a process for solvent welded rapid prototype tooling

Robotics and Computer Integrated Manufacturing 17 (2001) 151}157

A process for solvent welded rapid prototype tooling

Denis Cormier*, James TaylorDepartment of Industrial Engineering, North Carolina State University, Raleigh, NC 27695-7906, USA

Abstract

Rapid prototyping is widely seen as an e!ective tool for compressing time to market for new products. The typical process followedby industrial and mechanical design groups is to model a new product in a CAD system, rapidly prototype the component parts, usethe parts as patterns for RTV silicone molds, and then cast polyurethane prototype parts from the molds. These prototypecomponents are an integral part of the simultaneous engineering process. With prototype components, engineers are able to design,implement, test, and re"ne the assembly systems for a product while production tooling for the components is being made. In thispaper, we describe an experimental rapid prototyping process, known as solvent welding freeform fabrication (SWIFT), that is verywell suited to the production of short to medium run tooling. The advantage of the process is that it is very fast and inexpensiverelative to the traditional RTV silicone mold making process. The process also produces ABS or polystyrene molds which lastconsiderably longer than RTV silicone molds. Process development details are provided in the paper and its application to a powertool component is described. � 2001 Elsevier Science Ltd. All rights reserved.

Keywords: Rapid prototyping; Soft tooling; Solvent welding

1. Introduction

Rapid prototyping is widely seen as an e!ective toolfor compressing time to market for new products. Thetypical process followed by industrial and mechanicaldesign groups is to model a new product in a CADsystem, rapidly prototype the component parts, use theparts as patterns for RTV silicone molds, and then castpolyurethane prototype parts from the molds. Wax in-vestment casting patterns for metal parts may also beproduced from such molds.Functioning prototype components are an integral

part of the simultaneous engineering process. These com-ponents are particularly bene"cial to assembly systemsengineers. With prototype components, engineers areable to design, implement, test, and re"ne the assemblysystems for a product while production tooling for thecomponents is being made. The advantage of the conven-tional rapid prototyping and RTV mold making processis that since it can be performed in parallel with theproduction tool making process, the product's time-to-market can be signi"cantly compressed. In practice,

*Corresponding author. Tel.: #1-919-515-1549; fax: #1-919-515-1543.

E-mail address: [email protected] (D. Cormier).

however, the process is not quite as rapid as the namewould imply. Considerable expenses can also be incurred.This paper describes an experimental rapid proto-

typing process, known as solvent welding freeform fabri-cation (SWIFT), that is very well suited to the productionof short run tooling. The advantage of the process is thatit is very fast and inexpensive relative to the traditionalRTV silicone mold making process. The process pro-duces ABS or polystyrene molds which last considerablylonger than RTV silicone molds. This makes the processquite valuable for companies who would like to have testparts available during the design, implementation, andtesting of new manufacturing and assembly systems.

2. Related work

The traditional method for making multiple prototypeparts is to start by making a part pattern using one of thecommercially available rapid prototyping processes.Depending on part size and complexity as well as theprocess being used, the part can be prototyped in as littleas a day or as much as one to two weeks. Once thepattern is made, RTV silicone is often used to make singleor multiple stage #exible molds.With a single stage mold,the pattern is suspended in a box, and liquid RTV sili-cone is poured until it completely surrounds the pattern.

0736-5845/01/$ - see front matter � 2001 Elsevier Science Ltd. All rights reserved.PII: S 0 7 3 6 - 5 8 4 5 ( 0 0 ) 0 0 0 4 9 - 1

After the RTV silicone has cured (approximately 24h),a knife is used to cut a parting line into the soft mold.Because the part is not visible through the opaque RTVsilicone mold, it can be di$cult to know where to cut theparting line. Single stage molds are therefore used prim-arily with relatively simple parts. For more complexparts, a backing board is constructed along the part'sparting line. The RTV silicone is poured one section ata time and allowed to cure before the next section ispoured. Most parts can be built with a two-piece mold,so the mold making process often takes 2}3 days. Oncethe RTV silicone mold is built, a thermosetting resin suchas polyurethane is typically cast into the mold. One RTVsilicone mold will produce approximately 30 castings.Readers interested in papers which describe the processare referred to Dahl and Cleary [1].While polyurethane castings are the most common

