Anna Bellini

Department of Manufacturing Eng. and Management, Technical University of Denmark DTU bldg. 425, DK2800 Kgs. Lyngby e-mail: anna@ipl.dtu.dk

Liquefier Dynamics in Fused Deposition

Layered manufacturing (LM) is an evolution of rapid prototyping (RP) technology whereby a part is built in layers. Fused deposition modeling (FDM) is a particular LM technique in which each section is fabricated through vector style deposition of building blocks, called roads, which are then stacked layer by layer to fabricate the nal object. The latest improvements in this technology brought about the possibility of fabricating not only a model but even the nished product. This paper presents the analysis of the liqueer dynamics towards establishing control strategies for ow control during the extrusion phase, which is necessary to achieve the mentioned objective. DOI: 10.1115/1.1688377

Selcuk Guceri Maurizio Bertoldi

Mechanical Engineering & Mechanics, Drexel University, Philadelphia, PA 19104 e-mail: Guceri@drexel.edu

Introduction
Fused deposition modeling has become a widely used rapid prototyping technology. In the FDM utilized in this study a Stratasys 1650 Modeler a plastic lament is supplied on a reel and fed into a heated liqueer where it is melted. This melt is then extruded by a nozzle while the incoming lament, still in solid phase, acting as a plunger. The nozzle is mounted to a mechanical stage Fig. 1 a , which can be moved in the xy plane 1 . As the nozzle is moved over the table in a prescribed geometry, it deposits a thin bead of extruded plastic, referred to as roads which solidify quickly upon contact with substrate and/or roads deposited earlier. Solid layers are generated by following a rastering motion where the roads are deposited side by side within an enveloping domain boundary Fig. 1 b . Once a layer is completed, the platform is lowered in the z direction in order to start the next layer. This process continues until the fabrication of the object is completed. Successful bonding of the roads in the deposition process necessitates control of the thermal environment. Therefore, the entire system is contained within a chamber, which is maintained at a temperature just below the melting point of the material being deposited. Several materials are commercially available for this process including polymers such as ABS and investment casting wax. When needed, support structures are fabricated for overhanging geometries using different materials, which are later removed by breaking them away from the object. A water-soluble support material, which can simply be washed away, is also available. In FDM the part is built according to a pre-specied tool-path Fig. 2 b , usually determined during the design phase i.e. Quickslice 2,3 . Each tool-path is conceptually divided into ve regions see Fig. 2 a 4 : I. Pre-movement: a prescribed volumetric ow rate is started before the deposition begins. A glob of certain size is generated to compensate for the intrinsic deposition delay, due to the internal length of the liqueer, which is the distance between the point of application of the pressure and the point of the material delivery. II. Start-up: as soon as the motion starts, an absolute ow-rate, higher than the steady state ow rate, is established and maintained throughout the acceleration phase; III. Steady-state: once the acceleration phase is complete, a constant ow rate is specied; IV. Slow-down: the main ow rate is stopped and a certain
Contributed by the Manufacturing Engineering Division for publication in the JOURNAL OF MANUFACTURING SCIENCE AND ENGINEERING. Manuscript received Dec. 2003. Associate Editor: C. Yao.

