Tuesday morning, a three meter banner comes off the take-up and the twelve point logo on it looks printed twice, two edges maybe half a millimeter apart. The nozzle check is clean. The operator has already run two power cleans and is on the phone ordering a printhead. Stop him. Ghosting is a placement error, not an ink delivery error, and a clean nozzle check is exactly what you expect to see when the head itself is fine. Cleaning will not remove it, no matter how many cycles you stack up, and every cycle pushes ink into the waste bottle.
Before you open anything, put the spoiled piece under a loupe or hold your phone camera on it at 2x. Look at one vertical line and one hard horizontal edge. The direction the second copy sits in tells you which subsystem to open, and reading it costs nothing. Two minutes with the eyeball beats a week of parts ordering.
Read the offset direction before anything else
The second copy sits in one of two directions, and they lead to different parts of the machine. Find a vertical rule in the file. If the two edges are separated side to side, along the path the carriage travels, you have left to right ghosting. If the second edge sits above or below the first, along the media feed direction, you have front to back ghosting. Those two look almost identical on a finished job and they are completely different faults.
Print your own test strip instead of pulling a customer file. A strip with a few hairline vertical rules, one solid black bar and one solid color block gives you both readings in a single pass. The rules show left to right error, because any offset along the scan path splits a thin line into a pair. The hard edge of the solid block shows front to back error, because a feed error leaves a narrow light seam or a dark overlap right at the boundary between passes.
Then note whether the ghost lands in the same place every time. Print the same strip three times and hold all three against a window. If the doubled edge appears at the same position on the page each time, in the same band across the width, suspect position feedback or something loose in the carriage rather than a calibration number. If it moves around, or only shows up at one speed, the alignment value is the likelier suspect.
One more reading that saves an hour: check whether the ghost survives a switch to unidirectional. Run twenty centimeters of the same file in one direction only. Clean output in unidirectional is the strongest single sign that the head, the dampers and the ink supply are all healthy, and that whatever is left to fix lives in the bidirectional timing or in the carriage motion. Write that result on the work order before you open a single panel.
Left to right ghosting starts with the stored value
The first move is a bidirectional alignment run, and the first mistake is running it with the wrong settings. The stored correction value is tied to print speed and resolution. If you calibrate at 6 pass and then produce at 4 pass, the number you just stored is answering a question nobody asked. Set the RIP to the mode you actually sell, then run the pattern.
Reading the pattern is where most people go wrong. It prints several groups of lines, each group at a slightly different offset, and only one group forms a single solid bar with no step in it. Take the number under that group and enter it. Do not round up, do not split the difference between two groups that both look close, and never type a large number just to see what happens. Entering ten when the answer is two will move the error instead of removing it, and then you spend the afternoon calibrating against your own mistake.
Print the pattern again after entering the value. That second run is not optional. A correct entry shows as one clean bar across the full width. If the second run still shows a step, look at where the step sits. A step near one end only, with the rest of the bar clean, usually means the carriage is not square to the platen or the platen is not level, and no amount of re-entering numbers will fix geometry.
Two things change the required value with no part failing. The first is speed: any move to a different pass count or resolution moves the number. The second is ink and media. A different ink set, or a substrate with a much rougher surface, shifts where the drops land on the return pass even at the same nominal settings. Shops that bounce between banner and backlit film all week should keep two stored values and a note taped to the frame saying which one is loaded.
If a fresh calibration holds at 8 pass but falls apart at 4 pass, you own a machine that can no longer be trusted in its fast mode. That is information, not a failure. Either quote the job at the slower mode or book the repair, but stop running calibration in the hope that 4 pass comes back on its own.
Front to back ghosting is a feed problem
When the second edge sits above or below the first, the head put its drops in the right place and the media arrived at the wrong time. That is media advance, not printhead alignment. Running bidirectional calibration on this fault burns an hour and changes nothing.
Media advance calibration is substrate specific, and this is the part most shops skip. A machine dialed in on heavy PVC banner will show advance error on a lighter matte polypropylene, because the rollers grip differently and the material stretches differently. Change the stock, run the advance adjustment again, and write down which stock each stored profile belongs to.
