9/1/2000 | 16 MINUTE READ

Accuracy in Rapid Tooling

Facebook Share Icon LinkedIn Share Icon Twitter Share Icon Share by EMail icon Print Icon

With shortened product life cycles, increased pressure to get new products to market quickly and squeezed development budgets, it is no surprise that tooling methods which claim to be faster and less expensive than conventional machined injection mold tooling are catching the attention of product developers.

Share

Facebook Share Icon LinkedIn Share Icon Twitter Share Icon Share by EMail icon Print Icon

Over the last several years, we have seen the introduction of a number of such methods. Most use some form of rapid prototyping to either create tooling directly or a master pattern, which is in turn used to create tooling. Many claim that rapid tooling methods can be both faster and less expensive than conventional tooling. But is this tooling viable?

Viability is clearly application dependent. A tooling process that will be perfectly acceptable for a run of 200 polypropylene cereal bowls may not be viable for 20,000 filled polycarbonate instrument housings. There are a number of factors to consider in assessing tool viability including durability, cycle time, ability to modify or repair the tool, ability to automate the tool, etc. However, a potential buyer's first question about rapid tooling typically involves accuracy. If a tool can't provide molded parts within the tolerances needed for the application, the other factors don't matter much.

Unfortunately, there is little quantitative information about the accuracy of various rapid tooling methods to help users make informed decisions about which process, if any, can meet their needs. A few years ago, an effort to commercialize a new rapid tooling process was underway. Frustrated by the inability to obtain satisfactory results, an analysis of the process - based on work done by Dr. Paul Jacobs - was launched and the results were interesting.

What is Rapid Tooling?

Rapid tooling is generally considered to include a number of tool construction methods, which are not primarily dependent on machining. Most rapid tooling processes use a rapid prototyping process in some portion of the tool building process. Most rapid tooling processes are focused on injection molding, although a number have been used for wax pattern tooling for injection molding, blow molding and even die casting applications.

Rapid tooling processes can be grouped into two classes; direct tooling and pattern-based tooling. Direct tooling processes use a rapid prototyping process to create cavity and core geometry directly. Examples are the Direct AIM™ process from 3D Systems, the RapidTool™ process from DTM and the Laser Engineered Net Shaping (LENS) process from Sandia and Optomec.

Pattern-based tools start with a master pattern, which is in turn used to create the tool. The simplest processes use a pattern shaped like the part to be molded and scaled up to compensate both for shrinkage of both the tooling material and the plastic to be molded. The tooling material is formed directly against the pattern. This type of tooling is called single reverse pattern-based tooling because the geometry of the master pattern is reversed to create the tooling - the negative tool is formed against the positive pattern. Examples of single reverse tooling are cast epoxy tooling, cast rubber tooling and electroformed tooling.

Unfortunately, the tooling materials we would most like to use usually cannot be formed directly against rapid prototyping patterns because of heat. For example, we cannot pour molten tool steel directly against an SLA pattern. To obtain tooling in more desirable materials, multiple reverse methods are used. An intermediate tooling material, which has higher temperature capability, is cast against the master pattern. The tooling material is then cast against the intermediate material. For example, in cast aluminum tooling, plaster can be cast against the rapid prototyping pattern and aluminum is then cast against the plaster. Such tooling is called double or triple reverse tooling depending on whether a positive or negative master pattern is used.

This discussion will focus on accuracy in pattern-based tooling. However, many of the conclusions also will be applicable to direct tooling.

Factors Affecting Accuracy

Figure 1 illustrates the major steps in building the tool. The first step is to build the master pattern, typically using a rapid prototyping (RP) process. None of the RP processes currently available, however, can build a pattern with a surface finish adequate for use directly as a master pattern. There typically is some evidence of layering (stair stepping) that must be eliminated or it will be carried into the tool and make ejection difficult. In addition, the surface typically must be smoothed to obtain the desired finish on the molded part. The next step in the tool building process, therefore is to finish the surface of the pattern. Finishing usually involves sanding, filling and painting the surface of the pattern until the desired surface finish is achieved. Once the pattern is complete, the tool can be constructed. Accuracy of the tool is reduced by dimensional errors introduced in each of these steps. To understand the accuracy capability of the tooling process, we need to examine each of these steps independently.

