染整生产过程质量控制关键技术研究与应用

华侨大学硕士学位论文染整生产过程质量控制关键技术研究与应用姓名：顾宇峰申请学位级别：硕士专业：模式识别与智能系统指导教师：金福江20100601I1.2.PCASPE,3.SVR4.II2TSPEPCA5.GA_SVRAHPSVRabstractIIIAbstractInordertosolvethetechnicalproblemaboutthedyeingandfinishingproductsforqualitycontrol,tofurtherimprovetheefficiencyofdyeingandfinishingproductionandmanagement,thisthesiswouldapplythequalityofclosed-loopcontroltothequalitycontrolingoftheproductionprocess,designandimprovepredictionalgorithmandtheproductqualityanalysisalgorithmwhichcanbefittedfortheprintingbusinessfortheactualdyeingprocess,andachievetheearlystageofdyeingandfinishingproductqualitypredictionandQualityAnalysisfinally.Themainresearchissummarizedasfollows:1.Baseonthefeaturesofqualitymanagementduringproductionprocess,thethesisproposedamethodforclose-loopqualitycontrolofdyeingandfinishing,i.e.applyingtheideaofclosed-loopcontroltothequalitycontrolofthedyeingandfinishingproductionprocess.2.Analyzedthemathematicalmodeloftherelationbetweenqualityindexduringdyeingandfinishingandprocessparameters.Analyzedthestatisticalcharacteristicsofvariablesduringdyeing,filteredvariablesinaccordancewithmultivariatestatisticalanalysis,inordertoensuretheaccuracyofthePCAmodel.ProposedbasedonaimprovedPCAofprocessparameterscontroltheanalysis,weutilizeSPEstatisticdescribeproductionsamplesandacceptablecolorrangeofthedeviationfromthestatisticalmodel,quantitativeanalyseofprocessparametersonthecolourdifferenceeffect,ultimatelyobtainedtheprimaryfactorsaffectcolourdifferenceandtheprocessparameterstowhatextentarecolourdifferenceacceptablerange.3.Aqualitypredictingmethodbasedonimprovedsupportvectorregressionisproposedandisappliedtodyeingqualitycontrol.Throughdataminingforhistoryofthedyeingprocess,wecreatedthepredictmodelofthedyeingprocesswiththeSVR.ToaimattheproblemthatSVRdoesn’thavegoodpredictionaccuracy,weusedgeneticalgorithmtochoosetheparametersofabstractIthemodelandimprovedtheaccuracyofthequalitypredictionmodel.4.Analgorithmbasedonimprovedprincipalcomponentanalysisisproposedandisappliedtoqualitycausecontrol.Basedonprincipalcomponentanalysis,usingtwokindsofstatistics2TandSPE,tofindexceptionalvariablesthatcausedqualitytofluctuate.UsingkernelprincipalcomponentanalysistoimprovePCAandoptimizetheprincipalcomponentmodelfurther,tofindoutthefactorswhichcausedtheprintinganddyeingproductsunqualified.5.Designanddevelopmentofthefinishingproductqualitytestingandanalysissystem.WeapplytheGA_SVRalgorithm,improvedPCAalgorithmandmulti-criteriaevaluationmethodAHP,developeddyeingandfinishingproductsqualityinspectionandanalysissystem,whichalsoappliedintheactualdyeingprocesscontrolofproductquality.KeyWords:Qualityprocesscontrol;Dyeingprocess;Colourdifference;Principalcomponentanalysis;SVR.1“”2000ISO9000[1],1.11000[2],,“”2nnn()212nCnn=−102000ISO900019941994:1.2000[1]2.19942020ISO9000ISO9000:2000“”[3][4][5]1.,,32.3.,,4.1.21.2.11.42.