Accepted Manuscript

Title: Rate limits of additive manufacturing by fused filament

fabrication and guidelines for high-throughput system design

Authors: Jamison Go, Scott N. Schiffres, Adam G. Stevens, A.

John Hart

PII: S2214-8604(16)30283-4
DOI: http://dx.doi.org/doi:10.1016/j.addma.2017.03.007
Reference: ADDMA 160

To appear in:

Received date: 27-10-2016

Revised date: 2-3-2017
Accepted date: 13-3-2017

Please cite this article as: Jamison Go, Scott N.Schiffres, Adam G.Stevens, A.John Hart,
Rate limits of additive manufacturing by fused filament fabrication and guidelines for
high-throughput system design (2010), http://dx.doi.org/10.1016/j.addma.2017.03.007

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
Rate Limits of Additive Manufacturing by Fused Filament Fabrication and
Guidelines for High-Throughput System Design
Jamison Go1, Scott N. Schiffres1,2, Adam G. Stevens1, and A. John Hart1*
1
Department of Mechanical Engineering and Laboratory for Manufacturing and Productivity,
Massachusetts Institute of Technology, Cambridge, MA
2
Department of Mechanical Engineering, Binghamton University, Binghamton, NY

*Corresponding author: ajhart@mit.edu

Abstract
While additive manufacturing (AM) advances rapidly towards new materials and
applications, it is vital to understand the performance limits of AM process technologies and to
overcome these limits via improved machine design and process integration. Extrusion-based
AM (i.e., fused filament fabrication, FFF) is compatible with a wide variety of thermoplastic
polymer and composite materials, and can be deployed across a wide range of length scales.
However, the build rate of both desktop and professional FFF systems is comparable (~10s of
cm3/hr at ~0.2 mm layer thickness), suggesting that fundamental aspects of the machine design
and process physics limit system performance. We determine the rate limits to FFF by analysis
of machine modules: the filament extrusion mechanism, the heater and nozzle, and the motion
system. We determine, by direct measurements and numerical analysis, that FFF build rate is
influenced by the coincident module-level limits to traction force exerted on the filament,
conduction heat transfer to the filament core, and gantry velocity for positioning the printhead.
Our findings are validated by direct measurements of build rate versus part complexity using
desktop FFF systems. Last, we study the scaling of the rate limits using finite element
simulations of thermoplastic flow through the extruder. We map the scaling of extrusion force,
polymer exit temperature, and average printhead velocity onto a unifying trade-space of build
rate versus resolution. This approach validates the build rate performance of current FFF
systems, and suggests that significant enhancements in FFF build rate with targeted quality
specifications are possible via mutual improvements to the extrusion and heating mechanism
along with high-speed motion systems.

1
1. Introduction
Improvements in the throughput of additive manufacturing (AM) processes are essential to
realize the vision that AM will become a cost-effective, on-demand method of production.
Towards this vision, the growth of the AM industry is evidenced by the rising sales of AM
equipment, the significance of individual AM-enabled products such as custom orthodontics and
advanced medical implants, and the emergence of large service bureaus that utilize a growing
variety of industrial AM equipment. Notably, in 2015, production of end use parts represented
51% of the worldwide market for AM services (~US$2.7B) [1].
AM service bureaus seek to maximize equipment utilization while maintaining a short lead
time and enabling the customer to choose among a variety of materials that are compatible with
each respective process. One strategy to do so involves maximizing the packing within a
machines build volume such that multiple parts are printed in the same build [2], [3]. The build
rate of each AM machine is also critical to the economics of AM. -Advances in the rate of AM
processes decrease part cost by increasing machine productivity and reducing required capital
investment.
Among the several mainstream techniques, extrusion-based AM, typically called fused
deposition modeling (FDM) or fused filament fabrication (FFF) is arguably the most versatile
and widely used method. In extrusion AM, molten polymer roads are laid to complete
consecutively stacked cross-sections of a three-dimensional model. The material, which is
typically provided as a spooled polymer filament, is fed to the printhead where it is heated,
melted, and forced through a nozzle. The printhead assembly is mounted atop a gantry which
typically moves the extrusion nozzle within the horizontal plane, while the build platform is
moved vertically.
Extrusion AM can operate across a range of length scales. Desktop and industrial AM
equipment typically uses extrusion nozzles with ~0.1-1 mm diameter, yet micron-scale extrusion
of viscous inks has been used to build electrical interconnects and a variety of functional
microstructures [4], and extrusion of ~10 mm beads has been used to fabricate prototype vehicle
structures and composite tooling [5]. The latter processes uses robotic or large gantry
manipulators and pelletized polymer feedstock. With regard to filament extrusion AM, hobbyist
desktop machines can be purchased $500 or even less, while professional systems can exceed
$100,000 [4]. Despite this price disparity, both hobbyist and commercial systems build parts at
approximately the same rate (~10-100 cm3/hr), though professional systems achieve greater
dimensional accuracy and overall quality. PLA and ABS are the most common thermoplastic
materials for use in FFF/FDM, and there is considerable industrial interest in ULTEM,
polycarbonate, PPSF, and fiber-polymer composites [5].
Therefore, it is of interest to understand the machine and process attributes that determine the
build rate of FFF, and identify whether improvements can be made. In this paper we examine
the functional mechanisms of FFF and analyze its rate limiting attributes. First, we study the
main modules of FFF machines: the filament feed, filament heating/extrusion, and gantry
positioning systems. Experiments and modeling are performed for each module, leading to an
understanding of the module-level rate limits. Then, we characterize the trajectory of a
commercial FFF printer, and show that the system-level measurements are consistent with our
module-level analysis. We then assess the tradeoffs between extrusion build rate and resolution,
guided by finite element simulations of the coupling between extrusion force and heat transfer in
the printhead. This enables us to map the predicted rate-limiting performance to the

2
specifications of commercially available thermoplastic extrusion AM systems, and identify the
feasible operating range for FFF machines.

