Detecting Machine Running, Idle and Stopped States with Computer Vision

Choose an observable cue

Potential cues include a moving mechanism, product flow or a status lamp. Confirm their meanings with the machine owner: a rotating fan does not necessarily mean productive operation. Check whether the cue remains visible during setup and maintenance. Prefer an existing reliable machine signal when it provides clearer evidence.

Define a state model

Specify running, idle, stopped and unknown conditions with transition rules. Use dwell times to avoid rapid switching caused by brief motion gaps, but preserve short stops if they matter to the process. Keep planned breaks and changeovers as contextual classifications rather than assuming the camera can infer their reasons.

Handle lost evidence explicitly

A blocked lens, failed stream or overexposed image should lead to unknown coverage, not stopped or running by default. Record the observation timestamp separately from processing time. Provide an operator correction workflow with reason and audit history when the visible cue is misleading.

Validate transitions and duration

Compare a labelled timeline with PLC records or supervised observation. Measure incorrect state duration and transition delay, not only frame accuracy. Include cleaning, maintenance, empty running and blocked production. Agree how unknown intervals affect utilisation reports so camera faults do not distort equipment performance.

FAQ