[6]40[7][8]1.2.21.[9]5[10]pH“”(RFT)2.[6][11]3.[12],[6]6[13]4.[14]AHP,AHP,,,(C/R),[0,100],1000,1.31.3.11-171-11.3.2QQ−,,ε-SVRε-SVR82TSPE2TSPEPCAGA_SVRAHP1.3.31.[15]2.9[16],,,PrincipalcomponentanalysisPCAPCA,,2.12.2.1102-12.2.22-22-2()092-12-11142%2-1096509049825363201402412139392696265822872.2.3PH6-8,sclavosX82-312X81.20152.1/3201023.1/220204.0.6/min60155.50%201050%20602-3X8X890g/l2g/l3.8g/l()2-3,16020’151/320’101/320’1/320’1010’1/220’20401/21/100.6/min1550%50%109/1020’601312:2334156273Zn12,,,nxxx1y1.12,,,nzzzmmZiz,1,,iiiazbim≤≤=(2-1)iziiaiibi2.12,,,nxxx∧∧∧()12,,,nyfxxx=(2-2)()f∗y3.12,,,nxxx∧∧∧12,,,,nXgyxxx∆=(2-3)14y∆[]12,,,nXxxx=ixixy∆,(2-2)[17]2.2[17]m15……12,,kpppmkkkPCA[18]nmXR×∈nmxX[]1,,mxxx=12xxDσµ−=2-4[]12,,,mµµµµ=()22212,,,mDdiagσσσσ=2iσiΣ,()1TXXnΣ=−2-5()12,,,mDdiagλλλλ=iλi[]12,,,mmmPppp×=ipiλPCAaPa[]12,,,maaaPpppR×=∈Sma−[]12,,,aaamPppp++=Sa[19]1iimiicontλλ==∑(2-616i1aiicont=∑aaa85%axˆxxexCxC=+=+(2-7ˆxS∈eS∈xSSTaaCPP=TTaaaaCPPIPP==−maSma−S2.32.3.1:µDσΣDλPa2.3.2Hotelling2T;SPE(Q)17()1mxxR×∈2-4xtxe()aTTaaaTaatxPxtPxPPexxxIPP====−=−(2-8Hotelling2TSPE(1)Hotelling2T[20]211TTTaaaaTxPDPxtDt−−==(2-9()1,,aaDdiagλλ=a2Tnna−F(na)2Txα2T()()()221,1anTFanannαα−=−−(2-10()α1-αα=0.05α=0.012T(2)SPE[20]Q()()()TTTTTTTaaaamamamamaQeexIPPIPPxxPPxxPxP−−−−==−−==(2-11αSPE()0122020021211211hChhhQααθθθθθ−=++(2-1218()11,2,3miijjkiθλ=+==∑13202213hθθθ=−Cαα2.41.QQ−[21],()()()'12,,,1,2,,pXXXXnααααα==pX()01:,,:pHXNHXµΣ∼(),pNµΣ1[21]:()()'1,,ppXXXNµ=Σ∼()()'1XXηµµ−=−Σ−()2pηχ∼0HXµ()2,DXµ2D1()()()'212DXXpµµχ−=−Σ−∼2D()22Dpχ∼()Xα()21,,Dnαα=2Dα()()()22212nDDD≤≤≤2D()()()()220.5nttttFDpHDpn−==≈()()2tHDp()2pχ()2tD2χtp()()21nttDFp−=()()nHxpFx≈()22ttDχ≈华侨大学硕士研究生学位论文第二章染色过程质量指标与工艺参数的模型研究1988.28.48.68.899.29.49.69.8100.010.020.050.100.250.500.750.900.950.980.99浴比ProbabilityNormalProbabilityPlot10002000300040005000600070000.010.020.050.100.250.500.750.900.950.980.99用水量ProbabilityNormalProbabilityPlots绘制()()22,ttDχ的散列图，当X服从正态分布时，这些点应该散列在一条直线上，这种方法就是QQ−图检验法。接下来使用QQ−图检验法对上节总结出来的二十二个过程变量的分布特性进行分析，来考察他们是否服从正态分布。见图2-4，蓝色加号标志是变量的散列点，查看散列点是否在一条直线上分布。由图2-4左图可知，浴比的散列点仅仅在8、9、10这三处分布，明显不在一条直线上，散点分布已经严重偏离直线，不服从正态分布。所以需要转化，我们知道浴比是水与布重之比，做一个简单的线性转化，将浴比乘以布重就是用水量。显然，由图2-4右图可知，用水量的散点分布非常接近直线，因此