The New York City Metropolitan Transit Authority (MTA) has partnered with Google to conduct a short-term experiment exploring the use of cheap, near-off-the-shelf technology to aid track inspections. In this experiment, the smart Google phone Pixel was fixed inside and at the bottom of the subway car, and through sensors such as accelerometers, magnetometers, and gyroscopes, and even external microphones, it collected track status data in real time.Currently, the MTA's track inspections rely heavily on manual inspections, with inspectors patrolling the 665-mile track of the New York subway on foot to check for track breaks, signal failures, and flooding. In addition, three times a year, MTA collects more accurate infrastructure data using "track geometry inspection vehicles" equipped with advanced sensors. TrackInspect, an experimental project with the Google Public Sector, shows that by collecting audio, vibration and location information, combined with AI predictive model analysis, it can effectively assist manual inspections and improve efficiency.Experimental data showed that during the four-month test, the technology successfully identified 92% of the track defect locations. MTA said that this technology is expected to reduce the workload of manual fault identification in the future, allowing inspectors to focus more on repairs rather than just finding problems, thereby improving the efficiency of track maintenance."In the future, we want to build an automated, modern system that will allow for the rapid detection of track problems and the organization of maintenance work," said Demetrius Crichlow, chairman of the MTA. "For the New York subway, which serves 3.7 million passengers a day, early detection and repair of faults could mean that commuters can get to their destinations on time, rather than being trapped in their carriages due to sudden breakdowns."The goal of this experiment is to identify issues before they impact operations, thereby reducing service disruptions due to sudden failures," Ctricklow emphasized. The MTA announced that the project will enter a full pilot phase, with Google developing an official version of TrackInspect and making it available to track inspectors.
AI-empowered track inspection: Smart devices help track maintenance
Google's TrackInspect experiment is one of the latest examples of the gradual adoption of AI-assisted rail inspections in the global public transportation sector. Brian Poston, vice president of rail transit at WSP Consulting, said that while the New York MTA's innovation lies in using "vibration and audio signals" to detect track problems, subway systems in other cities are also experimenting with installing sensors and cameras to automatically measure track status and flag anomalies in real time.
This technological breakthrough relies not only on the development of machine learning, but also on the use of cheaper, smaller batteries and processors. However, U.S. federal regulators still require rail facilities to be manually inspected and maintained, and this standard will not change in the short term. "Until the technology is precise enough, manual inspection will remain indispensable," Boston said. ”
In the experiment at the MTA in New York, one of the key human inspectors, Robert Sarno, the MTA's assistant chief track officer, put on noise-canceling headphones and listened carefully to 5 to 10 minutes of track sound clips in addition to his daily work to distinguish between normal operating noise and potentially faulty signals. He manually marked anomalies such as "loose joints", "loose bolts", "track defects" and "sleeper problems", and compared the results with photos taken by the on-site inspectors.
In the process of analyzing the track audio data personally, Sarno achieved an accuracy rate of more than 80%, which is close to the level of a machine learning model. MTA insiders have even called him a "track whisperer", arguing that his sensitivity to track sounds is extraordinary. Through his annotation and feedback, TrackInspect collected a total of 335 million sensor data, covering 1,200 hours of track audio, and combined with the MTA track defect database, trained about 200 predictive models to further improve the detection accuracy.
Chris Hein, Google's public sector chief technology officer, said that while TrackInspect is currently being developed specifically for MTAs, the project is expected to be a "lower cost and more preventive" new technology driver for rail safety around the world.
Rail Transit Exhibition Booth Reservation
Rail Transit Exhibition Free registration for visit
