You are on page 1of 7

COMPOSITES SCIENCE AND TECHNOLOGY

Composites Science and Technology 66 (2006) 22652271 www.elsevier.com/locate/compscitech

Flow rate control during vacuum-assisted resin transfer molding (VARTM) processing
Dominik Bender a, Jens Schuster b, Dirk Heider
b

a,c,*

a University of Delaware, Center for Composite Materials, Newark DE 19716, United States University of Applied Sciences, Kaiserslautern, Site Pirmasens, Carl-Schulz Str. 1-9 66953 Pirmasens, Germany c Department of Electrical and Computer Engineering, Newark, DE 19716, United States

Received 27 April 2005; received in revised form 22 November 2005; accepted 7 December 2005 Available online 3 February 2006

Abstract This study investigates a ow rate control system for the Vacuum-Assisted Resin Transfer Molding (VARTM) process. The system allows vacuum pressure being applied to the injection bucket to generate a computer-controlled vacuum dierential between the injection and vent gate. A fuzzy-logic controller has been implemented in LabVIEW and the system has been optimized using virtual simulation. A parametric study shows the performance of the control system as a function of process and material parameters. An experimental setup has been implemented including automatic pressure control and ow rate feedback from computer-controlled scales to validate the setup. The work is part of a state-of-the-art automated VARTM system, which will minimize costs and reduce engineering supervision currently necessary during resin injection. 2005 Elsevier Ltd. All rights reserved.
Keywords: A. Polymer-matrix composites (PMCs); E. Resin Transfer Moulding (RTM); E. Vacuum Assisted Resin Transfer Molding (VARTM); Process control

1. Introduction The Vacuum-Assisted Resin Transfer Molding (VARTM) has become a cost-eective method for producing large-scale composite parts for civil and defense applications. The process allows the use of low-cost tooling while still producing high quality composite parts making it the preferred manufacturing technique for current and future naval top-side structures (e.g. hulls, decks, and enclosures). The mold lling under vacuum pressure becomes more dicult as the complexity or size of the part increases or when advanced ber preforms exhibiting low permeability are being applied. This can lead to dry-spots requiring repair or scrap of the component. Here, resin ow control becomes important to improve the yield and performance of the process.

Corresponding author. Tel.: +30 283 188 98; fax: +30 283 185 25. E-mail address: heider@ccm.udel.edu (D. Heider).

Such control systems have already been developed for other liquid molding techniques such as Resin Transfer Molding (RTM). Lawrence et al. [1] control the ow of an RTM process by opening and closing multiple injection lines based on the resin ow position feedback from toolmounted sensors. The system minimizes the race tracking eect typically seen in the RTM process. A ow control system for VARTM processes has been proposed by Johnson et al. [2]. Here, the localized ow rate is changed by heating of the resin via induction changing the local viscosity of the resin. This system is limited to resin systems with long gelation times at elevated temperatures to restrict any possibility of premature gelation. Other control of the ow position has been demonstrated by a technique developed by Nielsen et al. [35]. Here, the optimum number of open injection gates and the correct ow rates are calculated using a simulated annealing optimizer connected to a neural network-based process simulator. On-line estimation of the preform permeability further improves the controller. Patel et al. [68] showed that the control of the resin ow

0266-3538/$ - see front matter 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.compscitech.2005.12.008

2266

D. Bender et al. / Composites Science and Technology 66 (2006) 22652271

rate during RTM processing can improve overall void content. Flow rates were reduced and kept constant to decrease the in-plane ow velocity allowing an increase in wetting time of the ber bundles. These control examples improve the yield or performance of the part and could have direct application in a VARTM process if ow rate control could be implemented in this process. Flow rate control in the RTM techniques is typically achieved by using a piston-based injector which adjusts the displacement velocity of the piston. Another approach controls the inlet pressure to regulate the pressure dierence between inlet and outlet and thus the ow rate into the component. During VARTM infusion the vent pressure is to be kept at full vacuum to minimize any void formation in the composite part. This study proposes to add a pressure vessel into the system evacuating the resin supply bucket and allowing control of the pressure dierence between the injection and vent location. A fuzzy-logic based controller has been implemented using feedback from ow position sensors or infused resin weight to adjust the pressure dierence. Other approaches such as PID and model-based controller were considered during the design phase. Model-based control would require a good understanding of the system which at the present form for complex geometries is only available in FE-form. The execution time for these model implementations is prohibitive and thus was not implemented. PID controllers are inherently applicable in linear systems but have limitations in multi-variable systems without continuous feedback. Scale-up to larger and complex geometries would be easily implemented in fuzzy control by adjusting the fuzzy set but may require more complex changes in the other control approaches. The fuzzy logic ow rate control system has been validated using virtual simulation as well as experiments allowing the implementation of VARTM control scenarios similar to the RTM process control examples. 2. Description of the VARTM process VARTM is a composite manufacturing process to produce high-quality large-scale components. Multiple layers

