Making Big Data Risk Analysis

work for the GB Railways

Peter Hughes

Big Data Risk Assessment research team, Institute of Railway Research,

University of Huddersfield

GB railway

• More than 15,000 km of track

• More than 1.5 billion passenger journeys per year

• Passenger numbers have doubled in the past 20 years

• Currently holding a good record for safety

Safety Risk Model

• Fault-tree-based model underpins railways’ SMSs.

• Allows queries only within the structure of the model.

• We often don’t have data for failures that have not occurred.

• Keeping the model up-to-date is laborious: it can take one to two years to reissue the model.

Undesired outcome

Real-time data is what we expect the future will be

It’s easy to imagine an alternative...

With modern technology it is easy to envisage: – automatic collection of real-time data; – data from many sources: traditional sensors, video,

mobile phones, text-based reports, ...

How to do this in practice?

Combining many sources of data has problems, for example how to:

– combine conflicting data? – use data that goes out-of-date? – capture usable data from audio, video, text? – include new data sources as they become available? – deal with interruptions in data feeds? – abstract raw data into railway risk assessment knowledge? – avoid spurious correlations? – obtain knowledge from a blend of numeric and

non-numeric data? – ...?

The vision

What is the bit in the middle?

Weekday

Date of last repair

Driver experienc

Platform crowding

Track condition

Bridge number

Event: Train

approaches signal

Train ID: 6B82

Signal ID: EW34

Location: 53.642669, -1.777940

Train mvmt events

Freight train

Mainline signal

Time: 04:32

Date: 02 Apr 16

Time before sunrise

Train operator

00011111000000011001110001001001011100000100000111011101001001110001100011100000000001000011100010

00000111111010111010001110101110110001100000010001001010101110011110010111101101111101110011100100

10011011110001100100010011111000100000100100011111100001011001000111101001011000000001010100111111

Queries Queries

The vision

Data processing Ontology Visualisation

Enterprise architecture

Data pre-processing

Enterprise architecture

Can we build it?

Data processing Ontology Visualisation

Are we building it? Hughes, Peter, Figueres-Esteban, Miguel and Van Gulijk, Coen (2016) From negative statements to positive safety. Proceedings of the 26th European Safety and Reliability Conference, ESREL 2016. (In Press)

Figueres-Esteban, Miguel, Hughes, Peter and Van Gulijk, Coen (2015) The role of data visualization in Railway Big Data Risk Analysis. In: Safety and Reliability of Complex Engineered Systems: ESREL 2015. CRC Press / Balkema, pp. 2877-2882. ISBN 9781138028791

Figueres-Esteban, Miguel, Hughes, Peter and Van Gulijk, Coen (2015) Visualising Close Call in railways: a step towards Big Data Risk Analysis. Proceedings of the 5th International Rail Human Factors Conference. (In Press)

Van Gulijk, Coen, Figueres-Esteban, Miguel and Hughes, Peter (2015) Big Data Risk Assessment the 21st Century approach to safety science. In: International Railway Safety Council 2014, 4th - 9th October 2015, Johannesburg. (Unpublished)

Figueres-Esteban, Miguel, Hughes, Peter and Van Gulijk, Coen (2016) Ontology netword analysis for safety learning in the railway domain. In: ESREL 2016, 25th - 29th September 2016, University of Strathclyde, Glasgow. (In Press)

Van Gulijk, Coen, Hughes, Peter and Figueres-Esteban, Miguel (2016) The potential of ontology for safety and risk analysis. Proceedings of ESREL 2016. (In Press)

Koornneef, Floor and Van Gulijk, Coen (2015) Computer Safety, Reliability, and Security 34th International Conference, SAFECOMP 2015 Delft, The Netherlands, September 23–25, 2015. Proceedings. Lecture Notes in Computer Science, 9337 . Springer International Publishing. ISBN 978-3-319-24254-5

EL Rashidy, Rawia Ahmed and Van Gulijk, Coen (2016) Driver Competence Performance Indicators Using OTMR. In: CIT2016. Congreso de Ingeniería del Transporte (XII Congress of Transport Engineering, 7th - 9th June 2016, Valencia, Spain.

Koornneef, Floor and Van Gulijk, Coen (2015) Computer Safety, Reliability, and Security SAFECOMP 2015 Workshops, ASSURE, DECSoS. ISSE, ReSA4CI, and SASSUR, Delft, The Netherlands, September 22, 2015,. Lecture Notes in Computer Science, 9338 . Springer International Publishing. ISBN 978-3-319-24248-4

Hughes, Peter, Figueres-Esteban, Miguel and Van Gulijk, Coen (2015) Learning from Close Calls.Technical Report. SPARK, Huddersfield, UK.

Oostendorp, Yvette, Lemkowitz, Saul, Zwaard, Walter, Van Gulijk, Coen and Swuste, Paul (2016)Introduction of the concept of risk within safety science in The Netherlands focussing on the years 1970–1990. Safety Science, 85. pp. 205-219. ISSN 0925-7535

Hughes, Peter, Figueres-Esteban, Miguel and Van Gulijk, Coen (2015) Learning from text-based close call data. In: Safety and Reliability of Complex Engineered Systems. ESREL (2015). CRC Press: Taylor & Francis Group, Zürich, Switzerland, pp. 31-38. ISBN 978-1-138-02879-1

Van Gulijk, Coen, Hughes, Peter and Figueres-Esteban, Miguel (2016) Big Data Risk Analysis for Railway Safety. In: World Congress of Railway Research, 29th May - June 2nd 2016, Milan. (Unpublished)

Stow, Julian, Zhao, Yunshi and Harrison, Chris (2016) Estimating the frequency of trains approaching red signals: A case study for improving the understanding of SPAD risk. IET Intelligent Transport Systems. ISSN 1751-956X (In Press)

Figueres-Esteban, Miguel, Hughes, Peter and Van Gulijk, Coen (2016) Visual analytics for text-based railway incident reports. Safety Science, 89. pp. 72-76. ISSN 0925-7535

Van Gulijk, Coen, Hughes, Peter, Figueres-Esteban, Miguel, Dacre, Marcus and Harrison, Chris(2015) Big Data Risk Analysis for Rail Safety? In: Safety and Reliability of Complex Engineered Systems: ESREL 2015. CRC/Balkema. ISBN 9781138028791

Figueres-Esteban, Miguel, Hughes, Peter and Van Gulijk, Coen (2016) Big Data for Risk Analysis: the future of safe railways. In: XII Conference on Transport Engineering, 7th - 9th June 2016, Valencia, Spain. (Unpublished)

Does the industry want it?

• Potters Bar, North London, 10 May 2002 • Seven people died following a derailment of a passenger train • The derailment was caused by degraded points • In the preceding days, drivers had reported abnormal track conditions

• With hindsight we can see that information necessary to address the problem had been available before the accident

• When the world is full of data, how can we get the safety information we really need?

End