2. FFF machine modules

Desktop and professional extrusion AM systems share three common modules (Figure 1): the
filament feed mechanism, the liquefier, and the motion system. Each module has a build rate
performance characteristic governed by its operating mechanics, and each module must work in
tandem with other modules to ensure proper operation of the FFF process. This section explains
the typical design and operating mechanism of each module.

2.1 Filament Feed

The filament feed module is responsible for supplying thermoplastic filament and applying
force to the filament such that it can be smoothly pushed through the liquefier (heater) and
nozzle. In AM systems, this is achieved using a pinch roller mechanism that comprises a drive
wheel, pinch wheel, and spring arm (Figure 1b). The filament, typically 1.75 mm or 2.85 mm
diameter, is fed tangentially to the drive wheel surface. The pinch roller, attached to the spring
arm, provides a normal force against the filament to maintain traction with the drive wheel such
that the filament advances ideally without slip. Typically, perfect traction between the filament
and the drive roller is assumed [6]; as a result, the failure criterion is set by the rate at which the
pressure drop required by the liquefier/nozzle exceeds the maximum motor torque, or by the
force at which buckling of the unsupported length of the filament occurs [7], [8]. Failure can also
be caused by loss of traction between the drive wheel and filament. In earlier FFF systems,
rollers were made from a synthetic rubber such as hypalon or polyurethane [9], [10]. In present
day systems, the drive wheel is metal and typically knurled or grooved; this creates small notches
in the filament as it passes and therefore failure may occur due to shear rather than slip.
The filament feed unit is typically mounted directly on the gantry but some systems use a
Bowden style mechanism where the feed mechanism is attached to the stationary machine frame
and the filament is guided to the liquefier through a flexible tube. The Bowden configuration is
attractive because it reduces the mass of the moving gantry carriage; however, filament-tube
friction, and hysteresis between the feed mechanism and nozzle can influence the extrusion
dynamics and part quality [11].

2.2 Liquefier and nozzle

The liquefier is responsible for heating and melting the polymer before it reaches the nozzle;
this is achieved by conduction via contact between the moving filament and a metal conduit
whose temperature is maintained using a resistively heated coil embedded in a metal block
(Figure 1b). A temperature sensor placed in the metal block is used for feedback control of the
block temperature. The liquefier inlet diameter ideally matches the filament diameter and the
flow is constricted to a finer diameter by the outlet nozzle which, in combination with the layer
thickness, filament feed rate, and gantry speed, dictates the printing resolution.
In previous work, heat transfer between the liquefier and filament has been modeled as either
a constant heat flux or a constant wall temperature condition [9]. Constant heat flux assumes
uniform filament temperatures at the inlet and nozzle of the liquefier and is therefore applicable
only for low feed rates [12]. The constant wall temperature assumption may be more accurate

3
because it represents the operating conditions of the liquefier (i.e., a feedback controlled
isothermal wall) and recognizes that the filament is heated as it progresses through the liquefier
[13]. Also, significant axial and radial thermal gradients may be expected due to the low thermal
conductivity of the filament compared to the surrounding metal block; failure to reach the melt
temperature at the core of the filament is expected to result in greater force required to reduce the
cross section as the material enters the nozzle.

2.3. Motion System

The motion system positions the printhead and build platform in 3D space. In the most
common FFF machine configuration, the motion system comprises three linear actuators each
responsible for a Cartesian vector (X, Y, Z). Planar (XY) motion of the nozzle is achieved by
converting rotary motion of a DC motor to linear motion using either a belt drive or lead screw
and connecting two such orthogonal axes in series. Vertical (Z) motion of the build surface is
achieved using a linear actuator oriented vertically. In most systems, stepper motors are used,
because of their combination of low cost and open-loop angular position control; angular
resolution can be increased using microstepping with the drawback of decreased torque [14].
Most desktop systems, such as the machines from companies Makerbot, Ultimaker, and Printrbot
(Figure 1b), use a series gantry for X, Y position, and a lead screw actuator for Z positioning.
This positioning architecture is found on professional-grade printers as well, including the
Stratasys Dimension FDM machines. Other parallel motion architectures for FFF systems, such
as the H-frame gantry and the delta architecture are becoming increasingly popular for their
higher agility [15], [16]. These architectures have improved dynamic performance because their
output carriages are moved by multiple stationary actuators simultaneously affording higher
force and decreased inertia at the expense of more complex kinematics and controls. The
Stratasys Mojo printer shown in Figure 1a uses an H-frame gantry for (X, Y) positioning, and a
lead screw actuator for Z positioning. As discussed later, the quality of the motion system and
the toolpath are critical to the quality of printed parts.

3. Module rate limits

To gather an understanding of the mutual requirements and tradeoffs governing system
operation, we next quantified the performance of each machine module using hardware extracted
from desktop FFF machines. We hypothesized that a FFF machine cannot print at a higher rate
than dictated by its rate-limiting module, and that the failure of a single module would cause
failure of the process altogether. For a given build rate (i.e., volume per time), the extruder must
feed the filament at the commanded velocity, while providing sufficient force to overcome the
flow resistance in the liquefier. The liquefier, in turn, must be capable of heating the filament to
the target temperature (e.g., Tmelt), at the specified rate. The motion system must then move the
printhead (e.g., the extruder and liquefier) at a rate commensurate to the exit velocity of material
from the nozzle.