Pattern Build Error

Determining the accuracy of an RP system is not a straightforward task. No standards exist for measuring accuracy of RP systems. At the time of the study, published information was readily available only for a geometry commonly known as the SLA User Group part, shown in Figure 2. The part, 9.5 inches square and 1.5 inches high, was designed by an engineer at Dupont Somos (now DSM Somos) to have 46 unique measurement points.

Dupont offered a service to the rapid prototyping community to measure contributed parts with its CMM and report the results back to the user. 3D Systems published results from several systems and resins over the years. The SLA User Group Geometry was chosen so that the results to previously published information could be compared.

The disadvantage of using the geometry was that it measured accuracy only in the xy, or build, plane, the build plane. It does not measure any dimensions in the z direction perpendicular to the build plane. Unfortunately, for most RP systems, their worst accuracy is in the z direction, perpendicular to the plane of the build. Consequently, the data obtained by measuring the SLA User Group part is likely to overstate the true accuracy of the RP system.

Because stereolithography was used to make the master patterns for the pattern-based tooling, the accuracy of stereolithography systems was measured. Stereolithography is generally believed to be the most accurate of the general-purpose rapid prototyping systems (at least one system is more accurate than stereolithography; however, it is too limited in size and speed to be considered a general purpose system) using a number of different models of machines and running a number of different resins - all of which were used to make master patterns - presented a dilemma.

There was a reluctance to choose one for fear it would not represent the range of systems and resins used. To ensure a good estimate of system accuracy, four parts were built and measured using four SLA systems with four different resins. The User Group part was built on four different SLA systems; two different SLA 250s, an SLA 350 and an SLA 500. In addition, four different resins were used; three from Dupont and one from CIBA. Once the test parts were completed and cured, they were sent to Dupont for measurement. To determine accuracy, the differences in both x and y directions between the actual location of the point on the part and the dimension on the CAD file were calculated.

The standard deviation of those differences was then calculated. Table I lists the results for the four parts. Resins were not identified so as to avoid incorrect conclusions about resin accuracy being drawn. Part accuracy is a function of a number of variables in addition to resin. The worst results here, do not necessarily mean that the resin used was the least accurate of the four.

The values of standard deviation determined here are significantly higher than had previously been published. To some extent, this is to be expected. In most cases, previously reported data was from parts built under ideal conditions. Parts were built immediately after the system had been calibrated, and the parts submitted for measurement were second or third iterations, built after adjustments to run parameters were made to improve accuracy. Unfortunately, this level of attention is impossible in everyday practice. An SLA part built as a pattern for an alternative tool will almost certainly be built without first calibrating the system and without multiple iterations to tweak accuracy. The numerical average of the four values of standard deviation is 0.0071" - which was thought to be representative of patterns supplied by a service bureau.

Knowing the standard deviation allows one to determine the likelihood that any dimension will be within a specified tolerance. Figure 3 plots the probability of a given dimension being within tolerance versus the tolerance. The probability that a given dimension will be within 0.002" is 22 percent, 52 percent that it will be within 0.005" and 84 percent that it will be within 0.010".

Pattern Finishing Error

Remember, however, that a raw pattern is not usable for creating a tool; it must be finished before it can be used as a master pattern. To determine error incurred during finishing, one of the four test parts used for the accuracy tests was sent to each of our four plants with instructions to finish them as if they were going to be a pattern for a tool. Each of the finished patterns was then sent to Dupont who measured the same 46 points and then returned the data. Table II lists the results of that measurement.

These data represent total error - including both build error and errors due to finishing. The average of the standard deviations is 0.0109". With this information, the probability of a given dimension on a finished pattern being within tolerance can be plotted (see Figure 4). It is clear that finishing has introduced some error. The probability of a given dimension being within 0.002" has fallen from 22 percent on the raw pattern to about 15 percent on the finished pattern. The probability that a given dimension will be within 0.005" has fallen from 52 percent on the raw pattern to 36 percent on the finished pattern. Finally, the probability of a given dimension being within 0.010" has fallen from 84 percent on the raw pattern to 64 percent on the finished pattern.

