ff0708
)
which has 7 experimental factors each at two levels. The design
consists of 8 runs and is a 1/16 fraction of the full factorial. The
noise design is
ff0304
. The control and noise designs are combined to
form a robust design that has 32 runs.
Engel, J. (1992), Modeling variation in industrial experiments, Applied Statistics 41, 579-593.
Used by permission of Carfax Publishing, Inc.
# This design is already available in S-PLUS under # the name mold.df. The following commands were used # to create the data frame: cont.des <- oa.design(rep(2,7),min.resid.df=0) nois.des<-fac.design(rep(2,3),c('M','N','O'), fraction=1/2) mold.des <- robust.design(cont.des,nois.des) mold.shrink <- c(2.2,0.3,0.5,2.0,3.0,2.1,4.0, 2.0,2.3,0.3,2.8,2.0,3.0,3.1,2.2,1.8,2.3, 2.7,0.4,1.8,3.0,1.0,4.6,1.9,2.1,2.5,3.1, 1.9,3.1,4.2,1.9,1.9) mold.df <- cbind(mold.des,shrink=mold.shrink) # Sample analysis summary(mold.df) plot(mold.df) mold.sn <- robust.sn(mold.df) mold.sn plot(mold.sn,"shrink.mean",data.pts=F) plot(mold.sn,"shrink.target") mold.fac <- fac.aov(mold.sn) mold.fac summary(mold.fac) moldsd.fac <- fac.aov(response=shrink.sd,mold.sn) moldsn.fac <- fac.aov(response=shrink.target,mold.sn) moldsn2.fac <- fac.aov(response=shrink.meanl,mold.sn) pareto(mold.fac) pareto(moldsn.fac) qqnorm(mold.fac,label=3) qqnorm(moldsn.fac) acplot(mold.fac) acplot(moldsn.fac) mold.facs <- fac.aov(shrink.mean~A+C+H,mold.sn) pareto(mold.facs, "mse") summary(mold.facs) newdata <- data.frame(A = seq(-1,1,by=.1), C = rep(-1,21), H = rep(-1,21)) print(predict(mold.facs)) mold.pred <- predict(mold.facs, newdata, numeric.levels = list(A = c(-1, 1), C = c(-1, 1), H = c(-1, 1))) mold.pred