Blog

2026.11.11

Implementing Process Defect Reason Codes in Manufacturing: RFP and Acceptance

Implementing Process Defect Reason Codes in Manufacturing: RFP and Acceptance

When introducing a manufacturing process defect reason code management system, the first decision is not the layout of the entry screen. Production, quality, and planning teams must be able to interpret the same observation consistently: where a defect was found, where it may have originated, and which specification was used to judge it. This guide follows the work from the code master and MES entry through ERP integration, RFP wording, and FAT/SAT acceptance. It is written with multilingual plants in Thailand in mind. Quantities in the examples illustrate data design; they are not measured customer results or industry averages.

Why process-specific defect reason codes matter

A plant may know its total reject count but still be unable to compare what happened. “Appearance NG” in molding, “scratch” in assembly, and “appearance defect” at final inspection might describe the same phenomenon—or different ones. Likewise, “dimension defect” may blend tool wear, incorrect setup, and a different measurement method. A Pareto chart can be numerically correct while its categories remain semantically inconsistent.

A reason-code system provides controlled language for the observed defect, the process where it was detected, the process thought to have caused it, and the cause approved after investigation. It is more than a dropdown list. An inspector can reliably record a linear scratch before knowing whether a fixture, mold, or downstream handling caused it. The cause field should remain “undetermined” until the evidence and an authorized reviewer support a conclusion. This is a system for human observation and validated classification, not automatic normalization of equipment failure codes by AI.

Our three-layer guide to quality data management explains recording, linking, and analyzing quality data more broadly. The present article concentrates on the meaning, use, and controlled revisions of manually selected production defect categories. The distinction matters because changing a label after the fact can change the apparent trend without changing the underlying process.

Separate detection, occurrence, phenomenon, cause, and disposition

Before writing an RFP, define separate fields for the detection process, suspected occurrence process, observed phenomenon, investigated cause, and disposition. Detection is where the issue was discovered; occurrence is where it is believed to have arisen and may initially be unknown. Phenomenon describes what the inspector saw or measured. Cause is a later judgment. Disposition describes what happens to the physical item: hold, sort, rework, scrap, or a concession approved under the plant’s procedure. Mixing “scratch” and “replace mold” in one field destroys analysis and auditability.

Consider work order WO-1042, item A, lot L-21. An inspection after assembly discovers five scratched items. The first record can state detection process = post-assembly inspection, phenomenon = linear scratch, occurrence process = unknown, cause = under investigation, disposition = segregated. Later, evidence may show that a transport tray touched the parts. An authorized quality reviewer can record occurrence process = transport and cause = tray contact while retaining the previous values, evidence, reviewer, and time. A single “process” field would falsely assign every issue found at final inspection to final inspection.

ISO 9001 provides a quality-management framework, not a prescribed column layout for a defect-code master. The cited ISO/IAF auditing guidance concerns service organizations and discusses nonconformity records and feedback to the appropriate management level; it is not a manufacturing-specific obligation. The data fields below are implementation proposals for traceable records. Confirm actual requirements against your quality procedures and customer contracts.

Implementing Process Defect Reason Codes in Manufacturing: RFP and Acceptance - figure 1

Design the manufacturing defect-code master

At minimum, a code needs a stable ID, display names, a precise definition, applicability, effective dates, a version, and a lifecycle state. Keep the ID stable even if the label changes. Store Japanese, Thai, English, and, where needed, Vietnamese labels as separate attributes of the same ID. Define when to choose the code, when not to choose it, its boundary with similar codes, and links to sample photographs or measurement criteria. A short screen label cannot teach a new operator or night-shift employee every boundary. Codes that are too broad hide improvement opportunities; codes that are too fine produce hesitation and excessive “other” entries.

A practical starting point is a common top-level family with process- or product-limited subcategories. “Appearance” might contain linear scratch, dent, and contamination. The same phenomenon can share one ID across processes while an applicability table controls where it can be selected. Process-specific management does not require a different ID for every process. Conversely, apparently identical names need distinct definitions or IDs when acceptance criteria differ. The goal is to preserve meaningful comparison without making the shop-floor list unmanageable.