For a number of applications, RP patterns will be well within desired tolerances. However, if there are dimensions with tight tolerances of 0.005" or less, there is a less than one in three chance that the dimension on the pattern will be within spec. The probability of out-of-spec dimensions on the pattern can be significantly reduced if critical dimensions are measured during the pattern finishing process and adjusted to be within spec. It is obvious, however, that tweaking a dimension to bring it into tolerance costs both time and money. The more dimensions that must be tweaked, the less advantage rapid tooling has over conventional machined tooling.

Although direct methods, such as the Direct AIMª and the RapidToolª do not use a pattern, they are subject to similar issues since they use an RP system to build the tool directly and require finishing before they can be used.

Error in Tool Creation

The last step in the tool building process is to use the master pattern to create the tool. There is even less information available on dimensional errors incurred in the tool build process than there is on RP system accuracy. It is speculated; however, that three major sources of error in the tool building process are (1) errors in reversing, (2) errors in polishing the tool and (3) errors from non-uniform shrink.

Reversing Error
All pattern-based tooling methods require at least one reverse - a transfer of geometry from positive to negative or vice versa. How accurately dimensions are communicated through the reversing process depends on a number of factors including how well the tooling material is able to flow into tight crevices and sharp corners of the pattern, whether any tool surfaces are moved slightly during pattern removal and how cleanly the material will separate from the pattern when the pattern is removed.

Although there is no published data about the magnitude of dimensional errors incurred in reversing, it can be speculated that it is impossible to perfectly reproduce all dimensions of the pattern in the tool, thus there must be some error incurred in the reversing process. Furthermore, it is safe to predict that the more reverses involved in the tool construction process, the greater the potential for dimensional error. Processes such as epoxy tooling or electroformed tooling require only a single reverse. A number of processes, however, such as KelToolª or cast metal tooling require two or three reverses - greatly increasing the possibility of incurring error. In this analysis; however, reversing errors were not considered.

Errors in Polishing the Tool
With a number of rapid tooling processes, the surface of the tool is too rough to allow clean release of molded parts and must be polished prior to use. The amount of error incurred in the polishing is not known; however, some processes require a great deal more polishing than others. In this analysis, polishing errors were not considered.

Error from Non-Uniform Shrinkage
In 1997, Dr. Paul Jacobs did some groundbreaking work to show that non-uniform shrinkage in tooling materials can be a significant source of dimensional error in tooling. He started with the assumption that shrink in tooling materials is not uniform. This comes as no surprise to anyone who has been involved in molding or casting. It is common knowledge that thick sections shrink differently than thin sections, etc. In fact, different shrinkage compensation factors often need to be applied to different dimensions to ensure accurate molded parts.

Dr. Jacobs carried this reasoning a step further by claiming that the population of shrinkage values for all dimensions on a tool would be normally distributed. Basic statistics tells us that normally distributed variables can be characterized by a mean and a standard deviation. The mean value will closely correspond to the published shrinkage rates for the material.

Dr. Jacobs then used published data from several sources to show that standard deviation of the shrinkage values is directly proportional to the mean shrinkage rate. In other words, if the mean shrinkage rate for the material is high, we will see a wider range of actual shrinkage values in the tool. On the other hand, if the mean shrinkage rate is low, we will have a much narrower range of actual shrinkage values.

Any difference between the actual shrinkage and the shrinkage built into the pattern will result in dimensional error. For example, assume that a tooling material with a published shrink rate of 1.0 percent is being used. If a dimension of exactly, 5.000 inches on the tool is desired, you need to scale the pattern up to 5.051" so that it will result in a 5.000" dimension on the tool after shrinkage. However, if the tool material only shrinks 0.75 percent, the resulting tool dimension will be 5.013" - an error of more than 0.012" The biggest contribution of Dr. Jacobs' work is an equation to estimate the magnitude of errors resulting from non-uniform shrinkage:

σ = K*s*d, where:

  • σ = standard deviation of dimensional error, inches
  • K = a constant, experimentally determined to be 0.96
  • s = the mean shrink rate of the tooling material, inches per inch
  • d = the nominal dimension, inches
The standard deviation of dimensional error increases with both the mean shrink rate and the nominal dimension. In simple terms, big dimensions will have more error than small dimensions, and high shrink materials will have more error than low shrink materials. If a nominal dimension of six inches is assumed, Figure 5 shows the probability of being within tolerance for a number of common tooling materials. In the calculation, mean shrinkage values were assumed identical to published shrinkage values.

