Minitab basic.ppt

《Minitab basic.ppt》由会员分享，可在线阅读，更多相关《Minitab basic.ppt（98页珍藏版）》请在三一办公上搜索。

1、Minitab-BasicMinitab-基础,Business Excellence卓越运营,Our RequestWhile the trainingis in Progress,we seek your kindassistance tohave all Mobile Phones&PagersSwitched toSilent Mode,欢迎来到,Welcome,Minitab basictraining,Minitab 基础培训,What can Minitab do for us?,Purpose目的,Become experts in the Minitab software.熟

2、悉Minitab软件;2.Apply the tools to make your work more efficiently.应用工具使您的工作更加高效;3.Cascade the use of the tools into your organisation.把工具的使用带回您的部门.,Training agenda培训议程,Table of Contents目录,1.Minitab 视窗介绍 72.Data processing数据处理 103.Hypothesis test假设检验 184.DOE实验设计 355.Regression analysis回归分析 536.MSA测量系统分

3、析 637.SPC 管制图 728.Graph图表 89,ContentsSlide(s),Minitab Windows 视窗介绍,Minitab 15 shortcut,1.Shortcut and License 快捷方式与许可代码,Session Window:Analysis output,2.Session and Data window 报告和数据窗口,Data Window:A Worksheet,not a SpreadsheetColumn names are above first rowEverything in a column is considered to be

4、 the same variable,Minitab Windows 视窗介绍,报告生成窗口:分析输出.,数据窗口:工作表,而非简单的数据表工作表上第一行即为每列数据的名称同一行的数据被认为是同一个变量.,3.Common Toolbar introduction 常用工具栏介绍,Minitab Windows 视窗介绍,Data processing数据处理,1.Generate random data 生成随机数 Path:Calc/random data/normal,Generate 50 random data 生成50个随机数;Mean:100;均值:100;Standard de

5、viation:10;标准方差:10,2.Random sample 随机取样 path:Calc/random data/sample from columns,Generate 7 random sample;产生7个随机数据From:C1 来自:C1Store:C2 储存:C2,3.Calculator 计算器 path:Calc/Calculator,Store result in C3;结果存放在C3;Expression:c1*2+8 关系式:c1*2+8,4.Column statistic 列统计 Path:Calc/Column statistic,Calculate the

6、 standard deviation of C3计算C3列数据的标准方差.,We can input expression in the window,also we canSelect functions in the window.我们可以在此窗口中输入关系式也可以选用函数,Raw statistic is the same as column statistic.行统计的使用方法与列统计一样.,Data processing数据处理,5.Standardize标准化 path:Calc/Standardize,Input C1;Store result in C4Standardize

7、 data from-1 to+1,Of course,other 4 method may also fit some of you.当然了,另外4种方法部分同事也可能使用到.,6.Simple set of numbers 简单数据设置 path:Calc/Make patterned data/Simple set of numbers,First value:1;Last vaule:9Step:2,You can change“Number of times to list each value”and“Number of times list the sequence”to fin

8、d the difference.大家可以改变”每个值出现的次数”和”每个序列出现的次数”来发现二者有什么样不同.,Data processing数据处理,7.Probability distribution 概率分布 path:Calc/Probability distribution/normal,Calculate the probability from sigma level 通过西格玛水平计算概率;Calculate the sigma level from probability.通过概率计算西格玛水平.,Exercise练习:1.If a product have 3 sigm

9、a process capability,what is the defect rate if the data is normally distribution?如果一个产品具有3个西格玛的制程能力,假定数据正态分布,请问产品的不良率是多少?2.If a process have 99.5%yield rate,whats the sigma level can we say?如果制程良率99.5%,我们可以说西格玛水平等于多少?,Data processing数据处理,8.Stack data 堆叠数据 Path:Data/Stack/Columns,Generate 3 column d

10、ata firstly;首先生成3列数据;Stack the 3 columns to one column 然后把3列数堆叠成1列.,You can input column in the optional item or not,then find the difference in the data window.大家可以在可选项中输入列号或不输入,然后观察数据窗中的差异.,Data processing数据处理,9.Unstack data数据反堆叠 Path:Data/Unstack Columns,Unstack the c4 columns to three columns.大家

11、练习把C4列中的数据分成三列.,Data processing数据处理,CodeText1Monday2Tuesday3Wednesday1Monday2Tuesday3Wednesday2Tuesday2Tuesday3Wednesday3Wednesday,10.Code 编码 Path:Data/code/Number to text,How to如何做到:Change the code to text?代码转换为文本?Change the test to code?文本转换为代码?Change code to code?代码转换?Change text to text?文本转换?,Da