of bers are laid up in a one-sided mold and impregnated with liquid resin using vacuum pressure only (Fig. 1). Curing of the resin and nally demolding of the composite part follow the impregnation process. Often, a high permeability layer (distribution media) is placed on the surface to accelerate the resin ow. Here, the resin ows preferentially across the distribution media and then through the preform thickness. The proposed control system regulates the injection bucket pressure between zero and one atmosphere and the resulting changes in pressure dierence between the injection and vent gate allows control of the ow rate into the reinforcement. The ow rate controller performance is limited by the maximum ow rate achievable with this setup. Filling the mold under constant pressure leads to the well-known 1D Darcys ow condition [9] (Eq. (1)): Q K Dp 1 1 a g x x 1

with Q being the ow rate, K the eective preform permeability, g the resin viscosity, Dp the pressure dierence between vent and infusion line, A is the cross-sectional area and x is the ow position. The material and pressure parameters can be summarized in a* and the ow rate is proportional to 1/x as shown in Fig. 2. Recent research has shown that the permeability of the preform during a typical VARTM application is dynamically changing as a function of location due to the pressure gradient in the preform [10,11]. Nevertheless, Correia et al. [11] showed that the eective permeability of the preform is constant for a constant pressure dierence in the preform and thus Eq. (1) can be applied. Two scenarios for the ow rate controller are observed. Constant ow rate for the complete part length is achievable when the ow rate at the end of the preform under one atmospheric pressure dierence (Q1atm) is larger than the desired ow rate. Partial ow control occurs if Q1atm decreases below the desired ow rate during infusion. The challenge is to adjust the pressure dierence relative to the desired ow rate and ow front position. Here, a

Fig. 1. Schematic of the VARTM setup with Distribution Media.

D. Bender et al. / Composites Science and Technology 66 (2006) 22652271


Controllability 10 Desired Flow Rate

2267

a*=1.00E-04 1 a*=1.00E-03 a*=1.00E-02

0.1

0.01 0.0 0.1 0.2 0.3 0.4 0.5

Infustion Length [m]


Fig. 2. The desired ow rate has to be above Q1atm to enable ow rate control. Q1atm falls below 0.04 cm/s for a* = 0.0001 m/s2 at x = 0.18 m making ow rate control impossible.

fuzzy-logic controller implementation is proposed capable of regulating the process over a wide range of material and process variables. 3. Fuzzy logic controller implementation

to a membership probability. Three linguistic terms for the antecedence (see Fig. 3) have been dened in the VARTM control case: negative error, almost zero error and positive error. The error signal is correlated to the three linguistic terms of the consequence membership: decrease pressure, do nothing and increase pressure. Through the center of gravity function a pressure change is being calculated and provided to the pressure regulator. The fuzzy logic controller has been implemented in the VARTM system as shown in Fig. 4. In each cycle a pressure change is calculated by the controller and applied via the pressure regulator. The ow advances and the feedback system obtains the ow position or resin weight change in the injection bucket. An error is calculated based on the sensor feedback and the desired set point of the infusion system. The error is forwarded to the fuzzy controller and the cycle repeats until the part is completely infused. The system has been implemented in LabVIEW and allows feedback from both sensors as well as virtual simulation results. 4. Controller validation via virtual implementation

Zadeh developed the fuzzy set theory in 1965 [12]. Fuzzy logic or rather fuzzy control is commonly used to control machines, robots and household appliances [1315]. The VARTM fuzzy controller assigns the input signal (error)
ANTECED ENCE
neg. error almost zero 1
X

Flow Rate [cm/s]

The performance of the controller has been evaluated using the Liquid Injection Molding Simulation (LIMS) developed at the University of DelawareCenter for
CONSEQUENCE
decrease do nothing 1
X

pos. error

increase

membership

membership

Normalized Error

INPUT SIGNAL

CONTROL SIGNAL

0 delta P center of gravity

Fig. 3. Schematic of the membership function of the VARTM fuzzy logic Controller.

Fig. 4. Schematic of the overall control system.

