Enterprises and OEMs are drowning in a flood of real-time data from equipment, assets, suppliers, consumers, and business systems. Hidden insights have the potential to optimize production, transform efficiency, streamline flows of goods, vehicles and services, protect the environment and ensure safety. But finding those insights remains a huge challenge. Complex, big-data focused, cloud-hosted IoT solutions are inflexible and unsuited for processing real-time data. Software must move to the data source. Edge computing paired with local, self-training machine learning can cost-effectively filter and analyse massive volumes of streaming sensor data. Edge intelligence affordably transforms dark data from industrial systems into real-time, automatically learned insights about equipment, assets, maintenance needs, and operational performance. This presentation describes an architecture for edge learning on time-series data using edge devices and an efficient edge fabric of digital twins, based on the distributed actor model. The approach flies in the face of the received wisdom of cloud-based, big-data solutions. Attendees will learn that there is more than enough resource at “the edge” to cost-effectively analyze, learn and predict from streaming data on-the-fly, avoiding the need to transport it, store, clean, label and learn in the cloud. Edge learning delivers new insights fast, specific to the local context, enabling the infrastructure to adapt to changing conditions.