Mar., 02 /A novel system is described that forms three-dimensional (3D) molded nonwoven structures through proper integration of a laboratory scale meltblown unit with a small die and a six-axis robot. The 3D fiberweb structures can be formed by deposition of fibers from the die of the meltblown unit, which is manipulated by the robot, on any desired 3D mold. The mold rotational and surface speeds can be controlled by an additional external axis. The die is connected by two flexible hoses to the melt extruder of the meltblown unit and a hot air supply system. This system directly sprays fibers onto a 3D mannequin mold to produce structures from polypropylene polymers. With varying degrees of success, several robot manipulation algorithms of fiber deposition on the mold are developed to accurately control the basis weight uniformity of the fiberwebs. A rule-based control algorithm using a linear variable differential transducer to map the mold contour results in the greatest fiberweb basis weight uniformity.

Meltblowing is a technology that produces extremely lightweight fiberwebs directly from polymers, and it is currently used in manufacturing two-dimensional (2D) structures. In this technology, molten polymer is extruded through a set of orifices in a knife- edge die [22]. The die is jacketed on both sides by a supply of high- - speed hot air. The interaction of the laminar sheets of hot air with the polymer streams from the orifices provides elongated continuous or discrete fibers that are deposited on a collecting surface (a rotating drum or a moving belt). Because the meltblowing process employs high-- velocity air to impinge upon the polymer as it exits the orifices, it elongates the polymer strands from 500 micrometer ((mu)m) diameter to diameters as small as 0. 1(mu)m. Extreme entanglement of fibers. characterizing meltblown fibrous webs, produces coherent fiberwebs. The entanglement of these long fibers makes it impossible to remove one fiber from the web or to trace one fiber from beginning to end [3]. The density of the web is such that it can contain and retain particulate matter [8]. Unique micrometer characteristics of meltblown structures produce a high surface area per unit weight and fine porosity. Such structures are lightweight, and they display high insulating values and excellent filtering characteristics [1, 16]. Meltblown technology can be used to produce efficient filter materials that are able to filter particles as small as 0.5 (mu)m [17]. Overall, meltblown webs are characterized by their softness, bulk, absorbency, low porosity, and poor abrasion resistance [2].

Because meltblowing uses an attenuating air stream to draw and orient the fibers, the distance between the polymer orifice and the collecting surface influences the fiber characteristics and resulting fabric properties. This distance is commonly referred to as the die-to-collector distance (DCD). The period of time that forming fibers spend in flight before being collected influences fiber orientation, strength, and surface properties [9]. Appropriately, a varying DCD permits the production of structures with varying properties from stiff and brittle at close DCD to soft and bulky at large DCD. Information on material properties is available in the literature [7-9, 11, 12, 15-20].

One of the most important issues in producing protective garments is fabrication with minimal seaming or joining, which is the "weak link" in a protective system. The larger the number of seams and joins, the greater the potential the garment will split in pieces, which can be disastrous for users needing protection. Since the current practice of the technology is to manufacture 2D fiber-- webs that are used as components for constructing protective clothes, it is impossible to form such clothes without joins and/or seams. This drawback has motivated our team to undertake research on producing 3D seamless meltblown structures to be used as components of protective garment systems.

With this in mind, the idea of integrating meltblowing technology and robotics to produce such structures was born. In such integration, the fiber collector is a mold. Continuously rotating the mold further provides necessary manipulation in order to form fiberweb structures to cover the mold's entire surface uniformly. Additionally, if the mold has 3D features, these will be transposed into the web structures formed on it. Using a comparatively small die limits the width and stops the die from over-- shooting the fibers to undesired areas. Appropriately choosing a die width is necessary, and that width should be much smaller than the size of the mold on which the fibers are sprayed.

Integrating Meltblown and Robotic Technologies

To produce seamless 3D meltblown nonwovens, we have developed a robotic fiber assembly and control system (MACS) comprising a model meltblown unit with a 7.8 cm wide die with a total of ninety orifices, a commercial ABB six degrees of freedom (or six-axis) robot, an additional external seventh axis, and a controller. Figures 1 and 2 illustrate the system. The die of the meltblown unit is housed in a cage that is controlled by the robot manipulator. The meltblowing die is connected to two flexible hoses, one connected to the melt extruder of the meltblown unit to supply the molten polymer to the orifices of the die, and the second connected to the high- speed hot air supply. The seventh axis is used to rotate any desired 3D mold around a vertical axis at programmed rotational and surface speeds. The seventh axis may also be used to power a traditional drum or belt to manufacture 2D structures.

