Virtual and Physical Prototyping
ISSN: 1745-2759 (Print) 1745-2767 (Online) Journal homepage: http://www.tandfonline.com/loi/nvpp20
3D printer selection: A decision-making evaluation and ranking model D.A. Roberson, D. Espalin & R.B. Wicker To cite this article: D.A. Roberson, D. Espalin & R.B. Wicker (2013) 3D printer selection: A decision-making evaluation and ranking model, Virtual and Physical Prototyping, 8:3, 201-212, DOI: 10.1080/17452759.2013.830939 To link to this article: http://dx.doi.org/10.1080/17452759.2013.830939
© 2013 The Authors. Published by Taylor & Francis. Published online: 25 Oct 2013.
Submit your article to this journal
Article views: 3961
View related articles
Citing articles: 3 View citing articles
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=nvpp20 Download by: [37.44.207.174]
Date: 20 January 2017, At: 08:50
Virtual and Physical Prototyping, 2013 Vol. 8, No. 3, 201212, http://dx.doi.org/10.1080/17452759.2013.830939
3D printer selection: A decision-making evaluation and ranking model D.A. Robersona,b*, D. Espalina,c and R.B. Wickera,c a
W.M. Keck Center for 3D Innovation, The University of Texas at El Paso, El Paso, Texas, USA b Department of Metallurgical & Materials Engineering, College of Engineering, The University of Texas at El Paso, El Paso, Texas, USA c Department of Mechanical Engineering, College of Engineering, The University of Texas at El Paso, El Paso, Texas, USA (Received 13 March 2013; accepted 18 May 2013)
The purpose of this paper is to evaluate the capability of five desktop additive manufacturing (AM) machines based on the ability to produce a standard component. This work also developed a model/method for evaluating and ranking AM technologies based on select criteria that can facilitate purchasing decisions. A standard part was designed and printed on each machine, and evaluated based on dimensional accuracy and surface finish. Additionally, the machines were compared based on build time for single and multiple parts as well as material consumption and unit cost. The research highlights the differences between AM units and suggests a method by which to ascertain the differences. With the rapid proliferation of desktop additive manufacturing units, a quantitative ranking system was developed to rate the units so a comparison can be made. Although the focus of the work was on desktop systems, the approach can be applied across all AM technologies. Keywords: additive manufacturing; desktop 3D printer; material extrusion; sheet lamination; vat photopolymerisation
1. Introduction Three dimensional printing (3D printing) for the masses was a concept introduced by so-called open source 3D printers such as the RepRap project started in 2004 (Pearce et al. 2010, Cano 2011) and the fab@home (Malone and Lipson 2007). The advent of low-cost ($5000 or less) additive manufacturing equipment has further pushed the migration of this technology to the home user allowing for hobbyists and do-it-yourselfers to have access to 3D printing. This access has resulted in a market that has grown from 66 purchased/placed units in 2007 to 23,265 units in 2011 (Wohlers 2012). In addition, low-priced professional grade printers with a cost of more than $5000 have
the potential to open the door for widespread use amongst small business and entrepreneurial applications, and these low-priced systems have also increased the availability of 3D printing to educational institutions. Material extrusion-based units are the most common table-top variety due to the simplicity of operation and the relatively simple materials handling; namely a spool of thermoplastic filament compared to powders or photocurable resins utilised by fusion and vat photopolymerisation 3D printing equipment. The adaptability of the material extrusion additive manufacturing (AM) process to incorporate non-commercial or novel materials, as has been demonstrated with polymethylmethacrylate for example
*Corresponding author. Email:
[email protected] # 2013 The Authors. Published by Taylor & Francis. This is an Open Access article. Non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly attributed, cited, and is not altered, transformed, or built upon in any way, is permitted. The moral rights of the named author(s) have been asserted
202
D.A. Roberson et al.
(Espalin et al. 2010), also makes this process attractive. Additionally, the expiration of the original fused deposition modelling (FDM) patents (Crump 1988) has opened the material extrusion technology market. To that end, there are several open source designs and do-it-yourself kits to fabricate your own 3D printer utilising material extrusion. Also adding to the increase of the availability of 3D printing to society is the concept of the ‘FabLab’ such as www. shapeways.com where a customer can design or receive help with the design of an object, pick the material, and have the 3D printed part shipped to them (Schmidt et al. 2011). Furthermore, the use of desktop 3D printers along with open source software has proven to be an invaluable tool in engineering education allowing for a virtual community to be created providing the proliferation of hands-on learning (Gonzalez-Gomez et al. 2012, Valero-Gomez et al. 2012). With the increase in the number of low-cost 3D printers, evaluation of the capability of different systems is paramount to the proliferation of low-cost additive manufacturing in terms of consumer confidence in the technology. As AM technology becomes available to more facets of society, logical evaluation tools should be readily accessible to individuals to allow a fair comparison of the performance of a given unit to another. Furthermore, the rate at which home-use grade AM units enter the consumer market creates a need for a tangible metric to be established by which to evaluate the units in a comparative manner. While benchmarking methods have been used in the past to evaluate some aspects of AM units such as speed and dimensional accuracy (Mahesh et al. 2004, Brajlih et al. 2011), we have been unable to identify a tool by which to evaluate the consumer worthiness of an AM unit. In this paper, we evaluate the consumer worthiness of a total of five consumer and professional grade 3D printers including three material extrusion printers, one vat photopolymerisation unit, and one sheet lamination system. The terminology we are using to describe the equipment is based on American Society for Testing and Materials (ASTM) F2792-12a which standardises and describes the terminology for additive manufacturing (ASTM F2792 2012). The
units were evaluated based on dimensional accuracy as compared to computer aided design (CAD) drawings, surface finish, build time for one and multiple parts, and finally, material consumption and waste. Also, to provide a tool by which a consumer at any buying level can compare the performance of an AM unit, a ranking system is presented here which scores an individual unit based on consumer- or user-specified critical performance factors. We believe the method described in the following can be applied across the AM technology platforms depending on the needs of the consumer (from high-end industrial applications to desktop do-it-yourself users).