amount of material empirically determinated and dependent on the deceleration constant and the steady state ow rate is brought back revolving the motion of the rollers; V. Exit-move: the ow control is turned off and the nozzle is moved to a pre-specied distance from the last point of the toolpath in order to avoid any further interferences. Figure 2 a schematically depicts some of the irregularities that are typical during road deposition. An under-deposition zone can be observed during the acceleration phase and an overdeposition during deceleration. In order to reduce these transient effects a better correlation between deposition speed and ow rate is needed. The objective of this paper is to perform an analysis of the process taking place in the liqueer. A study based on mathematical modeling is presented rst. Then a transfer function approach is developed that can be integrated as part of an overall control strategy for an effective ow control and improved part integrity 5 . These two models are then compared with experimental observations that are performed by inducing step changes in deposition rates. The dynamic of the liqueer Fig. 1 is one of the most complicated phenomena to analyze in an FDM process. Inside the liqueer the physical system exhibits a complex behavior due especially to the unsteady movement of the viscous melting uid. Other possible causes compounding this complex behavior can be due to some and/or all of the following reasons: Compressibility of the uid; Slippery contact conditions between the liqueer walls and the melting ow; Possible slippage between the feed-rollers and the lament; Uneven distribution of heat ux, which is provided by electrical heaters that consist of metal coils wrapped around the aluminum liqueer. Hence, it heats the thermoplastic material once the power is applied. A temperature regulator adjusts the power to the heater, based on temperature measurement at a single point 0.5 away from the exit tip 4 ; Change of physical state of the melt. The thermoplastic lament heats up and gradually melts as it travels down the liqueer 4; Sticktion effect of the melt while the lament is pushed into the liqueer. Furthermore, the non-linear dependences of the material properties on the temperature and the shear rate the melt can indeed be modeled as a generalized Newtonian ow bring additional difculties to the modeling process. MAY 2004, Vol. 126 237

Journal of Manufacturing Science and Engineering Copyright 2004 by ASME

Downloaded 06 Sep 2011 to 138.38.63.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 1 Schematic of a liqueer in an FDM process Fig. 4 Regional decomposition of LiqueerExtruder

Liqueer Dynamics: Mathematical Model

The extrusion is central to the fused deposition process during which the thermoplastic lament is introduced, via mechanical pressure, into the liqueer, where it melts and is then extruded. Since the rollers are the only drive mechanism in the material delivery system, the lament is under tensile stress upstream to the roller and under compression 6 at the downstream side acting as a plunger. The compressive stress thus becomes the driving force behind the extrusion process Fig. 3 . The force required to extrude the melt must be sufcient to overcome the pressure drop across the system, which strictly depends on the viscous properties of the same melt as well as on the geometry of the liqueer and nozzle Fig. 4 . Because of the fact that melts adhere to the liqueer/die walls, the material is subjected to shear deformation during the ow. dv dr (1) Thermoplastic materials currently used in FD exhibit shear thinning behavior, i.e., the viscosity decreases with increasing shear rate 7 . Although more extensive and complex models i.e. Bird Carreau law, Cross law, etc. are available, it can be assumed that adopting a power law for generalized Newtonian uids for modeling of polymers such as ABS, is sufciently accurate for modeling FD melts, especially in light of other high-order uncertainties in the system. Accordingly, the power law states:
1/m

(2) where m and are material constants, is the shear stress and is the shear rate. The power law parameters m and indicate the ow exponent and the uidity respectively. The general ow characteristic of a material and its deviation from the Newtonian behavior is reected by the ow exponent m. Applying the Power Law to the momentum ux balance on a uid element 8 respectively in the zone I, II, and III Fig. 4 , it is possible to determine the total pressure drop: P1
v

2L 1 1

1/m

m 3 D 1 /2
m 1 2

1/m

(3)
1/m

P2

2m 3tan /2

1 D 3/m 1

D 3/m 2
v
1/m

D1 2

m 3 2 m
2 1/m

(4) P3
Fig. 2 Flow-control regions of a tool-path a and sample toolpath b
v

2L 2

m 3 D 1 /2 D 2 /2
m 3

(5)

Fig. 3 Driving force

where D 1 and L 1 are respectively the length and the diameter of the liqueer section I in Fig. 4 , D 2 and L 2 are the length and the diameter of the tip of the nozzle section III in Fig. 4 , is the convergence angle of liqueer-tip diameter transition. The velocity v is the velocity of the lament at the entrance of the liqueer. Since at this stage the material is still in solid form, this value is constant and uniform over the cross section plug ow . In Eqs. 2 5 the uid is assumed to be isothermal at a temperature T equal to the temperature of deposition. However, when a sudden change in the ow rate i.e., step function is applied at the entrance of the liqueer, the steady state condition of the ow is disturbed. With the introduction of new material in the system, the average velocity of the uid increases and the average temperature drops. In order to stabilize new steady state conditions, the heat ow rate must be increased. Since the temperature control is based on a single input from a thermocouple reading at the end of the liqueer 12 cm far from where the change has taken place , the response of the system is delayed. Furthermore, the temperature is regulated via heat input from electrical coil heaters. The system continuously adjusts the power supplied to the coils according to the temperature difference between the desired value Transactions of the ASME