The master must record who proposes a change, who approves it, and when it becomes effective. Retire rather than delete old codes; preserve a successor mapping when appropriate. If yesterday’s record is silently shown using today’s revised definition, an auditor cannot tell what the operator selected at the time. Keep the selected ID and version with the transaction and, if necessary, a snapshot of the displayed label. Define the site and time zone for a version cutover and decide how an offline terminal may submit a transaction against an older version.

Do not try to eliminate “other” by force. Require a provisional description and, when useful, a photo or measurement, an owner, a review deadline, and a decision on later reclassification. Otherwise operators may choose a nearby but false category. Review how many items remain unclassified, who reviewed them, and whether a new code truly cannot be represented by an existing definition. A new code proposal should include actual examples and review by both production and quality.

Make MES defect entry workable on the floor

A basic flow loads the work order, item, lot or serial, and process; reconciles good, defective, and held quantities; then offers applicable phenomenon codes. Hide codes not valid for that item, process, and effective date, but provide a controlled “unclassified” route if an unexpected observation occurs. Do not force the operator to guess a root cause. Attach a photograph, instrument result, or inspection-specification version as evidence when the process calls for it.

Check quantity semantics carefully. If 100 items entered and there are 94 good, five defective, and one held, the total is 100. Reinspection, rework, split lots, in-process scrap, or carryover from an earlier process can invalidate such a simple equation, so define the rule per process. If one item has both a scratch and an out-of-tolerance dimension, there are two phenomena but only one defective item. Counting both phenomena as two defective units understates yield.

Plan for network interruption. If a terminal stores entries locally, specify the deduplication key, device time versus server receipt time, and what happens when a code master has changed during the outage. Retain the same event ID on retries so sending the same five rejects twice never produces ten rejects. Corrections should preserve the old and new values, reason, actor, approver, and timestamp rather than quietly overwriting an earlier record. Decide whether unapproved records may be sent to ERP.

There is no universal acceptable number of seconds per entry. Gloves, screen size, barcode location, takt time, and inspection density differ. Have real operators try normal, exception, and rework scenarios on actual devices. Observe hesitation and errors. Verify that a Thai term used on the floor and a Japanese term used by headquarters reference the same ID and boundary; a literal translation is not necessarily an operationally correct one.

Implementing Process Defect Reason Codes in Manufacturing: RFP and Acceptance - figure 2

Connect MES and ERP quality data with clear ownership

ISA-95 is a reference model for exchange between manufacturing operations and business systems. ISA describes MES primarily at level 3 and ERP primarily at level 4. This does not dictate every vendor’s product boundary, but it is a useful starting point for assigning a system of record. For example, ERP or production planning may send item, work-order, BOM, and routing data to MES. MES may send process-level production events, phenomenon codes, and quantities to ERP or QMS. A quality system may return approved disposition to both.

An integration contract can include event ID, work order, item and revision, lot or serial, detection process, suspected occurrence process, phenomenon code and version, quantity and unit, detection and entry timestamps, approval state, and disposition state. Do not copy every field into every application. Select the fields the receiving system needs for a business decision and retain a link to the original record. ERP may need reject quantity and scrap value for costing, while the source photograph can remain in QMS.

Assume retries, reordered messages, and partial failures. If an investigated cause changes, do not erase the first observation and create an unrelated new transaction. Send a correction referencing the original event ID and version, and require the receiver to apply it only once. Define an error queue and owner when ERP has canceled the work order, lacks the lot, or has a different item revision. “Integration succeeded” should mean the intended business record exists at the destination, not merely that an HTTP request returned success.

The OPC Foundation’s ISA-95 information model provides a common approach to representing manufacturing and ERP information. It does not define your plant’s defect taxonomy for you. The plant must own semantics, versions, and approval. NIST guidance on trustworthy manufacturing data also emphasizes traceability of exchanges and metadata. Be able to answer which inspection record, work order, and timestamp were sent to which system.

Keep coding separate from nonconformance disposition

A phenomenon code does not itself approve rework, scrap, concession, or shipment. Siemens describes defect management as tracking deviations and corrective actions, and distinguishes a nonconformance from a resulting defect. Define detection, physical segregation, evaluation, disposition approval, reinspection, and release as separate states. A scratch entry may put a lot on hold; a quality reviewer may later authorize polishing and reinspection. The phenomenon remains a scratch even if the final quantity passes. Silently subtracting the original defect from the historical record would erase the improvement signal.

