Traceability Barcode Selection 2026 | QR, Direct Marking and RFID

How do you put an ID onto the physical part? When choosing a traceability barcode, many factories settle it on “it reads well” or “it is cheap”. Six months later the labels are peeling, the print has bled under oil, and when a customer auditor asks how code quality is controlled, nobody has an answer. What decides whether an identification method is right is not readability. It is the physical condition of the part, the number of read points, the granularity, and the radio regulations of the country you are installing in. This article sets out how to choose between four methods — printed labels, direct part marking, 2D codes and RFID — against those four conditions. Further down, the four methods are recast as the four combinations you will actually be choosing between.

A traceability barcode is not chosen on “can we read it”

The selection usually starts the same way. A customer asks you to guarantee traceability at lot level. Or in-process defect traceability does not exist internally, so every time a defect appears the line stops while somebody chases the part backwards through the process. Someone says “let’s put barcodes on it”, and quotations for label printers and handheld scanners land on the table.

At that point three things typically make it onto the table as decision criteria: read accuracy, equipment price, and “can our shop floor actually operate it”. All three matter. But a selection made on those three alone tends to come apart in roughly four ways.

It peels off after six months. A label went onto a part that passes through a heat treatment furnace, or through a process where cutting oil is sprayed. The adhesive gives up, the label lifts, curls from the edge, and eventually drops. The label that dropped gets picked up somewhere downstream and ends up lying next to a completely different part.

It stops reading after a year. A label went onto product stored in an outdoor yard. UV bleaches the print, contrast falls, the scanner stops decoding. The digits are still faintly visible to the eye, so operators start typing them in by hand to get around it. The moment manual entry begins, that traceability turns into data that is recorded but cannot be trusted.

The customer audit stalls. “How do you control the quality of your direct marks?” you are asked, and you answer “they read fine, so there is no problem”. The auditor is not satisfied. Whether it reads is an outcome, not an explanation of how the process is controlled.

The equipment moves and cannot legally transmit. An RFID gate reader that had been running in the parent plant in Japan is moved to a new plant in Thailand. Under local rules, that power class needs a licence. Or the allocated band is simply different. Discovering this on the day of installation costs you the whole schedule.

None of the four is a read accuracy problem, and none is an equipment price problem. The physical environment the part lives in, how many times it is read in the process, how finely you need to identify it, and what is permitted in that country — those four are what actually decide the selection. Readability only becomes meaningful among the options that survive those four.

One note on the operating environment in Thailand, since it changes how often this question arrives. The Thailand Board of Investment made supply chain strengthening — promoting linkages between foreign investors and Thai parts makers — one of the pillars of its 2026 policy direction, and on 23 June 2026 launched a new scheme, the Thailand Fast Pass, which shortens permit and licence lead times for advanced manufacturing investment projects worth over THB 700 billion in total (Thailand-specific). More plants and more trading partners means one thing for the people who own identification: the number of times a different customer hands you a different marking specification also goes up.

The four conditions that decide the method

Of the four failure modes above, “it peels off after six months” and “it stops reading after a year” both come back to the physical condition of the part, and “the equipment moves and cannot legally transmit” comes back to regulation. (“The customer audit stalls” is a question of standards and pass thresholds, handled separately later.) Add the two things that drive cost and daily operation — read points and granularity — and you have the four conditions.

The order matters too. If you do not work down them from the top, what you decided under a lower condition gets overturned by a higher one.

Condition 1 | Physical condition of the part

Start by looking at the thing you are putting the ID onto. Projects that skip this and pick a method first almost always end up re-marking.

The item most often overlooked on the shop floor is life. “It is in-process traceability, so it only needs to read until the part leaves the factory” and “we are heading towards unit-level management in the field, so it must read for ten years after shipment” lead to different methods. In the first case a label is often enough. In the second, labels drop out of the candidate list immediately.

For processes where washing and water exposure are the norm, the constraints as they appear in food plants are set out in traceability design for wet processes. The constraints surface differently from metalworking, so it is worth reading alongside this.

Condition 2 | Designing the read points

Next, mark the places where reading happens on the process flow. What gets decided here is the number of points and who does the reading.

