5/25/2016Launsby Consulting1 Process Validation and Design of Experiments The following PowerPoint presentation was presented by Robert Launsby at the MDM Conference in Anaheim, CA on February 12, For more information regarding Design of Experiments and Process Validation go to:
Process Validation and Design of Experiments Launsby Consulting 5/25/20162
Robert G Launsby President of Launsby Consulting in Colorado Springs MS in engineering Taught thousands about this topic APPLICATIONS focused Author of four books Co-developer of WISDOM software 10k gold medalist at National Senior Games /25/20163Launsby Consulting
Agenda Enhancing New Product and Process success rate How to link DOE (design of experiments) with Process Validation Quick DOE example How to link DOE analysis and Monte Carlo Analysis to PV activities A brief example using Statabot 5/25/20164Launsby Consulting
How To Improve New Product Success Rate Customer focus (how do we make the customer successful?) Management leads change, All Levels Implement a development process and have the discipline to use it Make data driven decisions, use tools Metrics to map progress 5/25/20165Launsby Consulting T≠C
Roadmap 5/25/2016Launsby Consulting From book “Straight Talk on Process Validation” available at Amazon.com Get your design inputs right at system/sub system/com ponent level early
Roadmap 5/25/2016Launsby Consulting
Roadmap 5/25/2016Launsby Consulting
Roadmap 5/25/2016Launsby Consulting
In a Nutshell IQ….equipment setup correct? OQ…can we make a good part? Can we make good parts at worst case? This is where Design of Experiments supports PV PQ…can we make many good parts under production conditions? 5/25/2016Launsby Consulting
What Is A Designed Experiment? Systematic, controlled changes of the inputs (factors) to a process in order to observe corresponding changes in the outputs (responses). 5/25/201611Launsby Consulting Why do this: 4 times the information with ½ the tests
What Is A P-diagram? PROCESS Inputs Outputs 5/25/201612Launsby Consulting
Engineering Experimental Design Not a substitute for knowledge of technology Incorporates current understanding Physics first It is all about good scientific understanding (with some math blended in) 5/25/201613Launsby Consulting
Steps In Conducting DOE Define Objective, Select Factors, Levels, Responses, etc Automated by software Have a plan, be there Graphs, statistical analysis, predict responses at best set points Demonstrate with data the prediction from transfer function is useful Keys Plan Select O.A. Conduct Analyze Confirm 5/25/201614Launsby Consulting Taken from the text “Engineering Today’s Designed Experiments” available at Amazon.com
An Example RunNaClEDTAActivity 15127, , , , 30 5/25/201615Launsby Consulting Suppose we have processed an enzyme and want to store in a buffered solution. We want to maintain highest activity level while in storage. What are best conditions for NACL and EDTA to achieve this goal?
Pareto Chart 5/25/201616Launsby Consulting
Main Effects Plot 5/25/201617Launsby Consulting
Interaction Plot 5/25/201618Launsby Consulting
What Is An Interaction? Refers to synergism between factors relative to a response. Two factors interact if the influence of one factor is impacted by the level of another factor 5/25/201619Launsby Consulting
Transfer Function The equation (algebraic) It comes from MLR Three important assumptions Two levels O.A Variables are on orthogonal scale Software packages use MLR to generate transfer function 5/25/201620Launsby Consulting
MLR Math includes factors (assumes 4 run previous example), and interaction effect 5/25/201621Launsby Consulting
Contour Plot 5/25/201622Launsby Consulting
RSM Plot 5/25/201623Launsby Consulting
Basic Statistical Analysis 5/25/201624Launsby Consulting
DOE and PV Steps 1. Select key factors/levels responses (based upon pre-PV characterization studies) 2. Conduct Orthogonal Array 3. Perform analysis 4. Predict best set-points to target response(s) 5. Confirm above predictions 6. Using math. model from DOE, conduct MC using worst case for inputs 5/25/2016Launsby Consulting
DOE and PV Steps (continued) 7. Plot variation from Monte Carlo analysis 8. Run process at actual worst case scenario 9. Ensure MC analysis and actual results at worst case provide equivalent and acceptable potential process capability 5/25/2016Launsby Consulting
DOE and Monte Carlo Using the Statabot 5/25/2016Launsby Consulting
Graphical Analysis of Statbot DOE 5/25/2016Launsby Consulting
Outputs From Prediction Model 5/25/2016Launsby Consulting Set motor A at 40, motor B at 42 to target 20 seconds for response
Confirmation of DOE at 20 Seconds 5/25/2016Launsby Consulting Great potential capability at predicted settings, but very short term Note: suppose we know batteries power can vary by +/- 5% before software in Statabot shuts down….what is potential impact of this phenomena?
Monte Carlo Analysis Using parsimonious equation from DOE Vary input power by +/-5 in Monte Carlo analysis and plot the output in response 5/25/2016Launsby Consulting This is a simulation of longer term variation
Recommendations Software for DOE Minitab or JMP if you are well versed in statistics…and need package with numerous capabilities DOE Wisdom if you have little statistics background and just need DOE support Fimmtech’s NAUTILUS software if you are injection molder 5/25/2016Launsby Consulting