application for RTV silicone molds, the use of RTVsilicone molds with injection molding machines has alsobeen investigated. Venus et al. [2] describe a series ofexperiments in which iron powder is added to the basicRTV silicone mixture during the mold making process.Results of experiments in which low-density polyethylene(LDPE), 30% glass-"lled polypropylene, and polyamideare injected into the inserts are then provided. Theauthors conclude that pure RTV tools su!er from longdemold times due to poor thermal conductivity, andpoor dimensional accuracy which results from the defor-mation of the soft silicone in reaction to the injectionpressure. Conversely, the addition of iron powder im-proves both demolding times and dimensional accuracy.Despite the popularity of the RTV silicone mold-

making process, one drawback of the process is that itrequires the time and expense of an additional step in theprototyping process. Each step along the way has thepotential to reduce dimensional accuracy of the "nishedpart. Some developers and researchers have, therefore,begun to producing tooling directly with rapid proto-typing machines. For example, Venus and van deCrommert [3] describe an application in which nyloninjection molding tools are built using the selective lasersintering (SLS) process. More speci"cally, a shell of eachtool half is made and is then back-"lled with steel-"lledepoxy. The steel powder in the epoxy backing increasesthermal conductivity of the tool and therefore decreasesdemolding times. Due to the fact that sintered materialsare characteristically porous, the authors coat the surfaceof the nylon tool face with Loctite and then sand thesurface smooth. The authors conclude that the resultingtools are able to withstand the temperatures reachedduring injection molding. No mention is made of thedegree to which the manual sealing and sanding processa!ects dimensional accuracy.Recent attempts to directly prototype the tool rather

than the pattern around which a tool is made showconsiderable promise in the sense that they reduce lead

time and typically improve dimensional accuracy. How-ever, rapid prototyping processes can be quite expensive.For example, the author's experiences with local stereo-lithography service bureaus suggests that the going rateto have a part prototyped on an SLA machine is between$50 and $70/h. Even for relatively small tools, this cantranslate into several thousand dollars for each mold. Iflot sizes of several thousand parts are required, then thisexpense may be perfectly reasonable. For assembly linedesign and testing purposes, however, an expense ofseveral thousand dollars per tool may not be acceptable.In these cases, a less expensive alternative would bedesirable.The remainder of this paper describes e!orts to

develop a fast and inexpensive alternative to RTV sili-cone tooling as well as conventional direct rapid proto-type tooling. The basis of these e!orts is an experimentalrapid prototyping process which is based upon solvent-welded thermoplastic sheets.

3. Solvent welded tooling

This section describes a SWIFT process [4], and itsapplication to rapid prototype tooling. The SWIFT pro-cess is based upon a cycle of solvent welding and com-puter numerically controlled (CNC) contour machiningwhich is repeated one sheet at a time until the desiredpart is complete. The process can be used with anythermoplastic material which is solvent weldable andwhich is available in sheet form. ABS and polystyrene aretwo widely used examples of suitable thermoplastics.Generally speaking, the rawmaterials are extremely inex-pensive compared with the materials used with mostcommercially available rapid prototyping processes. Forexample, a 4 ft�8 ft sheet of 0.060�� thick polystyrenecosts approximately $15 from a local supplier at thepresent time. This equates to a raw material cost of lessthan $0.06/in�. This is signi"cant, because it is severalorders of magnitude less than the raw material costs formost commercially available rapid prototyping pro-cesses. Furthermore, the thermoplastic sheets are com-monly available from any plastics supply house.

3.1. Details of the SWIFT process

A #ow chart of the SWIFT process is shown in Fig. 1.As is the case with most rapid prototyping processes, theSWIFT process builds parts one layer at a time. For eachlayer, a sheet is fed forward through a laser printer whichoptionally prints a high-density polyethylene (HDPE)image which serves as a solvent mask. A solvent mask isoccasionally needed to prevent a downward facing sur-face of a part from being welded to scrap material froma previous layer. HDPE is a thermoplastic materialwhich is not soluble in acetone, and which, therefore,

152 D. Cormier, J. Taylor / Robotics and Computer Integrated Manufacturing 17 (2001) 151}157

Fig. 1. Flow chart of the SWIFT process.