If there is only one read point in the process, a person with a handheld is enough. There is almost no additional capital spend. Once read points grow to five or ten, the picture changes. Keep a person reading and those seconds turn into a monthly labour cost. Go automatic instead and you install a reader at every point, so initial investment scales with the number of points.

The other thing to look at is the posture of reading. If you design a process where both hands are already full of parts or tools and then tell the operator to pick up a handheld and scan, at some point they will stop scanning. And if you have not defined the recovery procedure for a failed read — is manual entry allowed, and if so whose approval does it need — that is where operation starts to unravel. The operational design of the reading device itself is covered in detail in handheld terminals and picking operation design, so this article concentrates on the side being read.

Condition 3 | Granularity

Lot level, box or pallet level, or unit serial level. How that decision is made is not the subject of this article. Working granularity backwards from customer requirements is set out in meeting IATF 16949 for automotive parts, and the design of the linking keys in how to assemble a quality data management system.

What this article takes as its input is only the conclusion of that work. Granularity is decided; now how do you get it onto the physical part?

As granularity rises, the code the part has to carry gets longer. For lot number tracking, part number plus lot number is enough. Once you move into serial number management, the fields multiply — part number, lot, production date, serial. Keep using 1D barcodes at that point and the part tag ends up with several barcodes lined up on it. That lining up is what bites later.

Condition 4 | Regulation and borders

And then the fourth, which most selection material never mentions. RFID alone has permitted frequencies, power levels and licensing requirements that differ by country. Neither barcodes nor direct marks carry this constraint. A code printed on paper does not become illegal at a border.

This difference should be treated not as one method being better than another, but as a difference in preconditions. The more sites a company has across ASEAN, the more heavily it weighs. A full section is given to it further down.

Comparing the four methods side by side

It is worth being precise about something first. The four things people talk about in practice — printed labels, direct marking, 2D codes and RFID — do not sit on the same axis as stated. Printed labels and direct marking are about how the ID is applied. 1D versus 2D is about what kind of symbology is used. You can print a 2D code on a label, and you can mark a 2D code directly onto a part.

So we recast them as the four combinations you will actually be choosing between.

What this table is meant to show is that no single one of them wins. Real factories end up combining them. Direct marks on work in progress inside the process, label plus 2D on the outer carton, RFID on the pallet — that combination is not unusual.

With that said, the typical narrowing decisions look like this.

• If the part passes washing, heat treatment or cutting oil, or has to stay readable for years after shipment, direct part marking is the first candidate. Labels will not survive
• If the part stays inside the process, there is plenty of surface area, and granularity is at lot level, label + 1D or label + 2D is enough. Your existing label printer will do
• If granularity rises to serial and you want three or more fields on the part tag, which in 1D means three barcodes side by side, fold it into 2D. That is the subject of the next section
• If there are many read points and the manual work at each point is long — opening boxes, counting contents — RFID is worth evaluating. But Condition 4 (regulation) and the economics need checking first
• If the part is metal and small, constraints appear on both direct marking and RFID. Measure the available marking area and the tag dimensions before anything else

Getting granularity onto the part — GS1 Sunrise 2027 and the move to 2D

How many barcodes are on your part tag

Let us make the “lining up” problem from Condition 3 concrete. As granularity rises, the part tag tends to become this.

• A barcode for the part number
• A barcode for the lot number
• A barcode for the production date
• A barcode for the serial number

Four barcodes make the part tag physically bigger. On small parts there is no longer a surface to stick it to, so somebody says “let’s bag the parts and stick it on the bag”. The link to the individual unit loosens, and the moment the bag is opened, the basis for unit-level management disappears. Operators also have to remember which of the four to scan at which process step, and misreads start happening.

The answer to this problem is the 2D code. GS1 DataMatrix can hold structured data — the GTIN (item code), lot or batch number, expiry date and serial number — in a single symbol. Four barcodes become one. The scan becomes one operation. The area shrinks.

This is not a story about switching to a newer technology. It is a story about solving a physical problem of tag area and number of operations.