Track first-pass acceptance, final acceptance, rework, and scrap separately. Make the physical hold label agree with the digital hold. For shipment-gate design, see our related guide to preventing defect outflow through release decisions and lot tracing. Set the scope of a code-management project separately from a full material-review-board or shipping-gate implementation, so responsibilities and estimates remain clear.

In SAT, follow an actual held part from the station to the segregation area and compare its physical label with the MES hold and the ERP inventory state. A discrepancy is an acceptance issue even if the defect code itself was entered correctly. State separately in the RFP how far the first project will automate disposition, release, and shipment controls, then price those steps explicitly.

Govern multilingual, multisite categories

If headquarters and a Thai plant interpret “dent” differently, a translation table alone is inadequate. Use samples, acceptance boundaries, measurement methods, and local shop-floor terminology to connect both languages to one ID. English can help cross-site comparison without replacing Thai training. Govern the effective version of every approved translation and check that long labels fit on actual devices.

Define what is central and what is local. Common product phenomena can be shared; a site-specific condition can have restricted applicability. Unlimited local codes undermine group-wide comparison, while overly slow central approval pushes operators toward free text. A practical arrangement gives site quality a first review and corporate quality responsibility for common-code approval. Review new, retired, “other,” and reclassified entries regularly, and preserve mapping tables.

A master change request should describe affected products, the relationship to old codes, historical-data policy, training audience, and report and ERP effects. If an old “appearance NG” category is split into three new codes, do not allocate old transactions across them without individual evidence. Keep the old category for historical reporting, or reclassify only supported records. Mark the change date on defect trends; otherwise a taxonomy revision may appear to be a process improvement.

Write testable RFP requirements

“Can enter defect codes” is too vague for a comparable quotation. Specify processes and product families, lot versus serial granularity, separation of phenomenon and cause, multilingual labels, applicability, effective dates, approval workflow, offline retry, correction history, reporting granularity, and ERP field mapping. Decide whether photographs and instrument data live in MES or link to an existing system. Price historical free-text conversion separately from importing it as unclassified data.

For example: “After selecting a work order and process, an operator may choose only phenomenon codes valid for that process, item, and time. The cause may remain undetermined and may be finalized only by an authorized quality reviewer. A correction retains old and new values, reason, actor, approver, and time. Approved reject events carry a stable ID to ERP; retries must not double the quantity.” Adapt permissions and transmission timing to your plant’s procedures.

Ask bidders to separate standard functionality, configuration, custom development, integration, training, migration, and support in price and schedule. Clarify who prepares the initial master. Even if a vendor configures the screen, the plant must review historical defect forms, resolve terminology, and approve the boundaries. Include the number of floor reviews, Thai terminology validation, and legacy-data scope in the estimate assumptions.

Map the actual process before designing the code list: where is a defect found, where is quantity confirmed, and who releases a hold? The same scratch found during self-inspection and at outgoing inspection may require different notification and trace-back. Explain to operators that detection process is a factual location, not an automatic assignment of blame. A punitive workflow encourages vague codes. Treat disagreement about classification as evidence that definitions need work.

For migration, classify old-to-new mappings as one-to-one, one-to-many, or unmapped. Even a one-to-one mapping needs a definition check. A broad old code cannot be reliably distributed among new subcodes if individual photos or measurements do not exist. Preserve unmatched history and show the taxonomy cutover on management charts. Create a permissions matrix for operators, supervisors, quality reviewers, master administrators, and audit readers; include deputy approval, night-shift holds, and user deactivation.

Run vendor demonstrations with the plant’s own ambiguous records, not only the vendor’s prepared normal case. For example, manufacture one item on two lines while only one line has moved to a revised inspection specification. Check whether the terminal presents the correct code version on each line, whether reports separate unlike definitions, and whether an operator can explain the choice. Attractive screens and many configuration fields cannot compensate for a category whose historical meaning changes invisibly.

