A micro-downtime rarely lasts more than a minute. Taken in isolation, it seems insignificant: the machine restarts on its own or after a simple intervention, without any breakdown or damage. The problem is not their individual severity, but their repetition. On a production line, several dozen micro-downtime events each day can add up to a loss of production time far greater than that of a major shutdown, while remaining largely invisible in standard OEE (Overall Equipment Effectiveness) reports.
Industrial monitoring systems generally record significant stoppages, such as machine breakdowns, scheduled maintenance, or other events. But they do not systematically detail interruptions lasting from a few seconds to a few minutes—such as a missing part, manual adjustment, quick cleaning, or a changeover. Without detailed visibility, it’s impossible to identify which ones occur most frequently and why.
A micro-downtime event may have a mechanical cause (adjustment, jam), a flow-related cause (waiting for a part, upstream bottleneck), or a human cause (recurring manual intervention). Without granular data on the context of each stoppage, the production team can only act based on assumptions.
Unlike machine sensors, which report only a binary status (on/off), video analysis observes the scene: the presence or absence of an operator at the station, manual intervention, a buildup of parts waiting to be processed, or unusual movement on the line. This contextual analysis makes it possible to distinguish between a micro-stop caused by a recurring adjustment and one caused by a supply disruption.
By cross-referencing downtime detection with the visual context at the time of the event (human presence, affected area, duration), it becomes possible to group micro-downtimes by cause rather than treating them as an undifferentiated data stream.
Over a given period, aggregating micro-stops by workstation, shift, or time slot reveals recurring bottlenecks on the production line that, when combined, have the greatest impact on overall throughput.
Once high-risk workstations and time slots have been identified, process and production teams can prioritize their corrective actions: reorganizing a workstation, adjusting a work standard, or providing temporary reinforcement in supply. The goal is not to eliminate all human intervention, but to bring to light what, until now, was lost in the background noise of day-to-day operations.
CORE connects to the cameras already installed on the production line—without requiring any additional hardware investment—and continuously monitors each workstation. The platform identifies:
This data complements existing supervisory systems (MES, CAPM) by providing the contextual layer that purely mechanical sensors lack. The processing is based on the analysis of silhouettes and movements, without facial recognition or the identification of operators by name, in compliance with the GDPR.