The separate seventh axis is linked and interfaced with the manipulator of the robot. The rotational speed of the seventh axis and the motion of the manipulator can be programmed using ABB's RAPID programming language. The six degrees of freedom of the manipulator and rotation of the seventh axis combined with the control system provide the necessary motion to deposit the extruded fibers on the 3D mold to cover its entire surface. Figure 3 shows four rotational positions of a 3D mold with uniform elliptical cross section during fiber deposition on the mold. The top of the figure shows the top view, and the bottom shows the front view of the mold and the die. In this example, the die starts depositing the fiber stream on the bottom end of the mold. The die is moved at a controlled speed upward while the mold is rotated as shown. The mold will be covered with one layer of fiberweb when the die reaches the top of the mold. Then the process may cease or continue depending on the fiberweb basis weight required. A collapsible mold covered with cotton knit fabric is used to maintain the integrity of the meltblown structures and facilitate removing the fabric from the mold (Figure 2).

It is obvious that the die-to-collector distance, the angle of fiber deposition with respect to the axis of rotation of the mold, the basis weight of the fiberweb, and hence the fiberweb properties (fiber orientation, pore size, stiffness, etc.) can be controlled by our novel system with a high degree of precision through programming. Additional details on REACS components and design are provided elsewhere [4].

Control Algorithms of Fiber Deposition

To control the basis weight uniformity of the fiberwebs, we developed two control models of fiber deposition over the fiber collector (3D mold). The first is a two-variable (2V) model that adjusts for DCD and the height of the tool (the meltblowing die) on the mold. The second is a three-variable (3V) model that additionally adjusts tool orientation such that it aligns the row of polymer orifices of the die parallel to the mold's surface, in order to accommodate any curvature changes in a given mold. To develop contour-following algorithms for mold surface shapes using either the 2V or 3V model involves determining point coordinates on the mold shape at regular rotational increments. To determine these points in a 3D space for the 2V model, we constructed a pointer such that the pointer tip assumes the virtual position of the right most orifice in the meltblowing die body. For the 3V model. we constructed a pointer mirroring the positions of the polymer orifices. This pointer is used to mark points on a mold relative to the universal coordinate system (three orthogonal x-, y-, and z- axes). We adopted this procedure as the means of developing the position and speed of the die to the rotation of the mold.

TRIANGULATION FOR ROBOT TOOL MOVEMENT

We used point coordinates obtained for the two models to determine tool motion parameters, i.e., for correlating tool motion to mold geometry. The model employed for tool speed control is that of triangulation, which is the task of finding the intersection of two lines in a space, and it is widely used in computer vision applications to find the position of a point in space given its position in two images [6, 13]. Triangulation here uses coordinates from the universal coordinate system.

The difference of two adjacent points in the x-coordinate is used to calculate the x-displacement. The z- and y-displacements are then used to provide the additional coordinates in order to compute the appropriate 2V or 3V triangulation values. The Pyt\hagorean relation (c^sup 2^ = a^sup 2^ + b^sup 2^) is used in two dimensions [5]. Calculating the triangulation in three dimensions (3V) involves the rectangular Cartesian coordinates x, y, z and position vectors r = (x, y, z) to determine the length for the new robot path [10].

By appropriately using these trivial triangulation calculations, we can determine the hypotenuse, which represents the intersection or linear path the robot tool (die) must follow in order to reach the adjacent or next marked point on the mannequin. The time period in which the robot tool makes this move is then stipulated depending on the mold's 3D shape and rotational speed, and the respective tool speed for the motion along the hypotenuse is determined.

MOLD ROTATIONAL SPEED MANIPULATION

Previously, using a set of pre-defined experiments (not reported here), we determined the fiber flaring profile of the die, which helped us to determine the range of rotational speeds used in this research. This is because the height and width of the fiber flaring profile provide a uniform layering of the fibrous material, and the profile also helps form a coherent web structure when the seventh axis is rotated at a controlled speed. Here, the time intervals are those that rotate the mold relative to the vertical tool displacement. In experiments, we found that the orientation of the tool played a critical role in the quality of the 3D fabric structures produced. Subsequently, we also considered the orientation of the die axis in developing the 3V model, using a rule- based control algorithm to be described later in the Experimental Details section.