2. Equipment The three material extrusion 3D printers tested in this study were the uPrint Plus (Stratasys, Eden Prairie, MI, USA), 3D Touch (Bits from Bytes, Clevedon, Bristol, United Kingdom), and MakerBot Replicator, (MakerBot Industries, Brooklyn, NY, USA). The sheet lamination system tested in this study was a SD300 Pro (Solido, Manchester, NH, USA) while the vat photopolymerisation system tested was a V-Flash 3D Printer (3D Systems, Inc., Rock Hill, SC, USA). Though the V-Flash is a hybridisation of two platforms (sheet lamination and vat photopolymerisation) the authors designated it as a vat photopolymerisation system as this most closely matched the ASTM terminology. The cost per unit is listed in Table 1 and ranged from $1499 to $20,900. It is noted that both the SD300 Pro and the V-Flash are no longer available on the market, however we have included them in our study to demonstrate the ability of our evaluation model to compare 3D printers that are based on different technologies.
2.1 Method of evaluation The purpose of this paper is not meant to be an exhaustive review of standard test parts intended for the evaluation of AM platforms, but rather this paper presents a model by
Table 1. Cost and technical data for the units tested in this study. Developer
Model
Build size (L W H) mm
Dimension
uPrint Plus
203 203 135
5563.2
20,900
Solido Bits From Bytes
SD300 Pro 3D Touch
160 210 135 275 275 210
4536 15,881.3
4375 3930
3D Systems MakerBot
V-Flash FTI 230 Replicator
228 171 203 225 145 150
7914.6 4839.8
9900 2072
*Referred to as vat photopolymerisation in this study.
Envelope volume cm3
Cost USD
Available layer thicknesses mm (in) 0.254 0.330 0.168 0.102 0.125 0.250 0.500 0.102 0.200 0.300
(0.010) (0.013) (0.006) (0.004) (0.005) (0.010) (0.020) (0.004) (0.008) (0.012)
Technology
Material Extrusion Sheet Lamination Material Extrusion
Vat Photopolymerisation/Sheet Lamination* Material Extrusion
Virtual and Physical Prototyping
203
Figure 1. (a) The modified Grimm test structure used by Espalin et al. (2009) compared with (b) the test structure used in this study. which consumers can evaluate the quality of an AM machine to meet consumer specified metrics. However, there are examples of the utilisation of a standard test part to compare the print quality among a pool of AM machines. Jayaram et al. (1994) utilised a standard test part to establish a benchmarking procedure in the comparison of four different 3D printing platforms. Ghany and Moustafa (2006) utilised a standard part to compare the capability of four different laser-based metal powder bed fusion units. Mahesh et al. (2004) utilised a standard part to compare the print quality among four different polymeric-based AM platforms. There have also been several efforts to develop a standard test part for the evaluation of AM machines hallmarked by an effort by the National Institute of Standards (NIST) (Moylan et al. 2012). Espalin et al. (2009) demonstrated the use of a modified Grimm test structure (Grimm 2003) in the evaluation of the dimensional accuracy of FDM-created parts. In our head-to head evaluation, we used a further modified version of the Grimm test structure. Our modification entailed altering the test structure from essentially being flat to becoming a step-like test structure. The reason for this modification was to test the machines with a part that would require the use of support structures. Also, due to the lack of a temperature controlled build envelope on most of the material extrusion units tested (with the exception being the uPrint), warping of the Grimm test structure was a problem which could only be minimised by moving to the step-like test structure. The modified Grimm test structure used by Espalin et al. (2009) is compared to the modified Grimm test structure used in this study in Figure 1. The dimensional accuracy for the test structures was evaluated with a touch probe fitted to an OGP Smartscope Flash 250 (Optical Gaging Products, Rochester, NY, USA). The measured dimensions were compared to the CAD drawing for the test structure and the difference was calculated. The positions for the probe measurements are seen in Figure 2 where a total of 41 individual points were measured. The use of similar coordinate measuring machine (CMM) technology was also used by Mahesh et al. (2004) in a head-to-head evaluation of commercial grade AM equipment where a standard test structure was used to evaluate the dimensional accuracy of four AM technologies. Material usage estimates came from the specific software used by each individual equipment set. We classified waste material as that used by support structures for the material
Figure 2. Locations for dimensional measurement of the test structures printed in this study. * denotes a height measurement, while ** denotes a symmetry measurement. extrusion and vat photopolymerisation equipment and the excess material left over by the sheet lamination system. Actual material usage and waste material amounts were measured using a X540025 Delta Range Scale (Mettler Toledo, Columbus, OH, USA). Estimated build times for single and multiple part builds were generated by the software for each given piece of equipment while the actual build time was measured by a stopwatch.
