What AI on a press can actually do

# What AI on a press can actually do The vendor left on Tuesday. The sheet they printed is still pinned above the console, a coated job with the colour bar where the imposition said it would be, and a face that nobody had to rescue. Thursday night the same console offered the minder three

What AI on a press can actually do

The vendor left on Tuesday. The sheet they printed is still pinned above the console, a coated job with the colour bar where the imposition said it would be, and a face that nobody had to rescue. Thursday night the same console offered the minder three things the demo had already named. A preset for the ink keys. A first suggestion for the colour, drawn as small arrows beside the zones. A schedule for the rest of the shift. He pulled a sheet, walked it to the light, and left the arrows alone.

The picture was a cover. Heavy cyan ground, a face sitting in the same zones as the ground. The suggestion wanted more cyan, because the bar said the solid was light. The face was already going the way faces go when you chase a solid. He didn't add the ink. The unit was wet. The cyan was greying in the midtones, and a wet midtone will make a bar look light without the face needing another key. He eased the dampers, pulled again, and the bar came up on its own. The suggestion was still on the screen, waiting to be accepted, as if the last two minutes hadn't happened.

That is the work these systems actually do, and the work they don't. You can have a preset, a first suggestion, and a schedule proposal on a press that is otherwise ordinary. You can pay for the words on the brochure. The sheet is still signed by the person who can refuse the arrows. If that person is told the screen has already decided, Thursday night will print Tuesday's promise on Thursday's plates.

A preset opens keys for a file

An ink preset is older than the current fashion for calling everything on a console intelligent, and it is still the most useful of the three. Prepress sends a coverage file taken from the plates, or from the same RIP that imaged them. The console turns that coverage into a starting opening for each zone. On a decent job, with the right ink and the right stock, you are closer to a printable sheet than you would be from a washed duct and a guess. Makeready gets shorter when the starting point is honest. The last moves are still judgement. The first stretch was mostly arithmetic.

The floor can hand the software the coverage, and nothing more mystical than that. It can hand across the plate job name, the stock code somebody typed, the colour names, the quantity. It can't hand across the ink that is actually in the fountain. It can't hand across a blanket that still carries a low spot from Monday's smash. It can't hand across the news that the warehouse swapped the coated sheet for a cheaper one with the same name on the docket and a different surface. It can't hand across the previous job, a heavy black, still ghosting in the rollers because the wash was short.

Thursday's preset failed for a plainer reason. Prepress had replaced the cyan plate after a type change in the headline. The coverage file on the server was the one from the first send. The keys opened for a picture that was no longer on the cylinder. On the first pull the headline was light, and a box that had been removed was still "inked" in the zones, which dragged the neighbouring zones out with it. The minder looked at the plate, looked at the screen, and killed the preset. He set the cyan from the sheet. The software had been faithful to the file it was given. Faithful was the fault.

This is the check that disappears in a demo. In a demo the file, the plates and the stock were aligned before you walked up. On nights the alignment is a habit, or it is a waste pile. Read the job name on the plate against the job name on the preset. If prepress can't say they are the same send, don't let the keys move. A preset you haven't identified is somebody else's layout loaded onto your press.

Coverage can match the plates and the sheet can still be wrong. The file is a map of area. It isn't a map of a cold duct, of ink that has been standing, or of dampers left where the last stock wanted them. The preset gets you into that argument sooner. It doesn't take a side. Time the makeready through to a signed sheet, or you'll congratulate a fast first pull and ignore the water and ink that still belong to the minder.

The first suggestion is allowed to be wrong

The colour suggestion is the part the brochure lingers on. A spectrophotometer reads the bar. The software compares the reading with a target and proposes a key move. When the bar is clean, the target is the target you agreed, and the instrument is on the measurement condition you use for that stock, the proposal is often a sensible first move. It's a first move. The word suggestion is doing real work. Plants get into trouble when the console treats "accept" as the normal path and "leave it" as a kind of disobedience.

The bar is a thin sample of the sheet. It's usually solids, and sometimes a few tints, sitting in the tail or the grip, printed on whatever substrate the imposition put under it. The picture is the rest of the sheet. Those two come apart in ways a single reading can't settle. A solid and a flesh tone in one zone. A vignette that has to stay open beside a logo that has to be dense. A bar printed across a glue flap that isn't the face the buyer will look at. A bar with powder on it from the sheet above, read as if it were ink. The instrument will give you a number either way. The number reports what the instrument saw. It doesn't report what the buyer will sign.

Targets are the other half of the trap. The suggestion is only as good as the aim stored against the job. A contract proof on proofing paper is a different aim from a drawdown on the actual board. A reprint aim from last year is a different aim again if last year's paper has been discontinued and nobody changed the numbers. On Thursday the stored aim was the proof. The job was a tinted board the buyer had signed as a drawdown, and the drawdown was in the job bag, not in the console. The suggestion kept offering magenta. Against the proof, it was right. Against the signed drawdown, it was taking the sheet away from the thing the customer had already accepted. The minder followed the bag. He had the assistant write on the ticket that the aim was the drawdown. The next shift would otherwise accept the arrows and call it consistency.