Clearly, the ability to maintain accurate tool dimensions falls off quickly as the mean shrinkage increases. For example, even if the pattern had zero error, there is less than a 35 percent chance that a given six inch dimension on a cast P20 tool would be within 0.005".

Remember that these are only errors attributed to non-uniform shrinkage of the tooling material. No pattern errors are included. What happens if pattern build errors as well as finishing errors are thrown into the equation?

With a little algebra, it can be shown that:

σt= [(1-2s)*σp² + K²s²d²2]1/2, where:

  • σt = standard deviation of dimensional errors on the tool, inches
  • s = means shrinkage of the tooling material, inches per inch
  • σp = standard deviation of dimensional errors on the finished pattern, inches
  • K = constant, experimentally determined to be 0.096
  • d = nominal dimension, inches
Again, if a dimension of 6 inches is desired then the probability that the dimension will be in tolerance for a number of common tooling materials can be calculated. Figure 6 plots the results.

From Figure 6 it is shown that with low shrinkage materials, the chances of a six inch dimension being within 0.005" is about one in three - with high shrinkage materials about one in four. With low shrinkage materials, the chances of a six- inch dimension being within 0.010" is about two in three - with high shrinkage materials, about one in two.

Keep in mind that these results are probably optimistic. The analysis does not include (1) the effects of pattern build errors in the z-direction, the least accurate direction for rapid prototyping methods, (2) reversing errors incurred in building the tool and (3) errors incurred in polishing the tool. Including these errors in the analysis will further lower the curves in Figure 6.

Does this mean that rapid tooling methods will not yield accurate tooling? Absolutely not! It does mean, however, that building tools without monitoring critical dimensions on both the pattern and the tool will probably not yield acceptable accuracy. The results of this study do suggest some guidelines in tool construction to obtain the tool accuracy that you want.

1. Critical dimensions will need to be monitored on the pattern and adjusted if necessary to avoid carrying dimensional error into the tool. Fusion Engineering, a supplier of KelToolª injection mold tooling, has stated that it routinely tweaks pattern dimensions to ensure tool accuracy. Since additional dimensional error is likely to be encountered in the tool building process, the tolerance on the pattern needs to be tighter than that desired on the finished part.

For low shrinkage materials (mean shrink rates less than 0.8 percent), pattern dimensions should be maintained within 75 percent of the tolerance desired on the finished tool. For example, if the tolerance on the finished tool is 0.010", the pattern dimension should be maintained within 0.0075".

For high shrinkage tooling materials or dimensions larger than 6 inches, tolerances on the pattern should be 50 percent of that desired on the finished tool.

2. For dimensions with tight tolerances, the safest strategy may be to adjust CAD model dimensions prior to building the pattern, or to adjust pattern dimensions to ensure adequate machine stock on the tool. The dimension can then be machined to tolerance on the tool.

3. If there are a number of tight tolerances, it may be faster and less expensive to machine the master pattern rather than using RP to create it. Rapid Design & Tooling, the largest supplier of KelToolª tooling, machines the master pattern rather than using RP for tools with tight tolerances.

4. Even with careful attention to the master pattern, it will be necessary to check and possibly adjust dimensions on the tool to obtain desired tool accuracy. For tools with very tight tolerances, it may be wise to choose a tooling material that can be machined to bring dimensions into tolerance.

It is important to recognize, however, that the additional attention necessary to achieve accuracy in a rapid tool will increase both the cost and time required to create the tool and reduces the advantage that rapid tooling processes have over machined tooling. For tools with a number of tight tolerances, the cost and time required to build rapid tooling can be comparable or even exceed that of conventional machined tooling.

For more information contact Tom Mueller of Express Pattern (Buffalo Grove, IL) at (847) 215-0001.

Resources