3. Results The equipment cost, build envelope size, and capable layer thickness for each unit is listed in Table 1. The Solido SD300 Pro possesses the smallest build envelope size for all units tested in this study ( 4536 cm3) while the largest build envelope belongs to the 3D Touch system at 15,800 cm3. The 3D Touch and uPrint differ from the other
204
D.A. Roberson et al. Table 2. Material usage and build time comparison for single and multiple parts for the units tested in this study. Number of parts
Estimated build time (min)
Actual build time (min)
Estimated material use (g)
Actual material use (g)
Support/waste material (g)
Model mass (g)
3D Touch V-Flash uPrint SD300 Pro Replicator
1 1 1 1 1
421 203 142 239 202
329 275 147 242 220
37.3 52.0 43.3 317.2 -
35.5 54.9 39.3 1168.9 42.5
10.0 8.8 9.0 1120.4 6.1
25.6 46.1 30.3 48.5 36.4
3D Touch V-Flash uPrint SD300 Pro Replicator
5 5 5 5 3
1358 285 711 630 626
1373 275 692 643 613
187.2 255 215.9 1231.7 -
175.3 274.4 195.6 2453 129.4
36.9 44.2 44.4 2209.3 20.1
138.4 230.3 151.1 243.7 109.3
System
systems tested in this study due to the use by these units of a different material for support structure fabrication as the 3D Touch utilises polylactic acid (PLA) as a support structure while the uPrint uses awater-soluble support structure (in our case SR-30). PLA is a corn-based biodegradable polymer (Ho et al. 1999) while the soluble support is a proprietary material (Stratasys, Eden Prairie, MN, USA).
3.1. Comparative data The data for build time and material usage are presented in Table 2. Due to the build envelope size and layout of the test part, the Replicator could not build five parts at a time and was only able to build three parts at a time. In all cases the equipment provided an estimated build time which could then be compared to the actual build time. In terms of actual build time, the 3D Touch system required the most time 329 minutes for one part and 1373 minutes for five parts (Figure 3). The longer build time for the 3D Touch occurred because the system, equipped with multiple extruders for depositing a support and model material, allows one extrusion tip to cool down before heating the second extruder when switching materials, as opposed to heating and cooling the tips simultaneously or keeping the tips hot throughout the build sequence. On the other hand,
the Replicator machine had the fastest build time (220 minutes) when building one part. While most of the equipment tested in this study overestimated the build time, the estimate when building one part that most closely matched the measured build time was that of the SD300 Pro which overestimated the build time by 3 minutes or 1%. The greatest difference between estimated and measured build time for one part was that of the 3D Touch which overestimated the build time by 92 minutes or 22%. However, the estimated time for building five parts with the 3D Touch was more accurate within 1% which indicates a software issue. Other systems also suffered from large time estimation errors; however, these errors are expected to decrease or be removed as their software matures. A build time scaling factor can be calculated as the ratio of the build time of N parts to the build time of one part multiplied by N parts (Table 3) to provide an insight into the expected increase in time when building multiple parts. The same strategy was used to calculate a scaling factor for material usage. In both cases, if the value for the scaling factor was less than 1, the units performed more efficiently while printing multiple parts than when printing a single part. If the scaling factor was greater than one, the units were less efficient when producing multiple parts. If the value was equal to one then the efficiency scaled linearly. As an example, the uPrint system yielded a build time scaling factor of 0.94 indicating that the build time for five parts is approximately the same as for one part. On the other hand, the build time scaling factor of 0.20 reveals the V-Flash to become more efficient when building multiple parts the small scaling factor is a product of the Table 3. Scaling factors for build time and material usage between systems. System
Figure 3. Actual build time for the 3D printers. * data for the building of three parts.