12、ta processing数据处理,11.Change data type 转换数据类型 Path:Data/Change data type/Test to number,In our actual analysis,we may often encounter data type transfer problem.在实际分析中,我们会经常遇到数据类型 转换的问题.,How to 如何:Change number to text?数据格式转换文本格式?Change text to number?文本格式转换数据格式?and so on?等等.,Data processing数据处理,Hypo

13、thesis test假设检验,1.Graphical summary 概要图 Path:Stat/Basic statistic/Graphical summary,C118.807522.735213.443328.819611.539437.759531.975334.634351.391325.374623.888931.853617.827139.608938.0237,What can we conclude from the Data of C1 column?我们可以从右边C1列数据得出什么结论?,Can you guess what will happen if we Cha

14、nge the confidence level from 95 to 99?大家猜测一下,这里我们把95改成99会发生什么情况?,Normally test:P-value0.05|Skewness|1|Kurtosis|1,Display data显示数据:Mean平均值;StDev标准方差;Min最小值;Max最大值;Median中位数;1st Quartile 1/4分位数3rd Quartile 分位数Confidence interval置信区间,Hypothesis test假设检验,2 1-Sample Z test Path:Stat/Basic statistic/1-Sa

15、mple Z,Test Condition 测试条件:Generate 30 random data firstly,mean:10,stdev:1 先生成30个随机数,均值为10,标准方差为1;Ho:Mean is not bigger than 8;Ha:Sample mean is bigger than 8;原假设:样本均值小于等于8,对立假设:样本均值大于8,Hypothesis test假设检验,One-Sample Z:C1 Test of mu=8 vs 8The assumed standard deviation=1 95%LowerVariable N Mean StDe

16、v SE Mean Bound Z PC1 30 9.861 1.050 0.183 9.560 10.19 0.000,session window,P0.05,接收 Ha.,Ho is out of interval,Hypothesis test假设检验,3 2-Sample t test Path:Stat/Basic statistic/2-Sample t,Supplier ASupplier B8.491915.651710.392414.62989.868611.702511.395512.79619.201613.979110.686615.952611.043220.006

17、28.928713.81269.344114.99888.969014.71979.057115.028410.690712.16508.565615.04828.223616.16039.910917.8253,Assumed Condition假定条件:Supplier A:mean:10,stdev:1;Supplier B:Mean:15,sedev:2 Ho:AB,Hypothesis test假设检验,Two-sample T for Supplier A vs Supplier B N Mean StDev SE MeanSupplier A 15 9.65 1.00 0.26S

18、upplier B 15 14.97 2.11 0.54Difference=mu(Supplier A)-mu(Supplier B)Estimate for difference:-5.31495%lower bound for difference:-6.352T-Test of difference=0(vs):T-Value=-8.82 P-Value=1.000,session window,P0.05,accept Ho.,Hypothesis test假设检验,4 2-Proportion test Path:Stat/Basic statistic/2-Proportion,

19、Assumed Condition假设条件:Line A and Line B are normal production line.产线A,B为正常生产线;Ho:A=B,Ha:A=/B 原假设:A=B,对立假设:A=/B,Hypothesis test假设检验,Test and CI for Two Proportions Sample X N Sample p1 5 100 0.0500002 7 174 0.040230Difference=p(1)-p(2)Estimate for difference:0.0097701195%CI for difference:(-0.041970

20、9,0.0615111)Test for difference=0(vs not=0):Z=0.37 P-Value=0.711,P0.05,accept Ho.,session window,The same with 1-Proportion test,Path:Stat/Basic statistic/1-proportion test,You can exercise this after class.,Hypothesis test假设检验,5.2variance Test(F test)Path:Stat/Basic statistic/2-vairiance,Supplier A

21、Supplier B8.491915.651710.392414.62989.868611.702511.395512.79619.201613.979110.686615.952611.043220.00628.928713.81269.344114.99888.969014.71979.057115.028410.690712.16508.565615.04828.223616.16039.910917.8253,Assumed Condition假设条件:The data is normally distributed.数据正态分布;Ho:A=B,Ha:A=/B;原假设:A=B,对立假设

22、:A=/B;,Hypothesis test假设检验,Test for Equal Variances:Supplier A,Supplier B 95%Bonferroni confidence intervals for standard deviations N Lower StDev UpperSupplier A 15 0.92205 1.31376 2.22675Supplier B 15 1.64886 2.34934 3.98201F-Test(Normal Distribution)Test statistic=0.31,p-value=0.037Levenes Test(A

23、ny Continuous Distribution)Test statistic=5.16,p-value=0.031,P0.05,accept Ha.,Hypothesis test假设检验,6.One way ANOVA(Unstack)Path:Stat/ANOVA/One way ANOVA,ABC13.099312.686213.051512.370912.44068.781911.795312.07683.538913.983713.65096.300013.319815.477012.171113.846916.58089.687412.593213.30255.603813.

