Blog

2026.11.10

Manufacturing MSA and Gage R&R Management System: RFP and FAT/SAT

Manufacturing MSA and Gage R&R Management System: RFP and FAT/SAT

When introducing a manufacturing MSA and gage R&R management system, start with the decision the measurements must support. Digitizing a gage R&R worksheet is insufficient if the characteristic, part, operator, instrument, procedure, environment and decision rule cannot be reconstructed. This guide helps quality, production and IT teams in Thailand define an RFP and prove its operation in FAT and SAT.

Connect MSA and gage R&R to production decisions

Measurement systems analysis (MSA) examines the measurement process. Gage repeatability and reproducibility (GR&R) is a common study of two sources of variation. ASQ defines repeatability as variation when the same operator repeatedly measures the same characteristic on the same part with the same instrument; reproducibility concerns differences between operators. Separating them helps locate improvements in the instrument, fixture, method or training. A favorable R&R result alone does not establish accuracy, stability or conformance. Bias, linearity, stability over time and calibration status need separate attention.

AIAG describes its MSA fourth edition as a guide for assessing measurement data quality and improving measurement processes. Express the business objective more precisely than “automate the R&R report.” Decide whether a dimension can support pass/fail release, whether the variation is suitable for capability analysis, whether shifts or plants can be compared, or whether a repaired instrument may return to service. Different decisions call for different study designs and approvers.

Link each characteristic to the drawing revision, inspection specification, customer requirement and control plan. Define units, location, contact direction, force, fixture, part temperature and surface condition. An instrument ID cannot stand in for the measurement procedure. Quality should approve the study, and production should use only approved combinations. One generic “gage approved” label cannot establish fitness for every characteristic.

How do calibration and MSA records differ?

Calibration management in Thailand addresses the instrument register, due dates, certificates and calibration results. MSA addresses the measurement process with specified parts, people, procedures and environment. A current calibration certificate cannot cure operators’ inconsistent positioning. A good R&R study cannot excuse an expired calibration or damage. Connect the two records through instrument ID and validity dates, while showing their distinct approval basis.

NIST explains metrological traceability as a property of a measurement result: a documented unbroken calibration chain connects it to a reference, with uncertainty at each link. Traceability is not simply a property of a gage or PDF certificate. A system therefore needs a path from the result to the instrument, applicable calibration, reference chain, procedure and time. Merely attaching a PDF must not trigger an automatic “traceable” label.

Turn the gage R&R study design into data structures

For a nondestructive measurement where every operator can remeasure the same parts, consider a crossed design. Destructive testing may require a nested design because the identical part cannot be measured again. ASQ distinguishes crossed, nested and expanded studies. A product that accepts only a crossed worksheet may be inadequate for destructive tests, multiple gages, missing results or part-by-operator interaction. State both the selected design and what it cannot estimate in the RFP.

Select parts that represent the relevant manufacturing conditions and spread of the characteristic. Selecting nearly identical parts can inflate a ratio that uses part-to-part variation as the denominator. Combining an excessively broad population can hide a problem under a particular process condition. Record why each part was selected, its lot, mold, cavity, line and production period. State the scope to which the result applies. Tolerance width and actual process variation are different quantities; the report must identify its denominator.

Approve operators, order and repetitions before collecting results. Deleting inconvenient readings during the study undermines the evidence. A fixed order can confound learning, instrument warm-up, part deformation or time drift with other factors. Use randomization and blinding where practical. Preserve the original reading and reason for any repeat, and version the data set used in each analysis. Missing, disconnected, unmeasurable and deliberately excluded results must not all become zero.

For temperature or humidity sensitive characteristics, record the plant environment. Day and night shifts may differ in people, lighting or waiting time. Changes to fixtures, measurement programs, corrections and reference masters can change variation. Change control should determine whether a new study is needed. A statistical result without its operating conditions will be difficult to reuse months later.

Manufacturing MSA and Gage R&R Management System: RFP and FAT/SAT - figure 1

Explain what the system calculates

Specify whether the analysis uses average and range or ANOVA, whether interactions are modeled, and how missing data is handled. Two products may display “%GRR” but use different denominators, such as total variation or tolerance, or different standard-deviation multipliers. Show the method and version, calculation settings, part variation, repeatability, reproducibility, interactions, residuals and observation count. If a value cannot be estimated, show why; never substitute zero or “pass.”