2268

D. Bender et al. / Composites Science and Technology 66 (2006) 22652271


0 Pa starting pressure 100% Normalized Error 80% 60% 40% 20% 0% 1.00E-05 100 sensors 50 sensors 20 sensors 1.00E-04 1.00E-03 a* 1.00E-02 1.00E-01 1.00E+00

Composite Materials [16]. The simulation package has been developed for RTM and VARTM applications and has been successfully applied to large-scale composite applications [17]. An one-dimensional nite element model with 100 elements has been designed to simulate a 0.5 m long VARTM part. Each element acts as a sensor point and the model calculates the arrival time of the resin for each control iteration with an updated pressure dierence set point. The parametric study is varying (1) material parameters such as the fabric permeability and resin viscosity, (2) the fuzzy set parameter, (3) number of feedback iterations and (4) initial pressure values to optimize the controller performance and to understand limitations of the system. Fig. 5 illustrates the LIMS simulation for a controlled and uncontrolled case. The ow lines indicate resin position after 60 s intervals. The ow progression distance is varying in the uncontrolled case whereas the fuzzy-controlled example shows constant ow progress. An error is calculated by comparing the position of the constant ow rate set point (xconstant) versus the actual position of the controlled VARTM process (xcontrol) (see Fig. 4). The controller error can be normalized (Eq. 2) by dividing it with the error of the uncontrolled (constant 1 atmosphere pressure dierence) baseline VARTM case. R tfill xcontrol;t jdt jx 0 constant;t 2 Eabsolute tR tfill x xcontrol;t jdt t0 j 1atm;t The controller performance was evaluated for a variety of material parameters. Fig. 6 shows the absolute error of the fuzzy controller versus the material parameter a* (with Dp = 105 Pa), a desired ow rate of 2.4 cm/min, an initial pressure dierence of 1000 Pa at the start of the infusion and 20, 50, and 100 sensor iterations. For 100 sensor iterations and a* below approximately 0.0001 m2/s the ow rate controller does not function because the desired ow rate for most of the infusion is above Q1atm. Here, for lower a* the absolute error is increasing rapidly to 100% because the initial pressure set point of 1000 Pa does not maximize the required ow rate at the beginning of the infusion. A transition region above 0.0001 m2/s indicates ow control possible at the infusion start. The ow controller operates successfully until 0.01 m2/s when Q1atm becomes much larger than the desired ow rate and the controller accuracy (pressure set points where changed in 1000 Pa multiples to take into account experimental limitation of the pressure set points) limits the controller performance. Nevertheless, the controller reduces the absolute error below 5% over a

Fig. 6. Normalized error as a function of material parameter a*.

wide range of permeability and viscosity values. Another important control parameter is the number of nodes being used as feedback sensors. A small number of sensors does not provide many control adjustments and thus increases overall error. The starting pressure at the beginning of the infusion greatly inuences the performance of the control system. Three dierent starting pressures (103 Pa, 0.5 105 Pa, 105 Pa) have been applied with the 100 sensor node setup discussed in the previous paragraph. It can be seen in Fig. 7 that a low starting pressure reduces overall error in almost all cases except for small a* where a high starting pressure is required to allow for maximum ow rate at the start of the infusion. Otherwise the initial starting pressure has to be small, because the ratio of achievable ow rate versus desired ow rate is large at the infusion beginning and thus the pressure dierence has to be minimized at this point.
100 sensors 200% 160% 120% 80% 40% 0% 1.00E-05 1000 Pa 50000 Pa 100000 Pa 1.00E-04 1.00E-03 a* 1.00E-02 1.00E-01 1.00E+00

Normalized Error

Fig. 7. Absolute error for dierent starting pressures.

Fig. 5. LIMS simulation a 0.5 m long part at: (a) constant atmospheric pressure and (b) controlled pressure.

D. Bender et al. / Composites Science and Technology 66 (2006) 22652271


100 sensors, 0 Pa starting pressure 100% Normalized Error 80% 60% 40% 20% 0% 1.00E-05 100% 75% 50% 25% 0%

2269

The virtual implementation of the controller allows the optimization of the fuzzy set and evaluation of the controller performance prior to experimental implementation. The system has shown to be able to control the desired ow for a wide range of material values but requires a substantial amount of sensor feedback. The starting pressure is an important factor in the performance of the controller and is set to close to zero atmospheres to allow gradual control of the ow rate. 5. Experimental validation

1.00E-04

1.00E-03 a*

1.00E-02

1.00E-01

1.00E+00

Fig. 8. Changing the membership function shape can optimize controller performance.