DETERMINING MOLD GEOMETRY

For the 2V model, twenty-four points were marked around the waist of the mannequin at 150 intervals, giving one complete rotation (24 X 15 deg = 360 deg). For the 3V case, four positions were marked, one for front and back center, one each for .45 deg off-center, and one normal to the shoulder position. This translated to four positions per 180 deg. or eight per single rotation. We assumed the uniformity of the structure was sufficient along the complete height of the mannequin. The rotational speed for the mannequin was initially set at 50 deg/ second, resulting in 7.2 seconds per revolution. The fiber flaring profile experiment showed that when spraying the mannequin mold, a constant vertical tool speed of 15 mm/ revolution in the -direction was appropriate. This arrangement provided uniform spiral layering on the mannequin and did not produce fabric stripes. Linear point-to-point displacements of the tool were determined with the triangulation technique described earlier. Each displacement was programmed to be completed in 0.3 second (i.e., 7.2 seconds per twenty-four points or revolution) for the 2V model, or as appropriately determined for the 3V case. The nature of the displacements stayed the same for each rotation thereafter.

Before we could develop a rule-based control, we had to complete initial evaluations to provide the necessary knowledge base. These initial evaluations did not take into account tool orientation relative to the mold's forming surface, and those experiments are described later.

Experimental

Since this research deals with the formation of 3D structures on the mold, we conducted experiments to develop appropriate control algorithms in order to dispense a uniform fiber distribution onto the mold body. We evaluated the quality of fiber distribution by testing the fabrics formed for their basis weight (g/m^sup 2^) uniformity and distribution on the mold.

Polypropylene (PP) meltblown samples were made from a nominal 1200 melt flow rate (MFR) PP resin. The meltblowing process conditions were 249 deg C melt hose temperature, 316 deg C air hose temperature, 277 deg C die temperature, 0.04 g/min/hole throughput, 1.4 bar air pressure, and 20 cm DCD.

DEVELOPMENT OF 2V MODEL

For the 2V model, five motion patterns were evaluated and controlled by five control programs adjusting either DCD or mold rotational speed or both. They included no correction, DCD correction (keeping constant distance, x-coordinate, from the surface of the mold), linear rotation speed correction (linearly incrementing rotational speed of the mold), DCD correction and linear rotationspeed correction, and DCD correction and nonlinear rotation-speed correction. Figure 3 illustrates the basic positioning steps. The particulars of each scenario are given below.

No correction: The tool was moved at 2 mm/second only in the z- direction, and the external axis was rotated at 50 deg /second.

DcD and rotation-speed correction: In this case the tool was programmed to move at 2 mm/second in the z-- direction, and the x- coordinate was adjusted in the range 3.95-59.4 mm/second to keep the die at the same distance from the mold surface based on the mannequin dimensions. Again, the rotational speed of the external axis was 50 deg/second. The rotation-speed correction was programmed for two different models. A linear correction was programmed using linear speed increases and decreases. The linear model assumed that surface speeds doubled from navel to shoulder on the mannequin. This assumption was based on actual mannequin dimensions (width shoulder to shoulder, 43 cm, depth front to back, 20 cm). Subsequently, rotational speeds were highest for middle front and back, and lowest for left and right shoulder positions. Requiring a single total rotation time of 7.2 seconds when rotating at 50 deg /second, speeds were incremented in steps of 5 deg /second around the initial 50 deg/ second speed setting. Therefore the linear rotationspeed model ranged from 35 deg to 65 deg/second.

The nonlinear correction of the rotational speeds was determined from the oval dimensions of the mannequin mold. Diameters were measured every 15 deg, and surface speeds were calculated assuming a perfectly circular structure at each point in question. The direction of rotation was also adjusted for, displacing rotational speeds by three steps forward for the downward motion. This was done to correct for the 8 cm fiber spray width pattern, which reverses for the downward motion. Instead of starting speed steps for the external axis at 65 deg/second, 56.62 deg /second was used for the downward motion, and continuing decrements in the speeds involved the same steps as before. The programming code for five-motion control is given by Farer [4].

DEVELOPMENT OF 3V RULE-BASED CONTROL ALGORITHM

In order to develop a simple rule-based control algorithm, we integrated a linear variable differential transducer (LVDT) into the system. The LVDT was mounted on the support frame structure of the external axis, and a wheel was attached to the end of the displacement rod. The wheel of the LVDT was in direct contact with the surface of the mold. Thus the rotation of the mold caused the rod to have translational displacements. The data from the LVDT were sent to the robot controller as voltage outputs ranging from -5 to +5 volts.

