Technological innovation
Special Recognition

Expanding prognostics and integrating machine learning with bus telematics

MTA New York City Transit (Department of Buses)

  • Technological innovation
Where
United States
when
2022
Elevator pitch
The Department of Buses (DOB) is implementing machine learning algorithms and expanding prognostics with telematics in our maintenance strategy, all with the view to proactively identify failures before they occur and provide precise repair plans to efficiently and accurately repair defects.
Project description

With the largest municipal bus fleet in the United States, New York’s DOB’s main maintenance strategies include:

  • Scheduled bus inspections to identify defects
  • OEM recommended maintenance schedule
  • Diagnostics based on Bus Telematics

Even with a comprehensive maintenance plan, having a fleet of over 5,800 buses inevitably leads to critical component failures and road calls. Road calls immediately disrupt service to our customers, and each instance is costly due to the additional labour, materials, and loss of revenue.

The DOB has been utilising bus telematics for post failure alerts and diagnostics for a number of years. We are now implementing prognostics and machine learning to predict issues and resolve them before failures occur.

Innovative features

Having the largest bus fleet in North America means that MTA has massive amounts of telematics data and history to feed the machine learning algorithm. The project is able to leverage the telematics technology, hardware, and infrastructure MTA has invested in over the years.

This has allowed for the daily flow of telematics data from each bus when it returns to the depot and ensures the machine learning model is working with latest data. In fact, prognostic alerts have shown to be over 80% accurate.

Impact features

The project pilot resulted in:

  • Improved customer experience
  • Reduced road calls, repeaters, and service interruptions
  • Improved bus availability
  • Machine learning aids in troubleshooting complex issues
  • Reduced work stress
  • Optimised maintenance
  • Machine learning helps troubleshoot complex recurring issues
  • Case studies showed labour hours reduced by 43% and material cost by 24%
  • 100 million performance data points + 10,000 fault codes = 50 actionable repair plans