e!ectively prevents unwanted bonding wherever it isapplied.Once a sheet passes through the solvent masking

printer, acetone solvent is applied to the bottom side ofthe sheet. The sheet is then pressed on top of the existingstack of sheets for 5}10 s while solvent welding takesplace. Acetone dissolves the surface of each thermoplasticsheet by breaking apart the van der Waal's bonds thathold the polymer chains together. As the solvent eitherevaporates or is absorbed into the mass of each sheet, thedissolved interfaces of each sheet blend together, and newpolymer chains begin to form between the two sheets.Provided the surfaces of the two sheets are clean, studieshave shown that the resulting bond strength can ap-proach that of virgin material within 24 h [5].Following solvent welding, a shell milling cutter is

used to mill down the thickness of the sheet by a smallamount to correct for any minor variations in thicknessthat might be present in the raw material. Furthermore,in cases where the height of a part is not evenly divisibleby the thickness of the thermoplastic sheets being used,

the "nal sheet is milled down to the appropriate thick-ness in order to improve dimensional accuracy.The "nal processing step for each layer is to machine

the cross-sectional contour of the layer for that sheet.This step can be performed with a conventional three-axis CNC milling machine. At the present time, a 1/32��diameter #at end mill is used which can produce cornerswhose radii are as small as 0.015625��.The SWIFT process derives considerable speed from

the fact that each layer is fabricated by simply machiningthe required perimeter of that layer. In comparison,many rapid prototyping processes do not use solid sheetsas their raw materials, and must therefore trace the per-imeter of each layer and then "ll in the solid area. Theprocess of "lling in each layer can be quite time consum-ing. A potential disadvantage of the SWIFT process isthat a conventional end milling cutter cannot machineundercuts (down facing surfaces) for a given layer. Whilethis problem is easily solved with a "ve-axis machiningcon"guration, the topic of this paper is the rapid proto-typing of tooling components. With very few exceptions,one characteristic of molding and casting insert tools isthat they do not contain any down facing undercuts. Ifthis were not the case, then the molded parts could not beejected from the tool. Consequently, the SWIFT processis inherently well suited for the direct production of tools.

3.2. STL model slicing

In order to prototype a tool via the SWIFT process,the CNC G-code needed to machine the cross sectionalcontours for each thermoplastic sheet must "rst be gener-ated. This is done through a process known as `slicinga.The objective of the slicing process is to automaticallygenerate chordal approximations to the cross sectionalcontours for each layer in the CAD model.The STL "le format is the de facto industry standard"le format for describing the surfaces of solid modelswhich are to be prototyped. Virtually every major solidmodeling CAD system is capable of exporting STL "les.The STL "le format approximates the surface of a part asa series of small triangular facets. Fig. 2 shows a solidCAD model along with a triangulated surface model inthe STL "le format. Each STL "le contains the cartesiancoordinates of the three vertices for each facet as well asthe components of each facet's unit normal vector.In order to generate the cross sectional contour that

corresponds to a given layer (e.g. a slicing plane), thebasic steps below are followed:

1. Make a list of all edges which are intersected by thecurrent slicing plane. Fig. 3 illustrates the edges of anSTL model being intersected by a slice plane.

2. For each of these edges, "nd the X}Y coordinates ofthe point where the edge is intersected by the slicingplane.

D. Cormier, J. Taylor / Robotics and Computer Integrated Manufacturing 17 (2001) 151}157 153

Fig. 2. (a) Solid CAD model and (b) triangulated surface model.

Fig. 3. Facet edges intersected by a slicing plane.

3. Connect the intersection points in the correct order toform one or more sets of straight-line chords. Each setof connected line segments constitutes one or moredirected cross sectional loops.

4. Convert the loops for each layer into standard CNCG-code format.

The "rst step of the procedure calls for the identi"cationof edges in the STL "le which are intersected by thecurrent slice plane. Any given edge in the STL "le has twoendpoints which can be denoted as P

�"(X

�,>

�,Z

�)

and P�"(X

�,>

�,Z

�) respectively. For any given ther-

moplastic sheet in a SWIFT part, the slice plane isde"ned as the plane through the bottom of the sheet. Ifthe bottom surface of the part is assumed to be at Z"0,then the slice plane simply becomes Z"Z

�, where Z

�is

the distance from the bottom of the part to the bottom ofthe current thermoplastic sheet. Determining whether ornot an edge is intersected by the slice plane is simplya matter of checking to see if Z

��)Z

�)Z

��, where

Z��

"max[Z�,Z

�] and Z

��"min[Z

�,Z

�].