3.1 Extruder force capability

The critical measure of FFF extruder performance is the maximum force that can be exerted
on the moving filament by the feed mechanism. To assess this, we instrumented a pinch-wheel
mechanism, from a Printrbot Metal 3D printer, as shown in Figure 2a. A machined interface

4
plate with an attached strain gauge was fastened to the end of the extruder, and the filament path
was blocked, such that the force exerted by the filament would be measured by the calibrated
strain gauge.
Using this setup, we measured the force versus time over a range of feed rates; each selected
feed rate experiment was performed five times. In each case, the pinch wheel grip failed after the
filament stalled against the plug and the peak force was recorded as the maximum force for a
given feed rate; the error bars indicate the standard error among the five trials performed at each
feed rate. The average maximum force across the entire range of feed rates (Figure 2b) was
approximately 62 N at the nominal feed rate of 0.08 mm/s, and was less at greater feed rates.
Under normal operation, the knurled drive wheel will leave small indentations on the surface of
the thermoplastic filament as it is fed through the extruder. When a critical force is reached, here
simulated by blocking the output with the plug, the extrusion mechanism fails as the textured
drive wheel slips against the filament and erodes its surface (Figure 2c); at this point, the
filament no longer advances, and the mechanism cannot exert significant force. We attribute the
inverse relationship between stall force and feed rate to the depth of engagement of the pinch
wheel with the filament, which is greater at slower rates.
The indentations in the filament suggest that failure occurs when the engaged area reaches a
critical stress causing the filament to fail in shear. This occurs when the supplied force (Fextruder)
exceeds the force supported by the filament shear area (Afs) and shear strength of the material
(s(v)) but is less than the force required for extrusion defined by pressure drop (P) and the
filament cross-sectional area (Af).
> () (1)
() < (2)
Shear strength can be approximated using the distortion energy theory for ductile materials [18],
/3, yet is related to the shear rate (v) to account for the viscoelasticity of
thermoplastics [17].
The filament shear area describes the surface inscribed by the drive wheel surface profile
after it has formed indentations in the material. This surface and the maximum extrusion force, is
related to the indentation depth, filament mechanical properties, and pinch wheel tooth geometry.
We calculate the filament shear area using a 3D model of the interaction generated from optical
microscope measurements of the indentation depth. Using this value (2.8 mm2), and the
estimated filament shear strength (19.0 MPa) [19], we predict that the filament should fail at an
equivalent extrusion force of 53.2 N which compares well to the data at lower rates that are
comparable to those used during printing.

3.2 Heat transfer in the liquefier and nozzle

Downstream of the extruder, the thermoplastic polymer is heated to its processing
temperature, and its cross-sectional area is reduced to form the bead that exits from the nozzle
and is laid into each layer; the processing temperature is typically close to the melt temperature
(Tm) recommended for injection molding of the same thermoplastic. As it enters the liquefier,
the filament is assumed to be rigid until it reaches its glass transition temperature (Tg); its shear
modulus then decreases significantly and it may be treated as a shear thinning fluid with
viscosity that depends on temperature and shear rate. Heat is transferred by conduction from the
heated walls of the liquefier whose bulk temperature is controlled using feedback from an

5
embedded thermocouple. The low thermal conductivity of thermoplastics suggests that radial
temperature gradients in the material may be significant with increasing filament feed rates; this
increases the required force to move the polymer through the liquefier and nozzle.
Therefore, the coupling of heat transfer and fluid flow in the liquefier must be considered
alongside extruder capability. We performed an experiment to measure the relationship between
feed rate, extrusion force, and liquefier setpoint temperature. The extruder described above was
modified to include a conduction liquefier, taken from a Makerbot Replicator FFF printer; the
force transducer was placed between the feed mechanism and the liquefier (Figure 3a). With the
setpoint at 260 C, the extrusion force was measured as the feed rate of the filament was varied
in a quasi-static manner from 1-15 mm/s, three times each (Figure 3b); the error bars indicate the
standard error for the three trials at each feed rate. At each trial, 10 seconds of material was
extruded until force reached steady-state; the average value of the steady-state region was
recorded as the force requirement for the trial (Figure S1). The same series of measurements was
repeated with the setpoint at 230 C and 200 C. At each temperature, the extrusion force
increased with feed rate until the extruder force limit of ~60 N was reached, and the pinch wheel
mechanism failed by shear of the filament; subsequent trials at feed rates faster than the failure
feed rate demonstrated lower force capability which coincides with result observed in Figure 2b.
At 260 C setpoint, the failure occurred at a filament feed rate of ~11 mm/s. At 230 C and 200
C, extrusion was limited to ~7 mm/s and ~5 mm/s feed rates respectively. This shows the
dependence of temperature on feed rate for conduction heat transfer within FFF liquefiers.
Modeling of liquefier dynamics must consider that the melt properties of the filament are
non-linear relationships of both temperature and shear rate. In past work, extrusion filaments
have been modeled as a power-law fluid because of their shear-thinning behavior, and
temperature dependence is included using an Arrhenius correlation [12]. This power-law model
does not include yield stress and loses accuracy at high shear rates which may become important
with high-throughput systems [9]. The density and heat capacity are approximated to be constant
[9], [12]. Accordingly, the pressure drop across the liquefier for thermoplastics, has been
estimated as the superposition of relations corresponding to the power-law flow through three
geometrically distinct regions of the liquefier [12], [13] (Figure 4a).
In order to accurately analyze the rate-dependent thermal development, we constructed a
finite element model, using the dimensions given in Fig. S2. The model was implemented in
COMSOLTM and assumed a constant wall temperature boundary condition (260 C), along with
perfect contact between the filament (1.75 mm diameter) and the wall. The filament entry
temperature was 20 C, and the filament rheology was modeled using a temperature dependent
Carreau viscosity model with constants derived from a published material library [20]. The
numerical uncertainty of the simulation results is expected to be approximately 1%; the largest
source of uncertainty comes from the material property values used in the simulation, compared
to actual variations in polymer formulation, yet the relative trends will not change. Although the
liquefier walls are maintained at 260 C, the melt temperature does not reach the filament core at
higher feed rates; this is apparent in the color visualizations of the temperature distribution
(Figure 4b). In Figure 4c, we show simulated temperature profiles along the central axis of the
liquefier at nodes 0.5 mm, 5.0 mm, and 10.0 mm from the nozzle exit. Therefore, if the
thermoplastic is not sufficiently heated before it reaches the nozzle region, the force required for
extrusion increases significantly, as observed in our experiments described above. This
represents a critical bound on the achievable extrusion rate according to the rate of conduction

6
heat transfer, the filament properties, and the liquefier geometry, and will be revisited later in the
paper.