General geometric principles were used to establish four positions, or tool orientations, respective to an angular position of the mannequin mold. These positions were developed so that the row of polymer orifices spraying the meltblown fibers would be aligned approximately parallel to the surface of the continuously rotating mannequin. Figure 4 illustrates the four-position mechanism. The control rules were established by using the data being gathered from the LVDT. We adopted a qualitative approach with three curvatures (shallow, medium, and deep) to correspond with the variation of the mold curvature that depends on which part of the torso the LVDT is on at any instant. Figure 5 illustrates the control rules (tool orientation) along with the three qualitative curvatures we adopted. Since the rate of changing displacement indicated which position of the torso was being measured (e.g., the curvature of the front part of the torso is shallow), the meltblowing die orientation could be adjusted accordingly (Figure 4). The programming code for the rule-based control algorithm is given by Farer [4]. Adopting the rule-based control permitted significant reduction of programming code, since the number of tool orientation positions is only eight per one revolution of the mold (versus twenty-four in the case of 2V).

SAMPLING AND TESTING

The RFACS was used to form one-layer and two-layer structures on the mannequin mold (Figure 2) at time intervals corresponding to forty and eighty rotations of the external axis as described earlier. In all, eight different structures were formed with the different control algorithms. Upon fiber collection, a fabric was cut along the middle of the back and removed from the mold. Each collected fabric was divided into 24 columns parallel to the mold axis. Thus the center of each column was at an increment of the 15 deg angular positions measured around the axis of the mold. Each sample was identified by a number from 1-24 where the sample at the front center of the mold was number 1. Four circular samples of 3.5 cm diameter were taken from each fabric using a die cut, then weighed on a digital scale of 0.0001 g accuracy. The results of the basis weights are reported in Figures 6-8 using the sample identification numbers on the horizontal axis to illustrate the angular basis weight distributions of the eight fabrics.

Results and Discussion

The basis weight distributions of the twenty- four samples using the five patterns of motion variation for the 2V control algorithm mentioned in the experimental section. The differences in basis weight distributions are related to the interaction of the geometric features of the mold with the characteristics of the fiber flaring profile and the tool orientation. Note that the highest basis weights of the fabrics of Figures 6 and 7 are at positions I and 12 (front and back center of the mold). The basis w\eight coefficient of variation (cv) values for the twenty-four samples of the fabric formed with the no- correction model were 8% higher than cv values for the fabric obtained with the linear correction model. The results further indicate that the DCD correction caused a slight decrease in fiberweb basis weight uniformity compared to the fabric formed with the no-correction control algorithm (Figure 5)While fabric basis weight uniformity was improved by implementing the nonlinear correction model compared to fabric formed with no correction, the improvement was only 2% in cv values, as we see from the results of two-layer structures in Figures 6 and 7.

The difficulty in achieving more uniform basis weight distributions is inherently related to the die orientation during production. When the die is not reoriented in relation to the surface of the mold during fiber application, fiber overshoot occurs, causing more fibers to collect in the front and back center regions (samples I and 12) of the mold. This is obvious by observing positions 45 deg and 135 deg of Figure 3. When the mannequin turns from front to side orientation. the 8 cm wide fiber stream will continue to spray fibers into the center section of the mannequin, even as it is turning away. During this motion, the orientation of the surface of the mannequin in relation to the tool quickly changes from normal to almost parallel, causing some of the left of the fiber stream during upward motion and some of the right of the fiber stream during downward motion to spray at steep angles onto the surface of the mannequin. This causes more fibers to collect in the front and back center regions (samples I and 12) of the mannequin. Fiber overshoot caused the higher observed weights in regions 10-14 and 22-23 of Figures 6 and 7.

Through implementation of 3D triangulation positioning and feedback control of the mold positioning data with the LVDT, overall control performance improved. Using the rule-based control resulted in the lowest overall cv values (11%) for fabric basis weight uniformity, as we see from the results of Figure 8. The improvement is related to significant reduction of fiber overshoot, as we see in positions 2 and 4 of Figure 4. Further improvement is obviously possible if a smaller size die is used.

Conclusions

We have successfully developed a novel robotic fiber assembly and control system (RFACS) to form 3D fiberwebs on a mannequin mold. RFACS can be implemented for production of 3D seamless meltblown nonwovens to be used as components of protective garments. The setup developed here allows for arbitrary positioning of the meltblowing die in an initial reference frame to an arbitrary collecting surface without difficulty and with an accuracy of 0.1 mm.

Experiments have demonstrated that the rotational motion of a mold needs to be varied according to its three-dimensional shape- and characteristics-when uniform basis weight distribution is required. Various motion correction models (i.e., linear and nonlinear) may be successfully employed to some extent. Robot motion and inherent internal speed averaging must be considered carefully to achieve desirable results. When spraying a mold, it is desirable to have the die aligned parallel to the surface of the mold. Such orientation permits more uniform fiber application to the mold surface and results in highly significant improvement in the basis weight uniformity of meltblown fiberwebs.