238 Vol. 126, MAY 2004

Downloaded 06 Sep 2011 to 138.38.63.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

and the value detected by the thermocouple. The response time of the temperature controlling loops is thus of the order of seconds, due to the thermal inertia of the extrusion assembly. The necessary heat ow rate, supplied through the wall of the liqueer, S 2 D 1 /2L 1 , can be approximately determined as: q v Ac p T Ti D1 2 L 1 2 (6)

where R r is the radius and r is the angular velocity of a roller. The expressions presented above are next arranged and developed into a MatLab program. In the initial approach the following assumptions were made: 1. Perfect adhesion between rollers and lament, thus v r R r ; 2. The angular velocity of the rollers abruptly changes i.e., step function in order to provide the specied ow rate see the Experimental Set Up section ; 3. The heat ow rate is increased following a ramp function in order to model the closed loop control ; 4. There is no upper limit to power and torque that drivemotors can provide.

where T i is the temperature at the entrance of the liqueer, is the density of the melt, A is the cross section of the inner liqueer and c p is the specic heat capacity of the melt. In order to simplify the calculation, the melt is considered at the uniform temperature T T f , that is the temperature at the end of the liqueer, and c p has been considered constant in the range of temperatures in consideration. During the time of thermal adjustment, the ow is nonisothermal, so temperature dependence of viscosity must be taken into account along with the shear-rate dependence 9,10 . The viscosity expression can be factorized as follows: H T
0

Mathematical ModelAnalytical Results

In Fig. 5 the results obtained with the rst approach no limitation on power and torque supply , have been grouped. The average velocity Fig. 5 a of the melt is given according to the command provided by the SML le, thus it suddenly changes at half the length of the road. The heat ow rate is also given and it is assumed to change linearly Fig. 5 b dq 540 T 0 D 1L 1
v end v 0 Ac p

(7)

where H(T) accounts for temperature dependence and 0 ( ) is the expression for viscosity at some reference temperature T value of the temperature at which the parameters m and of the power law have been determined . In this study the Arrhenius relation has been adopted to model the behavior of ABS: H T exp 1 T T0 T 1 T0 (8)

1 80

(16)

where is the energy of activation, T is a reference temperature for which H(T) 1 and T 0 0 for absolute temperatures T and T . Substituting Eqs. 8 and 2 into 7 , various pressure drops in the system can be expressed as: P 1 2L 1 2m 3tan /2 e
1/T 1/T

1/m

m 3 D 1 /2
m 1

1/m

e D1 2
2

1/T 1/T

(9)
1/m

P2

1 D 3/m 2

1 D 3/m 1

m 3 2 m

(10)
1/m

P 3 2L 2

m 3 D 1 /2 D 2 /2
m 3

1/m

e P3

1/T 1/T

(11) (12)

P1

P2

From Eqs. 9 to 12 it can be observed that when a sudden change in the ow rate occurs, the average temperature T drops and the velocity v increases, thus resulting in considerable increase for the pressure drop in the system. Once total pressure drop is known, the compression force applied to the lament in order to extrude the material can be calculated as: F PA (13)

where A is the cross section of the lament which is equal to the cross section of the liqueer . Since the force is imposed by two drive-rollers see Fig. 3 driven by a pair of micro stepper motors, the torque and the power required to each motor for the extrusion become: F R 2 r P
r