3.3 Gantry Dynamics

Third, the rate of positioning of the printhead must be commensurate with the rate of material
extrusion from the nozzle. In principle, if the printhead is capable of extremely high speed
extrusion, limitation by the motion system would, for example, constitute inability to reach the
maximum desired extrusion velocity. Practically, failure in the motion system occurs when it
loses position; this occurs when the actuators are commanded beyond their maximum speed
and/or acceleration. Stepper motors driven without feedback can receive movement commands
beyond their operating performance range, causing the motors to skip steps resulting in loss of
accuracy. Therefore, it is important to understand gantry dynamics and design systems for
operation within the performance curves of the actuators.
In FFF systems, XY plane dynamics are considered to be the primary influence to build rate
because Z axis motion is required only between layers. In a series tangential belt drive gantry
architecture, each axis acts independently; a simple model for a single axis tangential belt drive
system is shown in Figure 5a. The driven and driving pulleys are identical (diameter, geometry,
and mass), the drive belt is massless and has a strict no-slip condition with respect to the pulley
surface, and the output carriage is constrained to single-axis motion. This model further assumes
the belt is non-elastic; in reality the belts are typically made from rubbers with reinforcing fibers
so belt elasticity may influence dynamic accuracy at higher speeds [21].
The output carriage of mass m is described by x, , and for position, velocity, and
acceleration respectively. Its positive direction is defined by positive rotation by the driven
pulley which is described by its radius r, torque m, angular elements , , and , and rotational
inertia Jp. Because the belts are assumed to be rigid and move in exact synchronization with
driving pulley, the driven pulley is also described by r, Jp, , , and . Bearings on the rotary
and linear elements provide viscous friction and have rotary and linear friction coefficients terms
B and b respectively. This model can be applied to the X and Y axes despite the heavier output of
the Y axis. In a series architecture the X axis is mounted atop the output of the Y axis and
therefore the Y axis carriage mass includes the mechanical components of the X axis and its
carriage mass. Taking the Laplace transform of the sum of torques and kinematic equations
provides the following first-order transfer function which relates linear velocity with motor
torque.

() = () (3)
(2 + ) + ( + )
2

The steady-state speed time constant can be extracted from the transfer function which
describes the time needed to achieve 63% of steady-state speed.
2 + 2
= (4)
+
This time constant describes the general relationship between design variables and gantry
performance. Friction elements reduce rise time but also decrease overall speed and therefore can
limit the rate. Inertial elements increase rise time by slowing acceleration. Therefore high-
performance gantries should be designed to minimize inertia and friction in addition to mass.

7
 A complementary means of improving motion system dynamics is to maximize motor
torque. Motor torque decreases with speed as described as a linear relationship [22]; A motor has
maximum torque at zero speed (stall torque ) and maximum speed at zero torque (no
load ). The values for each endpoint depend on motor electromechanical constants such
as phase inductance and resistance. The measured torque-speed (TS) curve of an NEMA 17
motor, similar to the model found in the Makerbot Replicator, is shown in Figure S3, and is
linearized for approximate analysis.
We estimate the performance of a FFF gantry by applying the linearized TS curve (Figure
S3) against measured values for the gantry of a Makerbot Replicator (Figure S4). We predict
(Figure 5b) that this gantry has a maximum speed of 394 mm/s, and can reach 90% of this speed
in 1.1 ms and 1.2 ms for the X and Y axis respectively. Approximately 0.55 mm of distance is
needed to reach this speed; this is equal to 1-2 nozzle widths. The dynamics of the y-axis are
slightly compromised compared to the x-axis due to the higher inertial load (Equation 4). A more
accurate model which introduces motor-side losses may predict further reduced gantry
performance, and closed-loop control may be an effective method of mitigating these losses.
However, it is observed that standard print settings on many commercial printers range
between 40-60 mm/s gantry print speed which is far short of the speeds indicated by simulation
results. Probable causes include factors of safety, thermal limits, and motor commutation limits
which fundamentally prevent these theoretical speeds from being reached. When voltage is
applied to a phase within a stepper motor, current in the windings rises according to the first-
order system described in Equation 5 [14]. Resistance decreases the maximum current in any
phase while inductance slows the rate of current change. Switching frequencies exceeding the
motors electrical time constant result in suboptimal torque and thus position loss (i.e. skipping
of steps). Importantly, this must be avoided in FFF systems where stepper motors are controlled
open-loop and gantry displacement relies on accurate following of commanded steps. The
calculation of the electrical time constant therefore serves as an estimate for motor speed
capability.

() = (1 ) (5)

The electrical time constants of NEMA 17 stepper motors (Table 1) were sampled to estimate
maximum speed. Motors of the smallest size are commonly found in desktop FFF machines; to
explore higher-torque options, two motors with longer stators were also examined. In these
estimates, rise times to 63% of phase current were considered; a full period was achieved by
doubling this time. According to these estimates, stepper motors are incapable of achieving the
gantry speeds that were calculated using a perfect torque-speed response (Table 1). The smallest
motor tested, which is physically similar to the Moons motor used in the Makerbot Replicator
(Figure S3), had the highest speed yet was limited to 228 RPM; using the Replicator gantry
parameters (Figure S4) this equals a linear speed of 39 mm/s, which is a more accurate estimate
of the actual gantry speed during machine operation (between 30-90 mm/s for Makerbot
standard setting). Importantly, maximum speeds are lower for larger motors that have greater
phase inductance. Therefore, feedback control is needed such that position is maintained at high
speeds.

