7466345961792322124225336833763114726243704562541851513116Sigma綠帶培訓教材5111551635115214164203727446Sigma綠帶培訓教材50975543929634073178496965574778282436263219711353846講師:易洪清2632197113538466056773080710675248587466345961Define79232212422533683376311472Define定義624370456254185151311Measure測量Control控制511155163511521416420372744509755439296340731784969Analysis分析Improve改進65574778282436263219711353846分析改進263219711353846605677308071067524858分析階段Analysis分析(Anli)阶段的目的分析(Analysis)阶段的目的z评价影响输出变量Y的潜在Xs影响度z删除不太重要的多数X找出影响大的“VitalFewXs”UpperSpec0.06z删除不太重要的多数X,找出影响大的VitalFewXs”m0.050.040.03002Impurity潜在X’seasuremTime1Time2Time3P1P2P3P4P5P6P7P8P9LowerSpec210.020.010.00PientTime1Time2Time3PieceVitalfewX’s分析階段Analysis主要內容•中心極限定理•多變量分析及方差組分多變分析及方分•置信區間與假設檢驗簡介•均值檢驗&方差檢驗均值檢驗&方差檢驗•比例檢驗&卡方檢驗相關性分析•相關性分析•一般線性回歸分析中心極限定理CentralLimitTheorem例中心極限定理是統計推斷的基本概念我們可以通過例•中心極限定理是統計推斷的基本概念,我們可以通過該定理用樣本的數據推斷總體的特性•生成100行9列隨機數據(常態)-CalcRandomDataNormal中心極限定理CentralLimitTheorem例例•計算每行的平均值,存在C10列-CalcRowStatistics•將C1-C9進行累疊,存在C11列-DataStackColumns中心極限定理CentralLimitTheorem例例用單值數據繪制圖•用單值數據繪制I圖–StatControlCharts•用均值數據繪制I圖–StatControlCharts15UCL=13.9611IChartofdata98UCL=8.234IChartofMeanndividualValue105_X=5.13IndividualValue765_X=5.134ObtiI8117216315414513612711819110-5LCL=-3.69ObservationI9181716151413121111432LCL=2.035ObservationObservationUCL=13.96UCL=8.23中心極限定理CentralLimitTheorem例例建立單值的直方圖SBiSii•建立單值的直方圖–StatBasicStatistic•建立均值的直方圖–StatBasicStatisticA-Squared0.19P-Value0.899Mean5.1344SD29923Anderson-DarlingNormalityTestSummaryforIndividualA-Squared0.49P-Value0.214Mean5.1344StD10412Anderson-DarlingNormalityTestSummaryforMean12.510.07.55.02.50.0-2.51stQuartile3.1756Median5.09853rdQuartile7.1585StDev2.9923Variance8.9541Skewness-0.0362245Kurtosis-0.0454249N900Minimum-3.6107765431stQuartile4.3061Median4.98493rdQuartile5.8270StDev1.0412Variance1.0840Skewness0.243653Kurtosis-0.239552N100Minimum2.7133MedianMean3rdQuartile7.1585Maximum14.58654.93875.33024.89645.33802.86023.137495%ConfidenceIntervalforMean95%ConfidenceIntervalforMedian95%ConfidenceIntervalforStDev95%ConfidenceIntervalsMedianMeanQMaximum7.69794.92785.34104.77105.46040.91411.209595%ConfidenceIntervalforMean95%ConfidenceIntervalforMedian95%ConfidenceIntervalforStDev95%ConfidenceIntervals5.45.35.25.15.04.95.555.405.255.104.954.80σ=2.99σ=1.04中心極限定理CentralLimitTheorem1.如果容量為n的隨機樣本取自一個均值為u標准差為σ的分布,則樣本的均值將形成一個新的分布,新分布的均分布則樣本的均值將形成個新的分布新分布的均值與原分布相同,但均值標准差將縮小為:nσ•如果σ未知,樣本量大於30,則樣本標准差S可代用至上述公式中,那麼,標准差的估計值為:s≈σˆnx≈σ中心極限定理CentralLimitTheorem例例擲骰子時單個骰子擲起後其1面朝上的概率均為1/但同時擲起個骰子後•擲骰子時,單個骰子擲起後,其1-6面朝上的概率均為1/6,但同時擲起n個骰子後,骰子朝上面的平均值或點數之和近似地服從正態分布。