA analysis of the square test article thickness data points –
keeping in mind that there were only four samples per treatment group – seems to indicate that the
factors I have considered in this study are not a significant contributor to that particular type of
error and the greater than 1% bias noted. The best indicator of some discernible performance
difference is shown in the confidence intervals of Figure 18, where a thick shell and high speed
printing do provide an appreciable, if statistically insignificant, decrease in dimensional accuracy.
Considering other possible sources of this imprecision, the following should be noted: The
printer’s ability to maintain the maximal requested extrusion rate – the volume of material we
expect the printer to be able to deposit in a given time – is a more complex consideration at these
small dimensional deviation values. Rather than simply focusing on a threshold value for
maximum extrusion rate, defined by the test demonstrated in Figure 4, the extrusion rate varying
with different features – jerk and temperature variation within the hot-end control loop, for instance
– may influence results. Rather than relying on the printer settings for material thickness in order
to set the nominal extrusion rate, it is recommended that empirical measurement of the extrusion
rate be performed to calibrate all contributions to extrusion in order to remove any bias.
From the Cura slicer software manual (Ultimaker B.V., 2015b):
Infill Speed [6]: The speed at which infill lines are printed. If set to zero then same speed
is used as for the rest of the print. A slight loss in outer quality can be expected if you use
this to print a fast infill due to changes in nozzle pressure when switching between
outside and infill parts.
Published by Scholarly Commons, 2015
35
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
Minimal layer time [7]: The minimal time spend on printing a single layer. If a layer
takes less time to print then this configured time, then the layer is slowed down. This
ensures that a layer is cooled down and solid enough before the next one is put on top.
These somewhat ambiguous statements about nozzle pressure differences and the requirements of
allowing layers to cool down certainly can be taken to imply that other variables not included in
this study and not discussed in the public literature, to date, may be sources of dimensional
variation. The slicer software includes settings for cooling fan speed, and the toolpath, itself,
affects how long the hot print head lingers over a particular area of the print, which, in turn, may
affect deformation.
Z-wobble
Filament Inconsistency
Figure 19. Illustration of z-wobble effect and filament inconsistency effect, respectively, on
measured dimension of fused deposition layered objects.
When all other contributions to dimensional imprecision which can be mitigated have been
accounted for, there remain certain inherent sources of error due to the design of fused deposition
modeling machinery. Linear motion of a print head moving in the x-y direction, followed by
subsequent, dependent layers added in the z-direction is subject to errors in repeatability, where
the print head simply cannot line up perfectly where the user intends with each subsequent layer
(Figure 19). This inherent variation has been categorized by related terminology – z-wobble and
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
36
Hernandez: Dimensional Precision 3D Printing
backlash – and the error may be contributed to by either the linear motion of the print head or the
characteristics of the raw filament material itself (“Taxonomy of Z axis”, n.d.).
As explained by Hodgson (n.d.), “Backlash is a mechanical defect of one or more axes that
basically reduces the amount of actual motion whenever a motor inverts its spinning direction”
(para. 11). Any loose connection between the driving motors and the print head – often caused by
incorrect drive belt tension – can contribute to backlash. Because the x-y motion of the print head
is not precisely what was requested, due to backlash, z-wobble can be expected to occur. As with
many other types of linear motion, it can be expected that the greater the amount of motion in a
particular layer – a function of the intricacy of the part or parts being printed and the slicer
software’s elected toolpath – the more pronounced the random variation that will be observed in
the z-wobble effect.
Similarly, filament diameter is assumed to be uniform throughout an entire roll of raw
material. Adjustments for this filament size are typically fixed settings in the printer. If the filament
exhibits inconsistency across its length, however, this will change the rate of material extrusion
throughout the print duration, resulting in dimensional imprecision. Both print head linearity
effects and filament inconsistency are left for a future study to investigate.
Conclusion
The results of this study have shown that dimensional precision is affected by a variety of
factors, many of which are within the control of the user. While insight was provided into certain
trends in printer performance, it should be noted that the dimensional variation of the test articles
created for this study was so small that the printer precision was comparable to other modern-day
manufacturing techniques. The small biases observed appeared to be consistent. For this reason,
the printing errors I removed from affecting the prints, as outlined in the method section, are of
Published by Scholarly Commons, 2015
37
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
great practical importance. In practice, trial and error could be applied to discover the nature of the
relatively consistent biases in order to remove them from future prints. Anecdotal evidence
suggests that measurement error in this study is likely as prominent a component of variation as
dimensional differences, in the small range of values observed. Other factors, such as arc
compensation, filament inconsistency, printing multiple components simultaneously – both
temperature/proximity effects and repeatability of complex linear motion such as z-wobble and
backlash - are left to a future study.
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
38
Hernandez: Dimensional Precision 3D Printing
References
Akande, S. O. (2015). Dimensional accuracy and surface finish optimization of fused deposition
modeling parts using desirability function analysis. International Journal of Engineering
Research & Technology, 4(4), 196-202.
Alexrj. (2013, Dec 12). Arc Compensation feature request #1613. [Online forum comment].
Retrieved from https://github.com/alexrj/Slic3r/issues/1613
American Society of Mechanical Engineers. (2004). ASME MFC-3M-2004: Measurement of
fluid flow in pipes using orifice, nozzle, and venturi. New York, NY: ASME
International.
Bakar, N. S. A., Alkahari, M. R., & Boejang, H. (2010). Analysis on fused deposition modelling
performance. Journal of Zhejiang University Science A, 11(12), 972-977.
