Link Search Menu Expand Document
UC Berkeley, Fall 2020

Reduce service outages of BART

Partner: Wendy Wheeler, BART (BAY AREA RAPID TRANSIT), Government

Overview

Project Description

Due to the drop in ridership from the COVID-19 pandemic, BART’s operating budget has been drastically reduced. We need to use the vast amount of data gathered by our systems daily to predict outages and prevent train delays and impacts to plant operations. Our data consists of detailed event logs from traction power, fire life safety, train control and passenger information systems, and human generated logs of maintenance activity and daily operations impacts. If we could find a way to utilize our data more effectively, we could reduce our maintenance costs and provide improved train service to the entire Bay Area. BART would be able to do more with the limited funds available to handle system outages.

Expected Deliverable

There are two major lines of development to tackle this semester:

  1. Using natural language processing of the operations logs and maintenance data to identify train delays and the associated locations/devices, and possibly improve input with auto-complete or other UI enhancements
  2. Use machine learning techniques on the huge volume of machine-generated field device events to identify unusual field behavior.

Both teams working together will inform the longer term objective of developing predictive maintenance alarms - by correlating unusual field events with operations impact and maintenance response.

What would a successful semester look like to you?

BART will have a better understanding of how Machine Learning (ML) techniques can be applied to BART’s data, students will learn the value of ML in large control systems, and the foundation for ML-generated predictive maintenance alarms will be laid.

Additional Skills from ideal candidates

We are looking for students that can think outside of the box to solve real world problems.

Data

Models

Conclusion