Commissioning is where every decision a project has put off falls due at the same time. Everything you waved away in the workshop with “we’ll sort it out on site” is waiting for you on site. Over the years we have listed the causes of our delays; the surprise is how short the list is, and how often it repeats.
Anatomy of a delay
01 / 04
Plan
The panels are ready, the program is written, the robot has arrived. On paper, everything is in place.
02 / 04
Surprises on site
Everything waved away in the workshop with “we’ll sort it out on site” falls due on site, all at once: I/O, safety, network, tolerances, training, acceptance.
03 / 04
Tested in the workshop
I/O, the safety circuit and the network plan are tested at FAT before anything ships; on site they are only repeated.
04 / 04
Measurable acceptance
A measuring day for tolerances, training with real product, acceptance on measurable items. Fewer questions travel to site.
1. The I/O list and the panel don’t match
The signal list is the latest revision; the panel was wired to the one before. Every hour on site spent hunting for the terminal a sensor lands on is the dullest part of any delay. Our countermeasure: the I/O list lives in one place, the panel and the program are both generated from that same list, and before the panel ships, every input and output is tested one by one in the workshop with a simulator (FAT). On site, I/O testing should take a day; if it takes three, something was left undone in the workshop.
2. The safety circuit was designed “later”
The emergency stop chain, door interlocks, light curtains: if it is not defined from the start which zone each one stops and which reset brings it back, the safety circuit gets reworked on site with every machine change. We make the safety PLC logic and its validation test part of the FAT; on site we only repeat them with the real doors.
3. The network: IP conflicts, no VLANs, someone’s laptop handing out DHCP
Arriving on site to discover that the robot, the camera and the PLC can’t see each other is a classic. The network plan (IP table, switch ports, OT/IT separation) is part of the drawings, not a note added after installation. And on site, nobody plugs their own laptop into the plant network as a DHCP server.
4. Mechanical tolerance is not the same as automation tolerance
The conveyor is two millimetres high, the jig is rotated by a degree: within tolerance for the mechanical fitter, a missed target for the robot. On every robot installation, day one is measuring day: reference points, jig heights, how the product sits. Projects that add a 3D camera or a reference sensor with this in mind meet far fewer surprises on site.
5. Operator training left until the last day
The system works, but the operator doesn’t know how to change a recipe, so service gets called out on the first shift. We run training in the middle of commissioning, with real product and real faults: “now open the door and see what happens”. The training document should be the HMI itself; nobody opens the binder.
6. No written acceptance criteria
“The system works” is not an acceptance criterion. How many hours without interruption, at what speed, with what error rate? Projects that don’t write this into the contract end not at the close of commissioning, but in an endless loop of “just a bit more”. We write the acceptance test (SAT) up front, with measurable items, and tick off that list together on site.
What shortens commissioning is not working faster; it is taking fewer questions to site.