生產的下一步：使用 AI 獲得更好的結果.pdf

《生產的下一步：使用 AI 獲得更好的結果.pdf》由會員分享，可在線閱讀，更多相關《生產的下一步：使用 AI 獲得更好的結果.pdf（26頁珍藏版）》請在三個皮匠報告上搜索。

1、The Next Step in Production:Using AIto Achieve Better ResultsRalph J.Woerheide-Metromation Inc.Coatings Trends&Technology Summit,Lombard 9/6/2024IntroductionSales and Business DevelopmentModular FactoryWhat is Modular Factory(MoFa)?Traditional ManufacturingModular ManufacturingRaw MaterialsPROCESSPr

2、oductRaw MaterialsDispersingGrindingMixingMasterSlurriesBinderCocktailPPPPPPPMoFa as a basis for production data Recipe Modularization Compact Setup Dispersion and Mixing PLC Controlled Sensors and interfaces AI Enabled Benefits in productionefficiency,stability andsustainabilityWhere are we and whe

3、re do we want to goto?Instable processes complex formulationsRaw material variation final productDifficult prediction of product qualityMoFaAI?Where are we and where do we want to goto?Case Study at a Paint FactoryProblem to solve:Heterogeneous recipe structures adapted to the customer Low productio

4、n frequencies Difficult continuous data collection Raw material and recipe information is not sufficient as a basis for a qualityprediction Is an AI implementation possible at all?Phase 1:Status AnalysisInternal Analysis:Screening Data Get to know the data Data collection Data Bundling Creation of a

5、 database structure Based on this,further analysis AI-based data analysis on a defined product group Evaluation of correlations and patterns from the database Mathematical modelling to validate solutions for O.K.and not O.K.productsInternal Analysis:AUC EvaluationIdeal classifier100%TPR and 0%FPRPre

6、diction is consistent with the observedresultsClear distinction between TN and TPRandom Classifier50%TPR and 50%FPRIt is not possible to distinguish between TNand TPPredictive model is inappropriateInsufficient data basis=+=+Area under Curve(AUC)Internal Analysis:Check Models=+=+Area under Curve(AUC

7、)AUC ValueQuality0.9 1.0Excellent0.8 0.9Very Good0.7 0.8Good0.6 0.7Fair0.5 0.6BadIntegration of sensors toincrease the data basisAI possible,data basis O.K.AI impossible,data basis not O.K.Internal Analysis:ResultAs-is analysis successfully completed AUC 70%Problems can be solved by classificationCl

8、ear significance for practical usePotential for data diversity and qualityBasis for the development of AI is in placePhase 2:AI DevelopmentProcess Analysis(Value Chain)Raw MaterialsRecipeMoFaApplicationQCFinal Product(internal)Data chaosNo connected value chainNo prediction of product quality possib

9、leRework,Returns and DisposalData Caption AnalysisSensorsWhat happens at which point?Where are sensors located?What data do the sensors provide?How are the sensors digitally stored and stored in the database?Raw MaterialsRecipeMoFaApplicationQCFinal Product(internal)Connect Data Caption and Process

10、Integration of sensor technology and analytical measurement methods for the collection of rawdata Networking of the entire process chain Building a Centralized Database InfrastructureSensorsRaw MaterialsRecipeMoFaApplicationQCFinal Product(internal)Data Caption SystemRaw Material KPI/VariancesRecipe

11、 KPI/Process StepsProcess DataAppearance,Surface,ColorNew+traditional testmethodsSensorsRaw MaterialsRecipeMoFaApplicationQCFinal Product(internal)Example:Lab Spray ApplicationAir Data Paint BoothTemperature/HumidityTemperature Data Curing OvenControl of Curing ProcessExample:Lab Spray ApplicationLa

12、b Painter:Application SetupAutomated Data Logging and DataExportOrganize Data FlowSensor data from theindividual process steps isautomatically transferredto the cloud database.SensorsRaw MaterialsRecipeMoFaApplicationQCFinal Product(internal)Raw Material KPI/VariancesRecipe KPI/Process StepsProcess

13、DataAppearance,Surface,ColorNew+traditional testmethodsDatabase StructureData of the complete value chain is:Centralized Available Retrievable Basis for development of the Artificial Intelligence ApplicationRoadmap for AI ImplementationPROCESS ANALYSISTAKE IIFIXWEAKNESSESCHOSE MLALGORITHMAI PREDICTI

14、ON SYSTEM-AUC 85%AI OptimizationAutomatic communicationbetween the AI and the data inthe databaseBefore starting a newproduction,the AI qualityprediction system makes astatement about the productqualityAlgorithm/AISensorsRaw MaterialsRecipeMoFaApplicationQCFinal Product(internal)Raw Material KPI/Var

15、iancesRecipe KPI/Process StepsProcess DataAppearance,Surface,ColorNew+traditional testmethodsSoftwareResults Development of an AI qualityprediction and decision support systemsuccessfully completed.Predictability for O.K./not O.K.products is 89%.Raw material quality has a majorinfluence on the quality of the product,with a stable production process(MoFa).Thank You!Visit us at booth#58Ralph J.WoerheideMetromation Inc.215 W Michigan AveYpsilanti,MI 48197 USA+1 248-687-6814www.metromation.co