Do not turn a commonly quoted numerical guideline into a universal acceptance rule. Customer requirements, purpose, risk of a wrong decision, capability, analytical method and population all matter. Version approved rules by customer, part, characteristic and use case. When a threshold changes, keep the historical decision and make recalculation under the new rule available instead of silently rewriting history.

A screen showing only “gage A passed; gage B failed” encourages a premature purchase decision. Look instead for operator differences, certain parts tilting in the fixture, or insufficient resolution. Show part-by-operator plots, repeat readings, range charts and exception records so an improvement owner can be assigned. The value of management software is connecting analysis to corrective action.

The smallest record a management system must retain

Make the individual measurement event the core record. It should carry part ID, characteristic ID, drawing and specification revisions, gage ID, fixture ID, operator ID, procedure revision, timestamp and time zone, value and unit, acquisition method and reference to raw data. A study groups those events and adds part selection, operators, repeats, sequence, analysis method, approver and application scope. Retaining raw values allows a later comparison of analysis methods.

The gage record needs calibration status, restrictions, repair, adjustment, before and after calibration results and replacement history. Tolerances and customer drawings require revisions and effective dates. Never overwrite an old result with the current drawing revision. For supplier CSV and machine output, keep receive time, source file, row number, transformation rule and deduplication key. More calculated results do not compensate for lost provenance.

If manual entry remains, distinguish it from machine capture. Warn on unit conversion, rounding, implausible digits, negative values and out-of-range results; log who may override each warning and why. A correction must retain old value, new value, reason and approver. If a source value changes after study approval, return the study to review or alert quality. Silent edits are dangerous.

Multi-plant operation requires more than translation. Record a consistent time basis and time zone, then show local time to users. A date without a time zone obscures the night-shift boundary when Japan reviews Thai results. Define distinct state codes for remeasurement, unable to measure, approval on hold and outside scope. Japanese and Thai labels should map to the same states so translation does not change disposition.

Manufacturing MSA and Gage R&R Management System: RFP and FAT/SAT - figure 2

Separate authority and recreate audit evidence

Operators can submit readings and shop-floor explanations, quality can approve plans and results, and master-data administrators can revise controlled definitions. Set the rule for whether someone may approve their own measurements. Specify whether an expired, stopped or repaired gage blocks entry entirely or permits an exception for quality disposition. Test duplicate and out-of-order synchronization after offline operation.

An auditor should be able to start with a product inspection result and reach the drawing, measurement procedure, calibration status, MSA approval and original reading effective at that time. The quality assurance record system RFP guide likewise focuses on reconstructing a chain of evidence. Evaluate filters, revisions, excluded values, corrections and approval history before the appearance of a PDF report.

Compare functions and responsibilities in the RFP

Start the RFP with plants, characteristics, measurement methods, existing gages and calibration register, users, languages, approval path and retention. State whether crossed designs suffice or nested and expanded methods are needed. Identify the verified scope of each algorithm. A polished demo is irrelevant if it cannot represent the plant’s study design. Prioritize critical characteristics and decision frequency instead of rolling out to every gage at once.

AreaSpecific RFP requirementAcceptance evidence
Study planCharacteristic, part, operator, repeats, order, crossed or nested designApproved plan matches acquired readings
AnalysisMethod and version, denominator, missing data, exclusions, interactionReconcile known data against an independent calculation
MeasurementRaw value, unit, time, instrument, procedure, correctionTrace source file to final result
DecisionCustomer rules, authority, revision, holdShow rule changes and approval history
IntegrationGage, calibration register, QMS and MESTest disconnection, retry, duplicate and clock drift
OperationLanguages, training, monitoring and backupDemonstrate shift handover and restore

Give vendors the same sample data and request intermediate values and calculation output, not merely a “%GRR” screen. Ask for statistical method documentation, revalidation when software changes, export formats and calculation logs. Assign responsibility for machine communication specifications and driver maintenance. CSV integration needs agreed naming, units, encoding and duplicate handling.

