【正文】 quantitative larger / smaller the better nominal the best Prioritise Levels Design Environment Inputs / factors Settings / levels Screening trials Optimisation trials Variables df S V F S ’ P A B Error Total 1 1 1 1 3 107 26 70 4 Analysis of Variance C Plan / do: what, when, who Repetitions, randomisation stability / documentation …… …… …… …… …… …... Follow the Route Map Response variables EN/FAD 109 0015 169。Temperature (75, 100 F)。 Ola Johansson, 1999 Ericsson Quality Management Institute A Two Level Factorial Design ? Analysing the design use the contrast method ? Add the responses, taking account of minus/plus signs (modulus) and then divide by number of runs divided by 2 Pressure + + + + Heat + + + + Time + + + + Pr*He + + + + Pr*Ti + + + + He*Ti + + + + Pr*He*Ti + + + + Pressure + + + + + Heat + + + + + Time + + + + Pr*He + + + + + Pr*Ti + + + + + He*Ti + + + + Pr*He*Ti + + + + + + + + Effect = contrast / (N/2) Response EN/FAD 109 0015 169。 Interpret Data Conduct amp。 Ola Johansson, 1999 Ericsson Quality Management Institute DOE The basics A Basic Model A body of explanatory data Measurement of Response (Potential Causes) (Effect) x1 x2 x3 y __ __ __ __ __ __ __ __ __ __ __ __ x y Independent Variable Y = a + bx parameters coefficients a b Dependent Variable A Straight Line A Statistical Model y = ?0 + ?1X1 + ?2X2 + ?3X1X2 + Error ? Linear 2 factor model ? Need to determine ? coefficients To find ? ?our estimate of real world All other variables of influence X3 X1 X1 H L H L H L X2 Basic Design Doubles No. runs EN/FAD 109 0015 169。 Ola Johansson, 1999 Ericsson Quality Management Institute Intro to Design of Experiments ? Is the process stable? You cannot accurately predict product quality (location or dispersion) without a stable process. Stability(assessed with control charts) ensures that the experimental results will provide an accurate process prediction ? What are the goals for the experiment? What factors are important? How do the factors work together to drive the process? How can you achieve optimal results from the process? These questions are actually sequential。 quality criteria Manageable evaluation and optimisation …integrating team experience and knowledge with benefits of applied statistical techniques Planning for Success the design environment EN/FAD 109 0015 169。 Ola Johansson, 1999 Ericsson Quality Management Institute Design of Experiments Common Mistakes ? Setting levels Too close means no difference in response will be detected Too wide means nonlinearities might pass undetected ? Considering Noise Failing to recognise and deal with nuisance variables ? Emphasis On Curvature Belief in plexity rather than simplicity ? Desire To Optimise Need to isolate “vital few” and identify key leverage variables before optimising