DEV Community

robot-pallettizzatori
robot-pallettizzatori

Posted on

The palletizing cell already knows why it stopped — if you log the right things

Every palletizing cell I have ever walked up to had an opinion about why it loses time. The operator had one ("the wrapper upstream is slow"), the shift leader had another ("changeovers eat the morning"), and the PLC, quietly, had the real one — recorded in a fault register nobody had opened in months. End-of-line data logging is not an Industry 4.0 luxury. It is the difference between arguing about downtime and fixing it.

This is the logging setup I now treat as the minimum on a palletizing cell, and what each record is actually for.

Log stops as events, not as a single downtime number

A cell that reports "47 minutes of downtime" has told you almost nothing. What you need is a list of stop events: timestamp, duration, and a cause code chosen at restart. The distinction that matters most is between long stops and minor stoppages — the 20-to-90-second interruptions for a misaligned case, a slipped layer, a photo-eye that did not see a transparent film. Minor stoppages rarely appear in anyone's memory at the end of a shift, yet on high-rate lines they routinely add up to more lost production than the one big breakdown everyone remembers.

Cause codes should be short, physical, and boring: infeed gap, layer fault, gripper vacuum low, pallet missing, downstream blocked, wrapper waiting. If the list needs a meeting to interpret, it is too clever. And "downstream blocked" deserves respect: a palletizer that stops because the stretch wrapper or the pallet conveyor cannot take the load is not a palletizer problem, and the log is the only thing that proves it.

Split the cell's OEE before you quote it

Overall equipment effectiveness is usually calculated at line level, which hides the cell. I calculate the three components for the palletizing cell alone. Availability: planned production time minus stops, from the event log above. Performance: actual cycle time against the rated cycle for the current pattern — a cell running a complex interlocked pattern at 9 cycles a minute may be at 100% of its real rate while looking slow next to the catalogue number for a simple pattern. Quality: here it means good pallets out — no re-stacked layers, no rejected loads, no cases damaged by the gripper.

Splitting the number changes the conversation. A cell with 96% availability and 78% performance does not need maintenance; it needs its patterns and infeed spacing looked at. A cell with great performance and poor availability needs the stop log, not a faster robot.

Treat changeover as data, not as folklore

On multi-SKU lines the changeover is where the shift quietly goes. Log it like a stop with its own cause code, and log what the changeover actually consisted of: pattern recall, gripper adjustment or tool change, infeed guide settings, first-pallet check. After a month you will know which SKUs are cheap to run and which ones cost you twenty minutes every time they appear on the schedule — which is scheduling information, not just engineering information. In my experience the first-pallet check is the step that gets skipped under time pressure and the step whose absence shows up later as a quality stop. If it is logged, it tends to happen.

Counts that must reconcile

Three counters should agree at the end of every batch: cases entering the cell, cases in completed layers, and pallets discharged, with reject counts recorded separately — a case diverted for a damaged corner, a layer rebuilt, a pallet rejected at the wrapper. When the counters do not reconcile, something physical is happening that nobody has noticed: cases accumulating in a dead zone, a sensor double-counting, a pattern that silently drops a case position. Tie every count to the lot or batch number running at the time. Traceability requests never arrive when it is convenient, and "which pallets came from that batch" is a question you either answer from the log in two minutes or from memory, badly, in two days.

Timestamps are the whole game

None of this works without a single clock. Stops, counts, changeovers and alarms must share timestamps, because the useful questions are all correlations: does the vacuum fault cluster after changeovers to one particular SKU? Do minor stoppages rise in the last hour of the shift, when the infeed runs faster to catch up? Does the downstream block always follow the wrapper's film change by ten minutes? You cannot correlate what you did not timestamp, and you cannot reconstruct it afterwards — the cell forgets everything the moment the shift ends.

The weekly fifteen minutes

A log that nobody reads is just storage. The ritual that makes this work is embarrassingly small: once a week, fifteen minutes, the stop events sorted by total duration — not by count, and not by how memorable they were. The Pareto that comes out is almost never the one the shift would have predicted. The big breakdown that dominated Monday's meeting usually sits third or fourth; the top of the list is some ninety-second fault that fired forty times, each occurrence too small to mention and too frequent to forgive.

Two disciplines keep the data honest over time. First, protect the cause codes from inflation: when a code named "other" starts collecting a fifth of all stop time, the taxonomy has failed and needs one afternoon of revision, not a new software package. Second, never let the log become a surveillance argument. The moment operators believe the stop codes are used to grade people instead of processes, the codes turn diplomatic — everything becomes "waiting material" — and the cell goes back to having opinions instead of records. The log is a maintenance and scheduling tool. Used that way, it stays truthful, and a truthful month of cell data is worth more than a year of line-level OEE printed on a slide.

A practical minimum, then: an event log with cause codes, per-cell OEE components, changeovers logged as first-class events, reconciling counters tied to batches, one clock. None of this needs a MES on day one; a disciplined PLC log exported weekly will carry a plant surprisingly far, and it makes the eventual MES integration a formatting exercise instead of an archaeological dig.

I write these notes from the end-of-line side of the fence — I work on palletizing cells and Cartesian robots, and the company side of that work is in our company profile on Gust. The pattern across plants is always the same: the data to find the real bottleneck already exists inside the cell. Somebody just has to start reading it before the meeting, not after.

Top comments (0)