24、298614.18008.689411.742216.00667.996813.227218.557813.983912.620415.414615.396512.463014.76969.763613.314113.35035.304913.317716.951210.623612.793414.062310.5685,假设条件:Ho:No obvious difference with the three factors.各因子之间没有明显差异;Ha:There is a factor at least different with others 至少有一个因子与其它因子存在明显差异.,H

25、ypothesis test假设检验,One-way ANOVA:A,B,C Source DF SS MS F PFactor 2 210.90 105.45 20.79 0.000Error 42 213.00 5.07Total 44 423.90S=2.252 R-Sq=49.75%R-Sq(adj)=47.36%Individual 95%CIs For Mean Based on Pooled StDevLevel N Mean StDev-+-+-+-+A 15 12.919 0.661(-*-)B 15 14.634 1.849(-*-)C 15 9.431 3.371(-*-

26、)-+-+-+-+10.0 12.0 14.0 16.0,Conclusion结论:1.P value=00.05,indicate at least have one factor is different with others;P value=00.05,说明至少有一个因子与其它因子有明显差异;2.From the box plot and confidence interval,we can see there is no obvious difference between A and B,but C have obvious difference with A and B;从箱形图

27、或置信区间可以看出,A,B之间没有明显差异,C与A,B有明显差异;,Extend 拓展:One Way ANOVA(数据堆叠),如何操作?,Hypothesis test假设检验,7.Two way ANOVA Path:Stat/ANOVA/Two way ANOVA,Assume condition 假设条件:Ho:No obvious difference with the three factors.各因子之间没有明显差异;Ha:There is a factor at least different with others 至少有一个因子与其它因子存在明显差异.,Hypothesis

28、 test假设检验,Two-way ANOVA:productivity versus Shift,line Source DF SS MS F PShift 1 5625 5625.0 31.45 0.001line 3 98461 32820.3 183.48 0.000Interaction 3 475 158.3 0.89 0.489Error 8 1431 178.9Total 15 105992S=13.37 R-Sq=98.65%R-Sq(adj)=97.47%,Conclusion结论:1.Shift P value=0.0010.05,Indicate have obviou

29、s variance between shifts,and from the plot,we can see D shifts productivity is higher than N.Shift P value=0.0010.05,说明班次之间有明显差异,从图中可以看出,D班比N班产能要高.2.Line P value=00.05,indicate have obvious variance between lines,from the plot,we see line1 is highest,line 3 is lowest.Line P value=00.05,说明产线之间有明显差世,

30、从图中可以看出,1线产能最高,3线最低.3.There is no obvious interaction between Shift and line.Shift 与line 之间没有明显交互作用.,Hypothesis test假设检验,8.Main effect plot 主效果图 Path:Stat/ANOVA/Main effect plot,Conclusion结论:1.From the shift main effect,we can see D shifts productivity is higher than N shift.从Shift 的主效果图可以看出,D班次比N班次

31、产能高;2.From the line main effect,we can see the sequence of productivity is:1,4,2,3;从Line 的主效果图可以看出,产能按从大到小的顺序为:1,4,2,3,Hypothesis test假设检验,9.interaction plot 交互作用图 Path:Stat/ANOVA/Interaction plot,Conclusion结论:From the interaction plot,we can see there are obvious interaction within A,B,C.从交互作用图可能看出

32、,A,B,C之间都存在明显的交互作用.,Hypothesis test假设检验,10 Common hypothesis test table常用假设检验对照表,Hypothesis test假设检验,1.Full Factorial DOE 全因子实验设计 优点:实验简单,所有因子之间交互作用都会考虑到;缺点:实验次数较多,成本高;适用:K5;,DOE 实验设计,1.1 实验次数设计 假设:3因子2水平,均值差异为5,Stdev为1;路径:Stat/Power and sample size/2level factorial design,Power and Sample Size 2-Le

