A robot cell that stops mid-shift costs more than the minutes it sits idle. Upstream stations pile up, downstream ones starve, and the people on the floor start improvising. Then comes the second cost: the argument about what happened. Operators remember one thing, maintenance another and the controls engineer a third. The remedy is not more meetings but a single-page fault summary written while the facts are fresh. This article diagnoses the problem of messy stoppage records and lays out the page that fixes it.
The problem: stoppages that leave no usable record
Most plants log downtime somewhere, but the entries are thin: “gripper fault, restarted”. That tells the next shift almost nothing. It does not say what the gripper was doing, what was tried or whether the fault is likely to return. Meanwhile, the detailed knowledge lives in someone’s head, and heads go home at the end of the shift.
Three symptoms signal that your records have this problem.
- The same fault returns and nobody connects the dots. Each occurrence is treated as new.
- Handovers rely on conversation. The incoming shift asks the outgoing one what happened, and the answer varies by who is asked.
- Fixes are rediscovered. A workaround found on Tuesday is unknown by Friday.
The diagnosis: why the records fail
The cause is rarely laziness. Stoppage forms are built for counting minutes, not for explaining events, so there is nowhere to put the useful parts. Under pressure, people write the shortest thing that closes the entry. Long reports, on the other hand, take too much effort to write, and so they are produced only after serious events, if at all. The result is a gap between the one-line log and the full investigation, and most stoppages fall into that gap.
The fix: a one-page fault summary
Fill the gap with a page that is short enough to complete in ten minutes and structured enough to be useful. Every summary has the same five blocks.
- What stopped and when. The cell, the station, the time and what the robot was doing at the moment it stopped.
- What was observed. Facts only: the fault code displayed, any unusual sound or movement, the last successful cycle. Photographs and controller screens can be linked.
- What was tried. Steps taken in order, including those that did not help. This prevents the next person from repeating dead ends.
- Cause or working theory. Label it clearly as confirmed or suspected. Pair a suspected cause with the check that would confirm it; that beats a loose note like “mechanical issue” by a long way.
- Decision and next step. Keep running with a watch, replace a part, schedule a deeper inspection. Name the role responsible and the time frame.
A worked example
Consider a pick-and-place cell whose arm halts intermittently with a position error. The summary might read: Stopped: packing cell, mid-shift, during a pick from the incoming tray. Observed: a position-limit fault appeared three times within an hour; the arm’s last move looked slightly short of the target; no damage visible. Tried: restarted the cycle, cleaned the vision lens, checked tray alignment. The first two did nothing; correcting the tray stop cured the fault for the rest of the shift. Theory: the tray locating pin has worn, allowing trays to sit a few millimetres off; suspected, not yet confirmed. Decision: maintenance replaces the pin at the next planned stop; operators check tray seating at each change until then.
It is a short read, and it turns an annoying interruption into something the whole plant can act on. If you want grounding in how the pieces involved work, the explainer on how industrial robots use sensors to know where things are is useful background for the “observed” block, and the walk-through of key measures managers watch on robotic lines shows where stoppage data eventually ends up.
Making it work on the floor
- Keep the form on a tablet or a clipboard at the cell. If it is far away, it will not get filled.
- Allow photos and screen captures. They replace paragraphs of description.
- Agree on the vocabulary. Use the same names for stations, parts and fault types across shifts.
- Review weekly. Read the pages together and look for repeats. The patterns are the payoff.
- Close the loop. When the next step is done, mark it, so open items do not pile up.
A word on photos, codes and wording
Small habits make the page far more useful. Copy fault codes exactly as displayed, because a single character can point to a different subsystem. Write times in the controller’s clock, not from memory. When describing what the arm did, use plain verbs such as “stopped short”, “drifted left” or “hesitated”, and avoid guessing at internal causes in the observed block. If a picture shows the problem better than a sentence, attach it and say what to look at. A reader six months later will thank you for each of these.
What to watch out for
Do not use the page to assign blame. If operators fear that an honest entry will be used against them, the pages will say very little. Frame the page as a tool for fixing the machine and protecting the next shift. Also resist the urge to extend the form. Every added field lowers the chance that it gets finished. If a stoppage truly needs a full investigation, link to it from the page instead of merging the two.
Borrowing habits from software operations
Plant teams sometimes find ideas in the way software teams record and review incidents, because the discipline is similar: concise records, a clear decision and an owner for the follow-up. Anyone curious about those habits could browse DevOpsSchool China, a site where learning and advisory services for software delivery and cloud running are presented to a Chinese-reading audience. The settings differ, since software has no tray locating pins, but the thinking behind a good write-up transfers readily.
The result you should expect
After a few weeks, you should see the same things happen in most plants that adopt this. Handovers get shorter, because the incoming shift reads instead of asking. Repeat faults show up in the weekly review, and the fix is applied once instead of rediscovered many times. Maintenance can plan work from real descriptions rather than guesses. None of it requires new equipment, only a sheet, a few honest minutes and the habit of writing down the point first.