3D Touch V-Flash uPrint SD300 Pro Replicator
Build Time Scaling Factor
Material Use Scaling Factor
0.83 0.20 0.94 0.53 0.93
0.99 1.00 1.00 0.42 1.01
Virtual and Physical Prototyping
205
Figure 4. Actual material use by the 3D printers. * data for the building of three parts. projection-based photopolymerisation process. Applying the same method to the material used by each system reveals a nearly linear relationship between number of parts built and materials used for all systems with exception of the SD300 Pro which improves in terms of material use efficiency when building multiple parts. Estimation of material usage was a capability of all the units tested with the exception of the Replicator. Material usage was considered to be that of the part and the support material. Support material was also separated into an individual category so the amount of waste material could be compared between units. The extrusion-based machines used less material when compared to the SD300 Pro (Figure 4). This difference between material extrusion and sheet lamination processes (which require a single sheet for every layer) is highlighted by noting that the greatest material usage was that of the SD300 Pro at 1168.9 g for one part while the 3D Touch used 35.5 g of material a difference of 1133.4g or 97% when referencing the SD300 Pro material use. Comparing extrusion-based units to one another revealed the Replicator to have consumed a slightly higher amount of material to build a single test structure 42.5 g as opposed to 35.5g when using the 3D Touch for example. When building five parts, the V-Flash used similar amounts of material as the extrusion-based machines. Finally, the material usage estimation errors were also observed via the relative difference between estimated and actual material use, but as with the build time estimates, the estimated values for material use are expected to improve as the systems’ software evolves. As expected, the SD300 Pro produced the most waste material due to the sheet lamination process which resulted in 1120.4 g of waste when building one part (Figure 5a). This meant that 96% of the material used to produce the test part was wasted as shown in Figure 5b. However, when building five parts the percentage of material wasted decreased to 90%. This decrease underscores the fact that the build process in a sheet lamination process is analogous
Figure 5. Support/waste material used by the 3D printers. (a) support/waste material. (b) relative difference between support/waste and model material.* data for the building of three parts. to machining a part from a billet. That is, if there are small features on the part, only a small fraction of the entire sheet of material is used for actual part fabrication, and more material (the entire sheet) must be laid down and then removed when the process is finished. Building small parts with the sheet lamination process results in high percentages of waste an undesirable characteristic that is not observed in the material extrusion and vat photopolymerisation processes. Conversely, as more of the build volume is occupied in a sheet lamination process, say by a solid block, the percentage of wasted material is decreased. The lowest amount of waste material was generated by the Replicator at 6.1 g or 14%. Nonetheless, the use of less support material may not be preferred as parts may exhibit poor dimensional accuracy when using fewer supporting structures. In this study, there was no support optimisation performed. Though the 3D Touch generated the most amount of waste material in terms of mass among the extrusion-based units, adjusting for the density of PLA (1.25 g/cm3) revealed that the volume of waste was slightly lower than the uPrint. Considering PLA is biodegradable, the 3D Touch may be
206
D.A. Roberson et al.
considered to have less impact on the environment in terms of waste generated. Among the extrusion-based units, the Replicator produced the heaviest part (36.4 g) meaning that the Replicator is creating the part nearest to fully dense. Based on the density of acrylonitrile butadiene styrene (ABS) (1.04 g/cm3), a fully dense test structure would have a mass of 39.02 g. The densities of the filaments were measured and compared to one another and revealed slight differences in the density of the order of two tenths of a gram, but these slight density differences cannot account for the large differences in part density. As a result, the difference in mass among the ABS parts is most likely due to the differences in the toolpaths each unit utilises to build a part. All toolpaths were created in an attempt to produce a fully dense part as allowed through each system’s build preparation software, and it should be noted that no toolpath optimisation was performed in this study. Based on these builds, mass difference can then be used as a tool to evaluate how well a build pattern is achieving a fully dense part.
3.2. Dimensional accuracy Dimensional accuracy was evaluated based on the average measurements of five individual parts built by each unit and then comparing those measurements to the original CAD drawings. The difference between the measured and anticipated measurement is seen in Figure 6. The data are
expressed in the form of a percentage difference while using the CAD dimensions as a reference. A negative value indicates the measurement was smaller or undersized as compared to the CAD drawing while a positive value indicates the measurement was larger or oversized as compared to the CAD drawing. It should be noted that each system was calibrated using the manufacturer’s recommended procedure. Analysis of the percentage difference of the measurements of critical feature size (measurements 116) reveals parts created with the V-Flash produce features that have a propensity to be undersized while test structures created with the Replicator have a propensity to create features which are oversized. Comparing height measurements (measurements 1737) reveals the V-Flash, Replicator, and 3D Touch produce test structures with height values greater than indicated by the CAD drawings. Comparing the symmetry measurements (measurements 3841) reveals that all units tested produce features closer together as compared to the CAD drawings; however all measurements were less than 5% undersized as compared to the CAD drawings. The absolute value of all difference measurements was added together and compared revealing the uPrint to produce parts, which overall, most closely matched the CAD drawings followed in order by the SD300 Pro, 3D Touch, Replicator, and V-Flash. The total absolute value difference for all features is seen in Table 4.
Figure 6. The percentage difference between produced parts and CAD drawings for the test structure printed in this study.
Virtual and Physical Prototyping Table 4. The sum of the absolute value of dimensional measurement differences. System 3D Touch V-Flash uPrint SD300 Pro Replicator
Sum of difference (mm) 12 25 4 6 18
3.3 Surface roughness measurement Surface roughness measurements were made with a Mitutoyo, model SJ-201P surface roughness tester (Mitutoyo America Corp., Aurora, IL, USA). The roughness at five locations along a vertical plane was measured and the average of these five measurements was compared. It can be seen in Figure 7 that the SD300 Pro and the V-Flash produced the smoothest parts, although the surface roughness of the parts produced by the V-Flash is expected to be worse at the bottom surfaces where the support structures leave bumps. Conversely, the Replicator and 3D Touch produced the roughest parts. Among the extrusion-based machines, the uPrint produced the parts with the smoothest surface finish. Although the vertical plane measured in this study revealed information regarding the layer height and surface roughness, important information about the stair-step effect and straightness of plane can be gathered from a set of inclined planes (e.g., planes inclined at 10, 15, 30, and 458 from vertical as described in Pandey et al. (2003) and Espalin et al. (2013)). To that end, the development of a standardised test part that can be used across all AM technologies is needed and has been recognised by other researchers (Mahesh et al. 2004, Ghany and Moustafa 2006, Moylan et al. 2012). Since the test part used in this study does not contain an inclined plane, surface roughness was not included as a factor in the model to be described below.
