By Terry Wohlers
Published in Vol. 11, No. 10, October 1992 issue of Computer-Aided Engineering
Copyright 1992 by Terry T. Wohlers
Evaluating rapid prototyping (RP) systems can be a daunting task. Not only is it important to measure the safety and durability of materials and the accuracy of the process, but it's also imperative to weigh the cost of installation, maintenance, and training.
In fact, system speed and cost may rank as the most important elements to consider, for without reasonable performance and price, the process would not be rapid or practical. These elements were so important to Chrysler's Jeep and Truck Engineering (JTE) that it set out to compare them.
Under the direction of Lavern Schmidt, manager of Design Aid & Packaging at JTE, the company chose to produce a single automobile part on five RP systems. It selected a speedometer adaptor as the benchmark part because of its small size (1.5 x 1.5 x 3 in.) and fine detail. The group's familiarity with the build characteristics of the part on its SLA-250 systems was also a selection factor. The SLA-250 StereoLithography Apparatus, from 3D Systems, Valencia, CA, is the most widely installed and used RP system in the world. JTE owns two of them and is presently shopping for a third RP system.
Since Chrysler does not own all of the RP machines compared in its study, the company sent an STL file of the speedometer adaptor to Cubital, DTM, Helisys, Stratasys, and 3D Systems. The STL file is a de facto standard format developed by 3D Systems and accepted by all major RP systems. STL files contain groups of three x-y-z coordinates, each defining a triangle. Together, the groups define a connected set of triangular facets that describe the shape of the original CAD model.
Stratasys' 3D Modeler and DTM's Sinterstation 2000 products were able to produce the part the fastest-more than twice that of Helisys' Laminated Object Manufacturing (LOM) and Cubital's Solider systems. 3D Systems' SLA-500 and SLA-250 came in third and fourth.
Interestingly, the faster and more expensive SLA-500 did not produce the part much faster than the SLA-250. Yet the SLA-500's laser draw speed is 100 in./sec, compared to the SLA-250's 15 in./sec.
Without benchmark data such as Chrysler's, one could conclude that the SLA-500 is many times faster than the SLA-250 when reading product spec sheets only. Focusing on the steps required before and after the actual solidification of the part, then, becomes an important element in the evaulation process because they measure actual throughput, not just machine build speed.
Machine build time usually refers only to the actual production of the individual layers that make up a part. Chrysler's benchmark work involved throughput-the total time required to make the part, which includes preprocessing and postprocessing.
Preprocessing includes the time it takes to prepare the computer files for processing on the RP machine and preparing the machine for operation. It also includes the creation of support structures (for RP systems that require them), slicing of the STL file, and merging of files prior to starting the build.
Postprocessing usually involves the time it takes to clean the part, remove support structures, post cure the resin, and finish the surface of the part. Because the pre/postprocessing steps can vary greatly from one RP system to the next, these steps should be carefully considered before purchasing a system.
However, they may or may not represent a significant part of the total time. For instance, Stratasys' 3D Modeler's pre/postprocessing times involved 2 hrs and 20 min, while the machine build time required only 2 hrs and 10 min. Meanwhile, pre/postprocessing time for the Helisys' LOM system consisted of only 1 hr and 11 min versus 10 hrs to build it.
While pre/postprocessing time is important, don't ignore the build time. Not only does it represent a large part of the total time, the RP machine cannot accept additional work while it is building parts. Therefore, machines with lengthy build times produce fewer parts in a single day. Note, however, that most preprocessing tasks can occur on jobs in the queue. Likewise, skilled technicians can postprocess built parts as the machine is constructing new ones.
It is especially important to note here that layer (slice) thickness impacts build time. Parts that have 0.005-in. layers contain twice as many layers as parts that have 0.010-in. layers, and, therefore, require more time to make. Also, layer thicknesses can differ from one test part to the next. For example, the part produced by Stratasys' 3D Modeler contains layers that are five times thicker than the layers produced by Cubital's Solider system. As a result, the surface finish of the Solider part is superior.
Layer thickness also affects the time it takes the computer to slice the layers. Slicing time is not a factor with the Cubital, DTM, and Helisys systems because their machines begin to build the part as soon as the computer has sliced the first layer. Not so with the 3D Systems and Stratasys machines. They begin to build the part after slicing is complete. So when you compare times in this and other benchmarks, pay close attention to differences in layer thickness.
Also, consider that the benchmark involved the production of one part only. Certain RP systems permit you to build several parts simultaneously, without paying a significant time penalty. Using the Cubital Solider system, for example, the time required to build multiple parts is the same as required to build just one. The outcome of the benchmark, therefore, would have been different had it involved the production of 5, 10, or even 40 parts.
DTM and 3D Systems equipment also enables you to build multiple parts at once. However, their machines would require more time per additional part than Cubital's system, which uses a flood of light. Systems from DTM and 3D use a laser that must work longer to solidify each additional part. Yet, a time savings would result from the fact that parts are able to share the time it takes to deposit a new layer of material. With any of the RP systems, you must tack on extra time for pre/postprocessing for each additional part you build.
As for costs, the benchmark considered the purchase cost of the system as well as maintenance, personnel, material, and depreciation. Using these, Chrysler estimated the actual cost of producing one prototype part of the speedometer adaptor, as if it had bought and operated each piece of equipment.
Whether the RP system can run by itself or requires an attendant is a considerable cost factor, according to the benchmark results. In fact, attended operation of the DTM and Cubital systems represents about 38% and 58%, respectively, of the total part cost. This may be, in large part, why these two systems are the most expensive to operate, based on the benchmark data.