Compare cost by separating licenses, instrument interfaces, master cleanup, migration, calculation verification, shop-floor training, languages, support, backup and additional plants. A cheap offer that excludes nested analysis or audit logs is not a comparable scope. Confirm customer report layouts and approval signatures before buying. For excluded functions, name the temporary procedure and owner rather than writing “later.”

Does every instrument need an individual MSA study?

The IATF official FAQ for IATF 16949 clause 7.1.5.1.1 says a complete statistical study is not required for every individual device. Instruments sharing characteristics such as range, resolution and repeatability may be grouped, with a representative instrument studied. The plant must still justify grouping and representativeness. Store the gage family, rationale, representative device, covered members, exceptions and reevaluation triggers. This is not permission to reuse one result for unrelated instruments. Check customer-specific requirements.

Reproduce the results in FAT and SAT

FAT at the supplier checks calculations, screens and basic integration against specification. SAT at the Thai plant checks actual instruments, procedures, users and network conditions. The aim is not simply to display a green “pass.” Another person must be able to reproduce the report from original readings under the same conditions. Quality, production and IT owners should sign the results and list open issues and interim controls.

In FAT, use known-answer data, boundary values, wrong units, missing and duplicate readings, exclusions and an operator-specific bias. Ensure the application does not fill gaps and calculate a crossed study whose assumptions are violated. Change the analysis method and decision rule; the old result should remain and a separately versioned new one should appear. Test that one customer’s rule is not applied to another customer’s study.

In SAT, conduct a representative study end to end with the actual caliper, micrometer, coordinate measuring machine or fixture in scope. Check calibration status, instructions, part labels, permissions and local time. Demonstrate reconnection and replay, duplicate imports, wrong part ID, wrong unit, shift change and correction after approval. Matching a legacy report’s number is insufficient if change history or application scope disappears.

Manufacturing MSA and Gage R&R Management System: RFP and FAT/SAT - figure 3

Define acceptance in terms of who can reconstruct which evidence and within what locally justified time. Do not present one response-time target as a universal standard. Test both sample volume and expected operating volume. After restoring a backup, verify raw data, calculation version, approval history and certificates. Make sure the test report records evidence, failures, repairs and retests.

Decide the rollout order and operational changes

Start with a characteristic whose customer requirement is clear, whose result directly affects release or process decisions, and whose current procedure exists. Fix part labels and gage IDs before software rollout if they are unreliable. Complete acquisition, analysis, approval and connection to inspection records on one line before expanding to shifts and parts. Old spreadsheets without provenance or study conditions should be marked as reference material, not imported as newly approved studies.

After launch, review whether a new study is needed when a gage is replaced or repaired, fails calibration, a fixture or procedure changes, a drawing is revised or a station moves. Let quality decide using risk rather than applying one automatic rule. A red overdue label should connect to potentially affected products and hold decisions. In monthly review, examine critical-characteristic coverage, studies with unknown conditions, raw-data corrections, discrepancies between calibration and MSA state, pending approvals and actual audit retrieval. A rising study count is not proof if operators maintain shadow spreadsheets.

Where do ISO 10012:2026 and ISO/IEC 17025 fit?

ISO 10012:2026 specifies requirements for measurement management systems and supersedes its 2003 edition. ISO/IEC 17025:2017 concerns the competence, impartiality and consistent operation of testing and calibration laboratories; ISO says the edition was confirmed in 2023. Management of a plant’s measurement process and assessment of an external laboratory are different scopes. Map requirements to records and controls before claiming compliance. Buying software cannot alone establish organizational conformity. An accredited laboratory’s certificate does not automatically make the plant’s R&R good.

Turn shop-floor failures into requirements

An instrument ID does not capture every measurement condition. The same caliper can measure internal and external dimensions with different contacts and operator forces. Record the drawing characteristic, location, fixture, procedure revision and part condition. If operators rarely open a separate instruction, present a concise diagram at the terminal. “Trained” alone does not prove that a measurement was performed correctly.

Study parts can also be treated differently from normal production parts. If an operator recognizes a near-limit part, they may unconsciously repeat or adjust a reading. Blind and reorder parts where practical, but appoint someone to control identification so parts are not mixed into production. Choose an executable, explainable design over a statistical ideal that the plant cannot perform safely.

