Commissioning

Problems are coming. That is not failure

Let's set expectations honestly: commissioning problems are 1000% going to happen. There will be things you didn't consider — a product variant nobody mentioned, a draught that trips a sensor, a pallet that arrives wrapped differently on Tuesdays. This is true of every cell ever built, including the ones commissioned by integrators at ten times your budget. The difference between a good commissioning and a miserable one is not the absence of problems; it is how you work through them.

The working rule: solve the problem as quickly as possible by any reasonable means, get the cell running again, then refine the fix and make it excellent. The cell needs to be running to find the next problem — a stopped cell teaches you nothing. Don't stop. Keep working the list until each problem is solved, then go back and turn the quick fixes into proper ones. The beauty of having built this in-house is that every problem is solvable by the people standing next to it: it is a matter of when, not if, and you are not waiting three weeks for an integrator's engineer to fly back.

Day one: safety again, before anything moves

The cell was moved, re-wired at the joints, and reassembled. That means the safety validation from Part 8 is repeated on site — every e-stop, every interlock, every light curtain, physically proven and recorded again. Transport and reinstallation are exactly where a safety circuit picks up a damaged core or a swapped terminal. Nothing else in this part happens until this is done, and the site validation record joins the Technical File alongside the FAT record.

Bring it up in the same layers you tested it

Power-up and supply checks first — rotation direction on three-phase motors, 24V rails healthy, controllers booting clean. Then confirm the robot's relationship to the world: check mastering survived transport, and re-verify the base frame against the real pallet position — the taught reference from Part 7 was taught somewhere else, and the whole point of the base-frame approach is that correcting it here means re-teaching one point, not forty. Then I/O spot-checks at the boundaries that were disconnected for the move, then slow full cycles with real product in real conditions, then speed raised in stages exactly as in Part 7. Resist the pressure to jump straight to full speed because the FAT passed — the FAT proved the cell; site conditions are what's on trial now.

Run the snag list like a discipline, not a memory

Every problem goes on a written list the moment it appears: what happened, when, what was observed, the quick fix applied, the permanent fix needed. Severity-order it daily. The list does three jobs — it stops 2am fixes being forgotten by 8am, it reveals patterns (three "random" stops that are all the same marginal sensor), and it is the honest record of what the cell still owes you. Commissioning is finished when the list is empty or every remaining item is genuinely cosmetic — not when everyone is tired of it.

The problems will cluster in familiar places

• Product reality — the actual product varies more than the samples you tested with. Dimensions, weight, surface finish, how it settles on a conveyor. Expect grip tuning and search-window adjustments
• Marginal sensing — a photocell that worked on the test floor at 90% margin fails at 60% margin under production lighting, dust and vibration. Reposition, shield, or re-select; don't just extend timers
• Timing at the boundaries — the upstream machine feeds faster or lumpier than agreed; the downstream one blocks. This is where the simple, dumb handshakes from Part 5 pay you back, because they can be reasoned about while standing at the panel
• Recovery paths — operators will stop the cell in states you never imagined and expect it to recover. Each new stuck state becomes a defined recovery in the program or procedure
• Speed-dependent behaviour — the last 20% of override reveals vacuum, inertia and settle-time issues the slow runs hid. Tune the named speed variables from Part 7, one change at a time

Involve the operators — they finish the design

You designed the cell around operators in Part 1; commissioning is where they take it over. Have them running it with you beside the cell, not you running it with them watching. Their observations in the first week — where product actually jams, which fault message is confusing, what they reach for and can't — are the final design inputs, and folding them in now is what turns "the robot" into "our cell". A team that helped commission a machine looks after it. A team that had a machine imposed on it phones you at 2am forever.

Declare the end, then hold a review

Set completion criteria just as you set FAT criteria: target rate sustained over defined runs, availability over a week, snag list clear, operators signed off, documentation updated to as-built — because the drawings, I/O schedule and program you finish with are not quite the ones you installed with, and Part 10 needs the truthful versions. Then hold a short, blame-free review: what would we do differently? Write it down. The second cell is where in-house automation gets dramatically cheaper, and it is built on that list.

Common mistakes

• Treating problems as failure instead of the job. The plan was always that there would be surprises; the plan is how you respond
• Perfect-fix paralysis — engineering the ideal solution to problem one while the cell stands still and problems two to ten stay hidden
• Quick fixes that never get refined. Fix fast, then make it excellent — both halves are the method
• No written snag list. Memory does not survive a commissioning week
• Bridging safety devices under pressure. No
• Skipping the as-built documentation update, leaving Part 10 to certify a machine that no longer matches its drawings

Next in the series

CE Marking — coming soon

→