Car sensors now have a real-world dataset to learn what engine trouble looks like.

Researchers have released a new dataset of real-world engine sensor data from commercial vehicles. This data includes expert labels for normal operation and early signs of engine faults, which will help train AI systems to detect problems before they happen.

For years, developing AI to predict engine failures in vehicles has been difficult because there wasn't enough real-world data. Most available data was either synthetic or too simple. This new dataset, EngineAD, provides a realistic benchmark. It shows that current AI methods struggle to work across different vehicles, and surprisingly, older, simpler methods often perform as well as complex deep learning models. This means the path to reliable AI for vehicle maintenance is more challenging than expected.