The MBTA uses a machine-learning system, developed in partnership with vendor 4Atmos Technologies, that analyzes oil samples taken from diesel commuter rail locomotives to predict engine failures 30 to 60 days before they occur. The program began as a pilot in spring 2018 and transitioned to full production across the commuter rail locomotive fleet.

The AI input is molecular oil sample data (measured in parts per million) from each locomotive, collected at the MBTA's on-site oil lab at the Somerville maintenance facility. The system analyzes levels of elements such as copper, iron, and zinc and applies machine learning to identify progressive patterns—called 'failure signatures'—across the entire fleet simultaneously. The output is a risk ranking of locomotives flagged as 'high fliers' most at risk of failure, along with prescription recommendations for inspection and repair. Prior to this program, MBTA technicians manually reviewed individual oil sample reports as a lagging indicator, only examining historical data after a failure occurred. The MBTA's chief operating officer confirmed that the AI can find a problem instantly that would previously take two to three hours to do manually.