RA TechED 2019 - IN10 - What Machine Learning can do for you using FactoryTalk InnovationSuite
3 likes592 views
This document discusses how machine learning can be used with FactoryTalk InnovationSuite. It outlines the machine learning adoption strategy and data science process. The data science process includes defining the business problem, mapping it to machine learning problems, data analysis, model building, evaluation, and operationalization. It discusses using FactoryTalk products like DataFlowML, DataView, and Edge at different steps. Critical success points are defining the machine learning problem, preparing training data, and creating an accurate model. The document also covers operationalizing predictions with pipelines for prediction, visualization, writing back to controllers, saving predictions, and automatic retraining.
RA TechED 2019 - IN10 - What Machine Learning can do for you using FactoryTalk InnovationSuite
- 1. What Machine Learning can do for you using FactoryTalk® InnovationSuite
- 3. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 3 Agenda Introduction Adoption Strategy Data Science Process ML on FactoryTalk® InnovationSuite ML Reference Architectures Data Pipelines / Big Data Pipelines Data Scientist / Citizen Data Scientist Experiences Citizen Data Scientist Demo 1 2 3 4 8765
- 4. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 4 ML on the Introduction
- 5. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 5 ML on the
- 6. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 7 Manufacturing is going digital Current L3 Production Execution (MES / MOM) L2 Process Monitoring (HMI-SCADA) L1 Process Sensing, Manipulating (PLC) L4/ L5 Business Systems (ERP, SCM, PLM) Governance & planning ISA-95 Break- Thru Innova- tion PLANT & CORP. MANAGEMENT MAINTENANCE QUALITY OPERATORS • Connected • Real-time / Streaming Data • Role-based • Predictive / Prescriptive • Mobile & augmented • Wrap and extend Innovation Industrial Innovation Platform Innovation Platform Source Contextualize Synthesize Orchestrate Engage SUPPLIERS ENVIRONMENTALLOGISTICS IOT GATEWAY %
- 7. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 8 FTIS IOT AR Fit for Purpose Data Analytics Edge Compute ML and AI Market-leading industrial innovation platform to drive digital transformation for increased operational performance and agility across all factories Industry-Leading augmented reality development tools to improve workforce efficiency and training CLOUD| HYBRID| ON-PREM Data Visualization / Data Discovery tool. Self-service Analytics with Insightful Storyboards that may be saved, shared and viewed on any form factor or device Streaming / Machine Data from your PLCs / Controllers, low latency, with the capability to store and forward data. Enables closed loop feedback applications and provides advanced analytics in the “hardware stack”. Operationalizes ML and AI at the Edge. Solve Engineering Challenges through Machine Learning and AI. Scalable from on-prem to the cloud – Connects to Data Lakes / Big Data, Structured / Unstructured Data, and Streaming / Machine Data. A set of tools built on top of our FactoryTalk® ProductionCentre® MES Platform that target specific needs of Manufacturing Operations: Performance, Quality, Production & Warehouse
- 8. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 9 ML on the Adoption Strategy
- 9. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 10 ML Adoption Strategy Product Path ML on the Available Roadmap * Edge can also assist with Data Pre-processing @ the Edge *
- 10. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 11 ML Adoption Strategy Steps ML on the Available Roadmap Digitization • FactoryTalk® Historian SE or equivalent Historian • FactoryTalk® Production Centre or equivalent MES Visualization and Streaming • DataView • Edge • ThingWorx {ThingWorx Mashups, ThingModel} ML – Part A • DataFlowML • Edge (if not already Purchased) • ThingWorx Analytics / Interactive ML Studio ML – Part B • Define Business Problem • Map to ML Problem • … ML Starts with Data! Understand your Data! Get ML-Ready! Dive In! Predict and Prescribe! •Model… • … • Operationalize… … More on ML – Part B Next…