The fuzzy set structure can have an important eect on the performance of the control system. The membership function shape can be complex and the number of membership functions can be varying as well. The following study varies the base width of the triangular shape antecedence membership function and evaluates the controller performance. Five dierent widths are being examined: 0% base width results in a Boolean operator of the input parameters and 25%, 50%, 75% and 100% are intermediate values with the negative and positive error functions having triangular shape (Fig. 8). The compression of the antecedences membership function reduces the overall error. However, using the Boolean logic for the 0% case increases the error and reduces overall stability of the control system. The Boolean function changes the pressure changes drastically even for small changes in the error signal. Fig. 9 shows the nal controller performance for three dierent a* material values: 0.0008 (200 mPas in experimental validation section), 0.00046 (350 mPa s), and 0.00032 (500 mPa s respectively. The ow position error is below 2% at all times. The pressure dierences increase from 1000 Pa (starting value of every control setup) to 22, 43, and 62 kPa reecting the required pressure increase to maintain the ow rate at the end of the experiment.

The experimental setup uses a computer-controlled high-precision scale (HP-20K from Precision Weighting Balances) as a feedback sensor. Every ten seconds the weight was updated and the controller executed. The vacuum is controlled by a pressure controller (TESCOM ER3000SI-1) connected to a venturi system (Vacoon HVP 200) with feedback from a vacuum sensor (Vacoon VSSA-18). The 25% triangular fuzzy set described in the previous paragraph is used in all experiments. Two layers of random mat were used as the preform with a total length of 50 cm. The fabric was bagged and infused using a mix of corn syrup and water with a viscosity in the range of typical VARTM infusible vinyl-ester and epoxy systems (100 500 cps). The viscosity of a resin sample was monitored with a viscometer (LVDVE115, Brookeld Engineering Laboratories, Inc.) until the preform was fully impregnated and did not signicantly change during the experiments. Three separate experiments were conducted to validate the controller performance. The resin viscosity was varied to 200, 350, and 500 mPas. The permeability of the preform was independently measured to be approximately 6.4 1010 m2 at 40% ber volume fraction under 105 Pa compaction pressure. The material values chosen resulted into an approximate a* of 0.0008 m2/s, 0.00046 m2/s, and 0.00032 m2/s respectively for the average permeability value. The ow rate was set to 18 gr/min to infuse the component in approximately 2025 min for all scenarios and to obtain similar ow rates as used in the virtual simulation experiments without relying on integrated or tool-mounted

Flow Location versus Time 50 45 40 35 30 25 20 15 10 5 0 0 200 400 600 800 Time [s]
70000 60000 Pressure [Pa] 50000 40000 30000 20000 10000 0

Pressure versus Location

500 mPas 350 mPas 200 mPas

Location [cm]

500 mPas 350 mPas 200 mPas Desired

1000

1200

10

20 30 Location [cm]

40

50

Fig. 9. (a) Arrival time plot shows very small dierences between desired and controlled ow location and (b) pressure plot shows gradual increase in pressure dierence.

2270

D. Bender et al. / Composites Science and Technology 66 (2006) 22652271

Fig. 10. Flow front position marked every 90 s for the (a) uncontrolled case and (b) controlled case.

Used Resin versus Time 450 400 Used Resin [gr] 350 300 250 200 150 100 50 0 0 500 Time [s] 1000 1500 500 mPas 350 mPas 200 mPas Desired Pressure [Pa] 90000 75000 60000 45000 30000 15000 0 0 500 Time [s] 1000 1500
500 mPas 350 mPas 200 mPas

Pressure versus Time

Fig. 11. (a) Arrival time plot shows very small dierences between desired and controlled ow location and (b) pressure plot shows gradual increase in pressure dierence.

ow position sensors. Fig. 10 shows the arrival time location marked at 90 s intervals on the bag for both the controlled and uncontrolled case for Case I (a* = 0.0008 m2/s). It can be clearly seen that the ow rate controller accurately regulates the pressure to maintain a constant resin ow rate into the part over a period of approximately 25 min. Fig. 11 illustrates the sensor feedback of the infused resin weight and the pressure sensor data as a function of infusion time. The tube was fully lled prior to infusion and when opened allowed ow into the reinforcement. Initially, the pressure data and weight data show a transition region up to approximately 250 s where the controller reaches a steady-state behavior. The experimental setup spreads the resin on the preform using a perforated infusion tube to quickly obtain 1-D ow conditions. This tube creates a disturbance but once lled does not aect subsequent resin ow. Then, the ow rate controller works as predicted by adjusting the pressure dierence and a maximum error of 5 gr between the desired and infused resin weight is measured. The pressure in the infusion bucket increases throughout the complete infusion. The nal pressure value depends on the viscosity and increases for the

