Predictive Analytics and the Industrial Internet of Manufacturing Things with Will Sobel

The IoT Inc Business Meetup Silicon Valley

Opening remarks and guest presentation

Bruce Sinclair (Organizer): [email protected]

Target of Meetup

For business people selling products and services into IoT

but of course everyone else is welcome: techies, end-users, …

Focus of presentations and discussions:

© Iot-Inc. 2015

Type of Products

All incremental value in an IoT product comes from transforming its data into useful information

• Trinity of value: model, app and analytics

Analytics transforms data into value

© Iot-Inc. 2015

Coming Up

© Iot-Inc. 2014

Next Meeting, Thursday, May 5, 2016

Mega Meetup2 will be in July

Presentation, recording of Meetup and announcements for today’s meeting will be sent in one week to everyone who provided their email upon signup for this meeting or any other past Meetup

Send announcements to me: My email: [email protected]

Feel free to upload pictures and tweet with hashtag #iotbusiness

Reviews would be great!

mailto:[email protected]



Sponsor spots still available!

Food & Drink is $600 per meeting x 10 meetings a year = $6,000

• Gold = $2,000 and Bronze = $500

We have Sponsors!

© Iot-Inc. 2014

Gold Bronze Shelter

Predictive AnalyicsIndustrial Internet of Things for discreet manufacturing

IOT for BusinessApril 7, 2016

Berkeley,CA|Chennai,India1

William SobelChief Strategy OfficerSystem Insights

MTConnect Chief Architect

VIMANA by SystemInsights Copyright © 2016 System Insights, All Rights Reserved

About Myself…

2

• ChiefStrategyOfficer–SystemInsights• ChiefArchitectoftheMTConnectStandardforthelast9years• ChairoftheMTConnectTechnicalSteeringCommittee

– OriginalAuthoroftheStandard• Co-ChairIICIndustrialAnalyticsTaskGroup

• MyBackground– FinancialSystemsArchitecture– Real-TimeAnalytics– DistributedArchitecture&FaultTolerance


System Insights

3

The predictive analytics platform for manufacturing intelligence

2009

Berkeley, CA

2010

India

40+

Customers

60+

Sites


IIOT for Manufacturing

4

Industrial Internet of Things


Internet of Things

5

• Industrial vs. Everything Else – Highest potential Impact – $1.2 - 3.7T

• Areas – Operations Optimization – Predictive Maintenance – Inventory Optimization – Health and Safety

McKinsey Global Institute THE INTERNET OF THINGS: MAPPING THE VALUE BEYOND THE HYPE JUNE 2015


New Economies

6

Present

Emergent

5-10 Years

10 - 25 Years

Manufacturing Services & Data Monetization

Outcome Based Economy

Autonomous On-Demand & Distributed

Operational Efficiency

World Economic Forum: Industrial Internet of Things: Unleashing the Potential of Connected Products and Services, Jan 2011


Intelligence Hierarchy

7

← 95% of shopsPrimitive - No software or control software only

Monitored - Data collection and backward looking reports

Real-Time Analytic - Determines why a process failed or productivity was lost

Proactive - Detect problems before they happen using CEP and learning

Self Optimizing - Learning models optimize processes

Predictive - Problems are solved before they impact process

Pres

ent

Disr

uptiv

e

Intelligence


Steel Production

8


Metal Cutting

9

Milling

Turning

TraditionalLaser

Water Jet

Wire EDMNon Traditional

Additive

HybridAdditive


Digital Thread

10

010010011


Products

11


Case Study: Predictive Analytics for Process Manufacturing

Karthik Ranganathan

12


Predictive Quality and Yield

13

• Complex multi-stage high-speed manufacturing process• Process is manually configured and adjusted• Utilize legacy data collection of low-level sensor data• Data Cleansing and Semantic Transformation

– Multidimensional Data Modeling• Contextualize Data with Device, Part and Process


Multi-dimensional data

14

1. StoreEverything–Can’tcreatedata2. Distributedanalysis:Slicedataacrossany

plane,including:time,machineorganization,parts–Findpatternsandcorrelations

3. Everythingmusthavecontext


Data Set

15

• Source– 33Tablesand800+variables/columns– Size:250+GB

• Challenges– Datawasorganizedper-operation,notacrossalloperations– Variedparametersacrossoperationsandtypeofequipment– Potentialmissingdataanddataentryerrors

• Thedatahadtobepreppedforanalysis,includingCleanupandContextualization


Data Cleansing

16

• Outliers– Observationswhichdeviatesignificantlyfromthe

norm,withsuspectveracity– Canremovetopreventmodelcorruption

• ProbableCauses– DataEntryErrors– MissingData– DifferingUnits– Measurementflaws

• Examples– UTS/LYSvalues>1000MPa,<10MPa– Outputslabweight>Inputslabweight(~1.85%)


Process-based Quality Prediction

17

Mean Deviation 2.0%

R-squared 81%

Actual Quality Parameter

Pred

icte

d Q

ualit

y Pa

ram

eter

Build process-based model to predict final quality


Quality Prediction

18

• Expectedqualityparametersprovidedbycustomer• However,itwasseenthattheseboundsarenotstrictlyfollowedwhileattributinggrade• Wedetectedthetruequalitylimitsautomaticallyfromthedata• Reductionoflowerboundaryby10~20MPaforbothLYSandUTS


