Conveyor barcode scanner: why a moving DataMatrix is a different problem

Stop the belt and the code reads first time. Run it at speed and it fails on some items, and the failure looks random until you measure it. What changed is that the object moved while the sensor integrated, the trigger fired at the wrong distance because belt speed differs from last week, and a read taken at station A has to be matched to a paddle at station B. Exposure, timing and identity — those three are the discipline of conveyor code reading.

Motion blur is an arithmetic constraint, not a quality complaint

During the exposure the part travels, and that distance mapped through magnification is the length of your blur kernel. The decode survives while the kernel stays well inside one module of the symbol: a decoder reads transitions between light and dark, blur widens and shifts each transition, and once the kernel approaches the module width adjacent modules merge and the grid is gone.

So the design inequality is exposure time short against module size divided by object speed. Shortening exposure is easy; the cost is light, because what you must resolve is contrast between code and background.

That is why this class uses a global-shutter sensor. A global shutter integrates every pixel in the same window, so the only artefact is a uniform blur you can trade against illumination and aperture. A rolling shutter exposes rows at staggered times, so a moving object is sheared — the module grid stretched and skewed by an amount that varies across the frame, which attacks the finder and alignment step rather than merely softening edges. Strobing, a short LED pulse synchronised to the exposure, buys shutter speed without heat, which is why these readers pulse their light rather than run it.

The hardest corner: tiny DPM, short working distance, high line speed

Direct part marking has no ink. A DataMatrix on a machined part is geometry — etched, ablated, stamped or embossed — so its contrast is a fraction of a printed label's and its modules are far smaller than a printer reliably makes on paper. Reading it needs magnification, and magnification is the expensive part of the budget: it shrinks depth of field, shrinks the field of view, and lens extension cuts the light reaching the sensor. Short working distance removes the room you would have used to bring illumination in at a useful angle, and turns belt vibration into image motion at exactly the magnification where that matters. Add line speed and the demands collide: shorter exposure, more light, tighter mechanics.

Triggering: where the read happens is not where you detected the part

A photo-eye tells you an item arrived; the exposure fires later, after sensor wake, evaluation latency and a controller scan cycle, and in that interval the item travels speed × delay. Put the eye upstream by exactly that distance and the code lands centred in the field; get it wrong and every read happens at the edge of the lens where the optics are weakest.

At constant speed a fixed delay is a fixed distance and this works. It fails the moment the line accumulates, slips or runs a recipe-dependent speed, and it fails quietly — the read rate degrades instead of stopping. The fix is encoder-follow: a shaft encoder or belt tracker supplies distance rather than time, so exposures fire at defined positions along the belt regardless of speed.

Then the worst field bug. The read happens at the reader and the reject at a paddle downstream; map one to the other by elapsed time and any speed change between the two points shifts the map, usually by one item, so the line rejects good parts and ships the one whose code was missing. Carry the read to the reject with a slot-based shift register in the controller, with the item's decoded identity, or with distance from the same encoder that fired the exposure — never with a wall clock alone.

Illumination geometry is a contrast decision

The mark has to be brighter or darker than its background in the image, and on metal that is mostly a question of where light goes after it hits the surface. Pick the geometry for the surface, not for the code.

Wavelength belongs in the same decision: red and infrared sit differently against inks and plastics than white light does.

Symbology is a process decision, not a scanner setting

For DPM, DataMatrix ECC200 is the working standard: square, no quiet zone demanding surrounding space, and Reed-Solomon correction that survives a mark partially destroyed by a later machining or cleaning step. For case labels the decision is not yours. GS1-128 remains the language a dock, a partner and a warehouse all read, because it is wide, tolerant of long working distance and readable by installed linear equipment; a 2D component buys batch, serial and expiry in less space but must be readable by every downstream party. Your own leverage is X-dimension, print darkness, label position and quiet zone — where most "reader cannot cope" investigations should start.

What an ISO/IEC 15415 grade does and does not prove

ISO/IEC 15415 grades a 2D symbol by components — decode, symbol contrast, module quality, non-uniformity and print growth — reported as a letter and an overall level, with instrumentation covered by the 15426 series and linear symbols by ISO/IEC 15416. It is a real, comparable measurement, taken on a static symbol at one position under the verifier's own illumination and geometry.

