The Grafana screen is on, but nobody uses it during the shift

A production display hangs above the entrance to a workshop. Charts move, counts rise, and a status tile turns red now and then. When a job falls behind, the shift lead still walks over to ask an operator what happened. The display may be working exactly as configured. It simply does not answer the question the shift lead needs answered.

Grafana can present measurements from connected data sources in a live dashboard. The difficult part in this case is not putting a chart on a television. It is making a screen people use during the shift. That requires a decision about audience, definitions, data age and what happens after someone notices an exception. A screen that is always visible is not necessarily part of the work.

Who is supposed to look at it?

An operator needs to know whether a stop was recorded against the right machine. A shift lead needs to spot jobs that may miss the next handoff. Maintenance wants to see whether the same fault keeps returning. Those questions overlap, but they are not interchangeable. Showing everyone the same crowded page because the data is available usually hides the one fact each person came to find.

Consider where the shift lead will stand. If the screen is read while walking through the cell, a concise list of jobs needing attention may be more useful than a detailed trend over several hours. A separate view can hold the history for investigation. On the live screen, the job identifier, the current state and the timestamp of the underlying observation should be readable at a glance.

Labels matter as much as charts. “Production performance” gives little guidance. “Assembly jobs currently waiting” is closer to a question someone can act on. But it also needs a definition: does waiting start when a machine stops, when an operator records a reason, or when a planned start time passes? If two teams use different definitions, a single figure can create disagreement rather than common understanding.

A shift in a hypothetical sheet metal shop

Imagine a small shop moving cut parts into a bending area. The screen near the press brakes shows the next jobs, the time each job’s status was last recorded and the state of each station. One order remains in setup longer than expected. The lead reads the job number from the display, then checks at the machine whether a tool change is taking longer or the material has not arrived.

The display has done something useful: it directed a conversation to the right job. It has not diagnosed the cause. “Long setup” should not quietly become “slow operator.” The next part may have an unusual geometry. Quality may be holding the first piece. The status may have been left unchanged after the setup ended. Used with the team, the screen also reveals which status fields are reliable and which definitions need repair.

A large total output counter might have missed this situation entirely. Overall throughput can rise while one urgent order waits. Conversely, painting every pause red makes planned maintenance and normal changeovers look like failures. People quickly learn to ignore an alarm that cannot distinguish an expected event from one requiring intervention. Design the display around an exception someone can check, not simply a number that looks impressive from across the room.

Does “live” mean the data is fresh?

Grafana reads what a data source provides. Refreshing a panel does not prove that the source has received a new observation. A connection could stop while the last known value remains visible. On a shift display, the time the measurement was captured should be distinguished from the time the page last refreshed. Repainting an old value every few seconds does not make it current.

Different questions tolerate different delays. Yesterday’s completed order totals do not need a second-by-second update. A panel claiming that a machine is currently running should not silently present a stale reading. If a source has stopped reporting, the display should make that gap apparent rather than continue to show an apparently healthy state. An explicit timestamp and a visible stale-data condition are more helpful than a rapidly changing animation.

The physical setup deserves the same scrutiny. Can people read the job number from the usual viewing distance? Does the color scheme work in the actual lighting? Does the browser session remain signed in, or does the television end up showing a login page after a restart? An internal test on a laptop will not answer these questions. Nor should a display in a shared area reveal personal or customer-sensitive details just because someone found them useful in a private planning view.

Connect the signal to a next step

A useful screen does not need to send a notification for every deviation. It does need an agreed check. If a bending job has been in setup unusually long, the first step might be to confirm when the status was recorded. Then check whether the material reached the machine. Only after the hold is understood does someone involve planning or maintenance. The team needs to discuss that sequence alongside any threshold chosen for the panel.

A restrained pilot can start with one cell, one shift and one operational question: are we noticing waiting jobs soon enough to check them? Show a rough view to the people who will use it before making it permanent. During the trial, note when the display helped, when it lacked enough context and when it raised a false concern. If the identifier is hard to read, make it larger. If “waiting” means different things to different people, fix the definition. If updates arrive late, display the delay rather than polishing the colors.

Screen uptime alone is a poor success measure. A more useful test is whether someone who notices an exception can verify it against a shop floor record and reach the person who can resolve it. If no one looks at the display, duplicating it across the plant will not make it more useful. It may simply sit outside the normal route through which the shift communicates.

The earlier piece on Grafana as a live view over the data layer covered what the dashboard can show. This question is narrower: what makes it worth glancing at during work? The best live display does not ask people to believe every number without question. It shows which observation is recent, which is uncertain and which conversation should happen next.

Sources: Luke van Enkhuizen’s overview of open source manufacturing tools and Grafana’s dashboard design guidance. The workshop example is illustrative.