(14) (15)

where 540 represents the exit temperature, in Kelvin, from the liqueer; T 0 343 K is the temperature at the entrance; D 1 1.83 10 3 m is the diameter of the liqueer; L 1 0.15 m is the length of the liqueer; 900 Kg/m3 is the density of the material; v end is the nal velocity of the melt; v 0 is the initial velocity of the melt; A /4D 2 is the cross section of the liqueer; 1 c p 1500 J/KgK is the specic heat for the material. The average temperature, that is considered independent of the location, is then calculated using Eq. 6 . From Fig. 5 c it is possible to note that a sudden change in the inow results in a sudden reduction in temperature. This causes, according to Eqs. 7 and 8 , an abrupt increase of the average viscosity of the melt, thus signicant additional pressure see Eqs. 9 12 and force Eq. 13 are needed to maintain the extrusion. Consequently, to allow continuity in the process, the motors have to provide excessive levels of torque and power. In Figs. 5 e and 5 f it is possible to see that the required power increases from 0.0008 W to 15000 W at t 2.3 seconds when the ow rate is increased . After this sudden change, the necessary power and torque decreases because the average temperature of the melt increases the ow rate is kept constant, but the heat ow augments . With the assumptions presented earlier, the results presented in Figs. 5 g and 5 h show that in order to provide a sudden increase of the angular velocity of the rollers, the stepper motors have to supply excessive levels of torque because of the increase in the pressure drop, due to the increase in the viscosity and power. The motors that drive the rollers are MicroMo Motors, type 2233T012S made by MicroMo Electronics Inc. The maximum output power that they can supply is 3.66 W, the stall torque is 1.633 10 2 N/m 11 and the most efcient area of operation is for 1.6310 3 4.910 3 N/m. This indicated that the model has to be modied in order to take into consideration a limitation in power and torque supply. In the modied approach the following assumptions have been adopted: MAY 2004, Vol. 126 239

Journal of Manufacturing Science and Engineering

Downloaded 06 Sep 2011 to 138.38.63.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 5 Results for the initial model for a step amplitude2.9851 mm3sec

1. Slip conditions between rollers and lament, according to the generalized Navier law 12 ; 2. The angular velocity of the rollers changes in the middle of the deposition process see the Experimental Set Up section as much as possible, but without requiring an excessive power supply i.e. grater than 3.66 W ; 240 Vol. 126, MAY 2004

3. The heat ow rate is increased following a ramp function in order to model the close loop control ; 4. There are prespecied upper bounds for the torque and the power supplied by the stepper motors. Figure 6 presents the results obtained with the modied model. In Transactions of the ASME

Downloaded 06 Sep 2011 to 138.38.63.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 6 Results from the modied model for step amplitude2.9851 mm3sec

Fig. 6 a the theoretical response of the velocity of the rollers, according to the command provided by the SML le is shown. However, because of the motor specications, the speed of the rollers increases until a point when the maximum torque and/or power limits are reached. Beyond this point the rollers maintain a constant velocity on an interval t Fig. 6 a , during which the Journal of Manufacturing Science and Engineering

desired temperature of the melt is reestablished due to the increase in the heat ow rate Fig. 6 c . The necessary heat ow is calculated by specifying the exit temperature which must be kept between 540 and 552 K Fig. 6 d . The slip condition is taken into account in calculating the melt velocity. The algorithm applied for this case is as follows: 1 the MAY 2004, Vol. 126 241

Downloaded 06 Sep 2011 to 138.38.63.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

speed of the rollers is increased of a constant value, thus an iteration is performed imposing that the melt is owing at the same rate no slip conditions ; 2 the force applied by the motor is then calculated using Eq. 13 and a check on the power and torque is made Eqs. 14 and 15 ; 3 if the limit on the power and torque is reached, the program proceeds considering that the applied force is also used to overcome the friction between the rollers and the lament: Fs vr where
e 1 0 v r

(17)

where 0 1, e 0.48 are material properties; 4 the force F applied to extrude the melt is thus reduced by the amount F s ; 5 the new value of F is consequently used to calculate the real average velocity of the melt see Fig. 6 b . These results also indicate that the power and torque necessary to maintain the extrusion process are kept within an acceptable range Figs. 6 g and 6 h . Thus, it can be concluded that the modied model is more realistic and applicable to physical systems. Furthermore, with these more realistic assumptions the model became capable to better reproduce the experimental observations as will be discussed in a later section.

