Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

DOE - Design Of Experiments

7,444 views

Published on

DOE - Design Of Experiments

Published in: Business
  • Be the first to comment

DOE - Design Of Experiments

  1. 1. Design of Experiments FF E A T U e E eatur R By Mark Anderson ndustrial physicists no longer can afford to experi- clearly indicate the cause of the paint’s limited perfor- ment in a trial-and-error manner, changing one factor mance, says supply management engineer Wayne Mills. at a time, the way Edison did in developing the light- John Deere first ran a screening design to identify the bulb. A far more effective method is to apply a comput- important variables among the 12 that its experimental er-enhanced, systematic approach to team had established, which included the casting DOE considers experimentation, one that considers method, pretreatment process, paint type, and the con- all factors si mul ta neousl y. That centration of adhesive ingredients. The results surprised all the variables approach is called design of experi- everyone. Chromate conversion had very little effect on ments (D OE), an d corpor at ion s paint adhesion, but paint type proved a very significant simultaneously and across the nation are adopting it as a factor. No one at John Deere had considered paint type predicts response cost-effective way to solve serious as a major player. More tests narrowed the variables to problems afflicting their operations. three: chromate conversion, paint type, and surface DOE provides information about the treatment. A three-dimensional-cube plot generated by over a wide range interaction of factors and the way the the DOE software clearly showed how the three vari- of values total system works, something not ables interacted. The results again identified paint type obtainable through testing one factor as the major problem. at a time (OFAT) while holding To confirm the software’s other factors constant. Another results, Mills and his col- advantage of DOE is that it leagues performed a tradi- shows how interconnected fac- tional one-variable test with tors respond over a wide range several data poi nts. This of values, without requiring the final test convinced all the testing of all possible values members of the experiment directly. te am. As a result, Joh n DOE fits response data to Deere solved its paint adhe- mathematical equations. Col- sion problem and eliminat- lectively, these equations serve ed the chromate-conversion as models to predict what will pretreatment process from happen for any given combina- aluminum parts—a change t ion of va lu es . Wit h these that has saved the company models, it is possible to opti- nearly $500,000 annually, mize critical responses and according to Mills. find the best combination of Better approach va lu es . DO E software is Figure 1. Software generates rotatable 3D tremendously fast, enabling When attacking a new plots of interacting variables. the testing of many factors in design challenge, an indus- just days. That’s drastically trial physicist must consider shorter than the months, if not years, spent testing only many factors simultaneously. Variations may occur, for a handful of factors using the traditional method. DOE example, in casting, machining, finishing, plant condi- takes some of the art out of experimentation, replaces it tions, parts assembly, machines, and operators. Besides with science, and yields better results. this group of obvious factors, there are subtler ones— Consider the experience of the John Deere Engine factors that only those very familiar with the process Works (Waterloo, Michigan) and its search for a way to can identify, such as the length of the production cycle. improve paint adhesion to aluminum components And still other complications stem from the interac- while eliminating the increasingly expensive chromate- tions of factors among themselves, which are generally conversion coating process as a pretreatment to paint- understood by no one. ing. It was difficult to find a basis for improving the The physicist who must justify experimentation by paint’s adhesion because John Deere’s data did not supplying a cost-benefit ratio is under pressure to gain SEPTEMBER © 1997 American Institute of Physics 24 The Industrial Physicist