Check the mechanical side at the same time, since it is free. Make sure every pinch roller is down, not just the ones under the middle of the roll. Wipe the rollers with a cloth dampened with isopropyl alcohol and let them dry fully. A glazed roller slips in a rhythm, and that rhythm shows up as a repeating seam. Load the roll square and confirm both edges carry equal tension before you lock the flanges.
Heat is the other common cause and it is easy to miss. Media being heated stretches as it runs. If the heater is still climbing when the job starts, the first meter prints at one advance and the rest at another. Let the platen reach its set temperature and sit for a few minutes before sending the file. On a cold morning this is often the entire story.
Finally, rule out the data path. If the ghosting starts suddenly mid job, jumps position, or arrives together with a stutter in the RIP, unplug and reseat the data cable at both ends, close the RIP, power the printer down and back up, then reopen the RIP and resend. Interrupted data produces doubled and misplaced bands that look mechanical and are not. Do this before you loosen a single screw.
Ghosting that began after a media or head height change
If the doubling started the same week somebody raised the head to stop a head strike, you already have your answer. Change the printhead height and the bidirectional relationship changes with it. HP states this plainly in the Designjet 9000s service documentation: once the head height is changed, a gap can appear in bidirectional printing, and the printhead position adjustment has to be run again. That is the manufacturer telling you the two settings are coupled, not a theory from a forum.
The mechanism is simple. Raising the head lengthens the flight time of every drop, so any air movement in the gap pushes the drop further off course before it lands. The error shows up along the scan direction, which reads as left to right ghosting and fuzzy edges on fine text. Roland's manual for the XG-640 makes the same point from the other side: keep the lever at Low, and only go to Medium or High when media is lifting or wrinkling, because print quality gets coarser as the gap grows.
So the order matters. If somebody raised the head for a thick substrate, put the lever back to its normal position once that job ships, then re-run the position adjustment. Leaving a machine in High for everyday banner work is a self inflicted ghost, and it is one of the few faults on this list that costs nothing at all to fix.
The same logic covers the carriage beam. On machines that let you set a custom beam position for thick stock, the working figure is substrate thickness plus two or three millimeters, not more. Extra clearance buys nothing and costs placement accuracy. Write the current lever position and the date of the last alignment on a strip of tape on the frame. The next operator will thank you, and you stop wondering why a machine that printed clean in March is doubling in June.
Work the list from cheapest to most expensive
Here is the order to work in. Each step costs less than the one after it, and each step either fixes the fault or removes a suspect for good.
- Print twenty centimeters in unidirectional. If it is clean, the head and the ink path are healthy and the fault is timing or motion.
- Re-run bidirectional alignment in the mode you actually produce in. Read the pattern properly and enter the number under the solid bar.
- If the ghost runs front to back, run media advance calibration on the stock that is in the machine right now.
- Reseat the data cable at both ends, restart the RIP, power cycle the printer, resend. Free, and it clears the data path.
- Wipe the encoder strip with a lint free cloth and a little isopropyl alcohol. One direction, light pressure, no scrubbing, and never bend the strip.
- Clean the carriage rail and lubricate it as the manual specifies. Check belt tension with a gauge or against the spec in the book, not by feel alone.
- Only now look at head height, carriage squareness and the X axis travel coefficient. On most machines those last three need a service visit.
One detail on the encoder strip, because it is the most over blamed part on any wide format machine. A dirty strip usually shows as ragged edges and misplaced verticals that repeat at the same spot in the print width, not as a clean offset copy running in one direction. If the ghost is a crisp second edge and the strip looks clean when you hold it against a light, put the cloth away and go back to the alignment value.
Know when to stop. Three alignment runs that will not converge, a ghost that sits in the same band across the width no matter what you change, or an offset that grows as the carriage travels further from home, all point at mechanical wear or a failing feedback signal. At that point more cleaning only adds waste ink and shortens the head. Book the service, and in the meantime run the job unidirectional at a slower mode so the work still ships.
None of this should cost a printhead. That is the point of reading the direction first. Most doubled prints that arrive at a parts counter as an order are really a stored number that no longer matches the mode being run, or a lever somebody moved and never calibrated back.