Hitachi Rail has announced the integration of NVIDIA's AI technology to enhance railway operations, reduce maintenance costs, and improve energy efficiency.
The company is incorporating NVIDIA IGX, an industrial-grade edge computing platform, into its new HMAX platform. This integration aims to enable real-time processing of sensor and camera data, allowing for quicker detection of track repairs, monitoring of overhead power lines, and assessment of train and signaling equipment health.
Koji Agatsuma, executive director and chief technology officer of rail vehicles at Hitachi Rail, highlighted the potential impact: "Using previous digital monitoring systems, it would take a few days to process the data and discover issues that need attention. If we can instead conduct real-time prediction using NVIDIA technology, that enables us to avoid service disruptions and significantly improve safety, reliability and operating costs."
Hitachi Rail estimates that proactive maintenance is approximately seven times less expensive than emergency repairs following unexpected infrastructure failures. The company's existing AI monitoring systems have already yielded significant benefits, including a reduction in service delays by up to 20%, a decrease in train maintenance costs by up to 15%, and a reduction in fuel costs at train depots by up to 40%.
In the UK, Hitachi Rail trains are equipped with sensors that report nearly 50,000 data points as frequently as every fifth of a second. The new AI infrastructure is designed to keep pace with this data flow, enabling timely alerts for maintenance needs and optimising maintenance schedules.
The technology also has applications beyond train and track monitoring. Hitachi Rail is exploring the use of train-mounted cameras to monitor overhead power lines, identifying degrading electric cables to prevent disruptive failures. This process, which traditionally takes up to 10 days to process one day's worth of video data, can now be performed in real-time at the edge using NVIDIA-accelerated sensor processing.