- 11. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 12 ML on the Data Science Process
- 12. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 13 Overview Data Science Process Define Business Problem Map to Machine Learning (ML) Problem(s) Exploratory Data Analysis ML Model Building (Train / Test) ML Model Evaluation Operationalize Data Preparation / Data Wrangling 80% of work 20% of work 100% of value
- 13. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 14 Detail Data Science Process Define Business Problem Map to Machine Learning (ML) Problem(s) Exploratory Data Analysis ML Model Building (Train / Test) ML Model Evaluation • Clearly defined Business Problem • Set Success Criteria • Define clear Data Science Objectives • Determine ROI • Determine Budget • Determine Executive Sponsorship • Understand data points and constraints • Formulate Data Science Strategy • Perform required Data Transformations / Data Wrangling / Data Preprocessing • Break Business Problems into Data Science (DS) Problems • Identify Machine Learning (ML) Problems • Perform Statistical Analysis / Data Science Analysis • Discover and handle Outliers/Errors • Shortlist Machine Learning (ML) Modeling Techniques Operationalize Data Preparation / Data Wrangling 80% of work 20% of work 100% of value • Operationalize your ML Model against Streaming Data from the PLC / Controller • Data Transformations / Data Wrangling / Data Preprocessing need to be applied • Maximum value realized Choose the Optimal Model • Experiment with multiple ML Models • Tune Hyperparameters
- 14. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 15 Breakdown Data Science Process Define Business Problem Map to Machine Learning (ML) Problem(s) Exploratory Data Analysis ML Model Building (Train / Test) ML Model Evaluation Operationalize Data Preparation / Data Wrangling 80% of work 20% of work 100% of value Pre-Sales ML on FactoryTalk® InnovationSuite
- 15. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 16 Options Data Science Process Define Business Problem Map to Machine Learning (ML) Problem(s) Exploratory Data Analysis ML Model Building (Train / Test) ML Model Evaluation Operationalize Data Preparation / Data Wrangling 80% of work 20% of work 100% of value Pre-Sales ML on Option 1 – Use FactoryTalk® FactoryTalk® InnovationSuite for ALL your ML Needs Option 2 – Use “BYODST” for Some of the Steps Prior to Operationalize Then Use to Operationalize
- 16. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 18 Products & People Data Science Process Define Business Problem Map to Machine Learning (ML) Problem(s) Exploratory Data Analysis ML Model Building (Train / Test) ML Model Evaluation Operationalize Data Preparation / Data Wrangling 40% of work 10% of work 100% of value ML on FactoryTalk® InnovationSuite DataFlowML DataView ThingWorx Analytics Edge Key People - Executive Sponsorship - Account Management / Sales - Data Scientist - {RA, Customer} - Domain Expert / Engineers - {RA, Customer} “BYODST”“BYODST” Data Explorer Interactive ML Studio ThingWorx - Mashup, - ThingModel Multiple Options Available Multiple Options Available DataFlowML Edge 50% of work Pre-Sales DataFlowML Data Scientist / Citizen Data Scientist Operationalizing Considerations - Prediction Pipeline - Prediction Itself - Retraining Pipeline Available Roadmap
- 17. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 19 Critical Success Points Data Science Process Define Business Problem Map to Machine Learning (ML) Problem(s) Exploratory Data Analysis ML Model Building (Train / Test) ML Model Evaluation Operationalize Data Preparation / Data Wrangling 80% of work 20% of work 100% of value Pre-Sales ML on FactoryTalk® InnovationSuite Critical Success Point # 1 – Successfully Defining the ML Problem Value Not yet realized until Operationalization on Critical Success Point # 2 – Successfully Preparing the Training Dataset Critical Success Point # 3 – Successfully Creating the ML Model # 1 # 2 # 3 Most Important Steps Yet to Come!