Boots Industries. (n.d.). The Importance of High Quality Filament in 3D Printing. Retrieved
from http://bootsindustries.com/portfolio-item/importance-of-good-filament/
Burns, M. (2015). Automated fabrication. Englewood Cliffs, NJ: Prentice Hall. In Ennex
Corporation. (2015). The StL Format: Standard Data Format for Fabbers. Retrieved from
http://www.fabbers.com/tech/STL_Format. (Original work published 1993)
Coban Engineering. (2015). ANSI limits and fits, ANSI standards. Retrieved from
http://www.cobanengineering.com/Tolerances/ANSILimitsAndFits.asp
Creastudiostore. (2014, October). Can the choice of a Slicer determine the quality of a print?
[Online forum comment]. https://www.3dhubs.com/talk/thread/can-choice-slicerdetermine-quality-print
CustomPartNet. (2009). Additive Fabrication. Retrieved from:
http://www.custompartnet.com/wu/additive-fabrication
Published by Scholarly Commons, 2015
39
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
Dassault Systemes. (2013). CATIA ICEM Class A Surface Design. Retrieved from
http://www.3ds.com/uploads/tx_3dsportfolio/2013-03-05-ica.pdf
Flack, D. (2014) Measurement good practice guide no. 40: callipers and micrometers. [PDF
document]. Middlesex: National Physical Laboratory. Retrieved from:
http://publications.npl.co.uk/npl_web/pdf/mgpg40.pdf
France, A. K. (2013). Make: 3D printing. Sebastopol, CA: Maker Media, Inc.
Grace-Martin, K. (2012, May). Checking the normality assumption for an ANOVA model. [Web
log post]. Retrieved from: http://www.theanalysisfactor.com/checking-normality-anovamodel/
Gr5, (2014, June 5). Reduce filament friction in the Bowden tube? [Online forum comment].
Retrieved from https://ultimaker.com/en/community/view/6454-ultimaker2-reducefilament-friction-in-the-bowden-tube?page=2
Hodgson, G. (n.d.). Slic3r manual. Retrieved from
http://manual.slic3r.org/troubleshooting/dimension-errors
Illuminarti, (2013, December 25). Infill speed and print speed. [Online forum comment].
Retrieved from https://ultimaker.com/en/community/view/4840-infill-speed-and-printspeed
Illuminarti (2014, January 17). Almost always missing layers / underextruding. [Online forum
comment]. Retrieved from https://ultimaker.com/en/community/view/4976-almostalways-missing-layers-underextruding?page=16
Illuminarti (2014, April 25). A tougher extrusion test :-o. [Online forum comment]. Retrieved
from http://umforum.ultimaker.com/index.php?/topic/5436atougherextrusiontesto/
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
40
Hernandez: Dimensional Precision 3D Printing
Illuminarti, (2014, June 5). Reduce filament friction in the Bowden tube? [Online forum
comment]. Retrieved from https://ultimaker.com/en/community/view/6454-ultimaker2reduce-filament-friction-in-the-bowden-tube?page=2
Juran, J. M. & De Feo, J. A. (2010). Juran’s quality handbook, sixth edition. New York:
McGraw-Hill.
Liing Shian Colorant Manufacturer Co., Ltd. (2013). ABS and PLA Comparison Table. Retrieved
from https://www.botfeeder.ca/abs-vs-pla/
Luzanin, O., Movrin, D., & Plancak, M. (2013). Experimental investigation of extrusion speed
and temperature effects on arithmetic mean surface roughness in FDM-built specimens.
Journal for Technology of Plasticity, 38(2), 179-190.
Mee, R.W. (2009). A comprehensive guide to factorial two-level experimentation. New York:
Springer-Verlag.
Mitutoyo America Corporation. (2009). Surface roughness measurement [PDF document].
Retrieved from: http://www.mitutoyo.com/wpcontent/uploads/2012/11/1984_Surf_Roughness_PG.pdf
Roland DGA Corporation. (2011). Subtractive Rapid Prototyping. [White Paper]. Retrieved
from: http://www.rolanddga.com/resources/pdf/srp/SRP_White_Paper_2011-07-27.pdf
Ruttimann, B. G. & Wegener, K. (2015). The Power of DOE: How to Increase Experimental
Design Success and Avoid Pitfalls. Journal of Service Science and Management, 8, 250258.
Published by Scholarly Commons, 2015
41
International Journal of Aviation, Aeronautics, and Aerospace, Vol. 2 [2015], Iss. 4, Art. 2
Sandvik AB. (2011). Small to medium diameter holes. Retrieved from
http://www.sandvik.coromant.com/engb/knowledge/drilling/application_overview/general_drilling/small_to_medium_diameter
_holes/pages/default.aspx
Taxonomy of z axis artifacts in extrusion-based 3d printing. (n.d.). Retrieved from
https://www.evernote.com/shard/s211/sh/701c36c4-ddd5-4669-a482953d8924c71d/1ef992988295487c98c268dcdd2d687e
Tmorris9. (2014, Sep 15). Undersized circles – Has this been solved yet? [Online forum
comment]. Retrieved from http://forums.reprap.org/read.php?262,407139,409002
Udroiu, R., & Mihail, L. A. (2009). Experimental determination of surface roughness of parts
obtained by rapid prototyping. Proceedings of the 8th WSEAS International Conference
on Circuits, Systems, Electronics, Control & Signal Processing: 283-286.
Ultimaker B.V. (2015a). Ultimaker products. https://ultimaker.com/en/products/ultimaker-2family/ultimaker-2
Ultimaker B.V. (2015b) Cura 13.11.2 user manual [PDF document]. Retrieved from:
https://ultimaker.com/download/170/Cura_User-Manual_v1.0.pdf
Ultimaker B.V. (2015c) Cura [software]. Available from
https://software.ultimaker.com/old/Cura_15.02.1.exe
http://commons.erau.edu/ijaaa/vol2/iss4/2
DOI: https://doi.org/10.15394/ijaaa.2015.1085
42
Download PDF
Similar pages