  2. 2. Figure 2. John Deere broke free from an expensive coating process the maximum amount of information from each experi- after this software mental run. After all, a detailed study of all factors is program design generally cost- and time-prohibitive if it requires many runs. Because of this, physicists must limit the number builder guided of factors tested to the most obvious. The contributions experimenters to a of the factors they assign as minor are rarely discovered. profitable two-level Later, after months of testing, the experiments end. The process shows improvement, but how much more factorial design. would the process have improved if other factors had to design the tool as one piece to further ensure high been studied? That’s usually not answered with the con- repeatability. This contrasted with the former tooling, ventional approach. which used components that had to be taken apart and In OFAT, the first factor is fixed as a “good” value, the reassembled for each new run. next factor is examined, and on and on to the last factor. By applying the DOE experimental method, Kodak Because each experimental run considers only one fac- saved the cost of a new stamping machine and through tor, many runs are needed to get sufficient information retooling, increased repeatability to 100%, reduced about the set of conditions contributing to the problem. scrap by a factor of 10, slashed inventory, and reduced This consumes a hefty amount of time and money, along setup time from 8 hours to 20 minutes. with running a high risk of error. Another limitation is Software speeds DOE application that when factors change, they generally change togeth- er, so it is impossible to understand the best solution by For decades, a few physicists have been manually pointing to a single, isolated factor. Traditional OFAT applying DOE principles by using algorithms found in experimentation frequently reduces itself to no method- handbooks. During the last 15 years, the development ology whatsoever—just trial and error and a reliance on of DOE software packages has made DOE more accessi- common sense. In contrast, the systematic DOE ble to physicists and engineers. Today’s software is approach provides information about the interaction of mostly menu-driven, and statistical jargon is minimized. variables and the way the total experimental system Developers have eased the complexity surrounding DOE works, something generally not attainable through the by increasing the emphasis on graphically displayed OFAT approach. results over numerical tables. The software is relatively Eastman Kodak (Rochester, New York) found DOE inexpensive, currently ranging from $100 to $1,500. quite cost-effective when its professional imaging finish- With a little training (typically a three- or four-day work- ing department had difficulty forming a small steel clip shop), people can integrate DOE into their experimenta- with consistent reliability. The 4×1×0.005-in. clip, used tion strategies. t o s eal o ne e nd of pr eloa ded p hotographi c film The first step in using DOE software is to define the envelopes, required a tight tolerance of a few thou- factors and the range in which each factor varies. Once sandths of an inch during a critical stamping operation the factors and their ranges are determined, the user in which three angles and a radius were formed. Accord- enters the values into the software. The first experimen- ing to Peter Runke, senior manufacturing engineer, tal run identifies important factors. Subsequent runs Kodak did not know whether it needed to improve the fine-tune the critical factors. To confirm the software’s tooling or spend several thousand dollars for a new results, the experimenter performs a traditional one-fac- stamping machine. tor test with several data points. After performing 32 experimental runs with DOE The advantage of DOE software is its ability to show software, the Kodak team found that of the six factors it the physicist real-world results. DOE helps physicists identified for examination, three were significant. Blade complete the testing process faster than the traditional clevis (ability to split) created the main effect. Blade experimental method. Just as important, DOE reveals shim (taper) was second, and tool pressure was third. solutions that would not be apparent without it. Runke says this evidence proved that the fault lay in the Ski manufacturer K2 Corporation (Vashon, Washing- tooling, not in the machine itself. This provided the jus- ton), a division of Anthony Industries, encountered a tification for retooling, and Kodak took the opportunity number of processing problems when the company 25 The Industrial Physicist