- 18. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 20 Operationalize Step Prediction Pipeline Data Science Process ML Model Execution (Prediction) Prediction Data Preparation / Data Wrangling 100% of value ML on FactoryTalk® InnovationSuite DataFlowML / Edge ThingWorx Analytics “BYODST”“BYODST” Data Explorer Interactive ML Studio Transforms pushed to DataFlowML or Edge 50% of work can operationalize ML Models created with “BYODST” Data Acquisition ? ML Model pushed to DataFlowML or Edge Question - What does your “BYODST” do? Available Roadmap FactoryTalk® can assist with these operationalization considerations FactoryTalk® I can assist with Training your ML Model and the Entirety of the Data Science Process Got a Prediction … Then What? Operationalize
- 19. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 21 Operationalize Prediction Pipeline - Prediction Data Science Process Prediction 100% of value ML on FactoryTalk® Fac DataFlowML Edge Operationalize - Prediction Visualization {Streaming Predictions} Question - What does your “BYODST” do? DataFlowML Edge WriteBack to Controller {CIP, FTLD, OPC-UA*} Saving Predictions {Data Lake, SQL Data Store} DataView Vuforia DataFlowML Edge Sending to Another Process {MQTT, Message Queue, IoT Hub} Edge DataFlowML Real-time Predictive Analytics Driving Human Decision Making ThingWorx - Mashup, - ThingModel Decision Making Done For You Prescriptively By ML and AI Available Roadmap FactoryTalk® I can assist with these operationalization considerations FactoryTalk® can assist with Training your ML Model and the Entirety of the Data Science Process * OPC-UA
- 20. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 22 Online Learning Auto Retraining Manual Retraining Operationalize Retraining Pipeline Data Science Process ML on Operationalize Question - What does your “BYODST” do? Legacy Approach Available Roadmap Available Roadmap Teaching AI to “Think like a Human” & “Learn like a Human” Automating…Manual… FactoryTalk® I can assist with these operationalization considerations FactoryTalk® I can assist with Training your ML Model and the Entirety of the Data Science Process
- 21. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 23 ML on the Then… and Now…
- 22. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 24 Then… Fall 2018 ML on the Available Roadmap That was Then…
- 23. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 25 Now… Summer, 2019 ML on the Available Roadmap FactoryTalk® I is rapidly expanding its capabilities, features, and interoperability related to Data Science, Machine Learning, and AI This is Now…
- 24. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 26 ML on the ML Reference Architecture
- 25. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 27 Reference Architecture Model Training Process ML on the Available Roadmap FactoryTalk® simplifies and streamlines the process for - Data Acquisition, - Data Wrangling, - Model Training … allowing multiple options for each
- 26. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 28 Reference Architecture - Model Execution Process - ML NOT @ the Edge ML on the Available Roadmap enables ML Model Execution NOT @ the Edge {Corporate Data Center, Public / Private Cloud} and facilitates - Data Acquisition, - Data Wrangling, - Model Execution … allowing multiple options for each
- 27. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 29 Reference Architecture - Model Execution Process - ML @ the Edge ML on the Available Roadmap enables ML Model Execution @ the Edge and facilitates - Data Acquisition, - Data Wrangling, - Model Execution … allowing multiple options for each
- 28. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 30 ML on the Data Pipelines
- 29. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 31 Legacy ML Approach Data Pipelines ML Select M1. a AS Temp1 M1. b AS Press1 ,M1. AS Temp1 ,M2.d AS Press2 ,M3.e AS Temp3 ,M3.f AS Press3 FROM M1 INNER JOIN M2 ON M2.time BETWEEN M1.time + 1s and M1.time -1s INNER JOIN M3 ON M3.time BETWEEN M2.time + 1s and M2.time – 1s # -- Comment Print(Temp1) # Adjust input by 5% Temp1 = Temp1 * 1.05 # -- Comment Print(Temp1) M1 M2 M3 Queries Various Tools PredictionCode
- 30. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 32 Legacy ML Approach Manual Steps Data Pipelines Difficulty Time- Aligning Time Series Data ML Select M1. a AS Temp1 M1. b AS Press1 ,M1. AS Temp1 ,M2.d AS Press2 ,M3.e AS Temp3 ,M3.f AS Press3 FROM M1 INNER JOIN M2 ON M2.time BETWEEN M1.time + 1s and M1.time -1s INNER JOIN M3 ON M3.time BETWEEN M2.time + 1s and M2.time – 1s # -- Comment Print(Temp1) # Adjust input by 5% Temp1 = Temp1 * 1.05 # -- Comment Print(Temp1) M1 M2 M3 Queries Various Tools PredictionCode Lots of CSVs! Lots of Imports / Exports! can eliminate these Manual Steps and streamline your Data Pipeline and your Data Science Process!
