Lee NIST PHM Fina Handout Revised.Pptx

NSF I/UCRC since 2001

Recent Advances and Transformation Direction of PHM

Jay Lee

Ohio Eminent Scholar, L.W. Scott Alter Chair, and Univ. Distinguished Professor Univ. of Cincinnati [email protected] & Director NSF Industry/University Cooperative Research Center on Intelligent Maintenance Systems (IMS) Intelligent Cyber Machine Systems Univ. of Cincinnati, Univ. of Michigan, Missouri Univ, of S&T, Univ. of Texas-Austin

Outline

► Evolution of Manufacturing and PHM

► Changing Big Data Environment and Emerging Technologies on Big Data Analytics, Cyber-Physical Systems, and Industry 4.0

► Lessons Learned from PHM and New Applications

► Conclusions

1 The IMS Center

Global Industry Partners (80+)

LEGEND: GERMANY KOREA JAPAN •Omron Corporation • Active Member • FORCAM BELGIUM • Samsung • Past Member •Nissan CANADA • Siemens • FMTC Semiconductor • Syncrude • Hitachi USA • Samsung (Electro- FINLAND Mechanics • Komatsu • P&G •Mitsubishi Heavy • Kone • KAU • NI Industry • Parker • Toshiba Corporation • Goodyear • Boeing TAIWAN • GE • HIWIN • Woodward FRANCE • Advantech • API • Alstom • III • GCWW • PMC • Applied SPAIN • MIRDC Materials (2) •Tekniker • PSi • LAM • ITRI • TI • Delta Electronics • Micron HONGKONG • Tongtai Machine Tool • Canrig • Metron Hongkong • Raytheon • SCK BRAZIL Ltd. • Emerson • HRL • CETA/ • TechSolv • Ford SENAI • Idaho Nat. Lab • Intel • Chevron CHINA CHINA • Ingersoll Rand • Caterpillar • ETAS • Siemens TTB • CEPREI • China State Ship Co. • Spirit Aerosystems• BorgWarner • Montronix • USPS • Genex • Bosch • CEI • Dongling Tech • 21st Century Systems • Daimler-Chrysler • Festo • Toyota • McKinsey & Co.• Sinovel • Shanghai Electric • Avetec • Harley-Davidson • Cisco • Inteligistics• United • Beijing Shenzhou Software• Haier • Eaton • Coherix • Johnson • Prometec Technologies • AITRI Shanghai • Bao Stell • Kistler • EDAptive Controls • Rockwell • We Energies • Sany Heavy Industry • Saanxi Heavy Truck • © Copyright IMS Center, 2014. All Rights Reserved. www.imscenter.net

2 White House Digital Manufacturing and Design Innovation (DMDI) —Led by UI Lab (Feb. 2014)

Outline

► Evolution of Manufacturing and PHM

► Changing Big Data Environment and Emerging Technologies on Big Data Analytics, Cyber-Physical Systems, and Industry 4.0

► Lessons Learned from PHM and New Applications

► Conclusions

3 Lean/ Smart Service Quality PLM and Value 6-Sigma Creation Focus Business

1980 1990 2000 2010 2020 TPM CBM PHM Predictive (Organization (Component/ Centric -Practice & (Data Centric) Analytics Machine Centric) Culture) (Customer-Centric) & Tools & Tools

Methods Methods • Smart Sensors • Web-enabled Monitoring/ • SPC and TQM • Big Data Analytics • Real-time Monitoring/eMainenance • Time Series Model • Autonomous Agent Control • On-Board Diagnostics / • Expert Systems- • Smart Cyber-Physical • Sensor Networks Prognostics based Diagnostics Systems for Fleet- • Machine Learning • Embedded Product Life Cycle • AI Tools based Systems • Data Mining Info for Closed-Loop Design • Fault Tolerant • Cloud-based PHM • Embedded Systems • RUL Prediction Control • APP-based PHM • 6-Sigma Design • Event-based PHM • Mechatronics/ • CPS-based PHM • Degradation • Multi-Regime PHM Robotics • Industry 4.0 Systems

Technologies Measurement • Virtual Metrology • Zero-Defect • Remote Monitoring • Decision Support System and Scheduling

Maintenance of the Future (2000)

Closed-Loop Near-Zero Life Cycle Downtime Design Product Health Monitoring Design for or Sensors & Embedded Just-in-Time Reliability and System Intelligence Serviceability Service

® • Web-enabled Monitoring Product Product Watchdog Agent & Prognostics Center Redesign Degradation Assessment (Feature Monitoring) • Decision Support Tools for Maintenance Smart Scheduling Design Self-Maintenance Communications • Redundancy • Tether-free • Active • Internet • Visualization • Passive • TCP/IP • Worry-Free Productivity Enhanced Six-Sigma Health Information Design