  3. 3. DOE Software Vendors Advanced Automation Corp. Experiment Strategies PQ Systems, Inc. retooled its existing molds for a new “cap” 16 Thomas Drive 1308 18th Street 10468 Miamisburg-Spring- ski design that featured a smooth, encased Rome, NY 13440 Anacortes, WA 98221 boro Road look. The new skis exhibited dimples, blis- Phone: 800-292-4519 Phone: 360-299-0700 Miamisburg, OH 45342 ters, and other troublesome surface prob- Fax: 315-337-3692 Fax: 360-299-0202 Phone: 800-777-3020 lems. The top layer of the skintight plastic Fax: 937-885-2252 Advanced Systems F&H Simulations, Inc. casing made every blemish visible and Consultants P.O. Box 658 Qualitron Systems, Inc. impossible to sand off. Scrap rates soared to P.O. Box 1176 Orem, UT 84069-0658 1208 East Maple Road, 30%, says Douglas O. Hubbell, director of Scottsdale, AZ 85252 Phone: 801-224-6914 #100 process improvement at Anthony Industries. Phone: 602-423-0081 Fax: 801-224-6984 Troy, MI 48083 To attack the problem, K2 ran a DOE Fax: 602-423-8217 Phone: 810-616-8001 IMAS, Ltd. Fax: 810-686-7408 screening design on 17 factors, four of Air Academy Associates 675 Tollgate Road, Suite P which turned out important to the ski 1155 Kelly Johnson Blvd., Elgin, IL 60123 SAS Institute, Inc. defects: core placement, core extension Suite 105 Phone: 800-977-3248 SAS Campus Drive material, top thickness, and base width. Colorado Springs, CO 80920 Fax: 847-608-7848 Cary, NC 27613 DOE revealed a surprising interaction— Phone: 719-531-07777 Phone: 919-677-8000 Fax: 719-531-0778 International Qual-Tech, Ltd. Fax: 919-677-8123 which OFAT experimentation would never 2820 Fountain Lane North have identified—between the top thickness Applied Technology Minneapolis, MN 55447 Stat-Ease Corp. and base width that pointed to the need for 12237 Clark Way Phone: 612-476-1449 2021 E. Hennepin Avenue, different dimensions, Hubbell says. K2 con- San Jose, CA 95125 Fax: 612-475-3823 #191 firmed the accuracy of the statistical predic- Phone: 408-266-5174 Minneapolis, MN 55413- Fax: 408-266-5174 JBL Systems, Inc. 2723 tions by conducting standardization proce- 41570 Hayes Road Phone: 800-325-9802 dures. The company made the process DataNet Quality Systems Clinton Township, MI 48038 Fax: 612-378-2152 adjustments recommended by the DOE 24567 Northwestern High- Phone: 810-286-3800 software, which eliminated the blemishes. way, Fax: 810-286-5446 SPSS, Inc. K2 then put DOE to work to solve other Suite 430 444 North Michigan Avenue Southfield. MI 48075 John A. Keane & Associates Chicago, IL 60611 processing problems. Phone: 248-357-2200 575 Ewing Street Phone: 800-543-5816 By a pply in g t he D OE exp erimen tal Fax: 248-357-4933 Princeton, NJ 08540 Fax: 800-841-0064 method, Hubbell says, K2 has dropped Phone: 609-924-7904 press downtime from nearly 250 labor hours Domain Solutions Corp. Fax: 609-92l-1078 StatSoft, Inc. a week to a mere 2.5 hours and has dramat- 150 Cambridge Park Drive 2300 East 14th Street Cambridge, MA 02140 Manugistics, Inc. Tulsa, OK 74104 ically reduced delamination scrap from 1.8 Phone: 617-873-5000 2115 East Jefferson Street Phone: 918-749-1119 to 0.12%. In addition, DOE helped K2 Fax: 617-873-6153 Rockville, MD 20852 Fax: 918-749-2217 streamline many processing steps and elimi- Phone: 800-592-0050 nate some costly materials, such as a syn- Dynacomp, Inc. Fax: 301-984-5157 Technicomp, Inc. thetic rubber filler. 4560 East Lake Road 1111 Chester Avenue Livonia, NY 14487 NAG, Inc. Cleveland, OH 44114 DOE offers industrial physicists and their Phone: 716-346-9788 1400 Opus Place, Suite 200 Phone: 800-735-4440 engineering colleagues the opportunity to Downers Grove, IL 60515- Fax: 216-687-0637 design better parts at lower cost and in less Exact GTQ, Inc. 5702 time. Its cost-effectiveness, greater speed, 1400 Boul, Parc Tech- Phone: 630-971-2337 Yura Technologies, Inc. and ability to reveal design solutions not nologique, Suite 206 Fax: 630-971-2706 16 Thomas Drive Quebec City, QB Rome, NY 13440 apparent with the traditional experimental Canada GlP 4R7 NCSS Phone: 800-292-4519 method make the DOE approach increas- Phone: 418-650-2723 329 North 1000 East Fax: 315-337-3692 ingly vital in maintaining American indus- Fax: 418-660-6901 Kaysville, UT 84037 try’s competitive edge. Phone: 801-546-0445 Experience in Software, Inc. Fax: 801-546-3907 2000 Hearst Avenue Mark Anderson is a principle of Berkeley, CA 94709 Stat-Ease Corp., a DOE software Phone: 510-644-0694 Fax: 510-644-3823 v en dor ba sed in M inne apol is (mark@statease.com). 26 The Industrial Physicist

×