Training technologists in experimental design 作者:James M Cupello 年代:1999 出處: Research Technology Management. Washington: 卷期 42, Iss. 5; pg. 47 報告者:蘇暘展 Outliner Introduction 1. Curriculum Content 2. Teaching Skill 3. Course Materials 4. In-Class Experiments 5. DOE Software 6. Integration with Other TQM Tools 7. Select Taguchi Methods In Conclusion Introduction Statistically designed experiments have proven to be one of the few reliable weapons in the 20th century arsenal of globally competitive firms. R&D managers would benefit greatly if they could identify the vital few outstanding courses from among the hundreds of academic and commercial ones available on this topic. Consequently, 7 best practices associated with such a world-class course are postulated. 1. Curriculum Content "Curriculum content" and "teaching skill" are truly first among equals when looking at all seven best practices. (1) Calculating factor effects for simple two-level designs, preferably via a student exercise. (2) Inclusion of a thorough explanation of interaction effects, their importance and how they are calculated. (3) Introducing the concept of orthogonal arrays and how they are constructed for two-level designs. (4) A thorough definition, explanation and demonstration of "confounding" in fractional factorial designs. (5) The benefits of DOE are compared with the traditional one-factor-at-a-time (OFAT) designs. (6) The concept of using center points in two-level designs is rigorously explained and demonstrated. (7) Normal and half-normal probability plots are explained and illustrated by example or in-class exercise. (8) Plackett-Burman screening designs are introduced and explained, and contrasted with traditional fractional factorial designs. (9) The concept of "foldover designs" is introduced and explained. (10) Mixture designs are briefly explained. (11) The technique known as the "method of steepest ascent" is introduced and explained. (12) The advanced topic of "response surface methodology" (RSM) is introduced by way of lecture. Total time allocated to RSM should not exceed two to three hours. The resulting course content is captured in the illustration, next page, which partitions the various course topics into four areas that mimic the Deming PDCA cycle: Planning (pre-experimental), (15%) Design (types), (30%) Conduct (of an experiment), (20%) Analysis (of the data). (35%) During course development, I identified ten key concepts or pedagogical best practices associated with a DOE course; they are listed below. Students should have at least one team exercise early in the course where they use their "pre-course" knowledge to plan and execute an in-class experiment. This experience then serves as motivation to learn the new DOE methodology. Because variation is a researcher's relentless enemy, the course should emphasize issues involving variability: blocking, randomization, measurement error, recording error, bias, confounding, residual plots, analysis of variance, etc. Experimentation should be sequential. Emphasize that no more than 25 percent of one's research budget should be spent on the first experiment. Multiple responses are a fact of life in today's complex world. Students should be shown how to simultaneously optimize a process involving more than one response variable. Software is essential for DOE; no one does calculations by hand. Two-level designs are the workhorse of DOE. At least half of the course schedule should be devoted to them. A transformation of the response variable is occasionally required. This is an advanced topic. Students should be introduced to the idea of transformations and when they might be needed, appropriate or essential. Sample size matters a lot. Students should be shown how to estimate the sample size required to detect an effect of a given size for a two-level factorial at various confidence (or, 1) levels Taguchi's concepts of variance reduction and robust design are not only important from a pedagogical standpoint; they are considered a best practice for a world-class DOE course. Students must learn how to explain their experimental results in the context of management briefings without using DOE jargon and buzzwords. 2. Teaching Skill Outstanding teachers are truly valuable human beings, more so because they are rare. A prospective DOE instructor should provide copies of student evaluations of his/her teaching of the DOE course under consideration. Evaluation forms (or rating scores) from all former students should be included to prevent biased reporting. A prospective DOE instructor should provide the names, addresses and phone numbers of at least two references who took his/her course, and who are willing to comment on its content and the quality of instruction. Managers should have at their disposal information relating to teaching skill from one or more of the following sources: student evaluations, reference checks, and/or course audits. With this information, it should be possible to effectively evaluate a course and its instructor using the following criteria: 1. The instructor's demonstrated subject-matter knowledge; 2. The instructor's demonstrated ability to respond effectively to student questions and comments; 3. The instructor's demonstrated ability to present information and concepts effectively; 4. The manager's overall assessment of the instructor's effectiveness as an educator. 3. Course Materials Essential course materials for a world-class introductory DOE course Student Textbook Course Notes Supplemental Readings Design Toolkit DOE Software Student textbook Only a few of the benchmarked courses provided a student text as part of the course offering. Students should be provided with an exemplary textbook to keep after the course is over. Course notes It has been claimed that 80 percent of learning occurs through the sense of sight . If true, this means that the use of overhead slides by the instructor can be a significant improvement over pure lecture without notes. Distribute copies of the instructor's lecture slides or overheads at the beginning of class, to facilitate notetaking. Supplemental readings Provide students with a useful and extensive selection of easy-to-understand journal articles relevant to the subject of experimental design. Design toolkit Provide students with an extensive collection of useful design tools/aids that simplify the design of an experiment and the analysis of data. DOE software Consider introducing any in-class activity designed to reduce student anxiety typically associated with mathematics, statistics, and computation. 4. In-Class Experiments With regard to in-class experiments, students should work alone, or preferably in small groups of three to five, to generate and analyze their own experimental data during class. Consider the use of one or more video presentations in class to present course material in an interesting and informative way. 5.DOE Software Students should be provided with user-friendly DOE software during the course for analyzing data. Students should be provided a free personal copy of the course software for use upon their return to work. 6. Integration with Other TQM Tools Two additional tools-flow charting and cause-and-effect diagrams-tend to be used together to solve problems. The flow chart documents the actual step-by-step process used in the conduct of an experiment; the cause-and-effect diagram is a vehicle for brainstorming actions that can be taken to reduce variation in each step on the flow chart. 7. Select Taguchi Methods Taguchi introduced the concept of robust design, which is a simple experimental design methodology that attempts to make products and processes insensitive to the degradation posed by uncontrollable factors (noise) in the environment. Any world-class DOE course should explain or illustrate Taguchi's approach to identifying variance-reducing factors. In Conclusion DOE is an important methodology for any company involved in product or process development. It is becoming increasingly difficult to compete in a global economy without the use of this and other select performance-enhancing tools. Considering that a firm's technical staff consists of some of the highest paid individuals on the payroll, their time and talents are too valuable to waste by sending them to less than stellar courses and workshops. The End