Do not assume a machine’s automatic output is the authoritative record. A value without unit or part ID can be associated with the wrong terminal selection. Device clocks and receive order may drift. Preserve raw signal or file, device and receive timestamps, label and operator action separately. Quarantine unlinked data instead of automatically feeding it into the study.

Finally, do not report only the best number. Replacing parts, changing operators or removing outliers can change a ratio. Some changes are justified, but require reasons and approval. Preserve the original plan and result, corrective action, and follow-up plan and result as separate revisions. Quality approves remaining limitations as well as a pass decision.

Example of a measurement system changing over time

Consider bore diameter measured across three shifts at a Thai machining plant. Quality defines the characteristic and drawing, then plans a study with operators representing the shifts. If operator variation is high, first check part positioning, cleanliness, measuring force and fixture seating before replacing the inside micrometer. Re-study after revising the procedure, retaining the old study. Compare before and after with comparable parts and the same denominator.

After a repair, a new calibration certificate arrives. Releasing the gage from the calibration register is a separate decision from continuing to rely on the earlier MSA. Quality decides the scope of reevaluation, including whether a gage family is affected. The system should find related characteristics and product lots so potential holds or extra inspections can be assessed.

If the night shift loses the machine connection and records on paper, distinguish measurement time from later entry time. Confirm the instrument and part, reconcile manual digits with the paper record, and deduplicate buffered device readings on recovery. If two readings conflict, route the exception to quality rather than selecting the convenient value. These scenarios belong in the RFP and SAT. Ask a vendor to demonstrate the state transitions and to answer an audit question without relying on one person’s memory, in each operating language.

Decide migration and exit requirements before purchase

Legacy spreadsheets may contain important context in another sheet or paper form. Classify complete studies, studies with only conclusions, and studies of uncertain provenance. Do not import all of them as approved evidence. Retain source-to-target mapping, counts, exclusions and verification signatures. Check units, decimal separators, names and date formats. Preserve sources read-only and record conversion scripts and logs. If the new engine differs from an old report, investigate equations, rounding and migration rather than deleting the old result.

Specify a machine-readable export of raw values, plans, analysis settings, decision rules, approvals and calibration references. PDFs alone cannot support recalculation or migration. Contract for data ownership, extraction at termination, attachments, audit logs, encryption keys and access after support ends. Keep an offline measurement and quality-decision procedure for outages.

Measure value beyond data-entry time: audit retrieval, investigation scope after a questionable gage, remeasurement, product holds and approval delay. Record a baseline for the same study and audit question, then compare under similar conditions. More completed studies are not the objective; more defensible measurement decisions are.

Decide what to do when numbers change after approval

Approval is not permanent. When new operators join, review training, competence and reading trends rather than automatically repeating every study. A one-direction bias in their readings calls for a check of posture and contact location. Define who decides when to restudy and the interim control. If calibration requires a large adjustment, search readings and lots since the last reliable check. Quality assesses risk using the characteristic, distance to limits, other inspections and ability to remeasure. Record unknown impact explicitly and set a hold scope.

A drawing revision may change the decision rule without changing a gage. Preserve the measured value, the original release decision and any later evaluation under the new specification separately. Application dates and customer change instructions determine which rule applies. A tolerance-based R&R ratio may need recalculation. When transferring production to another plant, compare benches, environment, fixtures and training rather than copying an approval. Use common parts and methods to test site differences before merging results.

If measurement stops, do not automatically condemn every part or assume unmeasured parts passed. Review alternate gages, downstream inspections and lot separation. Track product hold separately from measurement-system outage, and let quality approve restart. The screen should distinguish unmeasured, unable to measure, out of specification and decision on hold in the operator’s language. Keep missing results missing, with reasons. Rising exceptions can indicate equipment faults, unclear instructions, weak labels or unstable communication; investigate them jointly across quality, maintenance, production and IT.