8
4. System-level rate measurements
Next, to relate our understanding of module-level performance to that of a commercial FFF
system, we chose to analyze the Stratasys Mojo [23] (Figure 1a). The Mojo is an entry-level
professional FFF system that prints in ABS build material and soluble support, with a specified
layer thickness of 0.178 mm. The Mojo was instrumented with four optical encoders to measure
the rotation of the motors controlling the X, Y, and Z axes and the filament feed (Figure S5). All
four encoders were sampled at 1 kHz and the print head position was determined by using the
inverse kinematic equations of each mechanism.
We analyzed the trajectories during printing of a variety of test parts. The trajectory for a
layer of a triangular part is shown in Figure 6 where the rectilinear infill raster and the perimeter
contour can be seen. The steady-state printhead velocity during the infill path was measured to
be 90 mm/s, and during the perimeter was 51 mm/s. The slower velocity of the perimeter
enables improved accuracy of the outer surface of the part. Additionally, the filament feed rates
were 4.4 mm/s and 2.4 mm/s for the infill and perimeter, respectively, which were directly
proportional to the measured printhead velocities. At the nominal layer height, the infill and
perimeter volumetric deposition rates are 20.9 cm3/hr and 11.5 cm3/hr. The road widths were
measured to be 0.38 mm for the infill and 0.37 mm for the perimeters respectively.
Along the edges of the exemplary part (Figure 6b) we notice the most prominent defect
related the coupling of material flow and printhead motion in extrusion AM: at each hairpin turn
of the raster and perimeter the solidified track is wider [12], [24]. This occurs because the
filament feed rate is kept at the constant infill or perimeter rate, and excess material is laid when
the printhead velocity is momentarily decreased while changing direction. The slight waviness
of the infill trajectories matches waviness of the deposited ABS tracks, emphasizing the direct
relationship between motion dynamics and part accuracy.
We next tested whether print rate was related to the complexity of printed shapes. A
complexity index (CI), defined as the ratio of the part surface area to volume, was used to
compare parts ranging from solid cylinders (CI = 0.1) to multi-faceted shells (CI = 6.6). This
definition recognizes that additional features (e.g., holes, bosses, slots, etc) create more surface
area than material volume. The results (Figure S6) demonstrated a large disparity between the
print rate of low complexity and high complexity shapes; on the Mojo, solid shapes would print
at a volumetric rate up to 17.3 cm3/hr while shells would print at no greater than 4.3 cm3/hr.
Disassembly of the Mojo printhead (Figure 1a) revealed that it has a conventional
pinchwheel design, yet lower-cost components including the small motor are used because the
printhead is designated as disposable with each filament cartridge. The ABS filament (1.25 mm
diameter) is fed by a pair of knurled wheels spaced at a fixed distance. It enters a stainless steel
tube which is lined with a PTFE sleeve at the inlet and wrapped in a heating coil near the nozzle
which has 0.35 mm outlet diameter. The knurl profile and wheel spacing were measured using an
optical microscope to create a 3D model of the extrusion assembly. The total filament shear area
was estimated as 1.54 mm2; using equation 2, the maximum extrusion force was predicted to be
29.4 N for Stratasys ABS P430 filament. With a filament cross-section of 1.22 mm2, the
predicted maximum sustainable extrusion pressure is 23.9 MPa.
A finite element simulation of the heating and flow within Mojo liquefier (Figure S7)
predicted the extrusion force to be 5.6 N and the core temperature 10 mm from the entrance to be
225 C and 256 C at the liquefier exit at a build rate 20 cm3/hr (filament feed rate of 4.5 mm/s).
This design, having a longer liquefier and smaller filament diameter than the standard FFF

9
extruder, is predicted to heat the filament more effectively and demand less pressure at an
equivalent volumetric rate. Therefore, we conclude that the Mojo printer is performing below its
potential maximum extrusion rate, notwithstanding potential issues in motion control that are
apparent from close examination of the part microstructure. Further, we did not analyze the
dynamics of the H-frame gantry system but expect that performance is dictated by its actuators,
stiffness, and control loop; this may place a tighter limit on build rate than the volumetric rate
capacity of the printhead.

5. Scaling of rate limits

From the above, we may infer that the overall build rate of an extrusion AM system is
determined by the module-level rate limits; the module with the most restrictive rate limitation
should guide the design and performance of the overall machine. To understand how coupling
between module rate limits influences the FFF system build rate, we performed a parametric
series of FEM simulations to map the system requirements spanning a wide bounds of
volumetric build rate and resolution (here defined as the nozzle diameter, dn). The simulation
used the liquefier/nozzle geometry shown in Figure S2. The flow rate is prescribed at the
entrance, and the exit is given a zero pressure boundary condition. The walls of the liquefier are
assumed to be isothermal at 260 C, with a no-slip boundary condition.
In Figure 7a we show the relationship between required extrusion force and build rate for
nozzle diameters (resolution) ranging from 0.1-1.0 mm. For example, with 0.50 mm resolution,
achieving a 5X increase in build rate, from 20 to 100 cm3/hr, requires increased force from 5.2 to
21.1 N. Maintaining the same build rate of 20 cm3/hr while decreasing the nozzle diameter from
0.50 mm to 0.25 mm requires increased force from 5.2 to 8.4 N. Overall, the force grows
~linearly until approximately 120 cm3/hr, at which point the force rises exponentially with build
rate. The exponential rise occurs when cold polymer in the core reaches the nozzle reduction
(region II). Smaller nozzles require deeper thermal penetration of the melt, and therefore
experience the exponential rise at a slightly lower build rate than larger diameter nozzles.
The build rate at which the exponential rise occurs can be delayed by increasing the heater
length, thereby providing more time for the heat to penetrate through the filament by conduction.
As shown in Figure 7b, a longer liquefier requires less extrusion force above a build rate of ~40
cm3/hr, and has a negligible influence below this critical rate, indicating that the reduction and
nozzle regions (II, III) are the major contribution to the pressure drop. However, a longer heater
would require a more massive printhead, which would in turn require higher performance motion
actuators and greater structural stiffness.
In support of this argument, we show the relationship between the mean exit temperature of
region I, liquefier length, and build rate in Figure 7c. This indicates that, while the relationships
between force and build rate may permit successful extrusion of thermoplastic through the
nozzle, an important consequence to conduction-limited heating is that the exit temperature of
the polymer may not equal the liquefier setpoint. This lower exit temperature can be detrimental
to print quality and mechanics because of the greater viscosity of the polymer upon contact with
the build surface and/or the previous layer, which results in less spreading and weaker layer-
layer adhesion [9]. Here, for liquefier lengths of 1, 2, and 3 cm, the maximum build rates at
which the mean exit temperature is 90% of Tmelt, (236 C, defined as 90% of the temperature
difference between the heater wall and initial plastic inlet temperature), are 23, 46, and 69
cm3/hr, respectively.