GS1 Sunrise 2027 is not only a retail story

There is an industry-wide deadline attached to the move to 2D. It is commonly called GS1 Sunrise 2027 (Ambition 2027): the goal that by the end of 2027, retail point of sale worldwide should be able to read GS1 2D codes in addition to conventional 1D barcodes. The transition is under way in 48 countries, covering a scope said to represent about 88% of global GDP (global figure, not country-specific).

There are concrete moves already. In April 2026, Tesco in the UK replaced the conventional 1D barcode with a GS1-based QR code / Data Matrix on the main range of its own-brand sausages — reported as a first for a UK supermarket.

The 2D code options GS1 lists for retail point of sale are these three.

This gets read as “we do not sell to retail, so it does not concern us”. For a parts maker it concerns you in a different form, because it arrives as a customer specification. Once a finished-goods manufacturer moves to operations premised on 2D, that requirement flows down into the part tag and label specifications for incoming parts. Even if the requirement has not reached you yet, if you are about to select equipment, a configuration that can both read and produce 2D codes will cost you less later.

There are three things to judge.

1. Can your ERP or MES generate and hold the structured data that goes into the 2D code? If the fields you want in the symbol do not exist in the master data, replacing the printer alone achieves nothing
2. Can you print or mark at the quality the production line needs? As data content grows, cell size gets finer, and the accuracy demanded of printing and marking rises
3. Can you operate through a period where 1D and 2D coexist? A single cut-over is not realistic. It assumes a reading side that can decode both

“It reads fine” is not an audit answer — ISO/IEC 15415 and 29158

Whether you can discuss quality in terms of a standard

This picks up “the customer audit stalls” from the opening. When an auditor asks how you control the quality of your identification codes, “it reads fine on the shop floor, so there is no problem” is not an answer. Whether it reads is an outcome that depends on the scanner used at that moment, the lighting conditions, and part-to-part variation. It is not a description of a control mechanism.

What is an answer looks like this. “Printed labels are assessed to ISO/IEC 15415, with our internal pass threshold set at grade X or above. Direct marks are assessed to ISO/IEC 29158, with a pass threshold of DPM grade X or above. Verification is done with a verifier, at this frequency and this sample size, and the records are retained for this many years.”

In other words, which standard and which grade you have set as the pass threshold is the answer. A factory that has not decided those two will get stuck in an audit.

When to use 15415 and when to use 29158

Which standard applies is decided by how the code was applied.

The practical point is that you must not simply apply 15415 to direct marks. The revision history runs ISO/IEC TR 29158:2011 (published as a technical report) to ISO/IEC 29158:2020 to ISO/IEC 29158:2025. When you write marking quality into an internal standard, stating which edition you are referencing shortens the explanation at audit time.

Why DPM needs its own standard

“It is the same 2D code, so why a separate standard?” comes up often. Two differences matter in practice.

Difference 1 — how the measuring aperture is chosen.

Under 15415 it is common to select an aperture of roughly 80% of the module size. Under 29158 the procedure is to first vary the aperture diameter until the symbol decodes, and then re-grade under two conditions, 50% and 80%. Unlike printing, direct marking does not produce uniform module shapes, so measuring with a single aperture pushes the assessment away from what the mark actually is.

Difference 2 — how the threshold is set.

15415 sets a global threshold at the midpoint of maximum and minimum reflectance. On DPM on metal, however, glare makes the maximum reflectance spike. The midpoint-derived threshold then sits too high, and the modulation of a perfectly acceptable mark is scored unfairly low. 29158 is an assessment method that addresses this.

Knowing these two lets you separate a shop floor complaint of “we cannot get the mark quality up” into whether the problem is the marking itself, or the wrong assessment method. If a report comes in on metal parts saying the code reads but the grade will not come out, first check whether it is being graded to 15415.

Setting the pass threshold

Set the pass threshold by looking at three things.

1. The customer requirement. Some customers specify a grade explicitly. Where they do, that is your floor
2. Margin for the reading environment. If the grade is marginal at the point of shipment, the code stops reading as soon as the part gets dirty or worn. Conditions get worse further downstream, so build in margin upstream
3. What your equipment can actually achieve. Measure the grade your own marking and printing equipment produces, by material and by condition. Set a high threshold without measuring this and you generate failures every day until the control becomes a formality

RFID does not cross borders — the regulatory gap between Japan, Thailand and Vietnam

Of the four conditions in this framework, this is the one that selection material rarely covers.

