MLSEV. Use Case: The Data-Driven Factory
0 likes186 views
The document discusses the integration of machine learning into industrial processes, highlighting how it transforms operations through data-driven insights. It focuses on the digitalization of factories, predictive maintenance, and enhancing efficiency by improving data collection and analysis. Furthermore, it emphasizes the importance of democratizing access to machine learning across organizations to foster data-driven decisions.
MLSEV. Use Case: The Data-Driven Factory
- 1. 1st edition March 7-8, 2019
- 2. 2BigML, Inc | T2C S.L. The Data-Driven Factory Transforming Industries with Machine Learning César Henández Advanced Analytics & Account Manager, T2C
- 3. 3#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Outline Let’s talk about: Understanding why BIGML is the new paradigm. Who we are The Factory status Digitalizing an industrial process Automation in an industrial environment Applying Machine Learning 1 2 3 4 5
- 4. 4#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. T2C TALENT CONSISTENCY EFFICIENCY INNOVATION Barcelona Málaga • • Introduction
- 5. 5#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. T2 CL IE NT SAP Talent Efficiency Consistency Innovation Business Intelligence IT Systems Development NEW Advanced Analytics Introduction
- 6. 6#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Factories
- 7. 7#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. e Additive Manufacturing Such as 3D printing, to prototype and produce individual components. O Simulation Leverage real-time data and mirror the physical processes . E Integration Horizontal y vertical system integrations. 0Cybersecurity The need to protect critical industrial systems.b Augmented Reality The companies will make much broader use. H Autonomous Robots Laser, Automate, Self- Guided Vehicles (LGV, AGV,…) y Data Lakes The collection and comprehensive evaluation of data m IoT More devices will be enriched with embedded computing. 20% Current Implementation The development of this technologies is growing year to year. The advanced digital transformation is already used in manufacturing. The Industry 4.0 will change processes and productions. Industry 4.0
- 8. 8#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. First objective: Break the wall Current factory status • Papers and printing. • Manual checks. • Low traceability. • Processes based on average measures. • If it ain't broke, don't touch it! One Step Behind
- 9. 9#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Manufacturer’s adoption of ML to improve predictive maintenance will increase 38% in the next five years. PwC March’18 https://www.pwc.de/de/digitale-transformation/digital-factories-2020-shaping-the-future-of-manufacturing.pdf ML in the Industry
- 10. 10 Usually, only a 20% of the data owned by an enterprise is stored in traditional databases Without Data, We are blind Atriles By digitalizing processes we generate new data sources and steed up factory applications. Data Gathering Process
- 11. 11#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Digitalizing Processes FROM: • Checking production plan from paper. • Checking materials by barcode comparison from paper. • Recieving a call to register alternative materials. TO: • Visually checking materials. • Reading codebars with automatic validation. • Automatically recieving alternative materials. • COLLECTING DATA!
- 12. 12#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. 1. Faster material picking 2. Less production errors 3. Traceability on those errors 4. Reduced material waste 5. Data collection Benefits
- 13. 13 Usually, only a 20% of the data owned by an enterprise is stored in traditional databases Without Data, We are blind Atriles By digitalizing processes we generate new data sources and steed up factory applications. Data Gathering GrabIT Once we have the data, we can start creating new added-value tasks. Prove Data Value Process
- 14. 14#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Production plan from SAP Compute needs: MATERIAL QTY TIME Line operator: Request suggested mission Select package: Warehouse optimizatio n rules LGV’s serve materials to line prioritizing needs Automation: Material Delivery
- 15. 15#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. GrabIT Server WAREHOUSE SYSTEM SAP LGV SYSTEM MySql Web Movement service Cache service Movement queue service Calculations Calculations ATRILES GrabIT Architecture PRODUCTION SYSTEM
- 16. 16#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. 1. Efficient material buffering in lines. 2. Reduced delivery errors. 3. Higher efficiency on material management and consumption. 4. Reduced budget in shelf repairs. Benefits
