Unearthed San Francisco 201624 Sep - 26 Sep
From Idea to Prototype on Resource Challenges in just a Weekend
This competition has finished.
Predict Delays and Optimize the Worlds Largest Heavy Haul Railway
BHP Billiton is seeking innovative approaches in identifying, predicting and optimizing delays at the worlds largest heavy haul rail network
Download the full challenge description here
BHP Billiton operates the largest private heavy haul rail network in the world, within its Western Australian Iron Ore operations. Each ore train, on average, contains over 230 ore cars carrying 34 thousand+ tons of ore worth over $2 million dollars. Each minute of lost time on the rail network can cost over $28,000.
Rail delays are where a train has fallen outside an allowable threshold of a target time for travelling across a section of track. This could happen for a number of reasons including; speed restrictions, meets with another train, Maintenance on the track etc. Rail delays are currently automatically generated by a delay accounting system with no reason recorded, or manually entered by an operator and categorized to the best of their understanding. In May 2016 alone there was over 1900 classified delays.
The main challenge is assigning root causes for delays to the time periods that are normally automatically generated. Currently this is done manually by operators, as the automatic delays are flagged up in the train control system back in the remote operations centre. As we try to understand the root causes of delays and attempt to remove sub-optimal performance, the manual classification approach will simply not scale.
Once the delays are automatically classified, the potential for data driven improvement is massive and yet to be fully explored in both historical and real time applications.
Potential Areas to consider:
- Given historical delay and source data, can train delays be automatically classified, and predicted before they happen.
- Is there a way to improve the train schedule based on the historical delays and bottlenecks that can be inferred from the data?
- Is there a better way to visualize the train delays and track status?