Test claims of “no customization needed” against the complete workflow. Multilingual labels may be configurable while approvals, effective-dated versions, offline submission, or corrections require another module. Divide RFP items into mandatory, desirable, and future requirements, and attach a demonstration to every mandatory item. If unit-level traceability may be introduced later, ask how today’s lot-level events will relate to future serial IDs; an apparently optional future feature can affect the initial data model.

Implementing Process Defect Reason Codes in Manufacturing: RFP and Acceptance - figure 3

FAT and SAT acceptance scenarios

FAT checks the configured functions and master in a controlled test environment. SAT checks the actual line, devices, network, users, and connected systems. Even if your contract uses different terms, both levels of testing are needed. A code appearing on a screen is insufficient; the system must prevent wrong selections and route exceptions to the right owner.

FAT should cover at least: invalid process codes cannot be chosen; future and retired codes obey effective dates; translations resolve to one ID; a phenomenon can be entered with unknown cause; unauthorized users cannot finalize cause; corrections retain history; retries do not duplicate quantities; and submissions from a terminal with an old master follow the agreed policy. Give each scenario input data, expected results, and evidence. Compare the search view, API response, and receiver’s business record.

SAT should include real label scanning, gloves, language switching, lost and restored network, device replacement, interrupted work, and shift handover. Reconcile physical segregation with MES and ERP states. Test one item with multiple phenomena, rejects after a lot split, reinspection after rework, correction of a mistaken entry, and movement from “unclassified” to an approved code. Set sample sizes and tolerances by process risk and contract; no generic numerical threshold applies to every factory.

Name the sign-off roles. IT verifies API behavior; production verifies usability; quality verifies category meaning and evidence; planning verifies quantity and work-order consistency. Track defects by severity, reproduction steps, workaround, due date, and retest outcome. Decide in advance who can approve go-live with unresolved findings.

Roll out in a sequence that preserves data meaning

Start with a representative line, not the easiest line or every factory at once. Gather existing defect slips, spreadsheets, MES views, ERP work orders and inventory states, and quality-meeting reports. Follow a few real cases end to end. Classify legacy terms as duplicates, ambiguous, or unused; make an applicability matrix and multilingual glossary. Ask production and quality owners to approve boundaries using physical examples before configuring screens.

In a workshop, have several reviewers independently code past cases. When selections differ, identify whether the problem is an unclear definition, missing evidence, a different inspection revision, or a mix of phenomenon and cause. Rework the definition and test the same boundary cases again. During a limited parallel run, compare old and new counts, lots, process labels, and corrections daily. The objective is to explain differences, not merely to make totals match by accident.

Plan go-live around unresolved critical defects, training, backups, paper fallback, and preventing duplicate entry during recovery. If staff use paper during an outage, preserve both observation and later entry timestamps. Test the recovery process in SAT. After launch, monitor unclassified records, corrections, pending causes, master approval time, and failed ERP transfers alongside reject rate. Early increases may reflect improved visibility rather than worsening production.

A monthly Pareto chart alone cannot explain change. Check the production denominator, item mix, inspection scope, taxonomy revisions, and severity. Few but critical defects may deserve attention before high-count minor ones. Use code ID as the aggregation key across languages and document the code-version date. Train new sites on typical cases, boundary cases, and cases where a code must not be used.

Select the pilot line carefully. The easiest line may reveal too few exceptions; starting with every product in several plants creates an unmanageable master before definitions stabilize. Choose one line with a meaningful production volume, several types of defect, and production and quality owners who can join testing. Trace its items from the first operation to pre-shipment inspection, including every point where lots split or merge. Without that path, a defect cannot be linked reliably to the affected material.

In the first workshop, bring a sample of historical defect slips and ask several participants to assign codes independently. Do not settle disagreements by a majority vote. Determine whether the dispute comes from an ambiguous definition, missing photograph, different inspection specification, or confusion between phenomenon and cause. This is a method for finding boundary cases, not a prescribed sample count. Revise the definitions and ask people to classify the same cases again; agreement after clarification is evidence that the master has become usable.

After issuing the first master version, run old and new records in parallel for a limited period covering the line’s normal operating cycle. Compare quantities, lots, process, phenomena, and corrections each day. If duplicate entry would disrupt production, limit the work orders and time window. The goal is to understand why records differ rather than to force totals to match. A new system may show holds or multiple phenomena that the former paper form could not express. Document those as differences in counting rules.