For the set of edges which are intersected by the sliceplane, the next step is to "nd the exact coordinates wherethis intersection takes place. The general equation of thepoint (X,>,Z) which lies on the edge de"ned by pointsP�and P

�is

X!X�

X�!X

�

"

>!>�

>�!>

�

"

Z!Z�

Z�!Z

�

. (1)

Since the coordinates of points P�and P

�are known

from the STL "le, and the value of Z is equal to thecurrent slice height (i.e. Z"Z

�), Eq. (1) can be

formulated as a set of two equations with two unknowns.If the two equations are solved for the two unknowns,X and >, the coordinates of the trigger point (¹

��)

where the slice plane intersects the edge from P�

toP�are found to be

¹��

"(p(X�!X

�)#X

�; p(>

�!>

�)#>

�;Z

�), (2)

where p"(Z�!Z

�)/(Z

�!Z

�). If this process is repeat-

ed for each edge in the STL "le that is intersected by thecurrent slice plane, the result is a series of trigger pointswhich de"ne the layer's cross sectional contours. Thenext step is to sequence these points into directed loopswhich collectively de"ne a layer's cross sectional con-tours. Rock and Wozny [6] describe a marching algo-rithm for determining how the connectivity of points isdetermined from the facet data. Points are sequencedsuch that loops which lie on the part's exterior surface areoriented counterclockwise, while loops which lie on thepart's interior surface (i.e. holes or other cut-outs) areoriented clockwise. Fig. 3 shows a part which is intersec-ted by a slicing plane, while Fig. 4 illustrates a samplingof properly oriented slice contours for this part.The "nal step in the process is to generate an ASCII

text "le which contains formatted CNC machininginstructions (i.e. G-code) for each cross-sectional slice.Notice that every single cross-sectional contour is de-"ned as a series of straight lines which lie in a #at plane.The process of converting the cross-sectional contoursinto G-code is therefore a simple matter of outputtinga series of G01 (straight line interpolation) commands.


Fig. 4. Slice contours at various Z-axis heights.

Fig. 5. Solid model of mounting plate mold cavity.3.3. Generation of tooling models

One of the distinguishing details between the SWIFTtooling process and the conventional RTV siliconemold-making process is that the RTV process requiresthe initial production of a prototype part pattern whilethe SWIFT process produces the mold directly. Thechallenge is that while CAD models of the desired partsare readily available for export to the STL "le format,CADmodels of a mold cavity are not. As an example, theCAD and STL models of the part shown in Fig. 2 wouldnormally be prototyped in order to produce a patternaround which an RTV silicone mold would be poured.Conversely, direct tooling approaches such as the onedescribed in this paper require a model of the tool, suchas the one shown in Fig. 5.Fortunately, there are several software packages

available which address this problem. These softwarepackages take the STL model of the desired part andsemi-automatically generate the STL model of the coreand cavity inserts that would be used to mold the part.The Magics RP software package by Materialise, USAwas used to produce the tooling models for this researchproject.

4. Testing and results

In order to evaluate the suitability of molds producedvia the SWIFT process as an alternative to the tradi-tional RTV silicone mold-making process, several experi-ments where run. The experiment described in this paperwas run for an undergraduate manufacturing practicumlab course which provides students with their "rst expo-sure to fabrication and assembly processes. All lab activ-ities for this course are built around a single-case studyinvolving a battery-operated power tool. In the "rst-halfof the semester, students are introduced to a wide varietyof manufacturing processes while producing component

parts for the power tool. In the second-half of the semes-ter, students focus on assembly systems engineering.Using their prototype components, students constructassembly precedence relationship diagrams, performtime and motion studies, apply the results to line-balanc-ing algorithms, and build prototype assembly stations.The assembly stations are built around a Hytrol accum-mulating conveyor. Once the assembly line is ready, thestudents have a pilot production run in which they as-semble power tools on the assembly line for severalhours. Based upon performance results from the pilotrun, they are able to evaluate all aspects of their assemblyline designs. The remaining sections of this paperdescribe the use of a SWIFT mold which students useto produce the power tool's motor mounting plate (seeFigs. 2}5).

4.1. Generation of the tool

In order to generate a tool for the prototype partproduced by students, the CAD model shown in Fig. 2was modeled in the Unigraphics CAD system and wasexported as a binary STL "le. This STL "le was thenimported into Magics RP, and an STL model for the toolshown in Fig. 5 was generated semi-automatically. TheSTL "le for the tool was sliced by software which wasdeveloped in-house according to the slicing proceduredescribed in Section III. Using output from the slicingsoftware, the mold shown in the top-center of Fig. 6was prototyped using the SWIFT process outlined inSection 2.