Taking the 920 MHz band UHF RFID readers and tags you have been running in a Japanese plant and moving them as they are to a site in Thailand or Vietnam is a plan that sometimes does not work. The band, the power classes and the licensing regime all differ by country.

Comparing the three countries

On safety standards, Thailand references IEC 60950-1 or TIS 1561-2556. On the Vietnamese side, Circular 18/2018/TT-BTTTT is another relevant instrument.

Do not compare the numbers in that table directly

One technical caution first. The power figures in the table above are not on a common basis.

• The Thai figures are EIRP (equivalent isotropically radiated power, which includes antenna gain)
• The Vietnamese figure is ERP (effective radiated power, referenced to a dipole antenna)
• Japan’s 250 mW is a transmitter output class, on a different basis again from EIRP or ERP

There is about 2.15 dB (roughly a factor of 1.64) between ERP and EIRP. Converting Vietnam’s 0.5 W ERP to EIRP gives about 0.82 W. Thailand’s licensed ceiling is 4 W EIRP, so put on a common EIRP basis, the Thai ceiling is about 4.9 times the Vietnamese one. Put the other way round, if you confuse EIRP and ERP and read it as “Thailand is 4 W, Vietnam is 0.5 W, an eight-fold difference”, you have started your equipment selection from a wrong premise.

What “just move it over” actually fails on

Failure takes three forms.

1. The bands do not overlap.

Thailand is 920–925 MHz, Vietnam is 918–923 MHz. The overlap is only 920–923 MHz, so the 923–925 MHz channel settings you were using in Thailand fall outside the Vietnamese allocation. Conversely, the 918–920 MHz side you were using in Vietnam falls outside the Thai allocation. Japan’s “920 MHz band” is a similar name but its allocated range does not match those two countries exactly either. Concluding “it is the same 920 MHz band, so it will work” is already dangerous at this point.

2. The power classes differ, so the licensing requirement changes.

In Japan, 250 mW or below is specified low power and needs no application. In Thailand, the licence-exempt class handled by SDoC is 50 mW EIRP or below. A model selected in Japan on the grounds that no application is needed may fall into a class that requires a licence in Thailand. And using above 50 mW up to 4 W EIRP in Thailand requires the NBTC Category A licence. Noticing this after the equipment is bought hits the go-live schedule directly.

3. The device’s conformity approval is not for that country.

Japan’s technical conformity certification does not carry over to Thailand or Vietnam. In Thailand, conformity to NTC TS 1010-2560 is what is examined; in Vietnam, it is examined under the MIC framework. The principle is that the same model number is not usable unless it is the variant certified for that country. Some manufacturers offer country-specific models, so always confirm at the purchasing stage which part number is the Thai one.

How to work through it in practice

So the procedure, when RFID is on your candidate list, runs like this.

1. Fix the country of installation. “Try it in Japan first, and roll it out to Thailand if it works” is an approach that has to be split at the model selection stage
2. Work backwards from the read range you need to the transmit power you need. If that lands inside the licence-exempt class, everything gets much simpler
3. Confirm whether it fits inside that country’s licence-exempt class. If it does not, build the time needed to obtain a licence into the go-live plan
4. Get written confirmation from the manufacturer of the type approval status of candidate models for that country
5. Check radio interference at the installation site. Coexistence with factory wireless LAN and other systems is a separate problem. Design on the industrial network side is covered in factory wireless LAN and industrial networks

A point to confirm: the bands, power levels and licence classes given here are compiled from publicly available information. Actual selection and applications must be verified against the current notifications from the Ministry of Internal Affairs and Communications, the NBTC and the MIC, and against the type approval status of the specific device. These regimes get revised. We are also not in a position to act as your agent for radio equipment licence applications. Our involvement extends to separating out the method selection and identifying the points where regulatory confirmation is needed.

Look at the economics as mark it, read it, link it

Splitting cost into three

The cost of an identification method becomes comparable once you split it into three.

