1 / 19

Statistical Experimental Design

Statistical Experimental Design. A Primer by H. B. Oblad (Bruce). Getting Answers Easier - Overview. The Old Method The Better Method Simple Statistics for the Lab Let’s Try It Out!. The Old Method. Experiments one variable at a time in sequence. Effect of Temperature on Yield.

sahara
Télécharger la présentation

Statistical Experimental Design

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Statistical Experimental Design A Primer by H. B. Oblad (Bruce)

  2. Getting Answers Easier - Overview The Old Method The Better Method Simple Statistics for the Lab Let’s Try It Out!

  3. The Old Method Experiments one variable at a time in sequence. Effect of Temperature on Yield Pressure = 1000 psi Time = 20 min Yield, wt % Temperature, °C

  4. Next Set of Experiments • Effect of Pressure Temperature = 300 °C Time = 20 min Yield, wt % Pressure, psi

  5. More Experiments • Effect of Time on Yield Temperature = 300 °C Pressure = 1000 psi Yield, wt % Time, min

  6. What Have We Learned? • 13 Experiments in 3 Factors Pressure Time Temp

  7. What combinations of conditions have we covered? What’s still unknown? • Do we know anything about the repeatability of our lab technique? • Are the responses straight or curved? • Can we build a meaningful model that leads to a mechanism? • Could we have done less work and gotten more information? • Minor information about effects of factors. • Know nothing about interactions.

  8. A Smarter Way Pressure Time Temp

  9. 2-Level Factorial Design • 8 Tests (XY X = levels, Y= factors) • Now know what happens over a large experimental volume. • Now know the effects of factors at two surfaces. Effect of factors tested 4 x each • Some information about interactions between factors. • Repeatability is still unknown. • Curvature?

  10. An Even Smarter Way Pressure Time Temp

  11. 2-Level Factorial Design w/ Center Points • 11 Tests (3 cntr pts), 13 Tests (5 cntr pts) • Now know the effects of factors at two surfaces and within the volume. • More information about interactions between factors. • Repeatability is now estimated or known. • Curvature can be estimated. • Predictive model is easy to create.

  12. Box-Behnken Design 3 Factor, 3 Level A fractional factorial design Spherical, so extrapolation is less risky. 15 tests (3 cp), 17 tests (5 cp)

  13. Simple Statistics • Bell Curve = Normal Dist. = Gaussian Dist. • Total population or very large sample • Errors in lab methodology are assumed random and normally distributed except for time. Must randomize order to bury effect of time into error. • Repeated tests may be pooled to estimate std. dev. and variance.

  14. Bell Curve = Normal Dist. 68% of area is <>+/-1 std. dev. 94% of area us <>+/- 2 std. dev. 99% of area is <>+/- 3 std. dev.

  15. Means Testing • If the means and standard deviations of the measurements are equal, the things being measured are of the same population. Opposite is true also (null hyp.) Use Student’s t-test.

  16. Means Testing • If the means are the same, the things are of the same population. Use Welch’s t-test

  17. Analysis of Variance(ANOVA) • Variance (standard deviation2) of means of several sample groups is determined by F-test. Probability criterion is used for pass/fail or probability of F being equal is given.

  18. Factors, Responses and Interactions • Numeric Factors are variable inputs to a process e.g. feed rate, temperature, pressure, component concentration, knobs, levers • Categorical Factors are discrete inputs e.g. catalyst type, feed material, operator • Responses are effects of changes in factors e.g. Reaction rate increases w/ temp. • Factors that affect each other are said to interacte.g. drinking, driving, vs drunken driving

  19. Rubber Band Experiment • What affects the distance traveled? • Factors? How many? • Numeric or categorical? • Which design to use? • Can we make a predictive model? • Any interaction of factors? • Can we understand the problem better?

More Related