Make the go-live decision include a paper fallback and the return path after an outage. Production and quality owners must decide whether operations stop or continue on paper while the service is unavailable. On later entry, preserve both the actual detection time and the data-entry time, with a reference to the original sheet. Prevent the same defect from being counted twice when an automatic resend and manual entry coincide. Test this complete recovery in SAT, rather than only testing disconnection and reconnection.

Review monthly trends in context. A code’s count may fall because the process improved, inspection criteria changed, production volume declined, or people moved to a different code. Show the denominator and changes in product mix. A severe but rare defect may deserve attention before a frequent minor one. Reports for the floor should connect process, equipment, product, shift, and observed phenomenon to actions rather than rank individual workers. When sharing reports across languages, aggregate by code ID and state the version date so revised translations do not break comparisons.

Frequently asked questions

During product comparison, do not judge by screen appearance or the number of settings alone. The practical differences show up when a code definition changes: how earlier transactions remain visible, how language equivalents retain one meaning, and how a correction reaches ERP and reports. Bring difficult cases from your own plant to the demo. If two lines make the same item but only one uses a new inspection revision, check that each line receives its valid choices and that the report does not add unlike definitions as if they were equivalent.

“No customization” on a quotation may not cover the complete floor workflow. A product may offer translated labels while approval, effective-dated revisions, and offline retry are separate modules. Mark each RFP requirement as mandatory, desirable, or future, and provide an expected demonstration for the mandatory items. Ask about the future data model even for postponed features. A possible transition from lot-level to serial-level records is one example that can influence the initial keys.

Design the quality meeting that will use the data. A top-code list alone does not identify an action; drill into item, process, equipment, lot, and period. If many records remain with an unknown occurrence process, examine the linking data between operations before blaming the operator. Merge codes that repeatedly lead to the same countermeasure and split codes that combine unrelated countermeasures. The classification is an operational asset to refine against evidence, not a static list finished at launch.

For investment decisions, assess more than seconds saved on a terminal. Consider time spent finding investigation records, unresolved causes, rework, and correcting misleading reports. Define a comparison group and measurement method before deployment because product mix, inspection volume, and customer requirements may change. Do not convert benefits that cannot be measured into a fabricated monetary figure. A higher initial defect count can reflect previously hidden events becoming visible; separate record quality from the physical process trend.

Does each process need a separate defect code ID?

No. A shared phenomenon and acceptance rule can use one ID with process-specific applicability. If the same label has different decision criteria, distinguish it by definition or ID. The aim is comparable reporting and usable floor selection.

Should MES defect entry require a root cause?

Not when the operator cannot observe it. Record the phenomenon, quantity, object, and detection process first. Allow “undetermined” and route investigation to an authorized quality reviewer. Add urgent notification and hold rules for serious safety or customer issues.

Should “other” be prohibited in the defect-code master?

A controlled temporary entry is safer than a false nearby code. Capture description and evidence, assign a reviewer and deadline, and decide whether a formal code is needed. Reclassify history only when evidence and approval support it.

Which system is the source of truth for ERP quality-data integration?

Assign ownership separately for work orders and items, shop-floor events, quality dispositions, and inventory or costing. Document creator, approver, update rights, and recipients. Share stable event IDs and versions; test that a retry cannot double-count a defect.

What is most often missed in a defect-code RFP and FAT/SAT?

Teams often test normal entry but omit unknown causes, multiple phenomena, corrections, old versions, outages, retries, and retired codes. Link each RFP requirement to an expected result, test functionality in FAT, and verify the real workflow and receiving system in SAT.

Conclusion

A process defect reason code system works when it separates observed phenomenon, detection process, suspected occurrence process, investigated cause, and disposition; assigns ownership; and preserves change history. Usable MES entry, a clear ERP boundary, and exception-based FAT/SAT make defect numbers comparable. Define the code master using real samples and acceptance tests before selecting a product.

If you are still mapping your code list, MES screens, or ERP interface, you can discuss the scope with TOMAS TECH. We can help turn actual floor records into a controlled master, RFP requirements, and practical acceptance scenarios.

Sources