This three-way split is not specific to RFID. Switching from printed labels to direct part marking can be judged with exactly the same expression. Direct marking front-loads the capital spend in category 1, but the per-part cost is lower, and because re-marking and re-reading happen less often, category 2 falls as well. The worked example below uses RFID, because that is where the gap in categories 1 and 2 is widest, but the structure holds whichever method you are substituting in. When you run your own numbers, put the two methods you want to compare into categories 1 and 2.

The important thing here is that category 3 barely affects a comparison between methods. Whether the ID was read from a label or from an RFID tag, the system-side processing — write the ID into the production history, link it to the material lot — is the same. In absolute terms category 3 is often the largest line in the project, so for the overall budget see how to read the cost of building traceability. This section stays on categories 1 and 2, where the methods differ.

Only two variables set the break-even

RFID pays only when the labour saved in category 2 exceeds the unit price gap in category 1 — a tag costs more than a paper label. Comparing tag prices alone will not give you the answer.

As an expression, it is this.

Incremental cost of RFID per month = total tag price gap in category 1, plus the equipment gap per point in category 2 multiplied by the number of read points

Saving from RFID per month = seconds saved per read, multiplied by labour cost per second, multiplied by volume, multiplied by the number of read points

Category 1 does not scale with the number of read points — a tag is applied once per tagged object, however many points there are. Category 2 and the saving both scale with the number of read points. So as you add points, at some stage the saving overtakes the incremental cost. That crossing is the break-even.

And the saving side scales with the seconds saved per read. So the break-even is set almost entirely by two variables: the number of read points and the seconds of manual work per read.

Assumptions for the worked example

What follows is a worked example on the assumptions stated here, not market pricing. It is built on a Thailand cost base. If you are working in another country, what transfers is the structure of the calculation, not the amounts — substitute your own conditions.

Converting to a monthly basis.

• Category 1 tag price gap = (130 x 3,000 / 60) – (25 x 3,000 / 60) = 6,500 – 1,250 = THB 5,250 per month (constant regardless of the number of points)
• Category 2 equipment gap = (120,000 / 60) – (30,000 / 48) = 2,000 – 625 = THB 1,375 per month per point
• Saving = 30,000 pieces x 4 seconds x THB 0.027 = THB 3,240 per month per point

The net benefit of each additional point is 3,240 – 1,375 = THB 1,865. That has to cover the 5,250 from category 1, so the break-even is 5,250 / 1,865 = about 2.8 points.

Sensitivity A — varying the number of read points

The same factory, the same volume and the same equipment flip the conclusion depending on whether there is one read point or five. At one point you are out of pocket THB 3,385 a month; at five points you gain THB 4,075 a month, or THB 48,900 a year. Comparing tag unit prices will never reveal that reversal.

Note that the incremental cost column already includes the amortisation of equipment and tags. You must therefore not run a separate “payback period” calculation on top of this net figure. Because equipment cost is already carried as monthly amortisation, recomputing a payback period from the same equipment cost counts it twice. What you want to compare is the difference in total monthly cost.

Sensitivity B — varying the seconds of manual work per read

The other variable alongside read points is the seconds saved per read. Holding the assumptions above (30,000 pieces per month, category 1 gap of THB 5,250 per month, equipment gap of THB 1,375 per month per point), we vary only the seconds.

In a process where only one second is saved, no number of read points makes RFID work, because the equipment gap cannot be covered by the labour saving. Conversely, in a process taking ten seconds per read — opening a box, counting the contents, scanning several labels one at a time — it works at a single read point.

The reason the “RFID is expensive / RFID is cheap” argument goes nowhere is that it is conducted without fixing those two variables.

With disposable tags, this calculation rarely works

The example above assumes tags attached to carriers and reused. What happens if you attach a disposable RFID tag to every individual part? Same framework.

Take an RFID label tag at THB 12 and a paper label at THB 0.60, and the gap per piece is THB 11.4. At 30,000 pieces a month that is THB 342,000 per month. Covering that at THB 1,865 net benefit per point would need 342,000 / 1,865 = about 183 read points. No process has that many.

Which means that at Thai labour cost levels, in-process traceability with a disposable tag on every individual part is hard to justify on labour saving alone (Thailand-specific). This is the main reason an investment case approved at a head office in a higher-wage country does not produce the same result in Thailand. Lower labour cost also means a smaller denominator for any automation investment justified on labour saving.

