Highlight
Listening checks that rely on the subjective impression of an experienced technician have been a common practice in many different industries, including quality inspections on production lines and field inspections of infrastructure. Unfortunately, these in-person checks are becoming less practical as the shrinking of the workforce over recent years is causing a shortage of personnel with the necessary skills.
In response, Hitachi has been working to develop technology and offer solutions for automating sound-based inspection practices and making them more efficient. Past practice has been to input the sound of machine operation into a system that outputs a result indicating the extent to which it diverges from the normal sound, or that issues an alert if this divergence exceeds a threshold. This article describes an AI technique that, instead of just using data on machine operating sounds as a basis for issuing alerts, also generates text explaining the detected anomaly in a way that prompts maintenance actions.