33、vel Factorial DesignAlpha=0.05 Assumed standard deviation=1Factors:3 Base Design:3,8Blocks:noneCenter Total Target ActualPoints Effect Reps Runs Power Power 0 5 2 16 0.9 1.00000,因子个数,除去中心点的实验次数,重复次数,现在与目标的差值,统计能力,目前标准偏差,报告显示,要达到90%以上的统计能力,需要重复一次实验,运行16次实验.,1.2 实验参数表设计 路径:Stat/DOE/Factorial/Create fa

34、ctorial design,All DOE must ensure the random of experiment sequence.Because this can avoid noise.所有DOE必须确保实验顺序的随机性;因为保持实验顺序的 随机性,可以有效避免噪音因子的干扰.,DOE 实验设计,1.3 实验及分析 路径:Stat/DOE/Factorial/Analyze factorial design,Run experiment base on the run order;按Run order顺序进行实验;,DOE 实验设计,1.3 实验及分析 路径:Stat/DOE/Fac

35、torial/Analyze factorial design,DOE 实验设计,1.4 实验结果分析 路径:Stat/DOE/Factorial/Analyze factorial design,Estimated Effects and Coefficients for y(coded units)Term Effect Coef SE Coef T PConstant 67.63 2.132 31.71 0.000A-27.25-13.63 2.132-6.39 0.000B-31.50-15.75 2.132-7.39 0.000C-39.25-19.63 2.132-9.20 0.0

36、00A*B 6.00 3.00 2.132 1.41 0.197A*C-0.25-0.12 2.132-0.06 0.955B*C 16.00 8.00 2.132 3.75 0.006A*B*C 5.00 2.50 2.132 1.17 0.275S=8.52936 PRESS=2328R-Sq=96.11%R-Sq(pred)=84.43%R-Sq(adj)=92.70%,Pareto 柏拉图,Four in one(四图合一),结论:1.From the pareto and result,we can see C,B,A,B*C are significant.从柏拉图和分析结果可以看

37、出,C,B,A,B*C对输出影响显著;2.There is no abnormity in the four in one chart,so regression is available.四合一图中,没有异常,回归方程可用.,DOE 实验设计,1.5 减少模型 路径:Stat/DOE/Factorial/Create factorial design,In term option,we only select significant factors,then redo the analysis.在Term中只选择显著因子,然后重新进行分析.,DOE 实验设计,Factorial Fit:y

38、versus A,B,C Estimated Effects and Coefficients for y(coded units)Term Effect Coef SE Coef T PConstant 67.63 2.167 31.21 0.000A-27.25-13.63 2.167-6.29 0.000B-31.50-15.75 2.167-7.27 0.000C-39.25-19.63 2.167-9.06 0.000B*C 16.00 8.00 2.167 3.69 0.004S=8.66681 PRESS=1748.10R-Sq=94.47%R-Sq(pred)=88.31%R-

39、Sq(adj)=92.46%,Conclusion结论:1.From the pareto and result,we can see all factors are significant.从柏拉图和分析结果可以看出,现在所有因子都显著;2.In the four in one,find no abnormity,so regression available.四合一图中,没有异常,回归方程可用.,1.5 减少模型 路径:Stat/DOE/Factorial/Create factorial design,DOE 实验设计,1.6 主效果和交互作用分析 路径:Stat/DOE/Factori

40、al/factorial plots,From main effect plot,we can see A,B,C are significant.从主效果图上可以看出,A,B,C主效应显著;From the interaction plot,we can see there is some interaction between B and C;从交互作用图上可以看出,B,C 之间有一定交互作用.So for most optimized parameter,we will use optimizer to do further optimization.因此为了找出最优参数组合,下一步用优

41、化器进行统一优化.,DOE 实验设计,1.7 优化器 路径:Stat/DOE/Factorial/Response Optimizer,There are Maximum,Minimum,Target three options in the Goals menu,here we set Y as a target model,target:100,LSL:80,USL:120在Goal下拉菜单中有望大,望小,望目标三个选项,这里我们设置Y为望目标类型,目标值100,下限80,上限120.,DOE 实验设计,1.7 优化器 路径:Stat/DOE/Factorial/Response Opti

42、mizer,The red color remark parts are the most optimized parameter,base on the experiments have been done,we can obtain 100.24;红色圈注部分,为最优参数设置,基于之前的实验,理论上可把Y优化到100.04.,DOE 实验设计,2.Fractional Factorial DOE 部分因子实验设计 优点:实验次数相对较少,成本低;缺点:因子之间的交互作用,以及高阶复合因子的交互作用可能被忽略.适用:K5;,操作步骤及方法与全因子DOE是一样的,不同之处在于引入了混淆度的概念