RFID works more readily where the tag goes onto something that is reused — carriers, returnable boxes, pallets, jigs — or where the manual work per read is long. Stocktaking is precisely the latter case, and its cost structure is handled separately in the cost of RFID stocktaking.

A procedure for the selection, and the common failures

Eight steps

Here is everything above, reordered into something you can actually run.

1. Write out the physical conditions of the target items. Fill in one line per representative item across the six columns: material, surface, temperature, liquids, life, area. Do not proceed without this table
2. Mark the read points on the process flow diagram. How many there are, who reads, and whether both hands are free, all settled on the diagram
3. Fix granularity from the customer requirement. Lot, box and pallet, or unit serial. Follow the existing design here
4. Decide the symbology. Derive the digit count needed from the granularity and judge whether 1D is enough or whether it should be folded into 2D. Three or more fields on the part tag, which in 1D means three barcodes side by side, means 2D
5. Decide how it is applied. If the physical conditions exceed what a label can survive, go to direct part marking. If the part only has to be identified inside the process, a label is usually enough
6. If you are using RFID, check country-specific regulation before selecting equipment. Projects that do this in the reverse order almost always end up reworking
7. Decide the quality assessment standard and the pass threshold. 15415 for labels, 29158 for direct marks. Document it down to verification frequency, sample size and record retention period
8. Design the system-side linking. Which key, which table, at what moment the scanned ID gets written. The link to the material lot is secured here

Five common failures

Failure 1 — selecting equipment without counting the read points.

Starting with “let’s buy two handhelds and try it” hides the cost structure at higher point counts. As shown above, point count is one of the two variables that set the break-even. Trialling is fine, but only on the basis that you have already counted the points in the final configuration.

Failure 2 — raising granularity after the fact.

Take something designed at lot level and raise it to unit serial later, and the category 1 cost multiplies by the volume. What was one label per lot becomes one label per piece. If there is any prospect of the customer requirement moving towards unit-level management, build it in when you choose how the ID is applied.

Failure 3 — not running a durability trial on the label.

Do not decide on the basis that the manufacturer’s catalogue says heat resistant. Run real parts through the real process, leave them for the real duration, then read them. Doing this trial once dramatically reduces re-marking work later.

Failure 4 — transplanting head-office practice unchanged.

RFID regulation is covered above, but other things travel badly too: labels printed only in the parent company’s language, part tags built around the parent company’s form layouts. Design the human-readable face (what a person reads with their eyes) separately from the machine-readable face (the code).

Failure 5 — starting operation without setting a quality pass threshold.

Without it you not only get stuck at audit, you cannot make a judgement about the daily “it will not read”. There is no criterion for separating whether the marking equipment or the reader is what needs adjusting.

Frequently asked questions

Should I choose a traceability barcode or a QR code?

They serve different purposes, so it is not really an either/or. A 1D barcode carries little data and grows sideways in proportion to digit count. For a part number alone it is fine, but add lot number and unit serial and the part tag ends up with several barcodes on it. A 2D code (QR code or Data Matrix) fits the same data into a smaller area, and error correction lets it tolerate a degree of soiling and damage.

The criterion is reasonably clear: three or more fields on the part tag, which in 1D would be three barcodes side by side, points to 2D. Alongside that, customer requirements may lean towards 2D as GS1 Sunrise 2027 progresses, so if you are selecting equipment now, a reading side that can decode both 1D and 2D is the safer configuration. One further characteristic is that the quiet zone required around the symbol is smaller for Data Matrix, which makes it easier to fit the same data into a tight area. On parts where marking area is the constraint, that can be the deciding factor.

Is traceability RFID worth the cost?

It comes down to two things: the number of read points and the seconds of manual work per read. On the assumptions used in this article (30,000 pieces a month, tags attached to 3,000 carriers and reused, 4 seconds saved per read), the break-even was about 2.8 read points. In a process with only one point you are out of pocket THB 3,385 a month; with five points you gain THB 4,075 a month. The same conditions, opposite conclusions.

Assume instead a disposable tag on every individual part and the same calculation puts the break-even at about 183 points, which is not realistically achievable inside a process. The practical conclusion is that comparing tag unit prices alone will not give you an answer, and the discussion cannot begin until the point count and the seconds are fixed. These figures are a worked example on this article’s assumptions, on a Thailand cost base, and are not market pricing.