higher viscosity uids. The predicted nal pressure values using the LIMS simulation are lower compared to the experimental obtained values which could be explained due to lower permeability of the fabric, pipe losses and gravitational eects of the height of the injection bucket compared to the preform level. Overall, good ow control has been accomplished and the virtual simulation results have been validated. 6. Conclusions An experimental setup for ow rate control of the VARTM process has been demonstrated. The ow rate controller evacuates the infusion bucket and implements pressure control to adjust the pressure dierence between the injection and vent location. A fuzzy logic controller has been implemented to adjust the pressure of the injection bucket and the ow rate of the VARTM process. The system has been optimized oine using virtual ow simulation (LIMS) and physical limits of such a system have been calculated. The ow rate controller successfully regulates constant ow rate for a wide range of permeability and viscosity values.

D. Bender et al. / Composites Science and Technology 66 (2006) 22652271

2271

Experimentally, the ow rate controller has been validated for a random mat preform and three dierent uid viscosities. The experimental setup shows similar pressure history compared to the virtual simulation and accurate control of the ow position. The experimental validation in this paper shows that the control system can handle any of the process induced variations and thus demonstrates the robustness of the fuzzy-logic approach. Acknowledgements The authors would like to thank Pavel Simacek and Ajit R. Nalla for their LIMS support and gratefully acknowledge the assistance of the Oce of Naval Research through the National Testbed for Intelligent VARTM Processing of Polymer-Matrix Composites (N00014-99-1-0636) and Intelligent VARTM Processing of Composite Structures (N00014-00-C-0333). References
[1] Lawrence JM, Advani SG. Use of sensors and actuators to address ow disturbances during the resin transfer molding process. Polym Compos 2003;24(2):23748. [2] Johnson RJ, Pitchumani R. Active ow control in a VARTM Process using localized induction heating. In: Proceedings of the FPMC7, July 79, 2004, p. 24752. [3] Nielsen D, Pitchumani R. Closed-loop ow control in resin transfer molding using real-time numerical process simulations. Compos Sci Technol 2002;62(2):28398. [4] Nielsen D, Pitchumani R. Control of ow in resin transfer molding with real-time preform permeability estimation. Polym Compos 2002;23(6):1087110.

[5] Nielsen D, Pitchumani R. Intelligent model-based control of preform permeation in liquid composite molding processes with online optimization. ComposPart A: Appl Sci Manufact 2001;32(12): 17891803. [6] Patel N, Rohatgi V, Lee L James. Micro scale ow behavior and void formation mechanism during impregnation through a unidirectional stitched berglass mat. Polym Eng Sci 1995;35(10):83751. [7] Patel N, Rohatgi V, Lee L James. Experimental investigation of owinduced microvoids during impregnation of unidirectional stitched berglass mat. Polym Compos 1996;17(2):16170. [8] Patel N, Rohatgi V, Lee L James. Inuence of processing and material variables on resin-ber interface in liquid composite molding. Polym Compos 1993;14(2):16172. [9] Advani SG, Sozer E Murat. Process Modeling in Composites Manufacturing. Elsevier Marcel Dekker Inc.; 2003. p.151. [10] Lopatnikov S, Simacek P, Gillespie Jr J, Advani SG. A closed form solution to describe infusion of resin under vacuum in deformable brous porous media. Model Simulat Mater Sci Eng 2004;12(3): S191S204. [11] Correia NC, Robltaille F, Long AC, Rudd CD, Simacek P, Advani SG. Use of resin transfer molding simulation to predict ow saturation and compaction in the VARTM process. J Fluid Eng, Trans ASME, 2004;126(2):2105. [12] Zadeh LA. Fuzzy Sets. Inform Contr 1965;8:33858. [13] Wikipedia, die freie Enzyklopa die. Available from: http://de.wikipedia.org/wiki/Fuzzy_Logik. [14] Friedrich Alfred. Logik und Fuzzy-Logik: Eine leichtversta ndliche Einfu hrung mit Beispielen aus Technik und Wirtschaft. RenningenMalmsheim, Germany: Expert Verlag; 1997. [15] Jaanineh Georg, Maijohann Markus. Fuzzy Logik und Fuzzy Control, 1. Auage, Wu rzburg, Germany: Vogel Verlag; 1996. [16] Simacek P, Sozer EM, Advani SG. User Manual for LIMS 4.0, (Center for Composite Materials, University of Delaware, 1998). [17] Ali Gokce, Pavel Simacek, Hope Deor, Suresh Advani, Dirk Heider. Risk reduction via virtual ow simulation in manufacturing sandwich panels using VARTM process. In: Proceedings SAMPE 2005, Anaheim, CA.

You might also like