几何尺寸与公差论坛------致力于产品几何量公差标准GD&T (GDT:ASME)|New GPS(ISO)研究/CAD设计/CAM加工/CMM测量 - 查看单个帖子

multi variate question (doe)!
hi!
i received a large quantity of data from one of my students.
the variables (factors) are the final concentration in a bulk solution.
components a or b are mixed with different ratios of x or y or z...
water is also used 0 or 0,1%
a and b are present at 0,4 or 0,8 % when added into the bulk solution.
x, y or z is present in 0,4 or 0,8 % in the final bulk solution when added to a or b in the lower conc. and 0,8 or 1,6% when added to the high conc. of a or b.
that makes 48 experiments (+4 repetitive experiments for standard deviation).
only one response is measured in these experiments (a quality factor 0-100)
i鈥檓 familiar with 鈥渄esign of experiments鈥?programs but this time the experiments are already carried out and i don鈥檛 know who to get the different components contribution to the results?
could you give me some ideas how to sent up this program or links to similar problems?
maybe i can鈥檛 use multivariate analysis for this problem?
looking forward to your responses
if you look at the theory of how to set the experiments up it is relatively obvious how to analyse it for a given, not necessarily optimal, experimental design.
you may not be able to easily figure out the level of confounding in the higher order interactions, but that is also possible.
the key for analysing them is writing the the experiment out in a rational fashion.
you can use minitab or excel to analyse this.
cheers
greg locock
it seems that you have an experiment with three blocks of four factors at two levels : block 1 with a,b,water and x; block 2 with a,b water and y and block 3 with a,b, water and z, that makes 3x 2^4=48 experiments. for some reasons the experimenter used levels of x,y and z as a function of the levels of a,b and water- probably the response was not significant without it.
m777182
you end up with a confounded result, but, you suppress the independent contributions, so all you are left with is the combinations that were actually tested.
to be honest this stuff is harder to write about in text than to do in a test-matrix.
cheers
greg locock

2009-09-05, 09:00 PM	#1
huangyhg 超级版主注册日期: 04-03 帖子: 18592 精华: 36 现金: 249466 标准币资产: 1080358888 标准币	multi variate question 9doe01 multi variate question (doe)! hi! i received a large quantity of data from one of my students. the variables (factors) are the final concentration in a bulk solution. components a or b are mixed with different ratios of x or y or z... water is also used 0 or 0,1% a and b are present at 0,4 or 0,8 % when added into the bulk solution. x, y or z is present in 0,4 or 0,8 % in the final bulk solution when added to a or b in the lower conc. and 0,8 or 1,6% when added to the high conc. of a or b. that makes 48 experiments (+4 repetitive experiments for standard deviation). only one response is measured in these experiments (a quality factor 0-100) i鈥檓 familiar with 鈥渄esign of experiments鈥?programs but this time the experiments are already carried out and i don鈥檛 know who to get the different components contribution to the results? could you give me some ideas how to sent up this program or links to similar problems? maybe i can鈥檛 use multivariate analysis for this problem? looking forward to your responses if you look at the theory of how to set the experiments up it is relatively obvious how to analyse it for a given, not necessarily optimal, experimental design. you may not be able to easily figure out the level of confounding in the higher order interactions, but that is also possible. the key for analysing them is writing the the experiment out in a rational fashion. you can use minitab or excel to analyse this. cheers greg locock it seems that you have an experiment with three blocks of four factors at two levels : block 1 with a,b,water and x; block 2 with a,b water and y and block 3 with a,b, water and z, that makes 3x 2^4=48 experiments. for some reasons the experimenter used levels of x,y and z as a function of the levels of a,b and water- probably the response was not significant without it. m777182 you end up with a confounded result, but, you suppress the independent contributions, so all you are left with is the combinations that were actually tested. to be honest this stuff is harder to write about in text than to do in a test-matrix. cheers greg locock __________________ 借用达朗贝尔的名言：前进吧，你会得到信心! [url="http://www.dimcax.com"]几何尺寸与公差标准[/url]