Evaluating 3D printers for their dimensional accuracy using the same STL files.
The blending of lower costs and easier-to-use technology is still fueling rapid growth for 3D printers. Designers, engineers, and educators are using these printers in record numbers, contributing to the fastest growing segment of the rapid prototyping (RP) industry.3D printers are often used in the earliest stages of design where models become the tools for review, evaluation, and innovation. Though life spans of such design tools can be measured in minutes or hours, users still want reasonable accuracy to satisfy the needs of their applications.
Figure 1: End users of the three top-selling 3D printers were each asked to build four prototypes for this evaluation. One pair consisted of the top and bottom halves of a battery box, while the other was a light fixture consisting of a base and decorative cap. On the left, a battery box top created with the InVision SR. On the right, the battery box bottom created with a Dimension SST. Click on images to enlarge. 
The accuracy among printers differs. So how do you know whether the machine you are looking at will give you the accuracy you need? With that question in mind, we asked three users of the top-selling machines to create identical prototype parts for an evaluation. We then analyzed and quantified the dimensional accuracy of parts produced by the Dimension SST, InVision SR, and ZPrinter 310 and compare them here.
Ground Rules of the Process We asked end usersrather than manufacturers of the equipmentto build four prototypes. These end users were carefully selected for their experience with each of the three systemseach using different materials (see “Systems At a Glance” below)and to minimize bias. The participants were limited to one build for each part so that iterations could not be used to improve the results. The parts were postprocessed to the minimum standards, which included support removal, depowdering, and infiltration. Sanding and finishing were not permitted. Upon receipt, the parts were randomly labeled S1 through S12. By eliminating any reference to the systems, the labeling enabled a blind study. The parts were not matched to the system that produced them until all inspection analysis was complete.
The parts were then inspected with an LDI 150 laser scanner. For each part, the scanners output yielded a point cloud with an average of 237,000 individual points, which provides a truer representation of accuracy than would a CMM (coordinate measuring machine). Using PolyWorks software, the point cloud data was compared to the STL files from which the parts were built. This CAI (computer-aided inspection) data produced the color maps and accuracy summaries used to present this report.The results reveal some surprising information and, in some cases, the data is contradictory to typical preconceptions. Assembly Prototypes A pair of two-piece assemblies were created. The first was a battery box (Figure 1, above) consisting of a top and bottom joined with a hinge and snap enclosure. The components measure roughly 2.5 x 2.0 x 1.0 in. The second assembly was a light fixture comprised of a base, which holds a light bulb, and a decorative cap that screws to the base. These parts have diameters of about 1.5 in. and heights of 1.5 to 2.0 in.The color maps seen in Figure 3 (below) illustrate the deviation of the sample parts from their STL files. The color scale, which is different for each part, indicates the range of dimensional error. Green areas are nearest to the nominal dimension, while yellow/red shows the highs and cyan/blue shows the lows. Figure 2: This is a summary of the deviation of the sample parts from their STL files. The combination of the mean and the standard deviation documents the range of error for about 66 percent of all data points. Click on image to enlarge. 
A summary of these results is presented in Figure 2 (above) and in Table 1 (available here). Table 1 lists the mean (average) standard deviation and min/max error. The combination of the mean and the standard deviation documents the range of error for approximately 66 percent of all the data points. Figure 2 is a graphical representation of the data in Table 1. This chart shows the ±1 standard deviation as a solid bar. The vertical lines above and below indicate the maximum deviation for all data points. Battery Box Results For both the bottom and top parts of the battery box, the Dimension SST produced the best overall accuracy, as seen in the color maps. The 61 error is impressive, with both parts having a range of just 0.006 in. While the maximum errors are 20.017 and 0.020 in., 99 percent of all data points fall within a range of -0.010 to 0.010 in.The second best accuracy was posted by the ZPrinter 310. The ±1 error is 0.008 in. for the bottom and 0.010 in. for the top. For the bottom, the error range for 99 percent of the points approached that of the Dimension SST (-0.0010 to 0.014 in.), but the tops error band is twice the size (-0.020 to 0.020 in.). The top also has a wide maximum error range of -0.024 to 0.035 in.
Figure 3: These color maps illustrate the deviation of the sample battery boxes from their STL files.The color scale indicates the range of error. Green areas are nearest to the nominal dimension, while yellow/red shows the highs and cyan/blue the lows. Click on image to enlarge. The InVision SR came in third for accuracy among the three technologies. For both parts, the range of dimensional variance was quite large when compared to the other systems. The two parts have ±1 error bands of 0.021 and 0.020 in. Ninety-nine percent of the data points fall within a range of -0.042 to 0.028 in. The extent of this error band is significantly broader than that of either the Dimension SST or ZPrinter 310. Fixture Results The Dimension SST also had the best accuracy for the fixture base and fixture cap. As with the battery components, the ±1 error range is only 0.006 in. for both parts. Also consistent, the 99 percent range of values are between -0.009 and 0.009 in. Dimension SST did show an interesting difference in min/max error for the base and cap. The base had the biggest maximum error range of any Dimension part (-0.030 to 0.020 in.), while the cap had the best (-0.010 to 0.015 in). These Web Exclusive images show a fixture base created with a Dimension SST (left) and a fixture cap as created by the ZPrinter 301 (right). Click on images to enlarge. 