4.2. Molding results

Using a fast curing two-part polyurethane mixture,students in the Manufacturing Practicum class mademultiple castings of the motor mounting plate shown in


Fig. 6. Motor sub-assembly, SWIFT mold, and polyurethane casting.

Fig. 7. Prototype mounting plate in cutsaw power tool.

the bottom-center of Fig. 6. The motor-mounting plate ispart of the motor sub-assembly shown in the left- andright-hand sides of Fig. 6. A small DC motor is attachedto the plate with two screws. The small gear on themotor's shaft meshes with the larger gear which is visiblein the right side of Fig. 6. A pressed pin in the large gearrests in a small needle bearing in the mounting plate. Thecompleted motor sub-assembly is placed into the powertool housing, as shown in Fig. 7.

5. Discussion

Themold shown in Fig. 6 has "nished dimensions of 3��long�2�� wide�1�� tall. The mold was produced from4��4�� polystyrene sheets, so a total of 16 in� of polysty-rene was consumed. The polystyrene costs $0.06/in�, sothe primary raw material cost of this tool was just $0.96.The only other consumable material used by the processis a small amount of acetone solvent. It is estimated thatno more than $0.25 worth of acetone was required forthis tool. In total, the raw material cost was just over $1.With regards to the machine time costs, it took approx-imately 2 h to build the mold.Presently, students have made over 50 polyurethane

castings with the polystyrene SWIFT mold. This totalexceeds what would normally be expected from an RTVsilicone mold, and the SWIFT mold shows no signs offailing.

One area where RTV silicone molds still hold anadvantage over any of the rigid prototype tooling pro-cesses is in the production of parts with slight undercuts.Since RTV silicone molds are #exible, parts with slightundercuts can be easily de-molded by #exing the moldaround the undercut. This is not possible with rigid tools,and the only way to accommodate undercuts is to buildmore complex tools with sliding cores. Consequently,RTV silicone molds are still the preferred method forproducing prototype assembly components which haveundercuts.Another valuable lesson that has been learned during

experimentation with the production of SWIFT tools isthat parts with deep ribs or bosses should either havefairly generous draft angles of approximately 53, or ejec-tor pins, or both. The motor-mounting plate described inthis paper has a deep boss which is clearly visible inFig. 5. This boss was modeled in the CAD system witha draft angle of 23, and the cast parts were initially quitedi$cult to remove from the mold. To solve this problem,3/16a steel ejection pins were added to the mold.With theejection pins, cast polyurethane parts are now easilyremoved by gently tapping the steel pins until the partpops loose. When parts do not have deep ribs or bosses,experience with other SWIFT molds has shown that thecast polyurethane parts release quite easily from bothABS and polystyrene molds. Ejection pins are, therefore,not needed in these situations.

6. Summary

This paper has described the application of an experi-mental rapid prototyping process to the production ofprototype tooling. The process has been described, andits application to the production of a power tool com-ponent has been presented. Parts cast from the moldhave been used by students to facilitate the process ofdesigning, implementing, and testing an assembly line forthe power tool. Analysis of the process suggests that itcompares quite favorably with traditional part duplica-tion methods in terms of lead time, cost, and durability.The process is therefore potentially attractive to com-panies which develop products in a concurrent engineer-ing environment.

References

[1] Dahl D, Cleary B. Constructing multiple prototype parts fromcomputer-generated master models. In: Rapid prototyping:fast track to product realization. Dearborn, MI: SME, 1994.p. 35}548.

[2] Venus AD, van de Crommert SJ, O'Hagan S. The feasibility ofsilicone rubber as an injection mould tooling process using rapidprototyped patterns. Second National Conference on Develop-ments in Rapid Prototyping and Tooling, 1996. p. 105}10.


[3] Venus AD, van de Crommert SJ. Direct SLS nylon injectiontooling. Second National Conference on Developments in RapidPrototyping and Tooling, 1996. p. 111}7.

[4] Cormier D, Taylor J, West H. A solvent welding freeform fabrica-tion technique. IIE Trans, 1999, submitted for publication.

[5] Yue CY, Cherry BW. The structure and strength of solvent weldedjoints. In: Allen KW, editor. Adhesion 10. New York: ElsevierApplied Science Publishers, 1986 p. 147}77.

[6] Rock S, Wozny M. Utilizing topological information toincrease scan vector generation e$ciency. In: Marcus HL,et al., editors. Solid Freeform Fabrication SymposiumProceedings, The University of Texas at Austin, Austin, TX, 1991,p. 28}36.


a process for solvent welded rapid prototype tooling

Documents