Marking and reading parts in production: where identification actually breaks

The mark has to survive the process that follows it

Choose the marking method against every downstream operation, not against the first read. Printed labels are cheap and versatile until they meet solvent, heat, abrasion or a wash stage. Direct marking by laser, dot peen or electrochemical etch survives far more but requires surface access, adds a station and may be prohibited on stressed or coated areas. Moulded-in features survive everything and cannot carry a unique serial. The commonest error is applying an identity early for traceability and losing it at paint, plating or heat treatment, leaving a gap in exactly the section of the route where knowing the identity would have mattered most.

Two-dimensional codes, linear codes and the read geometry

Linear codes need a clean, flat, reasonably large area and a defined scan direction. Matrix codes carry more in less space, tolerate partial damage through built-in error correction and can be read at any rotation, which makes them the usual choice for direct-marked components. Both depend on contrast, and contrast on bare metal comes from surface finish rather than ink, so lighting matters as much as marking quality. Reading rates that look acceptable in a trial become a serious nuisance at production volume, because a small failure fraction on a fast line is a steady stream of parts an operator must handle by hand.

Radio identification and the physics that defeats it

Tags read without line of sight, through packaging, and in bulk, which is exactly why they get proposed. Metal detunes and reflects, liquid absorbs, and a tag lying flat on a steel fixture may not respond at all unless it is a type designed with a spacer for that duty. Bulk reading brings the opposite problem: the reader picks up the tray behind, the operator's badge, or the pallet in the next aisle, and you now have to bound the read zone physically. Tags also cost materially more per item than a printed label, so the case usually depends on reuse or on high-value assemblies rather than on unit traceability.

Reading is a station, with all that implies

An identification point needs presentation, illumination, mounting, cleaning and a defined outcome when the read fails. Treating it as a sensor bolted to a bracket produces intermittent behaviour nobody can reproduce. The no-read path deserves as much design as the read path: does the line stop, divert, or prompt an operator for manual entry? Manual entry is the pragmatic answer and the one that undermines the data, because a keyed number is a guess when the label is destroyed. If manual entry is allowed, record that it was manual so later analysis can tell a scanned identity from an asserted one.

An identity is only useful if the schemes agree

Plants accumulate several identification schemes: the supplier's code on incoming material, an internal work-order number, a customer part reference and a serial applied at final assembly. Each system in the plant tends to hold one of them, and the join between them lives in somebody's spreadsheet. Decide early which identifier is the primary key through production, where transitions between schemes occur, and how the relationship is recorded at that moment. Recall exercises and warranty investigations are where the gaps surface, and they surface under time pressure, which is precisely when reconstructing a broken chain of identity is least achievable.

Frequently asked questions

When is a radio tag worth the extra cost over a printed code?

When you need reading without line of sight, on items that pass through conditions destroying a printed mark, or on returnable carriers where the tag is reused many times. Reusable trays, tooling, fixtures and work-holding pallets are the strongest cases because the cost is amortised across thousands of cycles. For unit-level identification on a low-value part read once at a defined station, an optical code is usually more dependable and considerably less expensive to deploy and support.

How should we handle a part whose code cannot be read?

Decide the rule before it happens, and prefer diversion to a defined station over an operator keying a number from memory. If manual entry is permitted, require the source of the identity to be recorded and flag the record as manually asserted so later analysis can weight it appropriately. Then treat every no-read as a defect worth investigating, because a rising no-read rate usually indicates marking quality drifting or a reader that needs cleaning or realignment.

Should identity be applied at the start of production or at the end?

Apply it as early as the process allows if you need traceability through manufacture, and accept that you may need to re-mark after any operation that destroys the mark. Applying only at the end gives clean serialisation for the customer and no visibility of what happened during the build, which is exactly the information a failure investigation needs. Where a route includes a destructive stage, plan the carry-over deliberately: a carrier that holds the identity, or a re-mark station immediately afterwards.