3.4 Other aspects to consider In addition to the data-based comparisons, it is useful to compare other aspects of the desktop 3D printers tested in this study; post processing, portability, and safety. Post processing involves any steps needed complete the part after
207
the machine build process has stopped. Portability involves the ease of transporting the unit for use in other locations. Safety involves the presence of exposed components such as heating elements and moving parts. The post processing of parts created from the V-Flash involves the breaking away of support structures, cleaning the uncured resin with alcohol, and an additional post cure in an ultraviolet (UV) oven. Post processing of the three extrusion-based units tested in this study involves the breaking away of the support structures; however, the use of a soluble support structure by the uPrint adds the option of utilising the WaveWash support cleaner which becomes a necessity for the fabrication of extremely complicated parts. The post processing of the SD300 Pro and Replicator also involves the breaking away of the support and excess material by manual means. Affecting portability of the V-Flash, uPrint, and SD300 Pro is the requirement of these units to be connected to a computer in order to operate with the V-Flash having the added necessity of an internet connection as the software is web-based. Both the Replicator and 3D Touch have the ability to print a part by reading the .stl file directly off a secure digital (SD) card enabling these units to be more portable as compared to the others tested in this study. Also affecting portability is the weight of each system tested in this study. Table 5 lists the system weights for each of the tested units and it can be seen that the uPrint is the heaviest system tested in our study weighing 76 kg (168 lb). The lightest of the units tested in our study is the Replicator which weighs 16 kg (35 lb). Considering that there is no need for a PC to operate the system along with the relatively low weight, the Replicator is the most portable of the systems tested in our study. It should be noted that all units tested in this study use a 110 V/60 Hz power source. Evaluating the safety of the systems tested in this study based on the criteria discussed above, the uPrint, V-Flash, and SD300 Pro have enclosed build envelopes and all moving parts are enclosed during the part fabrication process. However, the operation of the V-Flash system requires the use of resins for building and solvents for postprocessing that may pose a safety hazard if not used properly. The Replicator and 3D Touch have exposed build envelopes with the Replicator having the added concern of exposed filament spools. Table 5. The system weight of the 3D printers tested in this paper. System
Figure 7. Average surface roughness, Ra, based on five measurements on a vertical wall.
3D Touch V-Flash uPrint SD300 Pro Replicator
Weight, kg (lb) 38 (84) 66 (146) 76 (168) 36 (79) 16 (32)
208
D.A. Roberson et al.
4. Development of a ranking system To provide a metric by which individual AM systems can be compared to one another, a ranking system was developed to allow a score to be awarded to each system based on four equally weighted factors: 1) a time factor (TF) time to build one test part, 2) unit cost of the printer (UC), 3) material cost to build a part (MC), and 4), a factor incorporating dimensional discrepancies from the original CAD drawing (SD), which is the sum of the absolute value of difference between measured and expected dimensional reference. The parameter for material cost (MC) is defined by multiplying the cost per gram of raw material by the total mass of material used to build the part. The best possible effort was made to allow for an apples-to-apples comparison across all the 3D printing platforms. In the case of the V-Flash, each build also requires the use of a build pad; here, the build cost is the cost of resin used to build a part plus the cost of a build pad. In the case of the Solido SD300 Pro, a kit is sold which includes a polyvinyl chloride (PVC) roll, adhesive, and a release agent. To our knowledge, in order to replace materials, a kit must be purchased so the MC parameter is cost of the kit per kilogram of build material in the kit multiplied by the mass of material used to build a part. In the case of the uPrint, a support material is used to build a part which has a different price point as compared to the build material. Here the MC is calculated by adding the material cost per part of the support and build materials. Another comment should be made regarding the vertical plane used to determine the surface roughness. While the factors chosen for this study are not all inclusive, these factors enabled a quantitative method for comparing AM systems that removed ambiguous, inconsistent, or potentially biased evaluations originating from qualitative factors (e.g., user satisfaction, ease of post-processing). The quantitative method described below can be easily modified to include additional factors as well as additional printers as the data become available, and therefore, the discussion that follows is intended to provide a methodology for comparing 3D printers. It is also our intention to provide a more comprehensive evaluation (factors and printers) in the future. The contribution from each factor to the total score of a printer was determined by following the steps below. (1) For each data set within a factor, the sample mean ( x) and standard deviation (s) were calculated using the following formulas: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P Xn ðxi x Þ2 xi s¼ x ¼ i¼1 n 1 n (2) A threshold value (q) was calculated by using one of the following formulas. þ s when a low value for x was preferred as was W¼x the case for TF, UC, MC, and SD.