Watchdog Agent® is a registered trademark of IMS Center. IMS 2000

4 Competitiveness Transformation Strategy

Utilize New Value Creation Knowledge/ using Avoid Technologies Smarter Information For Value-added For Unknown Knowledge Improvement

Problem Solving Through Continuous Improvement and Utilize New Methods/ Solve Standard Work Techniques to Solve The Unknown Problems 5S Surprise

Visible Invisible

Outline

► Evolution of Manufacturing and PHM

► Changing Big Data Environment and Emerging Technologies on Big Data Analytics, Cyber-Physical Systems, and Industry 4.0

► Lessons Learned from PHM and New Applications

► Conclusions

5 6Cs in Big Data System 1. Connection --RFID, Wireless, Sensor Networks

2. Cloud – Computing and Data on Demand

3. Cyber— Model and Memory

4. Content/Context – Correlation and Classification

5. Community -- Relationship and Sharing

6. Customization – Service and Value

Smart Manufacturing Analytics Information on Demand (meaning and content)

Machine Health Analytics (PHM) Factory Analytics Lee, J. Lapira, E. Predictive Factory, Manufacturing Leadership Journal, March 2013. Lee J, Lapira E, Yang S and Kao A. Predictive Manufacturing System – Trends of Next-Generation Production Systems. Intelligent Manufacturing System (IMS) 2013 Symposium, May 21-24, 2013. Sao Paolo, Brazil.

6 What are Cyber-Physical Systems?

► Physical – natural and human-made systems governed by the laws of physics and operating in continuous time ► Cyber – computation, communication, and control that are discrete, logical, and switched ► Cyber-Physical Systems – systems in which the cyber and physical components are tightly integrated at all scales and levels

Ref: NSF CPS Program, 2007

Cyber-Physical Interface

Physical World Cyber Physical Interface Cyber Space COMPONENT Degradation Model Risk Analysis Chart Performance Evaluation DATA 12 10 8 Threshold 6 4 Accelerated Testing 2 0 Performance (Health Value) Value) (Health Performance 0 40 80 120 160 Time (Days) Data- Expert Physical driven Knowledge Model Model Leave from Factory

MACHINEE Health Check

Maintenance # Time Machine Predicted Life

#2 Time Virtual Synthesis #1 Time Machine PRODUCTION Machine LINE Virtual Virtual Virtual Production Component Machine Machine Line #3 Time Cluster Machine

#n Time Machine INFORMATION #4 Time Machine

7 Architectures for 5 Levels (5C) of CPS IMS CPS V. 1.0 A F V. • Self-configure for resilience Configuration • Self-adjust for variation T U Level • Self-optimize for disturbance T N R C • Integrated simulation and synthesis I T IV. Cognition • Remote visualization for human • Collaborative diagnostics and decision B I Level making U O • Twin model for components and T N machines • Time machine for variation E S III. Cyber Level identification and memory • Clustering for similarity in data S mining • Smart analytics (smart agent) II. Data-to-Information for component and machine Conversion Level machine health assessment • Data correlation • Plug & Play • Tether-free I. Smart Connection Level communication • Sensor network

EU & Germany

Industry 4.0 Internet of Things and Services

8 Germany Global Mechanical Product Sales 2006-2011

Wake-up call for Germany

From Industry 1.0 to Industry 4.0

Acatech- German Academy of Science & Eng, Prof. Dr. Henning Kagermann

http://www.uberb2b.com/b4b-presents-the-first-industry-4-0-mini-conference/

9 Comparison of Industry 4.0 Factory vs. Today’s Factory

Today Factory Industry 4.0 Factory

Key Data Source Attributes Key Technologies Attributes Technologies

Degradation Component Sensor Precision Sensing Self-Aware Measurement

Self-Predict Quality & Monitoring & Health Machine Controller Self-Compare Performance Diagnostics Prognostics

Efficiency Lean & Production Networked Self-Reconfigure Worry-Free & Green Systems Systems Self-Optimize Production Productivity Manufacturing

Jay Lee, Germany Harting Tech New 26 ,2013

1St Germany-U.S. Workshop

10 Outline

► Evolution of Manufacturing and PHM

► Changing Big Data Environment and Emerging Technologies on Big Data Analytics, Cyber-Physical Systems, and Industry 4.0

► Lessons Learned from PHM and New Applications

► Conclusions

Watchdog Agent®  Self-Aware Machine Sensor Data Signal Processing, Vibration Current Acoustic Temperature Data Acquisition Feature Extraction & Selection Metrology Data