10
 We may approximate heat flow into the moving filament using the differential equation for a
heat exchanger

= ( ) (6)

where Ts is the surface temperature (assumed constant along liquefier), Tm is the flow-averaged
mean temperature, x is the distance along the liquefier in the direction of flow, P is the perimeter
of the cross-section, is the mass flow rate, cp is the specific heat, and h is the heat transfer
coefficient. However, because the local flow velocity, viscosity, and shear rate are coupled, the
common laminar flow correlations for the heat transfer coefficient cannot be used. A reasonable
prediction of the mean exit temperature can be obtained through modification of the
conventional entrance region heat transfer correlation coefficients reported in literature [25],
[26], to match the simulations of mean exit temperature versus flow rate (Figure 7c) for varying
liquefier lengths. Within the liquefier, the local heat transfer coefficient to the filament can be
approximated as.
0 0
= (500 + 6.5 ) (7)
0
Here, BR is the volumetric flow rate (build rate) in cm3/hr, x is the position from the entry of the
liquefier in millimeters, D is the filament diameter in millimeters, k is the polymer thermal
conductivity, and is the polymer thermal diffusivity. The subscript zero on these properties
represents the values of these properties used in the simulation. This correlation can therefore be
used to estimate the exit temperature as a function of build rate and filament diameter, and to
determine the heater length necessary to satisfy a minimum exit temperature criterion.
Above, our experiments determined that the pinch wheel grip area and material properties
determine the maximum extrusion force. As a result, notwithstanding thermal considerations,
system design may be guided by the relationship between build rate and resolution at a constant
force, which is shown in Figure 8a. For example, with a force constraint of 30 N (~half of the
maximum force measured for the pinch wheel mechanism, Fig. 4), the standard heater can
achieve 43 cm3/hr build rate at a resolution of 0.1 mm, and 120 cm3/hr build rate if the resolution
is reduced to 0.5mm (Figure 8b). However, at this build rate, the approximate average printhead
velocity (V ~ BR/dn2) is 1.5 m/s for the 0.1 mm resolution, and 0.17 m/s for the 0.5mm
resolution. Both velocities exceed the typical limits that we derived for a series gantry system in
section 3.3, indicating that motion system performance becomes a critical limit in addition to
extruder performance.
Now, we can capture the various module-level limits to FFF printingforce, mean outlet
temperature, and gantry speedin a single plot (Figure 9). For example, the maximum extrusion
force places a resolution-dependent limit on build rate, yet shows that significantly greater build
rates can be achieved by sacrificing resolution (i.e., increasing the nozzle diameter). Here, the
representative force values are shown for a liquefier length Ll = 2 cm. However, in spite of the
ability to extrude polymer, an imposed constraint that the polymer core exit temperature should
be no less than 90% of the melt temperature limits build rate to 46 cm3/hr when Ll = 2 cm. The
heat transfer limitation is independent of resolution because the nozzle (regions II, III)
contributes negligibly to heat transfer below build rates of ~100 cm3/s (Figure S8). Parabolic
curves (V ~ BR/dn2) representing constant average gantry speed further show the limitation
imposed by the motion system. The gantry speed, if limited to 0.1 m/s as characterized for the

11
desktop FFF systems such as the Stratasys Mojo, would further limit build rate when the nozzle
diameter is smaller than ~0.40 mm.
Therefore, an accessible space of rate and resolution for an FFF machine can be defined by
the curves for maximum extrusion force, maximum gantry speed, and maximum build rate for a
given liquefier length. Examining the operating points of selected current desktop FFF systems
with respect to the rate limiting curves, we suggest that conduction heat transfer is the primary
limitation to build rate, followed by gantry speed, and extrusion force. We make this statement
because the measured build rates (15-20 cm3/hr), and corresponding liquefier lengths (0.5-1.0
cm) of three desktop FFF machines (Stratasys Mojo, Ultimaker 2+, Zortrax M200) are close to
the predicted build rate limit of 23 cm3/hr for Ll = 1 cm. Comparatively, the industrial-scale
Stratasys Fortus 360mc achieves a greater build rate than the desktop machines (61 cm3/hr),
suggesting that it uses a longer liquefier, and (as observed) has a higher speed gantry system.
Moreover, to avoid inconsistent extrusion due to slip of the pinch-wheel mechanism or skipping
of the extrusion stepper motors, extruder mechanisms must be operated at a fraction of their
maximum force. For example, the measured performance of the Mojo printer (21 cm3/hr infill
build rate, 0.35 mm nozzle diameter) is far below its calculated extrusion force limit (29 N).
Therefore, to achieve significantly higher build rate along with high resolution, faster gantry
motion and greater extrusion force are required, along with adequate heating of the polymer.
The liquefier geometry and extrusion mechanism design influence the mass of the printhead,
which in turn must be accounted for in motion system design to achieve the required steady-state
speed and dynamic accuracy.
Importantly, the above analysis does not consider the extrusion dynamics [6], [24],
particularly die swell which will be more prominent at greater extrusion pressure. Due to
overextrusion of material at corners, at higher speeds it may be necessary to couple the extrusion
rate with the gantry speed. Nevertheless, our study provides important guidelines for tailored
design of FFF systems for improved production rate, and shows that significant increases in
build rate may be possible by integrated design of the system modules along with understanding
of the polymer extrusion dynamics. The practical significance of high-throughput FFF is also
supported by the growing availability of desktop, industrial, and larger-scale FFF systems, whose
build rate is increased significantly by enlargement of the system and use of a larger extrusion
bead [27][29]. In-depth analysis of these systems relative to their intrinsic rate-resolution
tradeoffs, and development of new high-throughput FFF equipment, is a topic of ongoing study
by our team.