So a grade cannot tell you that the code will read at your line speed, that your incidence angle collects enough return, that labels are applied off-square so a share of cases present the code at the field edge, or that bracket vibration is smearing the image. Use 15415 as a statistical monitor of the marking process; prove readiness separately, with the actual reader, illumination and belt speed over several hundred consecutive parts.

The I/O and protocol layer is what actually integrates the reader

multimeter — but the code's content must leave by another path, unless the application genuinely is pass/fail.

string becomes a structured tag, reader health becomes machine condition data, and timing follows the bus cycle.

means ends up somewhere nobody documented.

• Digital lines. Opto-isolated inputs and outputs: present, read, no-read, strobe to the lamp. Deterministic, no stack, debuggable with a
• Industrial Ethernet and fieldbus — PROFINET with its device description, EtherNet/IP with an EDS, EtherCAT, CC-Link IE. The decoded
• Plain TCP/UDP or serial-over-IP. Easy into a historian or MES, no PLC coupling; not deterministic, and the parsing of what the string

Ask two questions before choosing. Where is symbology parsed — reader or controller? And how does the controller distinguish no symbol presented from symbol present but unread, since collapsing them makes every label-supply problem invisible.

Reject and recover, so a no-read never stops the line

Cover the label zone first: more than one optical axis, or a second reader at a different geometry and height, which turns "the code was on the hidden face" into a normal event. Then a re-read station downstream at a different angle or a deliberately slower speed. Then the decision: divert to a review lane rather than stopping, so a person clears exceptions at the end of the run.

Put the stop threshold — a count of consecutive no-reads — in the controller rather than the reader's firmware, so the people who own the line can see it. Make signalling fail-safe: an absent response is indistinguishable from a no-read, so the actuator must default to reject on de-energised. Then classify every no-read by cause (label absent, unreadable, wrong position, damaged, trigger missed) and trend it per shift.

In regulated production the deliverable is the record

Pharmaceutical packaging and medical-device UDI invert the objective. Under the FDA's UDI programme (21 CFR Part 830 and associated parts) and EU MDR, the device identifier plus production identifiers — serial, lot, expiry — travel in a GS1 DataMatrix or HIBCC equivalent, and each read is the event that inserts an identity into an aggregation hierarchy: unit into carton, carton into case, case into shipper, shipper into pallet.

So the reader must retain evidence — decoded data plus image, keyed to the item — so a disputed code can be re-examined years later. Electronic records and signatures sit under the audit-trail and access expectations of regimes such as 21 CFR Part 11, which means an operator override must be attributable and a silent bypass of a reject physically impossible. A line that reads everything but cannot prove which item received which identity has failed the requirement that was contracted.

Specifying it

Send a fixture request, not a datasheet request: your part or printed case, your mark, at your speed, on a sample configured for that geometry. EDOO lists this class as ED3000 for fast-moving Data Matrix, ED5000 as an IP67 industrial fixed-mount reader for fast-moving codes on a conveyor belt, and ED-SC2000-04AM-02WBN as a smart navigation camera and code reader for AGV use, with OEM from 100 units, 15-day samples and 30-day mass production — enough calendar for a capability trial at line speed before the decision becomes a capital one.

FAQ

Why does my conveyor barcode scanner read codes fine when the belt is stopped but miss them at speed? Because a moving code travels during the exposure, and the read fails once that displacement in object space approaches the width of one symbol module, so a stopped-belt test proves nothing about performance at line speed.

Is a global-shutter sensor necessary for a moving DataMatrix? For short exposures on fast lines yes, because a global shutter integrates all pixels in one window so the only artefact is a uniform blur you can trade against light, while a rolling shutter shears the moving symbol by an amount that varies across the frame and damages the grid alignment the decoder needs.

What illumination do laser-etched codes on metal need? Usually low-angle dark-field lighting, which makes a polished surface scatter out of the field so it reads black while the roughened mark scatters back and reads bright, with coaxial optics for flat mirror-like surfaces and a cross-polariser for glossy labels.

Can I rely on an ISO/IEC 15415 grade to predict reads on my line? No — a grade is measured on a static symbol under the verifier's own illumination and geometry, so it monitors the marking process well but cannot see belt speed, incidence angle, vibration or label position; commission with the real reader at real speed.

Related EDOO resources

Industrialbarcodescanner · Fixed mount barcode scanners · IP65 product page · IP67 product page · Barcode scan engines · SolutionsForYourProjects · technical-recognition · Product documentation and SDKs · ODM / OEM services · Contact EDOO