Before Regularly Errors Inspect Happen Health Health Performance Assessment Diagnosis Prediction Act Logistic Support Vector Autoregressive Regression Machine (SVM) Moving Average Statistical Pattern Feature Map (ARMA) Recognition Pattern Matching Feature Map (Self-organizing Elman Recurrent Pattern Matching Controller Data Maps) Neural Network (Self-organizing Maps) Bayesian Belief Each Fuzzy Logic Neural Network Network (BBN) Gaussian Mixture Hidden Markov Match Matrix Model (GMM) Model (HMM)

Info- graphs

Health Map Risk Radar Degradation (Image courtesy: http://www.huazhongcnc.com/en/Product/CNC-M/473.aspx Chart (Confidence Value) http://www.apisensor.com/index.php/products-en/machine-tool-calibration-en/vec-en)

11 Implementation of PHM Analytics

Step 1:

Understand and Validate the Past Issues

GM E-Manufacturing Testbed St. Catherine Assembly Plant, Canada

12 GM E-Manufacturing Testbed (2002)

GM Factory Analytics for Data Base Validation

Machining Big Data and Stored by: • Use the past sensor data • Plant • CV Calculation • Machine • 0-1 Indicator • Sensor • Date Center for • Time IMS Historical Data © Copyright IMS Center, 2014. All Rights Reserved. www.imscenter.net

Analytics for Machine Tooling Health

• Spindle load sensor in a machining center. • 1218 boring process cycles.

Normal Operation

Degradation

Warn Tool

Tool Change

Using Logistics Regression algorithm to convent past current signals to tool life degradation

13 Data Challenge

1. NASA PHM Data Depository

2. PHM (Prognostics and Health Mgt) Data Challenges Competition:

2008 (1st and 3rd) 2009 (1st Professional Group) & (1st, 2nd, 3rd Student Group) 2010 (3rd) 2011 (1st and 3rd) 2012 (3rd)-- 2013 (3rd, 4th, 5th) 2014 (1st).

Implementation of PHM Analytics

Step 2: Don’t Just Listen to Customer but Do Understand the Issues & Prioritize the Issues Focus on the High-Impact Things

14 IMS, Toyota, NI, Rockwell Automation Partnership Toyota Motor Manufacturing, KY

Problems on Compressor Bearing

Varnishing - overheating due to metal to metal contact

15 Overview of Work Sensors installed Data acquisition installed • PXI hardware on loan to New IGV installed IMS center IGV & surge testing • Donation of time: Data collected Onsite &distributed troubleshooting Data analyzed • Expert support: DAQ internally & by IMS center Project documentation Key Areas of Concentration • Loaned Preliminary surge Hardware Data Collection model & Analysis • Expertise in sensor troubleshooting © Copyright IMS Center, 2014. All Rights Reserved. www.imscenter.net

16 What did we learn?

Work with maintenance expert and suppliers to understand the untold issues?

Surge Compression Ratio – 2nd Stage Point 20

We can prevent surge by monitoring the instability of the compression ratio. A control can be implemented to avoid the surge issue.

Amp. Transfer to supplier for implementation.

Predictive Robot Maintenance at Toyota Georgetown, KY MR50% in 2008

17 Robot Health Map by Location

► Show whether certain areas (of process) were failing more often than others. ► Display bottlenecks of the process.

Critical Components

Clamp Unknown Switch, other Switch, prox.

More Spare Robot 2 Parts 1 Pallet Switch, limit Matehen Cable, other Encoder Motor Regular 3 Bracket4 Predictive Maintenance Maintenance

18 Robot Health Monitoring Nissan Manufacturing Plant Smyrna, Tennessee

Robot Health Assessment -Example Health Assessment Result for Third Servo-motor Robot Joint

Degradation can be detected as early as 3 weeks before event!!! Actual failure event

► Applying the IMS logistic regression algorithm included using the moving average and RMS torque value of the low-speed regime segmented data set as the two features used in training the model using the unacceptable (degraded) state data and acceptable (healthy) state data.

► The results of applying this method for the third robot joint servo motor are shown above, which shows that early signs of degradation can be seen as early as 3 weeks in testing cycle 125 but failure does not actually occur until cycle 220.

19 Screenshot of Reliability --Before

Screenshot of Health Map--After

20 Factory Sentinel—Predictronics New Product

Ref: Predictronics

Implementation of PHM Analytics

Step 3:

Integrate Machine Data, Historical Data, and Expert Experiences for Improved PHM Value

21 Intelligent Maintenance of Komatsu Mining Equipment and Diesel Engines 2005-2007 (Komatsu Engineer Stay at IMS for 18 Months)

Prediction of Remaining Lifetime of Engines

ENGINEERING KNOWLEDGE AND DATA ACQUISITION FROM THE SYSTEM

Komatsu Customer Komatsu Engineers Appliance WebCARE

A B Group Methods Signal Processing Data Preparing Autoregressive Data Handling (Filtering Outlier & Modeling Based Feature estimation) Feature Extraction Extraction

C D E Health Assessment Fuzzy Logic (FL) Match Matrix & Bayesian Based Decision ARMA Modeling & Belief Making Prediction Network (BBN)

F AUTOMATIC AID AUTOMATIC AID TO Remaining Lifetime Human Machine TO DIAGNOSIS PROGNOSIS Prediction Comparison DECISION Interface DECISION and Predictive MAKING MAKING Maintenance Planning

22 IMS vs. Komatsu Approach

– to increase quality of data in Komatsu database – to enable more advanced predictive maintenance decision Sensors making base on analytical reliability models IMS – to achieve intelligent Watchdog maintenance excellence Agent® KOMATSU On - Board Computer

– and to implement knowledge Monitoring & Aid View to Decision based decision making Engine DT09 XXN Replacement, High priority Engine Engine DT09 XYH Replacement, High priority Comp .A ,C Maintenance required approach Engine DT43 XZA Plan replacement in 1500 h Comp .D Engine DT77 YVP Good, 5000 h remaining No change

IMS/KomatsuTestbed 2005-2007

IMS Watchdog Agent™ Toolbox

Sensors Valve Bearing

Cylinder

IMS Watchdog

IMS/Komatsu Prognostics Simulator

Phase I Phase II Phase III

23 Implementation of PHM Analytics

Step 4:

Embedded Smart Analytics with Any Systems (Controller, Server, Cloud, Components, etc.)

Reconfigurable Prognostics Systems Enable Zero-Breakdown Productivity

24 New Next of PHM Analytics

Using Cyber Physical Systems

IMS Self-Aware Machine Tools

Cyber Space PHM Apps Module-Based Controller “App Store”

Physical Space Data Machine Self-Awareness

Closed-Loop External: Feedback Sensor

Internal: Controller

(Image courtesy: http://www.huazhongcnc.com/en/Product/CNC-M/473.aspx)

25 Industry 4.0 Machine Demonstration at IMTS Chicago Sept. 2014

Case52 Study with Cosen Band Saw Machine

http://www.directindustry.com/prod/starrett/band-saw- blades-11639-534692.html Failure Modes Blade Types Material Domain • OEM • Geometry • Material • Cross-section Knowledge Cutting Parameters • Tooth configuration • Hardness

Signal Processing CPS objective Maintenance & utilization suggestion Health Assessment » IMS PHM • Adaptive clustering » Product quality control Tools • Statistical pattern recognition Prognostics • Utilization based prediction • Proportional hazard model

26 53 On-Machine Agent for Utilization based Prognostics

Rotating machine: High Low Load Load High Rate 1 Rate 2 Speed Low Rate 3 Rate 4 Speed Prediction Degradation Histories Knowledge Extraction Inputs: Current Condition Utilization Prediction Health Future Utilization pattern Model vs. Stress 1 degradation rate Stress 2 Predicted

Health Health Stress …

Time

Band Saw Degradation Analysis Results

► After each cut, features are extracted as health indicators: – Energy percentage of frequency range [3750 4000] Hz – Blade downward pressure ► After each feature extraction, adaptive clustering is used to perform machine health assessment in real-time.

Step 1: on-line adaptive local clustering Step 2: hierarchical clustering Feature2 Feature2

Time Feature 1 Step 3: prediction and decision making under each cluster Feature 1

27 Machine Mobile Agent (APP)- New Way of Machine Management

List of Machines in factory plant with abstract information about each machine including: working status, Latest health value and last timestamp of historical data

Cognition & Analytics on Demand

By clicking on each machine, detailed information of that machine will be displayed. In the first section (overview) a radar chart for overall health status of machine components is displayed

28 Cincinnati as Industry 4.0 City, Nov. 5, 2014

Conclusions

29 Transformation

Maintenance/PHM Transformation Map

Engineering Self- Preventive Immune Industry 4.0 Maintenance Maintenance Systems Cyber-Physical Systems Prognostics & Health Resilient eMaintenance Systems Management

Predictive Maintenance Robust CBM RCM Design

Ref: Jay Lee, Annual Reviews in Control, Volume 35, Issue 1, April, 2011.

30 Key Value of Predictive Big Data Analytics

Big Data is to Create No-Data Decision for Worry-Free Productivity

New Book on Cyber Physical Systems in Manufacturing, Industrial Press, 2015.

www.imscenter.net

Google Jay Lee, Prognostics, E-Manufacturing, E-Maintenance, Dominant Innovation, Cyber Physical Systems, Industrial Big Data Analytics