(3)
(4)
(5)
(6)
s when a high value for x was preferred (none W¼x of the factors used in this study employed this formula; however, attributes such as ultimate tensile strength would use this formula) Outliers were removed from the data set by comparing each value x to the threshold q. Outliers were removed if xi q when a low value for x was preferred xi Bq when a high value for x was preferred Steps 13 were repeated until all outliers were removed. The outliers were not an indication of measurement error, as is commonly the case, but rather a measure of the limitations of a particular system. As is discussed later, the iterative process highlighted the technological gaps between the three evaluated processes (material extrusion, sheet lamination, and vat photopolymerisation) in addition to identifying inadequacies in each machine. This iterative process ensured that only competitive printers within the pool were being rewarded and non-competitive printers were removed from the factor at hand. The factor’s contribution (FC) was calculated using the remaining values FC ¼ minxðxÞ when a low value for x was preferred i FC ¼ maxxiðxÞ when a high value for x was preferred The equally weighted contribution, f(x), was calculated for each machine. FC f ðxÞ ¼ P FC
(7) The weighted contribution was then scaled based on how many units survived the iteration process. For example, if three units survived a given iteration, the total amount of points available for that factor was 3/5 or 0.6. If all units survived a given iteration, then the total points available for that factor was 1. (8) The ranking score (R) was calculated for each printer by adding the contributions from each factor and averaging over the number of factors. R¼
sum of f ðxÞ for each printer sum of f ðxÞ for all factors
Step 6 ensured that the sum of all f(x) within a factor was equal to 1 if all units survived a given iteration and Step 7 was necessary to ensure that printers were not excessively rewarded (by having more points available for distribution) if other printers were removed from the particular FC under study. For example, in the limit of 4 of the 5 printers being removed in the FC analysis, instead of being rewarded with a score of 1 for the remaining printer, a weighted score of 0.2 would be assigned. This strategy assumed that the user valued each of the factors equally; however, the model can easily be modified to accommodate different users who may weigh each factor differently. To do this, one would
Virtual and Physical Prototyping
209
Table 6. Iterative process demonstration for the Built Time factor (TF).
System 3D Touch V-Flash uPrint SD300 Pro Replicator sum mean, x standard deviation, s threshold, u min [f(x)]
Iteration 1
Iteration 2
Contribution Calculation
Build Time, TF (min)
Build Time, TF (min)
Build Time, TF (min)
value (x)
FC
f(x)
value (x)
FC
f(x)
value (x)
FC
f(x)
329 275 147 242 220
329 275 147 242 220
329.00 275 147 242 220
0.00 0.53 1.00 0.61 0.67 2.81
0.00 0.15 0.29 0.17 0.19 0.80
243 67 310
simply multiply each f(x) by a weighting fraction equivalent to the user’s prioritised factors. For example, a user may prioritise and weigh the factors in the following fashion: UC (0.4), MC (0.2), SD (0.2) and TF (0.2) the parenthetic values represent the weighting fractions. In this case, the user feels that UC is the most important factor, and that MC, SD and TF are equally important but two times less important than UC. Even more profound is the customisation which can be applied by the user to this model. For example, if one desired to evaluate a machine based on the optimum mechanical strength of parts fabricated, a metric for strength can be added to the model. The end goal is to have a model with enough parameters and systems that Step 7 becomes unnecessary even if not all parameters survive a given iteration. Perhaps the iterative process for the determination of f(x) can be best explained by discussing an example. Here, the factor TF will be discussed using the data collected during this study (Table 6). After employing , s, and u using the collected data Steps 12 to calculate x (Table 6 - Iteration 1), it was noted that the value x for the 3D Touch (x 329 min) was larger than the threshold value u (310 min for iteration 1) and therefore this value x was removed from the dataset (noted by the shaded cell) as per , s, and u were Step 3. During iteration 2, the values for x recalculated omitting the value for the 3D Touch. When comparing the threshold value for iteration 2 (u 275) to the remaining x values, it was noted that none of the x values were larger than u. Therefore, FC and f(x) values were calculated as per Steps 56 for each of the printers except the 3D Touch (f(x) 0 for this system). The iterative process for TF is shown in Table 6, where the starting data and resulting data from the iteration process are shown. The iterative process for TF also shows that as systems were removed, the total number of available points dropped by 1/5. The same approach was performed with the other factors and the results are summarised in Table 7 while the results of the ranking process are seen in Figure 8. As mentioned before, our test part did not contain an
221 54 275
221 54 275 147
inclined surface, so the parameter for surface roughness (Ra) was not included in the ranking model described above. As previously mentioned, the iteration process highlighted the technological gaps between the material extrusion, sheet lamination, and vat photopolymerisation processes. This is most evident by referring to the results for the factors MC and SD. The material cost for the sheet lamination process ($79.22 for the SD300 Pro) is about 26 times greater than that of the most expensive material cost among the material extrusion processes ($3.08 for the uPrint). Similarly, the material cost for the vat photopolymerisation process ($30.62 V-Flash) is approximately 2.5 times more expensive than the material cost of building a part using the uPrint. Thus, the iterative method was sensitive to this difference in material cost and therefore the V-Flash and SD300 Pro values were omitted when calculating the weighted contribution. Similarly, the 3D Touch, V-Flash, and Replicator were removed from the pool of machines via the iterative process when considering SD. In our evaluation, the ranking system identified the Replicator as the top ranked machine followed by the uPrint and the SD300 Pro. The Replicator did not receive any contributions to its ranking score from the SD factor (Table 8). Similarly, the SD300 Pro did not receive contributions from the MC factor and the uPrint did not receive contributions from the UC factor. Though the uPrint is an order of magnitude more costly than the replicator, the quality of the part produced and fast build time allowed it to compete with the other less expensive models tested in our pool. It should be noted that the ranking scores for these three machines were very similar and fell within a narrow range (0.210.31). The fourth ranked machine was the 3D Touch, which received about half the ranking score of the top three machines. Its lower ranking score was due in part to the lack of contributions from the TF and SD factors. The V-Flash did not receive contributions from the UC, MC, and SD factors and therefore was the lowest ranked machine. These results