Maintenance cost per part can also be a significant portion of part production cost. Chrysler found the maintenance cost of Cubital's Solider system to be about $56 for the single part, representing 15% of the total cost of the part. 3D Systems' SLA-500 was the second highest, $46, representing about 31% of the part cost. Meanwhile, the system maintenance cost of Stratasys' 3D Modeler was calculated as the lowest, costing only $1.73 for the speedometer adaptor part. This represented only about 1% of the total part cost.
According to the benchmark results, material cost represents only a small fraction of the cost to produce the speedometer adaptor. Ranging from about $4 to $6 for all parts built, it is only 2-3% of the total cost per part, on average.
This means that even moderately expensive material should not be a major factor when evaluating systems and considering individual part production costs. However, don't overlook material costs entirely. The speedometer adaptor is very small. At $250 to more than $700 per gallon, an organization with one SLA-250 can easily go through $16,000 worth of resin in one year, according to Tom Sorovetz, a designer and stereolithography applications and systems engineer at JTE.
Helisys' LOM-1015 was the least expensive system to use to make the Chrysler part. At $85,000, it is also the least expensive to buy. But while Helisys' device was the least expensive to use, it was among the slowest. Only the Cubital Solider system was slower.
3D Systems' popular SLA-250 came in second as the least expensive system to produce the speedometer adaptor. Cubital's Solider 5600 was by far the most costly. Using Cubital's system to build the part costs 3-4 times more than either the SLA-250 or LOM system. At $490,000, the Solider system is also the most expensive system to purchase and operate.
Stratasys' 3D Modeler produced the part faster than any other system tested, and its maintenance cost is the lowest, by a large margin. In fact, maintenance is 11 times lower than the LOM system, which is the second lowest, according to Chrysler. DTM's Sinterstation 2000 scored well in the total time category but was the second highest in total cost. Still, this cost is not substantially more than either the SLA-500 or 3D Modeler.
Chrysler did not attempt to evaluate the strength and surface finish of the benchmark parts. This was not the purpose of the study, reports Schmidt, even though these aspects of a prototype part can be very important.
Schmidt's group did measure the parts for accuracy using a Sheffield Measurement CMM, but Schmidt downplays the value of this data. "All parts were well within our requirements," says Schmidt.
So, many factors come into play when trying to compare accuracies of machines. If three different machine operators were to make the same part using identical machines, the dimensions of the three parts would be different. Skill and experience can affect part accuracy as much as the machine itself.
If the results of this benchmark tempt you to conclude that one system is substantially better than another, think again. Benchmark results can differ widely, and they measure only certain elements of a system. If you were to benchmark the same systems, your findings would probably disagree with those presented here.
Results depend on the part selected and the guidelines followed in testing. A large flat object with lots of surface area, for example, tends to build very fast on the Helisys LOM system, yet it was one of the slowest to build the Chrysler part. Also, prices and system specifications change regularly. What may represent the worst price/performance ratio today may be among the best tomorrow.
When evaluating systems, take into consideration the following factors:
System complexity. Moving components and replacement parts increase the chances of additional expenses and downtime. Complex systems also require additional hours of training for those who will be using and maintaining them.
Size, weight, and portability. You may want to move the system to another room or building. Consider whether the equipment is reasonably safe to use and what facility changes, such as wall partitions and ceilings, the installation may require.
Operating paraphenalia. In addition to the investment of between $85,000 and $490,000 for a rapid prototyping system, budget an additional 30-90% of this amount for other costs. Stereolithography systems, for example, include a post cure oven, cleaning tools such as an ultrasonic cleaner, and protective equipment.
Material waste. The Cubital and Helisys systems both waste considerable material. The Stratasys and DTM systems waste the least amount. 3D Systems' SLAs waste only the resin material that clings to the part when you remove it from the build chamber. n Producing CAD data. If you're not creating solid models or fully closed, water-tight surface models, this will need to change if you plan to install an RP system.
Personnel. Finally, be careful when choosing the individuals that will investigate, implement, and manage the technology. Involve those with a "can-do" attitude. Find another job for people who constantly find reasons why something cannot be done. They will only delay implementation, and maybe even bring it to a dead stop. If possible, you should find one individual to champion the use of RP at your organization. The companies that have been most successful at implementing rapid prototyping almost always have such a person.
Chrysler's Jeep and Truck Engineering (JTE) owns two SLA-250 systems and operates them about 20 hrs/day, 7 days/week. In addition to Schmidt, three full-time employees staff the rapid prototyping operation. Since January 1990, the company has produced more than 2,225 parts representing over 840 geometries.
The models produced by JTE give Chrysler engineers the opportunity to evaluate the design of a physical part, demonstrate its feasibility, and sell the concept to management. The company may employ an SLA model to check appearances, such as the styling of a gear shift lever, or to check for interference problems in subassemblies.
SLA parts can also function as masters for tooling. One project, for example, involved the creation of a V-10 exhaust manifold for the Viper automobile. The SLA system saved 8 weeks in prototyping time and more than $50,000. JTE was able to achieve a dimensional tolerance of % 0.0025 in. over the entire manifold, which was accurate enough for production tooling-resulting in an additional time savings of 10 weeks.
Terry Wohlers is President of Wohlers Associates, Fort Collins, CO.
Copyright 1992 by Terry T. Wohlers