Fig. 7 The analog electric circuit for dynamic analysis of liqueer

the heat ux that needs to be adapted to the change in the ow rate. For a better understanding of the dynamics and development of a control strategy, an analog electronic circuit has been developed as shown in Fig. 7. The specied ow rate can be represented as the applied current i or voltage u, due to the proportional relation between the two entities at the node A; the slip condition between the rollers and the lament can be modeled as a resistor R; while the material properties and the heat ux can be represented as an impedance L they can be seen as an opposition to sudden changes in the ow rate . The response of the liqueer can thus be mathematically expressed using the following equations 13 : L di t R i t dt y t i exp t i t i exp t u t (18)

Liqueer Dynamics: Transfer Function Approach

While the mathematical model gives an inside of the complex dynamic taking place in the liqueer and is able to give an explanation to some observed phenomena, such as time-delay and steady-state error, it is of difcult implementation for further developments. Since the nal objective of this research is indeed to improve the quality of the end product, it is the focus of this project to study the possible development of a close-loop control system that, actively incorporated in the FDM software, can predict and consequently avoid defects, such as under-deposition, start-stop errors etc. For this reason a more direct approach was also considered, as explained in this section. In the simplest case the liqueer can be modeled as a dynamic system subjected to the digital SML command, which indicates the desired ow rate, as the input signal and the real ow rate as the output from the system. In the present work the following assumptions have been made: the communication between the PC and the controller are instantaneous; the Asymtek digital controller is innitely fast; the ABS lament at the entrance of the liqueer is perfectly round, with a diameter of 1.78 mm and the material density is constant; the lament is perfectly rigid and does not buckle between the rollers and the liqueer entrance. In order to develop a model for the transfer function of the liqueer, the response to a step input function has been studied and analyzed. The main objective is that once the time constant, time delay and gain of the liqueer system are known, the ow rate can be more accurately regulated according to the deceleration and acceleration of the deposition speed. With this capability it will be possible to signicantly reduce some of the irregularities depicted in Fig. 2, particularly in zones II and IV. In this section the experimental and analytical studies conducted to determinate the liqueer transfer function are presented. In performing the dynamic modeling for the liqueer, the following variables have been taken in consideration: the time delay between the signal change in the material ow rate at the entrance of the liqueer and the observation of the results change of the ow rare at the end of the liqueer ; the slip conditions between the rollers and the lament are accounted for as a delay term; the material properties of the ABS i.e., viscosity, melting ow, thermal conductivity, thermal capacity ; 242 Vol. 126, MAY 2004

where i exp(t) is the experienced response of the system, i.e. the current registered at time t by the amperometer. It can be noticed that, due to the time delay between the data acquisition of i current in the node C and the imposed voltage u thus the current i between the nodes A and B, the analyzed response is: y t Ly t Ly t LsY s i t (19)

Substituting 19 into the rst equation of the system 18 : Ry t u t (20) to Eq. 20 : (21) (22)

Applying the Laplace transform indicated with Ry t RY s u t u t

where Y (s) is the Laplace transform of y(t). In order to develop a transfer function of the liqueer, a step function has been applied as follows u t
1 2

t a t a (23)

Dening H(t k) the Heaviside step function: 0 H t k one can write 14 : u t and similarly u t
1 H 1 H

t k t k H t a H t a (24)

t 0

(25)

(26)

Transactions of the ASME

Downloaded 06 Sep 2011 to 138.38.63.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Remembering the Laplace transform of an Heaviside step function H(t) H t k e s

ks

(27)

and applying the property of linearity for the Laplace transform of Eqs. 25 and 26 , it is possible to derive: u t Similarly u t e e
1 s

U s

1 s

e
2 1

as

s
a s

(28)

e
2 1

s
s 1

s e
as 1

Fig. 8

a Top view of the road; b Side view of the road

1 s

s
as

(29)