- 17. 17 Usually, only a 20% of the data owned by an enterprise is stored in traditional databases Without Data, We are blind Atriles By digitalizing processes we generate new data sources and steed up factory applications. Data Gathering GrabIT Once we have the data, we can start creating new added-value tasks. Prove Data Value Prediction At the maturity stage, new possibilities arise with BIGML as enabler to unlock for potential on your data. Machine Learning Process
- 18. 18#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Movement Data ‘movements_queue’: • Transactional log of all the missions status. • Main fields: • MissionID • Status • Source • Destination • MaterialCode • PackageID • ProcessedTime • ErrorReason • DueDate • Volume: • Data from October 2018 until February 2019. • 47.147 rows. ‘status_code’: • Describes the different status codes. • Main fields: • Code • Description ‘master_materials_info’: • Contains information about the materials. • Main fields: • MaterialID • Unit • Type • Valid_from • Valid_to • Last_update ‘master_package_info’: • Contains information on warehouse packages. • Main fields: • PackageID • MaterialID • PalletType • Stability • Amount • DueDate • LastUpdateFromMes
- 19. 19#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Predict Missions Outcome Line operator requests material A mission is created Predict result Set priority, origin and destination OK ERROR
- 20. 20#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Data Wrangling RAW DATA FROM DIFFERENT TABLES ML READY DATA
- 21. 21#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Date Feature Engineering AUTOMATIC FEATURE ENGINEERING ON DATE-TIME FIELDS
- 22. 22#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Dataset Features • MissionID • Status • ErrorReason • Source • Destination • MaterialCode • PackageID • PalletType • Stability 5 Months | ~14k rows | 44 Features • Amount • DueDate • LastUpdateFromMes • LastUpdateFromLGV • ProcessedTime • SourceDisabled • DestinationDisabled • 28 Automatic Date-Time Features
- 23. 23#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Data Leakage Wait… error reason? • If we specify there was an error reason we are already giving the result. • At the prediction time we won’t know the error reason so It’s not a valid field.
- 24. 24#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Defining Target • Flatline always comes in handy for data transformations. • We also remove ErrorReason and Status from our dataset to prevent Data Leakage.
- 25. 25#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Split train and test SOURCE RAW_DATASET FLATLINE DATASET TEST TRAINING
- 26. 26#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. • We proceed to remove anomalies with an anomaly score higher than 60%. • 50 anomalies removed. Anomaly Detection Let’s remove anomalies from our training dataset
- 27. 27#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Creating our First Model SOURCE RAW_DATASET FLATLINE DATASET TEST TRAINING ANOMALY MODEL EVALUATION
- 28. 28#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Simple Model Evaluation
- 29. 29#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Improving with OptiML
- 30. 30#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Deepnet Evaluation
- 31. 31#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Improvement
- 32. 32#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Complete ML Workflow SOURCE RAW_DATASET FLATLINE DATASET TEST TRAINING ANOMALY DEEPNET EVALUATION PREDICTION
- 33. 33#MLSEV: THE DATA-DRIVEN FACTORYBigML, Inc | T2C S.L. Production Which missions are likely to fail? PREDICTIONMISION GENERATION OK ERROR EXECUTE MISSION REVIEW MISSION
- 34. 34 Usually, only a 20% of the data owned by an enterprise is stored in traditional databases Without Data, We are blind Atriles By digitalizing processes we generate new data sources and steed up factory applications. Data Gathering GrabIT Once we have the data, we can start creating new added-value tasks. Prove Data Value Prediction At the maturity stage, new possibilities arise with BIGML as enabler to unlock for potential on your data. Machine Learning Prescription ( Process
- 35. BigML, Inc | T2C S.L. 35 Predictive maintenance of industry will generate a 10% reduction in annual maintenance costs, up to a 20% downtime reduction and 25% reduction in inspection costs. McKinsey & Company October’18 Predictive Maintenance
- 36. Today 10% Tomorrow 90% Specific use cases The access to the ML is closed, difficult and specific. Democratize ML Focus on the data-driven decisions, providing access to BIGML through the whole organization. How we see the Machine Learning
- 37. What we can do together INDUSTRIAL Gathering Data, Cleaning, Standardization, Feature Engineering, Machine Learning. SALES Gathering Data, Cleaning, Standardization, Feature Engineering, Machine Learning CUSTOMER SERVICE Gathering Data, Cleaning, Standardization, Feature Engineering, Machine Learning … AND TRANSFORM YOUR BUSINESS!