Hitachi uses AI on Databricks to detect railway overhead line defects from onboard train video in near real-time

Overview

Hitachi installed cameras on trains to capture video of 40,000 km of overhead rail lines and built an AI pipeline on Databricks Lakehouse, using Delta Lake for storage and MLflow for model management, to analyze the footage and alert rail network operators to defects or displacement in near real-time, enabling predictive rather than reactive maintenance.

The challenge

Monitoring and tracking potential overhead-line issues was traditionally a manual task: network operators walked the tracks, visually inspected lines from moving trains, or relied on infrequent measurement trains limited to detecting certain issue types. Any overhead line break caused delays and costly unplanned disruptions.

The solution

Hitachi installed cameras on trains and uploaded the video to the cloud, building an end-to-end AI pipeline on Databricks Lakehouse. It uses Delta Lake for data storage and pipelines, MLflow for model management, training, deployment and experiment tracking, and Databricks SQL for monitoring dashboards, analyzing footage from 40,000 km of overhead lines to generate near-real-time alerts on defects or displacement.

Reported business value

Databricks Lakehouse sped up Hitachi's time to market for new ML models and improved the efficiency of its data engineers and scientists, letting a smaller, more flexible team create innovations benefiting thousands of customers daily. Hitachi has discovered thousands of equipment risks and line displacements that clients could fix proactively, with the net result estimated at millions of pounds in savings for Hitachi's customers, turning a days-long manual process into a near-real-time, remote one.