Substituting Eq. 29 into 22 : Ls R Y s e

s 1

1 s
s

e
2 1

(30)

Comparing Eqs. 28 and 30 : Ls R Y s e U s (31)

The transfer function G(s) of the system can thus be expressed as: Y s U s 1 R L s 1 R

G s

(32)

For simplicity, the following constant can be dened: 1 R T L R (33) (34)

Equation 31 can thus be rewritten in the more conventional form 15 : G s Ts 1 e

s

(35)

where T is called equivalent time constant, is called equivalent delay and is the gain. As it is shown in Eq. 35 , the transfer function of the liqueer is modeled as a rst order differential equation. This limits the validity of the studied model to low frequency responses. The investigation of the second order effects is not indeed part of the goals of the present paper.

deposited on a Plexiglas base. The measurements of the sections of the roads have been performed using Raman spectrometer equipment by determining the road width and thickness respectively at various locations. More specically, the CC/CD camera attached to the microscope connected to the Raman equipment and a computer equipped with Renishaw WIRE software have been used, while the green and red lasers of the RAMAN have been not utilized. One of the Renishaw features, the WiRE Video Viewer, allows not only to take pictures of the object set in the microscope stage, but also denes a Cartesian coordinate reference system to read the coordinates of a particular point, to accurately move the stage in x and y directions using coordinate inputs, etc. A two-speeds joystick is also connected to the microscope in order to manually adjust the position of the stage while observing under the microscope. To generate the sectional data, the roads are rst cut very near to the ends with a razor blade and are placed on the microscope stage. Using a 5 magnication lens, the origin of the reference system is dened at one end of the ber. Following this, the stage can be accurately moved within 1/10 of a m by entering the desired x-value distance from one end of the road in the program. At each desired location the y coordinates of two opposite points, i.e. points A and B in Fig. 8, are determined by simply moving the cursor to follow the edges of the road. The width of the ber is thus calculated as a difference between the two y values. These steps are repeated for different point locations i.e., points A , B , A , B etc. as well as for the side of the road Fig. 8 b , in order to also determine its height. For the purpose of clarity and comparison with the schematic drawing of Fig. 8, in Fig. 9 the actual micrograph of a deposited road shows the increment in width and high due to the suddenly imposed change of the ow rate. For accurate observation of the system response to the step function, the sampling rate has been considerably increased at x l/2 which is the location corresponding to the instant of sudden change in the ow rate.

Mathematical Model: Comparison of Results

In order to verify the validity of the mathematical model, three simulations have been performed for different step amplitude of the input ow rate command, as described in the section dedicated to the experimental results. In Figs. 10, 11 and 12, the theoretical change in the ow rate according to the SML le , the experimental curve determined as described in the previous section as well as the results obtained with the mathematical model have been assembled. Following observations can be made from these graphs: there is a good agreement between the applied ow rate i.e. theoretical curve and the physical response of the system i.e. experimental curve for small magnitudes; there is a steady state error of approximately 12% between the input command and the resulting physical ow rate; MAY 2004, Vol. 126 243

Liqueer Dynamics: Experimental Studies

In order to determine the dynamic response of the liqueer, the relation between the applied value and the actual response in the ow rate during the deposition of a road has been studied. More specically, a SML le has been written so to deposit a single 10 cm-long-road, at a constant speed of 2 cm/sec. In the same le the command to change the material ow rate is given at the halflength point during building of the road. With accurate measuring of the sectional areas along the extruded lament it was possible to calculate the physical ow rate at the exit of the nozzle, thus enabling comparison with the theoretical predictions. In order to minimize measurement errors due to the imperfect contact between the extruded material and rough, porous, foam platforms typically used in FD processes, the roads have been Journal of Manufacturing Science and Engineering