43、:,举例来说:如果混淆度设为3,则意味着:A+BCB+ACC+ABI+ABC,DOE 实验设计,3.Response surface DOE 部分因子实验设计 优点:可以精确模拟出响应曲线,从而找出最优点 缺点:实验次数较多,费用较高 适用:K5;,3.1 设计实验表 路径:Stat/DOE/Response surface/Create response surface,In the options window,we select randomize run,others use default setup;选项中,选中随机运行,其它使用默认设置.,DOE 实验设计,3.2 实验及分析 路

44、径:Stat/DOE/Response surface/Analyze response surface design,按Run order顺序进行实验;进行Terms和Graph设置,完成后进行分析.,DOE 实验设计,Term Coef SE Coef T PConstant 56.568 2.867 19.729 0.000A 15.383 1.902 8.087 0.000B 18.589 1.902 9.771 0.000C 13.799 1.902 7.254 0.000A*A-1.006 1.852-0.543 0.599B*B 1.292 1.852 0.698 0.501C*

45、C 2.176 1.852 1.175 0.267A*B 2.250 2.486 0.905 0.387A*C 4.750 2.486 1.911 0.085B*C 1.750 2.486 0.704 0.497S=7.03017 PRESS=3355.34R-Sq=95.67%R-Sq(pred)=70.57%R-Sq(adj)=91.76%,3.2 实验及分析 路径:Stat/DOE/Response surface/Analyze response surface design,4合一图没有发现异常,回归方程可以信任;观察P值,发现A,B,C,A*C显著;R-sq(adj)=91.76%

46、,DOE 实验设计,3.3 减小模型 重复上步操作,在Terms中去掉不显著因子,操作如下:,DOE 实验设计,Term Coef SE Coef T PConstant 58.250 1.494 38.981 0.000A 15.383 1.808 8.507 0.000B 18.589 1.808 10.279 0.000C 13.799 1.808 7.631 0.000A*C 4.750 2.363 2.010 0.063S=6.68272 PRESS=1427.89R-Sq=94.12%R-Sq(pred)=87.48%R-Sq(adj)=92.56%,4合一图没有发现异常,回归方程

47、可以信任;观察P值,发现A,B,C,A*C均显著;R-sq(adj)=92.56%,比之前模型回归效果更好.回归方程:Y=58.25+15.383*A+18.589*B+13.799*C+4.74*A*C,DOE 实验设计,3.4 Contour/Surface plot 等高/曲面图 路径:Stat/DOE/response surface/Contour surface plots,通过等高线,可以快速给我们指明方向,我们可以沿着与等高线垂直方向上升或下降,从而快速达到目标;,通过曲面图,我们可以看出两两因子之间的交互作用.,DOE 实验设计,3.5 优化参数 路径:Stat/DOE/Re

48、sponse surface/Response optimizer,这里设置Y为望目型,目标值50,下限40,上限60通过优化器模拟,最终可以达成目标,最优参数设置如图中红色圈注部分.,DOE 实验设计,Regression analysis 回归分析,1.多元回归 路径:Stat/Regression/,The regression equation isY=62.4+1.55 X1+0.510 X2+0.102 X3-0.144 X4Predictor Coef SE Coef T PConstant 62.41 70.07 0.89 0.399X1 1.5511 0.7448 2.08

49、0.071X2 0.5102 0.7238 0.70 0.501X3 0.1019 0.7547 0.14 0.896X4-0.1441 0.7091-0.20 0.844S=2.44601 R-Sq=98.2%R-Sq(adj)=97.4%,报告中,R-Sq(adj)=97.4%,四合一图没有发同异常,但是否是最好的回归模型呢?Minitab可以帮我们来确认.在此之前,我们先学习几个判定原则:R-Sq(adj)越大越好;Cp越小越好,CpK+1S越小越好.,Regression analysis 回归分析,Best Subsets Regression:Y versus X1,X2,X3,X

50、4 Response is Y Mallows X X X XVars R-Sq R-Sq(adj)C-p S 1 2 3 4 1 67.5 64.5 138.7 8.9639 X 1 66.6 63.6 142.5 9.0771 X 2 97.9 97.4 2.7 2.4063 X X 2 97.2 96.7 5.5 2.7343 X X 3 98.2 97.6 3.0 2.3087 X X X 3 98.2 97.6 3.0 2.3121 X X X 4 98.2 97.4 5.0 2.4460 X X X X,根据刚刚学到的三个判定原则,大家判断如何得到最好的回归方程?,去掉x3,重新回