210
D.A. Roberson et al. Table 7. Summary of the results obtained via the iterative process for all the factors. Build Time, TF (min)
System
x
3D Touch V-Flash uPrint SD300 Pro Replicator sum mean, x standard deviation, s threshold, u min [f(x)]
329 275 147 242 220
FC
Unit Cost, UC ($)
f(x)
x 3930 9900 20900 4375 2072
3D Touch V-Flash uPrint SD300 Pro Replicator sum mean, x standard deviation, s threshold, u min [f(x)] * ** ***
f(x)
8235.40 7658.16 15893.56
FC
f(x)
329* 275 147 242 220
0.00 0.53 1.00 0.61 0.67 2.81
0.00 0.15 0.29 0.17 0.19 0.80
221.00 54.26 275.26 147.00
x 3930 9900** 20900* 4375 2072
FC
f(x)
Material Cost, MC ($) f(x)
x
FC
f(x)
0.53 0.00 0.00 0.47 1.00 2.00
0.16 0.00 0.00 0.14 0.30 0.60
2.81 30.63** 3.08 79.22* 2.04
0.73 0.00 0.66 0.00 1.00 2.39
0.18 0.00 0.17 0.00 0.25 0.60
2.64 0.54 3.18 2.04
FC
f(x)
12 25 4 6 18
13.00 8.66 21.66
FC
3459.00 1221.61 4680.61 2072.00
x
24.16 33.58 57.74
Unit Cost, UC ($)
x
x 2.81 30.63 3.08 79.22 2.04
242.60 67.39 309.99
Build Time, TF (min) System
FC
Dimensional Discrepancies, SD (mm)
Material Cost, MC ($)
Dimensional Discrepancies, SD (mm) x 12*** 25* 4 6 18**
FC
f(x)
0.00 0.00 1.00 0.67 0.00 1.67
0.00 0.00 0.21 0.19 0.00 0.40
2.10 1.27 3.37 1.2
values removed after first iteration values removed after second iteration values removed after third iteration
Figure 8. The results of the ranking system for equally weighted factors. Table 8. Factors for which no contributions were made to the ranking score. System
Non-contributing factors
3D Touch V-Flash uPrint SD300 Pro Replicator
TF, SD UC, MC, SD UC MC SD
demonstrate that the ranking method is sensitive to the factors used in the model, and the model can arguably be used to detect differences in technologies that led to the differences in the factors (such as build layer thickness, the effect of layer thickness on part dimensional accuracy, the effect of a temperature controlled environment, etc.). The method of developing a ranking system to evaluate a population of 3D printers is also useful in that it can be customised based on the needs of the consumer by simply removing parameters that are not important in a given situation. For example, if unit cost is not a concern, the parameter can be removed or weighted differently and the pool of evaluated 3D printers can be re-ranked.
5. Recommendations Quantitatively, based on our ranking system, the Replicator ranks highest in our comparison of desktop units, but it may also be important to consider other qualitative aspects depending on the intended application. For an educational tool intended to proliferate knowledge pertaining to additive manufacturing, the extrusion-based units tested in this study are a logical choice as it is easy for one to
Virtual and Physical Prototyping
observe the build process in these technologies. The V-Flash and SD300 Pro, for example, present difficulties in observing the build process as the V-Flash has no window into the build chamber and the SD300 Pro only offers top-down visibility to the build envelope. The Replicator, as mentioned before, is the most portable of the units tested in this study based on the reasonably low weight and ability to fabricate a part with the use of only a SD card. Though the dimensional accuracy measurements for our test part revealed a tendency to oversize some of the features, the ease of use, and build time make the Replicator a suitable choice for the home user as well as for use as an educational tool. The caveat for use of the Replicator in an educational setting is the exposed filament spools, build envelope, and extruder component, though the newly released Replicator 2 has an enclosed build envelope. Although the 3D Touch ranked fourth in our analysis, it is also an option for use as a portable learning tool and for home 3D printer use. The main drawbacks to the 3D Touch are the slow build time and portability of the system due to its weight. However, advantages of the 3D Touch include ease of use due to the ability to fabricate a component by reading information from a SD card and the fact that it ranked second in terms of dimensional accuracy among the extrusion-based units tested in this study. The slow build times could potentially be remedied leading to an improvement in its ranking in our study. The filament spools and the extruder component are also shrouded by the housing of the unit increasing the overall safety of the unit. In terms of dimensional accuracy, the uPrint includes a temperature controlled build envelope and provides the best performance. The use of a soluble support structure enables the creation of more complex components as compared to the other extrusion-based units tested in this study. The enclosed build envelope and material spools also make the unit stand out in terms of safety. The high weight, potential need to utilise a secondary process to remove the support structure, and comparatively high cost may limit use by the casual home user or as a learning tool which could be taken to various Elementary and Secondary educational institutions. For the consumer with the necessary financial resources to purchase the uPrint, this unit may be the best for technical applications such as small engineering firms. The multitude of AM techniques which are being applied in a desktop format is a testament to society’s acceptance and need for the technology. Robust evaluation and comparison of the various AM systems is necessary to fulfil AM’s benefit to society. The ranking system presented in this paper has demonstrated the ability to discriminate between different AM processes and rank these systems based on quantitative measures. The process described in this paper is a work in progress; however the iterative model described in this paper can be
211
used by a potential AM machine user to determine which unit will best meet a given need. The end goal is to allow a user to send a file for a standard test part to a pool of manufacturers who would then send the user a 3D-printed part. Data derived from physical measurements of the user’s choosing would then be subjected to the iterative process and allow the user to determine which AM machine would best suit a particular application. Future work by the authors includes the development of a more robust test structure which will include inclined planes, the testing of an additional pool of AM machines, and the development of an online process by which a potential AM machine user can download the.stl file for a standard test file and perform the iterative ranking process based on data they deem to be important using an online calculator.