Downloaded 06 Sep 2011 to 138.38.63.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 11 Step response, step amplitude2.9851 mm3sec

operative conditions i.e. ow rate . This action, that results in the adjustment of the SML command so to annihilate the friction force between the rollers and the lament, avoids the steady state error when the machine operates at constant conditions, but can not eliminate it when the conditions changes during the construction of the part; the limitation of power and torque of the motors, and the strong temperature dependence of the viscosity of the material can be considered the cause for the existence of a certain time delay and of an equivalent time constant in the execution of the SML command.
Fig. 9 a Top view of the road; b Side view of the road

Transfer Function Approach: Comparison of Results

The results of the experiments described in the sections above have been also assembled in Figs. 13, 14 and 15. In these plots, the curves for the step input for the ow rate as well as the curves of the actually measured ow rates can be observed. The only difference between the three experiments depicted in these Figures is the amplitude of the input step functions, while all the other operating parameters are kept the same. Based on the experimental data, conveniently summarized in Table 1, the transfer function for the liqueer can been obtained see Eq. 36 as G s 0.8865 e 1 0.45s
0.04s

there is a good agreement between the experimental curve and the curve generated by the application of the mathematical model. The capability of the mathematical model to reproduce the experimental results is representative of the fact that: slip conditions between the rollers and the lament can be considered the cause for the existence of a steady state error when the ow rate is abruptly changed. It should be added at this point that the FDM machine Stratasys 1650, used for the experimental study, requires to be calibrated at the beginning according to the

(36)

Fig. 10 Step response, step amplitude3.5645 mm3sec

Fig. 12 Step response, step amplitude2.4332 mm3sec

244 Vol. 126, MAY 2004

Transactions of the ASME

Downloaded 06 Sep 2011 to 138.38.63.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Table 1 Parameters of the liqueer transfer function for a step input Test n. Step Amplitude mm3/sec gain sec T sec 1 3.5645 0.8994 0.04 0.47 2 2.9851 0.8762 0.06 0.38 3 2.4332 0.8839 0.02 0.5 Average 2.9946 0.8865 0.04 0.45

with the use of the studied transfer function, the response of the system can be predicted with good accuracy; however the presence of higher order dynamics will likely improve the dynamic modeling while increasing the complexity of the analysis substantially. The good prediction of the real ow, obtained when applying the transfer function approach, suggests that the integration of this function as part of an overall control strategy will provide an effective ow control and consequently will improve the part integrity 5 . Moreover, a further development of a close-loop feedback system will lead to the following advantages: increase of the accuracy; reduction of the sensitivity to disturbance; reduction of the system induced noise and distortion.

Fig. 13 Step response, step amplitude3.5645 mm3sec

Substituting Eqs. 36 and 28 into 32 , the function Y (s) was easily calculated. Consequently, through the inverse of Laplace transform, y(t) i.e., the observed ow rate was determined and plotted as well in each of the graphs shown in Figs. 13, 14 and 15. Following observations can be made from these graphs:

Conclusions
Customer-driven product customization and continued demand for cost and time savings have generated a renewed interest in agile manufacturing based on improvements on RP technologies and in particularly of fused deposition modeling FDM . A shift from prototyping to manufacturing necessitates however the following improvements: part integrity and built-in characteristics to meet performance requirements; improved surface quality. These objectives can be achieved only through accurate process modeling, especially in the category of ow control during deposition. For this reason an extensive study of the dynamics of the FDM liqueer was performed and explained in this paper. In order to have an insight of the complex phenomena that occur in the liqueer, a mathematical model based on physical assumptions was developed at rst. After comparison of the results with the experimental data see Figs. 10, 11 and 12 it was concluded that: slip conditions between lament and rollers can be considered the cause of a steady state error when sudden changes are applied to the ow rate; the limitation of power and torque of the motors as well as the strong dependency of the viscosity of the material on the temperature can be considered the cause of the time delay in the response of the system. The mathematical model, which gives a physically explanation to the process taking place in the liqueer, demonstrated nevertheless to be of difcult implementation for further developments. For the purpose of designing a close loop control system that, reducing the steady state error, will improve the accuracy of the deposited ow, hence the nal part integrity, a study based on the investigation of the transfer function of the liqueer was developed as well. In particularly, since the objective was primarily the investigation of rst order effects, the system was subjected to a step function input and its response was accurately studied see Figs. 13, 14 and 15 . The comparison of the experimental data with the results obtained using the transfer function of Eq. 36 shows a good agreement for the purpose of integrating a control function describing the dynamic response of a liqueer to changing input commands. MAY 2004, Vol. 126 245