The ZPrinter 310 came in second here, too, and the fixture base was the best of all ZPrinter 310 parts. The ±1 error for these parts are 0.009 and 0.010 in. The base has a standard deviation range of -0.003 to 0.006 in., and the cap has a range of -0.002 to 0.008 in. Of all data points for these parts, 99 percent fall between -0.013 and 0.019 in. While the accuracy of the fixture components from the InVision SR is better than that for the battery box, it came in third. The ±1 error bands are 0.0013 and 0.015 in. Thats 50 percent larger than that of the ZPrinter 310 and more than double that of Dimension SST. Although the accuracy of the fixture cap approaches that of the ZPrinter 310, the 99 percent error range is broader (-0.023 and 0.016 in.).
This Web Exclusive image shows color maps of the dimensional accuracy of the fixture base (left) and the fixture cap (right). Click on image to enlarge. 
Dimension SST The Dimension SST outperformed both the ZPrinter 310 and InVision SR in all but one measure. The analysis shows that Dimension SST has a narrower range of tolerance deviation and that the high and low deviations are centered about the nominal value. For prototype parts of similar size, this analysis shows that ±0.003 in. is a reasonable expectation for many measurements and that most should fall between -0.010 and 0.010 in. One factor that impacts overall accuracy and quality is that the Dimension SST may leave a narrow gap between the faces of small features. These gaps arise when the system cannot fill the space without adding excess material. The gaps are included in the accuracy data and can be seen in the color maps.Yet, the level of dimensional accuracy and the consistency of the results rival that of rapid prototyping systems that cost much more. ZPrinter 310 The accuracy results for the ZPrinter 310 seem contradictory to previous accuracy studies and general perceptions. This system performed well, and the results are reasonable for a system in the 3D printer class of rapid prototyping technology. Users of the technology, when constructing parts of similar size, can reasonably expect to see dimensional tolerances that are on the order of ±0.005 in. Overall, the test parts show that most dimensions should fall between -0.015 and 0.015 in.However, the ZPrinter 310 did show a tendency to produce dimensional inaccuracies that exceed the ±0.015 in. range. A contributing factor might be part postprocessing. Unlike the other two technologies, the ZPrinter 310 requires manual depowdering and infiltration of the prototypes. When enlarged, the point cloud shows that most surfaces have high and low spots that deviated significantly from the bulk of the surface. This is likely the result of an excess or shortage of powder and infiltrant.It is important to note that the battery bottom, which was infiltrated with cyanoacrylate, was broken in three places. A snap fit was broken off when the part was removed from the powder bed. If allowed to build the part a second time, the end user indicated that this could have been remedied. Additionally, in transit to the inspection company, two corners were broken off. While these missing features were not included in the dimensional analysis, they would detract from the overall part accuracy when used as a concept model or prototype. InVision SR It is difficult to state what can be expected from the InVision SR system, since the dimensional accuracy varied significantly by part. However, it appears that a reasonable range is -0.010 to 0.005 in. and that most dimensions will fall between -0.020 and 0.020 in. With the exception of the battery top, the features tend to be undersized.These results show that the InVision SR may be acceptable as a concept modeler, but for more demanding applications, the technology might not satisfy all accuracy requirements. Conclusion From this study of four parts, the Dimension SST has shown that it can deliver the accuracy needed for prototype components. The ZPrinter 310 also performed well, although the range of tolerance error is broader than that of the Dimension. Surprisingly, the InVision SR was not as accurate in spite of its high resolution.Caution is merited when using these results. As any user will state, the quality of a rapid prototype, including dimensional accuracy, is a function of the part, the system, the build parameters, the material, and the operator. Therefore, use this data only as an indicator of the possible accuracy for small prototype components.Todd Grimm is president of T. A. Grimm & Associates, a consulting firm that focuses on rapid prototyping and reverse engineering. He is also chairman of the Society of Manufacturing Engineers 3D Data Capture/Reverse Engineering technology group and is the author of Users Guide to Rapid Prototyping. Contact him at tgrimm@tagrimm.com or send an e-mail about this article by clicking here. Please reference “3D Printers, January 2006” in your message.
In March, 2004 Oskar came to visit me and discuss the production of the Oskar’s Exotic collection. Here we are in the Puzzle Palace workshop discussing some of the finer points of the collection. Once Oskar learned how to use the 3-D Printer, he was up all hours of the night feeding it new ideas and reels of material. We hope some day to have a vast collection of very unique puzzles. Click ; "Image")
; "Image")
; "Image")
; "Image")
; "Image")
; "Image")
;){kind=link}