Can I use the RFID readers we run in Japan at our plant in Thailand as they are?

Not necessarily. There are three reasons. First, the allocated bands differ: Thailand is 920–925 MHz, Vietnam is 918–923 MHz, and Japan’s “920 MHz band” does not match either exactly. Second, the licence classes differ: in Japan 250 mW or below is specified low power and needs no application, whereas the licence-exempt class in Thailand (SDoC) is 50 mW EIRP or below, and above 50 mW up to 4 W EIRP requires the NBTC Category A licence. Third, conformity approval is country-specific: Japan’s technical conformity certification does not carry over to Thailand, where conformity to NTC TS 1010-2560 is what is examined.

So before purchasing, get written confirmation from the manufacturer on whether the model is the country-specific variant and whether it holds type approval. Because these regimes get revised, the final decision has to rest on the current notifications from the Ministry of Internal Affairs and Communications, the NBTC and the MIC. Note also that within Japan, RFID in the 950 MHz band reached its usage deadline on 31 March 2018 and use after that date breaches the Radio Act (Japan-specific). If there is any plan to move older equipment to an overseas site, that is another thing to check.

Can material lots and product serials be linked the same way with any method?

The linking mechanism itself does not depend on the method. Using the scanned ID as a key to write into production history data is the same for a label as for RFID. Where the methods differ is in whether you can reliably identify the part at the moment the link is created.

Material lot linking usually happens at the moment the material is loaded. If at that moment the operator’s hands are full, or the material is inside a bag with the label out of sight, the read gets skipped. A skipped link cannot be reconstructed afterwards. So the order is: confirm the physical conditions at the loading step first (are both hands available, can the code be positioned so it is visible), and if those cannot be met, consider a method that does not need line of sight, such as RFID. The design of the linking keys themselves is covered in how to assemble a quality data management system.

For direct marking on metal parts, which standard and how far should the control go?

Quality assessment of direct part marking uses ISO/IEC 29158. Applying ISO/IEC 15415, which is for printed labels, directly to a direct mark can push the threshold up because of glare from the metal surface, so a mark that is actually fine may be scored unfairly low. 29158 addresses this by varying the aperture diameter until the symbol decodes and then re-grading under two conditions, 50% and 80%. Results are reported as a “DPM grade”.

As for scope of control, writing four things into your internal standard will cover most customer audits: which standard and which edition you use, the grade set as the pass threshold, the verification frequency and sample size, and the record retention period. Put the other way round, without those four you have nothing to say beyond “it reads fine”.

If the only goal is in-process defect traceability, how far do we need to go?

If you limit yourself to in-process defect traceability — that is, internal traceability — the condition becomes the identification only has to last until the part leaves the factory. That makes selection considerably easier. Because it does not need to read for ten years in the field, the durability requirement drops and labels stay in the running.

Concretely, design around three points: keep granularity to the unit needed to isolate the defect (if lot level identifies the cause, there is no need to go to unit serial), limit read points to before and after the processes where defects can arise, and require the identification medium only to survive the in-process environment.

There is one caution. A judgement that internal traceability is sufficient can break when the customer requirement changes. The moment a customer asks for unit-level recall capability, granularity has to be raised, and the category 1 cost multiplies by the volume. Even if internal is enough today, considering once at selection time which method would survive a rise in granularity keeps the later rework small.

Summary

Selecting an identification method is not something you decide on readability or equipment price. Here is the framework again.