Fig. 14 Step response, step amplitude2.9851 mm3sec

Fig. 15 Step response, step amplitude2.4332 mm3sec

Journal of Manufacturing Science and Engineering

Downloaded 06 Sep 2011 to 138.38.63.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Acknowledgments
This project was sponsored, in part, by the Ofce of Naval Research ONR in MURI project, #N0014-96-11175. Many constructive discussions with other MURI team members, notable with Drs Danforth, Safari and Jafari of Rutgers University, are also acknowledged.
7

References
1 Kai, C. C., and Fai, L. K., 1997, Rapid Prototyping: Principles and Applications in Manufacturing, Wiley ed., New York, ISBN 0-471-19004-7. 2 Bouhal, A., Jafari, M. A., Han, W., and Fang, T., 1999, Tracking Control and Trajectory Planning in Layered Manufacturing Applications, IEEE Trans. Ind. Electron., 46 2 . 3 Langrana, N. A., Qiu, D., Bossett, E., Danforth, S. C., Safari, M., and Safari, A., 2000, Virtual Simulation and Video Microscopy for Fused Deposition Methods, Mater. Des., 22, pp. 75 82. 4 Yardmci, A., 1999, Process Analysis and Development for Fused Deposition, PhD Thesis, University of Illinois at Chicago, Chicago. 5 Bellini, A., and Guceri, S. I., 2003, Mechanical Characterization of Parts Fabricated Using Fused Deposition Modeling, Rapid Prototyping Journal, 9 4 , pp. 252264. 6 Venkataraman, N., Rangarajan, S., Matthewson, M. J., Harper, B., Safari, A., Danforth, S. C., Wu, G., Langrana, N., Guceri, S., and Yardimci, A., 2000, 10

11 12 13 14 15

Feedstock Material PropertyProcess Relationship in Fused Deposition of Ceramics FDC , Rapid Prototyping Journal, 6 4 , pp. 244 252. Bird, B. R., Armstrong, R. C., and Hassager, O., 1987, Dynamics of Polymeric Liquid: Second Edition-Volume 1: Fluid Mechanics, Wiley-Interscience, New York, ISBN 0-471-80245-X v.1 . Michaeli, W., 1992, Extrusion Dies for Plastics and Rubber: Design and Engineering ComputationSecond Edition, Hanser ed., New York, ISBN 3-44616190-2. Bertoldi, M., 1998, From Rapid Prototyping to Rapid Manufacturing: Analysis of the Fused Deposition Process, Tesi di laurea del dipartimento di Ingegneria Aerospaziale del Politecnico di Milano, Milan. Venkataraman, N., 2000, The Process-Property-Performance Relationships of Feedstock Material Used for Fused Deposition of Ceramic FDC , Department of Ceramic and Materials Engineering, Rutgers University, New Brunswick. DC MicromotorsBrushless DC SevomotorsGearheadsTachogeneratorsEncodersControllers, MicroMo catalog 19992000. Fluent Incorporation, Polyow 3.8 DocumentationTutorial Guide. Melsa, J. L., and Donald, S. G., 1969, Linear Control Systems, McGraw-Hill Book Company, New York, ISBN 07-041481-5. Greenber, M. D., 1998, Advanced Engineering MathematicsSecond Edition, Prentice Hall, Upper Saddle River, NJ, ISBN 0-13-321431-1. Dean, F. K., and Chow, J. H., 2000, Feedback Control Problems: Using MATLAB and the Control System Toolbox, Brooks/Cole Thomson Learning, Pacic Grove, ISBN 0-534-37175-2.

246 Vol. 126, MAY 2004

Transactions of the ASME

Downloaded 06 Sep 2011 to 138.38.63.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm