Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data...

13
Large Scale Data Analysis Techniques in Grinding Planning and Management Joseph W. Palese, MCE, PE HARSCO Rail 1 1

Transcript of Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data...

Page 1: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Large Scale Data Analysis Techniques in Grinding

Planning and Management

Joseph W. Palese, MCE, PE

HARSCO Rail

1

1

Page 2: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Data Acquisition PlatformsGeometry Car Pre‐Grind Insp VehicleGrinder

UT Insp Vehicle Hand/Visual Measurements

2

Page 3: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Diagnostic Systems

Rail Profile Rail Surface

Corrugation Rail Defect

Track Layout

Rail Imaging

3

Page 4: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Data Elements

Rail Profile Rail SurfaceRail Imaging

Rail Defect Track LayoutCorrugation

4

Page 5: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

General Extent of Data – BIG Data

1000 kB kilobyte 10002 MB megabyte 10003 GB gigabyte 10004 TB terabyte 

10005 PB petabyte

Per Mi 20,000 mile RRMbytes Mbytes

Rail Defects 0.002 41                       Track Layout 0.7 13,517               Rail Profile Data 3                           50,688               Corrugation Data 14                        282,962             Dense Profiles 304                      6,082,560         Rail Images 3,244                  64,880,640       

Total 3,566                  71,310,407       

3 Insp/Yr (Mb) 10,697                213,931,222    for 20 years 213,931              4,278,624,441 

Petabytes of Data 4.3                      

5

Page 6: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Traditional Grinding Planning Activities

• Develop Grinding Plan– Performed offline

• Perform Grinding Pre‐Inspection– Using hy‐rail vehicle

• Perform Onboard Monitoring and QC– Onboard Grinder

• Monitor Effectiveness– Offline

6

Page 7: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Traditional Analytical Techniques• Physical and Deterministic Techniques Employed• Computationally Intensive• Make Use of Readily Available Data• Develop Indices for Large Data Sets for Data Reduction• Provide Information to Railways from Large Data

Rail Defects

Track Layout

Rail Profile Data

Corrugation Data

Dense Profiles

Rail Images

PlanningPre‐InspectionReal Time and QCMonitoring

7

Page 8: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Grinding and BIG Data

• The Five V’s of BIG data– Volume  5 Petabytes– Variety  profiles, corrugation, track layout, defects, images, crack density, machine performance, etc.

– Velocity  up to 2,500 Hz– Veracity  consistent data with some spread in accuracy– Value  North American rail budget = $2B

• While Big Data analysis is being performed, emerging techniques offer the ability to improve the current deterministic methods through advanced stochastic processes

8

Page 9: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Current Areas for Improvement • Develop Grinding Plan

– Use limited amounts of data– Simple statistical techniques– Only a rough plan

• Pre‐Grind Inspection– Use limited aspects of data– Simple statistical techniques– Missing condition elements– Missing asset properties

• Real Time Analysis– Computational limits– Disregard machine data– Missing condition elements– Missing asset properties

• Effectiveness Monitoring– Use limited amounts of data– Simple statistical techniques– Not correlated to other 

inspection data

9

Page 10: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Future of BIG Data Analysis in Grinding

5 Petabytes of data available for mining and analysis

Utilize advanced techniques such as:MARSMarkov Chain Monte Carlo simulationBayesian InferenceMachine LearningMany, many more

Expected improvements:Improved grinding pattern predictionRefined rail template definitionImproved makeup of grinding stonesGrinding cycle managementImproved grinder utilizationMany, many more

10

Page 11: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Improvements to Existing Techniques

Advanced Statistical Models‐MARS‐MCML‐BI

Self Learning System‐Improved patterns‐Machine speed‐Dynamic patterns for changing profile in curves‐etc

Important PredictorsPredicator RelevanceCoefficient Correlation

11

Page 12: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Strategic Initiatives• BIG Data available to define key relationships unknown to date

Data Over LifeAll Sources

MARSWeighted Predictors

Lubrication Effectiveness Grinding Cycle

Size of Grinder

Grinding Approach

Others to Learn

Rail Degradation

OperationalOptimization

Wheel/Rail InterfaceOptimization

12

Page 13: Joseph W. Palese, MCE, PE HARSCO Rail · Joseph W. Palese, MCE, PE HARSCO Rail 1 1. Data Acquisition Platforms Geometry Car Grinder Pre‐Grind Insp Vehicle UT Insp Vehicle Hand/Visual

Conclusions

• Rail grinding planning and management is a Big Data exercise• Current techniques are effective but do not make use of large 

amounts of data• Emerging techniques can be applied to provide:

– More effective results in the planning and management process

– Further useful information regarding grinding and the rail asset in a more strategic manner

13