• Four conditions decide it. The physical condition of the part, the design of the read points, the granularity, and regulation and borders. Work down them in that order
• You are choosing between four combinations. Label + 1D, label + 2D, direct part marking (DPM), and UHF RFID. No single one wins; combining them by process is the realistic answer
• When granularity rises, fold it into 2D. GS1 DataMatrix holds GTIN, lot, expiry date and serial in one symbol. GS1 Sunrise 2027 is under way across 48 countries, covering a scope said to represent about 88% of global GDP, and in April 2026 Tesco in the UK replaced 1D on the main range of its own-brand sausages. Making the reading side handle both formats before the requirement arrives from a customer is the safe move
• Discuss quality in terms of standards. ISO/IEC 15415 for printed labels, ISO/IEC 29158 for direct marks. “It reads fine” is not an audit answer. Which standard and which grade you set as the pass threshold is the answer
• RFID does not cross borders. Japan’s 920 MHz band, Thailand’s 920–925 MHz and Vietnam’s 918–923 MHz do not coincide. The licence-exempt power class differs by country, and so does conformity approval. These are things to confirm before buying equipment
• Split the economics into three and read it on two variables. Mark it, read it, link it — and the break-even is set almost entirely by the number of read points and the seconds of manual work per read. On this article’s assumptions, the conclusion reversed between one point and five

One point is worth repeating. The identification method is decided by the part, the process and the regulations, not by preference or familiarity. A method that has worked well for years at a parent plant elsewhere carries no guarantee of working the same way on the same product line in Thailand. Before you decide anything, write down the physical conditions of the part and the number of read points on paper. Doing just those two narrows the options worth considering quite sharply, and makes everything after that concrete.

If you would like a view on which method fits your own situation, we are happy to talk even at the stage of simply separating the question out. Should this part be direct marked or is a label enough, how many read points justify investing in automated reading, can the same equipment be used at our Thai and Vietnamese sites — we can help with the framing that comes before the selection itself. There is no need for the conversation to be premised on a project. Please get in touch via our contact page.

References

GS1 Sunrise 2027 and 2D codes

• GS1 Digital Link (GS1): https://www.gs1.org/standards/gs1-digital-link
• Preparing for the GS1 Sunrise 2027 Initiative (Cognex): https://www.cognex.com/blogs/machine-vision/preparing-for-the-gs1-sunrise-2027-initiative-shifting-to-2d-barcodes
• GS1 Sunrise 2027 (Scandit): https://www.scandit.com/blog/gs1-sunrise-2027-smart-data-capture-revolution/
• GS1 Sunrise 2027 (Videojet): https://www.videojet.com/us/homepage/resources/learn/gs1-sunrise-2027.html

Code quality assessment standards (ISO/IEC 15415 / 29158)

• ISO/IEC 29158:2025 (ISO): https://www.iso.org/standard/87123.html
• ISO/IEC 29158:2020 (ISO): https://www.iso.org/standard/69411.html
• Differences between grading standards 15415 and 29158 (Cognex): https://docs.cognex.com/dmst_6112/web/EN/DM475V_Manual/Content/Topics/DM360_370_470_475_Manual/differences-grading-standards-15415-29158.htm
• Commentary on ISO/IEC TR 29158 (Keyence): https://www.keyence.com/ss/products/auto_id/codereader/code_verification/tr-29158.jsp

RFID frequency, power and licensing (Japan / Thailand / Vietnam)

• Usage deadline for the 950 MHz band and migration to the 920 MHz band (Ministry of Internal Affairs and Communications): https://www.soumu.go.jp/menu_news/s-news/01kiban14_02000329.html
• Guidelines for 920 MHz band RFID radio station applications (Japan Automatic Identification Systems Association): https://www.jaisa.or.jp/pdfs/191227/001.pdf
• Thailand NBTC updates 920–925MHz usage, RFID / non-RFID (360 Compliance): https://360compliance.co/global-market-access/thailand-nbtc-updates-920-925mhz-usage-rfid-non-rfid/
• Guidelines for RFID usage in Thailand (Activio): https://www2.activio.asia/2025/06/20/guidelines-for-rfid-usage-in-thailand/
• Vietnam MIC Circular 18/2018/TT-BTTTT (CSA Group): https://www.csagroup.org/global-certification-regulatory-update/vietnam-mic-circular-18-2018-tt-btttt/

Thailand investment environment (BOI)

• Bangkok Shuho: https://bangkokshuho.com/thainews-731/
• Bangkok Shuho: https://bangkokshuho.com/thainews-1215/
• MarkLines: https://www.marklines.com/ja/news/338411

Standards and regulations are revised over time. What is set out here is a summary as at the time of writing, based on the sources above. For actual equipment selection and applications, please confirm the current notifications and the type approval status of the specific device.