We successfully implemented a rule-based control algorithm that compensates for the mold position and speed as well as the tool position. The approach improves basis weight uniformity and significantly simplifies the robot programming operation. It is obvious that additional improvement can be achieved if a smaller size die is employed to reduce the overshoot of fibers to the center of the mold.

AcKNOWLEDGMENTS

This research was supported in part by an ARO-MURI grant from the Army Research Office, and in part by support from the Nonwovens Cooperative Research Center, North Carolina State University. We gratefully acknowledge their generous support of this project.

Literature Cited

1. Bergmann, L., Microfiber and Electrostatically Charged Nonwovens in Filtration, Nonwovens Ind. 22 (2), 28-34 (1991).

2. Currie, K. J. G., and Mogel, P. J., Apertured Abrasive Absorbent Composite Nonwoven Web, U.S. patent 5,429,854, 1995.

3. Eian, G., Durable Melt-blown Fibrous Sheet Material, U.S. patent 4,681,801, 1987.

4. Farer, R., Formation of 3D Meltblown Structures Using Robotic Control of Fiber Deposition, Doctoral thesis, NC State University, 1999.

5. Fogiel, M., "Handbook of Mathematical, Scientific, and Engineering Formulas, Tables, Functions, Graphs, Transforms," Research and Education Association, NY, 1984, p. 136.

6. Hartley, R. L, and Sturm, P., Triangulation, Comput. Vision Image Understand. 68 (2), 146-157 (1997).

7. Hodgsen, W. J., Jr., Melt Blown Nonwoven Web from

Fiber Comprising an Elastomer, European patent application 88300241.2, 1988.

8. Insley, T., Sorbent Sheet Material, European patent application 0 156 649 A2, 1985.

9. Johnston, M., Resin to Web Process-Meltblowns, in "Proc. International Nonwoven Fabrics Conference," 1988. no. 67-68.

10. Kom, G. A., and Kom, T. M., "Mathematical Handbook for Scientists and Engineers-Definitions, Theorems, and Formulas for Reference and Review," McGraw-Hill, NY, 1961, pp. 56-57.

11. Lau, J. C., Meltblown Die and Method, U.S. patent 4,526,733, 1985.

12. Lau, J. C., Non-woven Web Formation, U.K. patent application GB 2 130 260, 1982.

13. Mayer, J. R., and Parker, G. A., A Portable Instrument for 3- D Dynamic Robot Measuring Using Triangulation and Laser Tracking, IEEE Trans. Robot. Automat. 10 (4), 504-- 516 (1994).

14. McAmish, L. H., Addy, T. O., and Lee, G. F., Nonwoven Medical Fabric, U.S. patent 4,622,259, 1986.

15. Milligan, M., Meltblown Fabric, U.S. patent 5,273,565, 1993.

16. Narasimham, K. M., and Shambaugh, R. L., The Process of Melt Blowing, in "Inch-Tec 1987-Book of Papers," 1987.

17. Nowicka, C., Rakowski, W., and Wcislo, P., Incorporation of Activated Carbon into Melt Blown Webs, Nonwovens Ind. 21(12), 50-58 (1995).

18. Okada, H., and Asano, S., Polyethylene Terephthalate Based Meltblown Nonwoven Fabric and Process for Producing the Same, U.S. patent 5,364,694, 1994.

19. Schwarz, E. C. A., Apparatus and Process for Melt-Blowing a Fiberforming Thermoplastic Polymer and Product Produced Thereby, U.S. patent 4,380,570, 1983.

20. Wadsworth, L. C., Novel Melt Blown Research Findings, Nonwovens Ind. 17 (11), 44-51 (1986).

21. Wente, Van A., Superfine Thermoplastic Fibers, Ind. Eng. Chem. 48 (8), 1342-1346 (1956).

22. Wisneski, T. J., and Morman, M. T., Polyolefin Containing Extrudable Compositions and Methods for Their Formation into Elastomeric Products Including Microfibers, U.S. patent 4,663,220, 1987.

Manuscript received September 26, 2000; accepted April 3, 2002.

RAOUL FARER1, ABDELFATTAH M. SEYAM. 1 TUSHAR K. GHOSH,1 SUBHASH K. BATRA,1 EDDIE GRANT,2 AND GORDON LEE2

North Carolina State University; Raleigh, North Carolina 27695, U.S.A.