- Improve process yield Ex. Raw material type/ supplier - Reduce variability Selecting Factor Level Values - Reduce development time - Spanning entire region likely to yield the - Reduce overall costs most understanding Applications of Experimental Design If factor’s level are close, measured - Evaluate and compare alternatives effect may be statistically - Evaluate material alternatives insignificant - Product robustness - Moving off currenr operating points - Determine key design parameters presents a risk - Formulation of New Products Probing techniques: Response Basic Principles Surface Analysis Evolutionary - Replication Operation (EVOP): converge on best 1. Error Estimation Solution 2. Accuracy Effects of Aliasing : Confounding - Blocking - Aliased Parameters are CONFOUNDED 1. Unimportant Signigicant Error Cannot be estimated independtly of 2. Precision one another - Randomization Estimates are linear combination of 1. Independence confounded parameters. 2. Even out uncontrollable factors - Aliasing creates other confounded pairs Steps If ABC = D, then A=BCD ; B=ACD; - Problem statement C=ABD; AB=AD; AC=BD; AD=AC - Choice of factors,levels,and ranges Desirable Designs - Choice of response variable - Provide sufficient distribution of - Choice of experimental design information throughout region of interest - Performing the Experiment - Provide model that predicts the response, - Statistical Analysis as close as possible to true response, at all - Conclusions and Recommendation points within region of interest Resource Allocation - Provide ability to detect model lack of fit - Don’t commit all resources to one design Design Performance Considerations 1. Start with Screening Design - Number of runs 2. Only 25% of resources on any one - Design Resolution experiment - Minimum Detectable Effect - Learn from each design - Orthogonality & Balance 1. What did you do wrong? - Other : D-Optimal, A-Optimal & G-Optimal Excluded factors,wrong conditions , Orthogonality etc. - Orthogonality refers to the property if a 2. What to do next? design that assures that all specified Sometimes next stage of parameters may be estimated improvement isn’t worth cost of independently of any other another experiment - Some wrutes lump balance as part of Selecting Factors orthogonality - For each response , brainstorming likely Balance factors - Balance implies data is properly distributed - For screening , if more than 5-7 factors over design spare 1. Reduce factor list through ranking Uniform physical distribution Nominal Group Technique, An equal number of levels of each Prioritization Matrix factor - Some design sacrifice balance to achieve These design are used to investigate better distribution of variance or predicted which factors are important error - “ Response surface designs “ Ex: Central Composite These designs are used to determine - Balance may be sacrificed by avoiding the optimal settings of the extreme combinations of factors significant factors Ex: Box-Behnken Sample Designs - Box Behnken - Placket Burman - 2^k designs (fractional, confounding, fold over, projection) - 3^k designs - Mixed level Designs - Latin Squares - Central Composite( with axial Points) - John’s ¾ - Nested Designs - Split Plots - Simplex Lattice Design - Simplex Centroid Design - D-Optimal - A-Optimal
Modern approach: DOE
- DOE= Design of Experiments - Key Ideas Change several factors simultaneously Carefully choose which runs to perform Use regression analysis to obtain effect estimates - Statistical Software (Statgraphics,JMP,SAS,) allows to Choose or construct designs Anlayse experimental results Short History of Statistics and Experimentation - 1920’s – introduction of statistical methods in agriculture by Fisher and co-workers - 1950’s – introduction in chemical engineering - 1980’s – introduction in Western industry of Japanese approach ( Taguchi,Robust design) - 1990’s – Combinatorial chemistry , high throughput processing Types of Experimental Designs - Screening Designs