ABB’s Head of Digital Industry Value Engineering and Data Science for Industrial AI, Anindya Chatterjee, says machine learning is changing industrial maintenance by continuously analysing operational data to catch failures before they occur. In customer deployments, ABB has used multivariate anomaly detection models to identify potential failures “up to six to 10 days before they occur,” which can mean the difference between an orderly repair and an emergency shutdown. The same approach predicts remaining useful life based on actual condition, allowing work to be planned around real risk instead of a fixed schedule. Chatterjee notes that most industrial maintenance budgets are still consumed by corrective repairs after failures, and machine learning offers a route towards greater reliability and less production disruption. He identifies data trust as a key challenge, stating that maintenance histories, failure records and reliability data are often less structured than operational readings, and gaps in this history make it harder to build trustworthy models. ABB works with customers to validate, contextualise and enrich data before model development. His advice is not to wait for perfect data, arguing that doing so risks turning data quality into an excuse for delaying technologies that can deliver meaningful operational benefits today.