Acknowledgements The research presented here was performed at The University of Texas at El Paso (UTEP) within the W.M. Keck Center for 3D Innovation (Keck Center), expanding recently to over 13,000 sq. ft. as a result of funding from the State of Texas Emerging Technology Fund and providing access to state-of-the-art facilities and equipment. The authors extend appreciation to Alfonso Fernandez, undergraduate student researcher in the Keck Center, for contributing in various ways to this work.
Funding This work was supported by the National Science Foundation (NSF) [grant number HRD-1139929], [grant number HRD-0703584]; Mr. and Mrs. MacIntosh Murchison Chair I in Engineering Endowment.
References American Society for Testing and Materials, 2012. ASTM F2792-12a. Standard terminology for additive manufacturing technologies. ASTM International, West Conshohocken, PA. Brajlih, T., et al., 2011. Speed and accuracy evaluation of additive manufacturing machines. Rapid Prototyping Journal, 17 (1), 6475. Cano, J.L.C., 2011. The Cambrian explosion of popular 3D printing. International Journal of Interactive Multimedia and Artificial Intelligence, 1 (4), 3032. Crump, S.S., 1988. The fused deposition modeling (FDM). United States Patent Applications 5,121,329 and 5,340,433. Espalin, D., et al., 2010. Fused deposition modeling of patient-specific polymethylmethacrylate implants. Rapid Prototyping Journal, 16 (3), 164173. Espalin, D., et al., 2009. Effects of vapor smoothing on ABS part dimensions. In: Proceedings of the RAPID 2009 Conference & Exposition, May 1214, 2009, Schaumburg, IL, USA. Espalin, D., et al., 2013. Multi-material, multi-technology FDM: Exploring build process variations. Rapid Prototyping Journal, in press.
212
D.A. Roberson et al.
Ghany, K.A. and Moustafa, S.F., 2006. Comparison between the products of four RPM systems for metals. Rapid Prototyping Journal, 12 (2), 8694. Gonzalez-Gomez, J., et al., 2012. A new open source 3D-printable mobile robotic platform for education. In: U. Ru¨ckert, S. Joaquin, and W. Felix, eds. Advances in autonomous mini robots. Berlin Heidelberg: Springer, 4962. Grimm, T., 2003. Fused deposition modeling: A technology evaluation. Time-Compression Technologies, 2 (2), 16. Ho, K.-L.G., et al., 1999. Degradation of polylactic acid (PLA) plastic in Costa Rican soil and Iowa State University compost rows. Journal of Polymers and the Environment, 7 (4), 173177. Jayaram, D., et al., 1994. Benchmarking of rapid prototyping systems Beginning to set standards. In: Proceedings of the 1994 Annual International Solid Freeform Fabrication Symposium, August 810, 1994, Austin, TX, USA. Mahesh, M., et al., 2004. Benchmarking for comparative evaluation of RP systems and processes. Rapid Prototyping Journal, 10 (2), 123135. Malone, E. and Lipson, H., 2007. Fab@Home: The personal desktop fabricator kit. Rapid Prototyping Journal, 13 (4), 245255.
Moylan, S., et al., 2012. Proposal for a standardized test artifact for additive manufacturing machines and processes. In: Proceedings of the 2012 Annual International Solid Freeform Fabrication Symposium, August 68, 2012, Austin, TX, USA. Pandey, P.M., Reddy, V., and Dhande, S.G., 2003. Improvement of surface finish by stair case machining in fused deposition modeling. Journal of Materials Processing Technology, 132 (13), 323331. Pearce, J.M., et al., 2010. 3-D printing of open source appropriate technologies for self-directed sustainable development. Journal of Sustainable Development, 3 (4), 1729. Schmidt, A., Doring, T., and Sylvester, A., 2011. Changing how we make and deliver smart devices: When can i print out my new phone? IEEE Pervasive Computing, 10 (4), 69. Valero-Gomez, A., et al., 2012. Printable creativity in plastic valley UC3M. In: 2012 IEEE Global Engineering Education Conference (EDUCON), 1720 April 2012 Marrakesh, Morocco. 19. Wohlers, T.T., 2012. Wohlers report 2012, rapid prototyping, tooling and manufacturing: State of the industry. Annual Worldwide Progress Report. Fort Collins, CO: Wohlers Associates Inc.