You can hand the software a reading, if a device measured this sheet and the bar was fit to measure. You can hand it a stored target. You can't hand it the argument about which object is the contract. That argument is commercial as much as it is technical. Sometimes the proof is what you sold. Sometimes the drawdown is what you sold. Sometimes the buyer is holding a swatch that was never printable in this process, and the honest answer is a phone call before anyone touches a key. A model that picks among those has pretended a commercial argument was a measurement.

One move, then look at the picture, then decide whether the next correction is a key at all. Water, a dirty bar, a wrong aim and a bad preset all look like a key problem from the screen's side of the glass. If the console walks the keys while the crew is at the feeder, the press is chasing a bar nobody is watching. That is a bad way to run a first job on a night shift.

A suggestion drawn from "similar jobs" needs the same test. Similar how: same stock, same inks, same press, same plate curves, or similar because the title looked alike. A stored setting helps when the conditions match. An average of old compromises is how a face inherits the ink from somebody else's solid. If nobody can show you the jobs behind the suggestion, you don't have a memory. You have a blend.

The schedule proposal doesn't know the room

The third offer is the schedule. The planning screen looks at due times, quantities, the makeready allowance in the estimate, and which press is free. It proposes an order. Used as a proposal, this is a gift. A planner with a dozen jobs and two presses will miss a grouping the sort can see: two jobs sharing a special colour, a long run that should start while plates are still being imaged so the press doesn't stand waiting. Used as an order, it's how you wash up three times in a shift for want of a conversation.

The software can see the fields. A due time, a colour count, a sheet size, a press name. It can't see that the finisher who can run this die is off, so printing the job now only builds a pallet in the aisle until Monday. It can't see that goods-in rejected the board and the replacement isn't unwrapped. It can't see that this customer sends a corrected file late, every time, so an early start prints the obsolete version. It can't see that this press misfeeds a lightweight sheet after a heavy aqueous job, and that the minder wanted the heavy work on the machine serviced last week.

Thursday's proposal had a logic you could defend in a meeting. It sorted by promise time. It put a light text form on the black unit straight after a cover with a full-bleed black. The ghost was built into the sequence. The planner had gone home. The minder could see it, because minders have been caught by this before, and he swapped the order with a job that could live with the black already in the rollers. He wrote the swap on the printed list. In the morning the planner's screen still showed the original proposal. The jobs were done in the other order. For a little while the argument was about the floor ignoring the system. The floor had used the system. The proposal was a proposal. The room held information the fields didn't.

If you want the schedule to get better, the correction has to be allowed to travel back. A swap that lives on a pencil list and dies in the morning is a correction you'll pay for again next week. A swap that updates the job, with a reason a planner can read, becomes the next proposal's starting point. That's a discipline about data. It's also a permission. Night shifts that get told off for marking the plan will stop marking it. They'll run the ghost, and the ghost will be blamed on the press.

Tuesday will not turn up at two in the morning

Tuesday was arranged. The stock was the one in the trial. Prepress had checked the colour bar before the plates were made. The specialist knew which screen was a proposal and which screen had already moved a key. There was time to stop and talk, and nobody was waiting on a lorry. The job had been chosen because it would behave. That's a fair way to show a feature. It's a poor picture of a Thursday, when the specialist has gone, the next job is late, and a refused suggestion feels like delay. The preset loads as the plates clamp. The aim in the console and the aim in the bag can disagree. The schedule can contain a ghost. A crew that has never practised saying no will follow the brightest instruction in the room.

Rehearse that refusal in the morning, with sheets on the table. One suggestion the night accepted, one they refused. Did the coverage file match the plates. Which object was the aim. Did a marked schedule get back into the plan. If you only look at the arrows that were right, you train the crew to accept the next ones. The misses are the training. So is a signed sheet that ignored a bad arrow and still matched what the buyer had approved.

The floor can pass across a coverage file that matches the plates, a reading from a bar somebody has looked at, and the quantities and due times a person typed and another person can challenge. It can pass a counter you've checked against a pile. It can't pass the decision to sign, a sample ageing in a drawer, a roller that's due, or the sound of a feeder about to double. A preset tied to the right file still saves makeready you'll feel by the end of the week. A suggestion you're free to refuse will catch a zone you'd miss when you're tired. They earn their keep on those terms, and they stop when "the system said so" replaces the signature.

You can buy the preset, the suggestion and the proposal. You can leave Tuesday's sheet above the console. You can't buy the conditions of the demo and expect them at two in the morning. The minder owns the sheet. The software owns a file, a reading and a sort. Leave those where they are.

Written by

Priya Shah

Prepress

Makes plates and proofs, and sends the file back when the press cannot fix it.

Related articles

Process

Flexo still fits packaging

# Flexo still fits packaging The buyer had been to a presentation, and he arrived wanting the range moved. Film, board, a family of labels, all of it, onto a di...

Process

Print on demand moves the stock and the cost

# Print on demand moves the stock and the cost The publisher showed me an empty bay and he was proud of it. A year ago the bay had held a title that sold steadi...

Process

Versioned packaging is a production problem

# Versioned packaging is a production problem Marketing called it a celebration. A commemorative sleeve, one brand, a dozen pictures, a few lines of type that c...

Process

Print that can be scanned, used and recovered

# Print that can be scanned, used and recovered The code scanned on the flat sheet. I watched the buyer's phone pick it up from the press table, quick and pleas...