- 31. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 33 ML on Big Data Pipelines Available Roadmap ML Select M1. a AS Temp1 M1. b AS Press1 ,M1. AS Temp1 ,M2.d AS Press2 ,M3.e AS Temp3 ,M3.f AS Press3 FROM M1 INNER JOIN M2 ON M2.time = M1.time INNER JOIN M3 ON M3.time = M2.time # -- Comment Print(Temp1) # Adjust input by 5% Temp1 = Temp1 * 1.05 # -- Comment Print(Temp1) M1 M2 M3 Queries Various Tools PredictionCode
- 32. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 34 ML on Streamlined Approach! Big Data Pipelines Available Roadmap ML M1 M2 M3 Queries Various Tools PredictionCode Accounts for Time Aligning of Data No CSVs! Inline Data! Select M1. a AS Temp1 M1. b AS Press1 ,M1. AS Temp1 ,M2.d AS Press2 ,M3.e AS Temp3 ,M3.f AS Press3 FROM M1 INNER JOIN M2 ON M2.time = M1.time INNER JOIN M3 ON M3.time = M2.time # -- Comment Print(Temp1) # Adjust input by 5% Temp1 = Temp1 * 1.05 # -- Comment Print(Temp1)
- 33. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 35 ML on the Data Scientist / Citizen Data Scientist Experiences
- 34. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 36 ML on Citizen Data Scientist Experience Available Roadmap GUI Based Data Wrangling GUI Based Model Execution ThingWorx Analytics Machine Learning PredictionData DataFlowML GUI / Livy Data Wrangling Model ExecutionData Preparation FT InnovSuite simplifies the Data Science Process … making harnessing the Power of Spark and the Scalability of Big Data as simple to leverage as “Point and Click” in Excel
- 35. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 37 ML on Data Scientist Experience Available Roadmap SQL Based Data Wrangling Machine Learning Prediction Data Spark SQL Data Wrangling Model ExecutionData Preparation M1 M2 M3 DataFlowML GUI / Livy Data Explorer GUI Based Data Wrangling Python R Custom Code Data Wrangling ThingWorx Analytics Interactive ML Studio GUI Based Model Execution Select M1. a AS Temp1 M1. b AS Press1 ,M1. AS Temp1 ,M2.d AS Press2 ,M3.e AS Temp3 ,M3.f AS Press3 FROM M1 INNER JOIN M2 ON M2.time = M1.time INNER JOIN M3 ON M3.time = M2.time # -- Comment Print(Temp1) # Adjust input by 5% Temp1 = Temp1 * 1.05 # -- Comment Print(Temp1) Java * Jupyter - Python, - R - PySpark - Scala, - Julia Scala Jupyter* See Visual on Previous Slide for Enlarged Image Custom Code Model Execution Python R Jupyter* C/C++/C#.Net Java Scala C/C++/C#.Net “BYODST” Model Execution Python R Jupyter* C/C++/C#.Net Scala Java SparkML H2O PMML Interoperability with your Data Science Tools!
- 36. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 38 ML Adoption Strategy Citizen Data Scientist Demo
- 37. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 39 Share your feedback Please complete the session survey on the mobile app Select TechEd and login Use your email and last name that you used to register for the event. Click on Schedule on the main menu • Select the session you are attending • Click on the survey tab • Complete the survey and submit 2 3 Download the Events ROK mobile app 1
- 38. PUBLIC | TechEd | #ROKLive | Copyright ©2019 Rockwell Automation, Inc. 40 www.rockwellautomation.com Thank you for attending! <Singularity /> ROKAI