示意圖如下:123456個骰子點數123456單個骰子點數的分布n個骰子點數和的分布平均值的分布n個骰子點數中心極限定理CentralLimitTheorem例例A-Squared0.46P-Value0.254Mean3.5660StDev0.7281Variance0.5301Skewness-0.155371Kurtosis-0.059717N100Anderson-DarlingNormalityTestSummaryformeanA-Squared14.72P-Value0.005Mean3.5660StDev1.7308Variance2.9956Skewness-0.07213Kurtosis-1.28428N500Anderson-DarlingNormalityTestSummaryforIndividual5.254.503.753.002.25Mean1stQuartile3.0500Median3.60003rdQuartile4.0000Maximum5.20003.42153.71053.40003.8000Minimum1.800095%ConfidenceIntervalforMean95%ConfidenceIntervalforMedian95%ConfidenceIntervalforStDev95%ConfidenceIntervals654321Mean1stQuartile2.0000Median4.00003rdQuartile5.0000Maximum6.00003.41393.71813.00004.0000Minimum1.000095%ConfidenceIntervalforMean95%ConfidenceIntervalforMedian95%ConfidenceIntervalforStDev95%ConfidenceIntervalsMedian3.83.73.63.53.40.63930.8458Median4.03.83.63.43.23.01.62981.845399.999Mean3.566StDev1.731ProbabilityPlotofIndividualNormal99.9Mean3.566StDev0.7281ProbabilityPlotofmeanNormalPercent9995908070605040302010N500AD14.720P-Value0.005Percent9995908070605040302010N100AD0.462P-Value0.254Individual10.07.55.02.50.0510.1mean65432110510.1P-Value005P-Value=0254PValue0.05PValue=0.254中心極限定理CentralLimitTheorem2.中心極限定理:無論單個獨立變量服從何種分布,隨著n增加,樣本均值的分布越趨向正態分布。(對任何分布n增加,樣本均值的分布越趨向正態分布。(對任何分布要求n30,樣本均值分布都近似於正態分布)•休哈特博士通過實驗總結:無論總體服從什麼分布,只要樣本量n≧5試驗次數足夠多樣本均值的分布總要樣本量n≧5,試驗次數足夠多,樣本均值的分布總是趨近於正態分布中心極限定理CentralLimitTheorem例例•SamplingDistributionofXforVariousSampleSizes•SamplingDistributionofXforVariousSampleSizes不同樣本容量的X抽樣分布SamplingdistributionSamplingdistribution(n=5)(n=20)Populationdistributionx=150x=150X=150Samplingdistribution(n=30)Samplingdistribution(n=5)Populationdistributionx=150x=150X=150x=150x=15050中心極限定理CentralLimitTheorem例例•SamplingDistributionofXforVariousSampleSizes•SamplingDistributionofXforVariousSampleSizes不同樣本容量的X抽樣分布(a)(b)(c)(d)(a)Normal(b)Uniform(c)Exponential(d)Parabolicn=1ParentPopulationSliDiibiff2n=2SamplingDistributionsofxforn=2SamplingDistributionsofxforn=5n=5SamplingDistributionsofxforn=5n=30SamplingDistributionsofxforn=30多變量分析&方差組分主要內容(Multi-Vari&ComponentofVariation)(MultiVari&ComponentofVariation)•多變量分析的概念•多變量圖•交叉式多變量分析&方差組分•嵌入式多變量分析&方差組分多變量分析&方差組分,例“交叉式”例例:某黑帶希望了解培訓和工作經歷對員工生產效率的影響根據與項目團隊的交流發現員例例:某黑帶希望了解培訓和工作經歷對員工生產效率的影響。根據與項目團隊的交流發現員工在岡時間(1-5年)和培訓項目(有基礎培訓與專家培訓兩種,分別為40和80小時),對工件的作業時間用來衡量生產效率?NoTraininghoursExperienceTimeNoTraininghoursExperienceTimeNoTraininghou