Predicting Grade

19

• Predictgradebasedontheinputparameters

• Analysis:– 5GradesAnalyzed– TotalObservations–5008– TrainingData–80%

Production of Coils that match Expected Grade

92 +/- 1%

Matched Plant Predictions 99 +/- 0.5% Improvement on Missed Predictions 45 +/- 10%New Prediction Methodology 95 +/- 1%


Predictive Services

20

• Usingthemodels,wecanpredicttheUTSwithto95%Accuracybasedonchemistry

• Derivedfromempiricalanalysisofmanufacturingprocessanddataanalytics


Value

21

• Reducecostofquality– Controlqualitywithcriticalparameters– Understandprocessparameterstocontroltohavebiggestimpactonpartquality– Setparameterlimits–highergranularityandcontroloverprocess

• CaptureTribalKnowledge– Captureandquantifyknow-howofthe“humansintheloop”– Improveknowledgetransferandmanagementoversight

• Targetareasforprocessimprovement– Significantprocessvariation–identifyareaswhereithassignificantimpactonquality


Market

22

• GlobalSteelIndustry– Production(2015)–1690milliontonnes– Market–$1.3Trillion

• OneCompanyinStudy– Market:$3billion– Operationalprofit:$700million

• EstimatedSavings:– Basedona3-5%reductioninoperationalcosts– $150millioninpotentialprofit– $250billioninpotentialprofitforentiremarket


Big Data Services

23


Present: Feed - Forward Processes

24

• Part • Machine • Tool • Processes

• Best Practices • Improvement Opportunities • Deviations from Plan

• Was the part produced to plan? – Faster or slower? – Why?

• Part Program • Setup Sheet • Work Instructions

• Tool Breakage • Poor Surface Finish • Productivity Gains

No Robust Feedback mechanism to planning functions.

Programmer

Operator

• Process Actuals – Feed/Speed – Cycle Time – Operator Feedback


Current Feedback

25

Machining VerificationEngineeringDesign


Digital Manufacturing Feedback

26

Machining VerificationEngineeringDesign


Crowd Sourcing for Manufacturing

27

FacilityCommunity

• Knowledge-Based Recommendations – Optimized Conditions – Optimized Asset Selection – Optimized Throughput

Community Instructions, Plans

Process Data

Programs Instructions, & Plans

• Rules • Analytics

& Process Data

Analytics = Data → Knowledge


Tool and Process Analytics

28

Analyze process parameters and tooling usage to reduce tool costs and improve tool life

Detailed understanding of cycle time and process parameters Automatically identify best practices for process planning


Increases in Efficiency

29

Conservative Estimates:10% Increase in Production Efficiency + 10% Reduction of Cycle Time

Total Time: 100%

Producing: 40%

ICT: 60% of Prod.

Total Time: 100%

Producing: 50%

ICT: 70% of Producing

Total Time: 100%

Producing: 50%

ICT: 70% of Producing

CT Reduction: 10% Effectively: 62% higher production

5000 Hours

5000 Hours

5000 Hours

2000 Hours

1200 Hours

2500 Hours

1750 Hours

2500 Hours

1750 Hours

1950 Hours

Baseline These improvements are what was referenced in the report – remember?

1.2 - 3.7 Trillion


Predictive Analytics

30

Why would we do anything else?


Semantic Transformation

31

0 10

20

30

40 50 60 70 80 90 100

010100100a = 5000b = 12.43c = 87.22d = 2

Device Linear X Position: 196.54mm Load: 12.43% Rotary C Rotary Velocity: 87.22 RPM Controller Execution: ACTIVE

Meaning

Syntax SemanticsRaw Data

Propriatary Structure Wisdom

Bad

Good

:01 :02 :03 :04 :05 :06

V Mill 1

Lathe 1

Robot

CMM

Cell

Day

Line

Part 1Part 2

Part 6

......Batch

Execution: ACTIVELoad[X]: 120%Feedrate[Override]: 80%

Time

Equi

pmen

t

Prod

uct

Analysis Predictive

Prognositcs

Value

Information

Data Enrichment

part XY32 – part AB56 –

Producing part AB56at avg. SS of 2500 RPM

Producingunknown part

Producing part XY32at avg. SS of 2160 RPM

–

Not Producing Producing

Producing State – Device: When All Paths Producing

Path 1

Path 2

MULTI-PATH Device

Not Producing Producing

Producing State – Device: When Any Paths Producing


Continuous Improvement

32

Continuous Improvement

Baseline

Analyze

Evaluate

Production

Implement

Execution optimization Process Improvement Workforce Training

Tooling Optimization Predictive Quality Manufacturing Strategy Design for Manufacturing


Questions?

33

Predictive Analytics and the Industrial Internet of Manufacturing Things with Will Sobel

Technology

Transcript of Predictive Analytics and the Industrial Internet of Manufacturing Things with Will Sobel