6. Conclusions
In this study we identified that each of the main modules of FFF systems---the material
extrusion, material heating, and positioning modulescan limit the overall build rate. Each
module was examined using analytical and experimental methods to determine its performance
limit, defined by its primary function in the FFF process, for typical design parameters. It was
found that the pinch wheel mechanism in the material feed module is limited to ~60 N of linear
force, the feed rate is capped at ~9 mm/s to achieve full melt within the liquefier, and the linear
speed of desktop motion systems are limited by their actuators at ~0.1 m/s or less. The
performance of these modules was validated by measurements of a complete system, the
Stratasys Mojo. Finally, a scaling analysis was performed which examined the relationship
between build rate and resolution. This scaling analysis is summarized in a single map that

12
shows how module-level rate limits can be engineered to achieve a target level of system
performance. We expect that a firm understanding of fundamental limitations of FFF can
motivate new machine designs that achieve breakthrough improvements in build rate, and
thereby accelerate the adoption and applications of extrusion AM technology.

7. Acknowledgements
Funding for this research was provided by a grant from Lockheed Martin Corporation,
managed by Dr. Padraig Moloney. Adam G. Stevens was supported by the Department of
Defense (DoD) through the National Defense Science & Engineering Graduate Fellowship
(NDSEG) Program. We thank the MIT Department of Mechanical Engineering for providing the
Mojo printer, and the MIT International Design Centre (IDC) and MIT MakerWorks for
providing experimentation and fabrication spaces. We also thank undergraduate research
assistants Maxwell Malinowski and Amelia Helmick for assistance with build rate measurements
and materials characterization.

8. References
[1] T. Wohlers, Wohlers Report 2016: 3D Printing and Additive Manufacturing State of the
Industry Annual Worldwide Progress Report, Fort Collins, CO, 2016.
[2] Raphael, How much does it cost when you 3D print a thousand different parts all at
once?, 3D Printing, Shapeways Blog, 03-Oct-2014. [Online]. Available:
http://www.shapeways.com/blog/archives/18174-how-much-does-it-cost-when-you-3d-
print-a-thousand-different-parts-all-at-once.html. [Accessed: 20-Aug-2015].
[3] A. Liszewski, How Shapeways Squeezes Every Last Bit Out Of A 3D-Printing Run,
Gizmodo Australia, 16-Aug-2013. [Online]. Available:
http://www.gizmodo.com.au/2013/08/how-shapeways-squeezes-every-last-bit-out-of-a-
3d-printing-run/. [Accessed: 20-Aug-2015].
[4] T. T. Wohlers and T. Caffrey, Wohlers Report 2015: 3D Printing and Additive
Manufacturing State of the Industry Annual Worldwide Progress Report. Fort Collins,
CO: Wohlers Associates, 2015.
[5] Stratasys, Additive Manufacturing, 2015. [Online]. Available:
http://www.stratasys.com/solutions/additive-manufacturing. [Accessed: 20-Aug-2015].
[6] A. Bellini, S. Guceri, and M. Bertoldi, Liquefier Dynamics in Fused Deposition, J.
Manuf. Sci. Eng., vol. 126, no. 2, p. 237, May 2004.
[7] M. a Yardimci, T. Hattori, S. I. Guceri, and S. C. Danforth, Thermal analysis of Fused
Deposition, Solid Free. Fabr. Proceedings, Sept. 1997, pp. 689698, 1997.
[8] M. K. Agarwala, V. R. Jamalabad, N. A. Langrana, A. Safari, P. J. Whalen, and S. C.
Danforth, Structural quality of parts processed by fused deposition, Rapid Prototyp. J.,
vol. 2, no. 4, pp. 419, Dec. 1996.
[9] B. N. Turner, R. Strong, and S. A. Gold, A review of melt extrusion additive
manufacturing processes: I. Process design and modeling, Rapid Prototyp. J., vol. 20, no.
3, pp. 192204, Apr. 2014.
[10] M. K. Agarwala, V. R. Jamalabad, N. A. Langrana, A. Safari, P. J. Whalen, and S. C.
Danforth, Structural quality of parts processed by fused deposition, Rapid Prototyp. J.,
vol. 2, no. 4, pp. 419, Dec. 1996.

