2. Agenda
B-LOGISTICS WAGON MAINTENANCE - 23 JULY 20152
1. About me & my company
2. Project introduction
3. Process raw data
4. Iterative improvement
5. Final results
3. • Bart Maes
• Background in Computer Science (A.I.)
• Data coordinator at SNCB Logistics
• New to data science/big data
• Traveler (on a shoestring)
B-LOGISTICS WAGON MAINTENANCE - 23 JULY 20153
A bit about myself
4. • 200 locs
• 7000 wagons
• 1900 employees
• 30 million transported tons
• 5 billion ton-kilometers
• 2000 trains per week
B-LOGISTICS WAGON MAINTENANCE - 23 JULY 20154
A little more about SNCB Logistics
5. • .. 2011: Ruination
• State owned, part of NMBS
• 2009: +100 million loss/year (EBITDA)
• 2011: private company, NMBS 100% shareholder
• 2011 2015: Transformation
• 2014: 11 million profit (EBITDA)
• 2015: private investor acquires 66,6% of shares
• 2015 ∞: Innovation
• eLearning
• MIA/RITA
• Big Data
• …
B-LOGISTICS WAGON MAINTENANCE - 23 JULY 20155
A brief history of SNCB Logistics
6. • Maintenance optimization for (own) wagons
• 4800 wagons
• Wagon breakdown
• Each wagon checked before each train departure
• Will probably result in lost train
• Hence, expensive
• Current maintenance process
• Time based: every 6 years, simplified or major overhaul
• Cost: €4-16k, independent of wagon condition
• Hypotheses
• Wagon usage varies widely
• Can tolerate less frequent maintenance of less used wagons
B-LOGISTICS WAGON MAINTENANCE - 23 JULY 20156
Project summary
7. B-LOGISTICS WAGON MAINTENANCE - 23 JULY 20157
Project steps
1. Reconstruct each wagon’s exact history to
deduce:
• Travelled distance
• #Loads/unloads (‘cycle’)
• Weight transported
• Damage reports
2. Reconstruct estimated wear from maintenance
reports
3. Correlate wear to wagon history
4. Given usage, estimate time until maintenance
is necessary
Wear
km travelled
Need
Overhaul
8. • Initial rough estimate
• Uneven distribution
• Large number of low-usage
wagons
• Smaller number of high-usage
wagons
• Indicate potential for
maintenance policy improvement
B-LOGISTICS WAGON MAINTENANCE - 23 JULY 20158
Analysis: distance travelled
9. • Source: CargoWeb
• Collected events for wagons
• Potentially 100s events / week
• Generated by operational systems
• Manually inserted
• ~200 types of events: departure, arrival, ready for X, take
over, hand over, passed through, …
B-LOGISTICS WAGON MAINTENANCE - 23 JULY 20159
Raw Data: wagon events
11. B-LOGISTICS WAGON MAINTENANCE - 23 JULY 201511
Summarize raw events
Time Place Event
… … …
30-Apr-2012 10:06 Antwerpen Berendrecht Cirkeldijk Physical restitution — empty wagon
30-Apr-2012 10:35 Antwerpen D.S. Berendrecht Train composition OK
30-Apr-2012 12:29 Antwerpen D.S. Berendrecht Train composition OK
30-Apr-2012 13:08 Antwerpen D.S. Berendrecht First Departure
30-Apr-2012 17:00 Antwerpen Waasl-Zuid van Moer Arrival at Destination
30-Apr-2012 17:20 Antwerpen Waasl-Zuid van Moer At Disposal
… … …
From To Action
… 10:06 At rest at ABC
10:06 10:35 Travel ABC – ADSB
10:35 13:08 At rest at ADSB
13:08 17:00 Travel ADSB – AWZvM
17:00 … At rest at AWZvM
12. B-LOGISTICS WAGON MAINTENANCE - 23 JULY 201512
Summarized data: still challenging
arrival
travelling: 231
at rest: 176
other (after arrival)
travelling: 571
at rest: 991
departure
travelling: 33
at rest: 658
travelling: 2
at rest: 29
travelling: 1126
at rest: 8048
travelling: 113
at rest: 1417
travelling: 2126
at rest: 10
travelling: 229
at rest: 217
other (after departure)
travelling: 66
at rest: 30
travelling: 58
at rest: 28
travelling: 2
at rest: 8
travelling: 7
at rest: 19
Even after abstracting and simplifying,
interpreting the data is challenging
13. B-LOGISTICS WAGON MAINTENANCE - 23 JULY 201513
Distance travelled: a first reconstruction
• Room for improvement:
• Noise
• Gap in Business Understanding
• Goal:
• Improve until ‘good enough’
14. • Ignore manually inserted events
• Unreliable
• Often provide little extra information
• Cluster stations into groups
• Maps with business activities
• Eliminates false positive mismatch between arrival & departure
station
• Improve interpretation of event sequences
B-LOGISTICS WAGON MAINTENANCE - 23 JULY 201514
Distance travelled: interative improvement
15. • ‘Travel’ action defines movement between 2 points A & B
• No central source for all distances between every 2
points
• Distance matrix constructed using different sources:
• Actual RAILDATA measurements
• Extrapolation (cluster nearby stations)
• Google maps
• Haversine distance
• Manual input
B-LOGISTICS WAGON MAINTENANCE - 23 JULY 201515
Distance travelled: another hurdle
17. B-LOGISTICS WAGON MAINTENANCE - 23 JULY 201517
Some wagons have very distinct usage patterns
Cycles & average load: results
18. B-LOGISTICS WAGON MAINTENANCE - 23 JULY 201518
Some wagons have very distinct usage patterns
Cycles & average load: results
Genk-Chatelet
Gent-La Louviere
Gent-Clabecq
19. • Business satisfied by presented results
• New maintenance plan uses ‘wagon groups’
• Focus on distance travelled
• Other elements may be added in the future
• #Cycles
• Average load
• Main achievements:
• Cost reduction for maintenance
• Risk/cost reduction for breakdown
B-LOGISTICS WAGON MAINTENANCE - 23 JULY 201519
Final results & business actions
20. SNCB Logistics I Koning Albert II-laan 37 1030 Schaarbeek I Boulevard du Roi Albert II 37 1030 Schaerbeek
T +32 2 432 90 00 I F +32 2 432 90 05 I info@sncblogistics.be I www.sncblogistics.com
Thank you