When results feed process capability, record whether gage, location, frequency, part temperature and exclusion rules stayed comparable across process changes. Fewer control-chart alerts could mean either a more stable process or a less sensitive measurement. For certificates of inspection, verify that the approved study covers the actual gage and procedure. Preserve prior certificate revisions if a later correction changes the result. Assign production to methods, maintenance to gages, quality to plans and decisions, and IT to data integrity and recovery, including deputies and approval deadlines.

The average across operators can hide a person-specific bias that changes pass/fail decisions for parts near a limit. Define the intervention threshold, responsible reviewer and temporary inspection control. After a major gage adjustment, trace the period from the last reliable check to the adjustment, the readings made in that window and their product lots. If impact cannot yet be bounded, record that uncertainty and decide the smallest defensible hold unit instead of marking the scope clear. A drawing revision should trigger review of linked tolerance-based R&R studies. Keep the decision made at release separate from an informational re-evaluation under the new drawing.

When transferring production, check whether transport altered the common test parts and whether fixtures clamp them the same way at each site. An apparent plant effect may come from those differences. During an outage, distinguish the actual time a paper reading was taken from the later entry time; on recovery, reconcile it with buffered readings without turning missing values into zero. Later customer explanations must distinguish decisions supported by measured evidence from decisions made under temporary controls. If a study or reading is corrected after certificates of inspection have been issued, find the affected certificate versions and lots, preserve the previous versions, and record the date and content of any customer notice against quality approval.

Design alerts as a workflow, not a message counter. An imminent calibration due date does not mean a replacement gage is ready. An owner must identify the affected instrument and operation, schedule its replacement, quality must approve the substitute method, and production must label the old instrument stopped. Record each state separately. Escalate an instrument still on the shop floor at expiry even if the notification was marked read. If exceptional use is approved, limit it by part, period, extra inspection and approver.

Apply the same discipline to study renewal. A past due study does not automatically make historical products nonconforming. Find current operations relying on it and let quality assess impact and interim controls. Schedule operators and representative parts before the due date; data collection is not completion until analysis and approval finish. Record which inspection decisions temporarily relied on the old study. Changes to renewal dates require reasons and approval. FAT/SAT should exercise notification, action, approval and release, including deputy coverage during leave, duplicate and delayed alerts after outages, entry with an expired gage and detection of warnings nobody acts on. Review unresolved alert age after launch.

Monthly review must also observe shop-floor burden: when shadow spreadsheets persist, reduce duplicate entry or improve direct capture. Administrators must not be able to delete an unfavorable study without controlled history. During migration, reconcile source and target row counts, minima, maxima, blanks and duplicates, and independently recompute a representative sample. Keep the mapping and verifier. Before issuing inspection certificates automatically, confirm that the approved study covers the actual instrument and method; if it does not, route the case to quality instead of quietly releasing a report. Alerts need a named recipient, the screen they open, the decision they make and the next state, including a deputy and approval deadline.

Frequently asked questions

Does good gage R&R eliminate calibration?

No. R&R evaluates repeatability and reproducibility under stated conditions; calibration addresses relation to a reference. Check both status and validity for a quality decision.

Should software fix one universal GR&R pass threshold?

No. Version approved rules against customer requirements, characteristic, tolerance, purpose, decision risk and analysis method. Show the denominator and rule version on every report.

Can a destructive test use the same crossed R&R design?

Not if the same part cannot be measured repeatedly. Select a suitable nested or other design with statistical support, and confirm software support in the RFP and FAT.

Must a plant study every instrument?

The IATF FAQ allows a representative gage for a justifiable family with shared characteristics. Document the rationale, members and any customer-specific requirements.

What should the first deliverable be?

An identified characteristic and decision, approved plan, raw-event model, decision rule, link to calibration, independent calculation check and site SAT evidence. Make one decision traceable end to end before polishing dashboards.

Conclusion

A manufacturing MSA and gage R&R system is more than an automated statistical worksheet. Retain parts, people, instruments, methods, environment and raw readings; version study design, calculation and rules; connect calibration without conflating it with MSA. Compare these abilities in the RFP and demonstrate calculations and exceptions in FAT/SAT.

If the target characteristic or connection to legacy measuring equipment is still being defined, contact TOMAS TECH. We can help map the shop-floor decision into an RFP scope and verifiable acceptance evidence.

References