13
[11] J. F. Veneman, R. Ekkelenkamp, R. Kruidhof, F. C. T. van der Helm, and H. van der
Kooij, A Series Elastic- and Bowden-Cable-Based Actuation System for Use as Torque
Actuator in Exoskeleton-Type Robots, Int. J. Rob. Res., vol. 25, no. 3, pp. 261281, Mar.
2006.
[12] A. Bellini, S. Guceri, and M. Bertoldi, Liquefier Dynamics in Fused Deposition, J.
Manuf. Sci. Eng., vol. 126, no. 2, p. 237, May 2004.
[13] M. a Yardimci, T. Hattori, S. I. Guceri, and S. C. Danforth, Thermal analysis of Fused
Deposition, Solid Free. Fabr. Proceedings, Sept. 1997, pp. 689698, 1997.
[14] R. Condit and D. W. Jones, Stepping Motors Fundamentals, Microchip Technology Inc,
AN907, 2004. [Online]. Available: http://homepage.cs.uiowa.edu/~jones/step/an907a.pdf.
[Accessed: 20-Aug-2015].
[15] K. S. Sollmann, M. K. Jouaneh, and D. Lavender, Dynamic Modeling of a Two-Axis,
Parallel, H-Frame-Type <emphasis emphasistype=italic>XY</emphasis> Positioning
System, IEEE/ASME Trans. Mechatronics, vol. 15, no. 2, pp. 280290, Apr. 2010.
[16] S. Staicu and D. C. Carp-Ciocardia, Dynamic analysis of Clavels Delta parallel robot,
in 2003 IEEE International Conference on Robotics and Automation (Cat.
No.03CH37422), 2003, vol. 3, pp. 41164121.
[17] J. D. Ferry, Viscoelastic Properties of Polymers, Third. New York: Wiley, 1980.
[18] R. G. Budynas and J. K. Nisbett, Shigleys Mechanical Engineering Design, 9th ed. New
York: McGraw-Hill, 2011.
[19] Stratasys, ABSplus-P430, Eden Prairie, MN, 2014.
[20] D. P. R. SolidWorks, General Values for ABS Plastic. 2015.
[21] K. S. Sollmann, M. K. Jouaneh, and D. Lavender, Dynamic modeling of a two-axis,
parallel, H-frame-type XY positioning system, IEEE/ASME Trans. Mechatronics, vol.
15, no. 2, pp. 280290, 2010.
[22] Center for Innovation in Product Development, Understanding D.C. Motor
Characteristics, 1999. [Online]. Available: http://lancet.mit.edu/motors/motors3.html.
[23] Stratasys, About the Mojo Desktop 3D Printer, 2015. [Online]. Available:
http://www.stratasys.com/3d-printers/idea-series/mojo. [Accessed: 21-Aug-2015].
[24] B. N. Turner and S. A. Gold, A review of melt extrusion additive manufacturing
processes: II. Materials, dimensional accuracy, and surface roughness, Rapid Prototyp.
J., vol. 21, no. 3, pp. 250261, 2015.
[25] T. L. Bergman, A. S. Lavine, F. P. Incropera, and D. P. DeWitt, Introduction of Heat
Transfer, 6th Edition. 2011.
[26] W. M. Kays, Numerical Solutions for Laminar Flow Heat Transfer in Circular Tubes, J.
Heat Transf. ASME, vol. 77, p. 1265, 1955.
[27] Cincinnati Incorporated, Big Area Additive Manufacturing Fact Sheet, 2014. [Online].
Available: http://wwwassets.e-ci.com/PDF/Products/baam-fact-sheet.pdf. [Accessed: 20-
Aug-2015].
[28] Cosine Additive, AM1 Specifications Cosine Additive, 2016. [Online]. Available:
http://www.cosineadditive.com/am1/. [Accessed: 07-Oct-2016].
[29] 3DP: 3D Platform, EXCEL Additive and Subtractive Manufacturing Series. 3D
Platform, Roscoe, 2015.

14
Figure 1. Desktop FFF systems used in the experimental study: (a) Stratasys Mojo; (b) Printrbot
Metal. The systems have three primary modules: the filament feed mechanism, the liquefier
(heater) and nozzle, and the gantry motion system. Both are actuated by DC stepper motors. The
Stratasys Mojo has a belt-driven H-frame gantry, and the Printrbot has a series gantry with lead
screw drives.

15
Figure 2. Force capability and failure of pinch-wheel filament feed mechanism: (a) instrumented
mechanism used to measure force under stall condition; (b) inverse relationship between filament
feed rate and maximum (stall) force before filament fails in shear; (c) image of shear failure and
model of shear engagement area between filament and textured drive wheel.

16
Figure 3. Coupling of extrusion force with heat transfer to the filament: (a) Instrumented
extruder-liquefier assembly; (b) measured relationships between filament feed rate and extrusion
force at three liquefier setpoint temperatures, under steady-state operation. The rapid rise in
force with feed rate, and the saturation of the force values, respectively, indicate the coupling
between flow mechanics and heat transfer, and the shear failure of the pinch wheel mechanism.

17
Figure 4. Finite element analysis of heat transfer within the liquefier: (a) diagram of the three
geometric regions of the liquefier: (I) heater tube, (II) constriction, and (III) nozzle; (b)
temperature distribution within the liquefier at volumetric rates of 30 cm3/hr, 60 cm3/hr, and 90
cm3/hr, showing limited thermal penetration at higher rates; (c) polymer temperature along the
liquefier centerline, versus volumetric flow rate (build rate), at noted distances from the inlet.
The dimensions match those for the instrumented liquefier used in the force measurements (Fig.
3).

18
Figure 5. (a) Model of a single axis tangential belt drive and (b) simulated dynamic performance
of a Makerbot Replicator gantry.

19
Figure 6. Mapping of the nozzle trajectory of the Stratasys Mojo while printing a layer of a
triangular part: (a) portion of toolpaths for consecutive layers, showing perimeter and infill; (b)
optical image of corresponding material near bottom left corner, along with scaled plot of
trajectory, showing deviations from ideally straight paths and sharp corners.

20
Figure 7. Simulations of: (a) required extruder force versus build rate for standard FFF liquefier,
with noted nozzle diameters and equal liquefier length (Ll = 2 cm); (b) required extruder force
versus build rate with noted liquefier lengths and equal nozzle diameter (dn =0.25 mm); and (c)
mean temperature at the exit of region I versus build rate for noted liquefier lengths and equal
nozzle diameter (dn =0.25 mm) where black lines are use laminar flow correlations and blue lines
are modified correlation coefficients to reflect coupling of flow parameters.

21
Figure 8. Simulated curves that bound the tradeoff between build rate and resolution (nozzle
diameter) at (a) noted extrusion forces, for Ll = 10 mm; and (b) noted liquefier lengths, for F =
30 N.

22
Figure 9. Performance space of FFF system build rate versus resolution as constrained by
maximum gantry speed, extrusion force (corresponding to Ll = 2 cm), and liquefier length (with
Tcore,exit = 0.9Tmelt). The common area under the three curves corresponding to a set of system
parameters (F, Ll, V), defines the accessible performance for the system, i.e., the possible
combinations of build rate and resolution that can be achieved. Measured values for selected
desktop FFF systems, as noted in the legend, are compared to the curves that have been created
from our calculations and simulations.

23
Table 1. Summary of measured motor specifications, calculated electrical time constants, and
estimated maximum speeds for three stepper motors used in desktop FFF systems.

42H33H-1334A

Kysan 1124090

17Y402S-LW4
Phase inductance (mH) 1.5 2.9 22.0
Phase resistance () 2.3 2.9 13.5
Electrical time constant (ms) 0.6 1.0 1.6
Estimated speed at 63% 3.8 2.4 1.5
phase current (rps)
Equivalent linear velocity 38 25 15
(mm/s)

24