Specialist educational guide for translators, manufacturing engineers, digital-twin developers, maintenance teams and technical editors. Original teaching examples; not copied standards text.
How do you translate manufacturing digital twin documentation without turning a simulated result into a measurement from a real machine? Translating digital twin systems, equipment-state dashboards, production data dictionaries and engineering handover records requires careful distinctions between physical assets, digital representations, observations, predictions and commands. This guide uses ISO 23247-1:2021 as its technical context and concentrates on preserving those distinctions across languages.
A plant engineer may read “predicted spindle temperature” while a localized dashboard displays “spindle temperature.” That one missing adjective can make a forecast look like a sensor reading. A maintenance team might stop the wrong line; an auditor might treat a model estimate as test evidence. Effective digital twin translation identifies the data source, timestamp, model version, uncertainty, event type, machine identity and owner before choosing fluent target-language wording.
In brief: distinguish physical equipment from its digital representation; identify real-time, delayed, simulated and predicted values; protect identifiers, units, timestamps and version states; never translate recommendation as automatic control; keep uncertainty and data quality visible; test the localized interface with engineering reviewers; and use this specialised guide alongside the established translation master rather than as a replacement for it.
What ISO 23247-1:2021 actually addresses
ISO 23247-1:2021 is the first part of a digital twin framework for manufacturing. The official ISO catalogue describes it as an overview and statement of general principles, including terms, definitions and framework requirements. It is not a declaration that every commercial “digital twin” has the same sensor architecture, latency, cybersecurity controls or validated predictive capability. Those details depend on a particular implementation and other applicable specifications.
This difference matters to translators. A technical brochure can say a project is “aligned with the ISO 23247 framework” without proving that an algorithm is validated for safety-critical action. A translation must not promote descriptive framework alignment into certification. Nor should it invent a standard-required value for refresh frequency, accuracy, model fidelity or intervention timing when the source does not supply one. Read the project specification, not the standard’s name alone.
Think of the standard as a map for discussion. It helps teams ask whether they are discussing a manufacturing element, a digital representation, data connections or services around that representation. The map cannot tell a linguist whether “current” in a particular data column means live at this second, most recently received or approved as the current engineering revision. That meaning must come from the source system and its owner.
Start with the thing represented, not the attractive term
A digital twin discussion may involve the machine on the workshop floor, a product moving through production, the production process itself, a workstation, an instrument or a line. Translators should ask what the model represents before translating a noun such as “asset,” “component,” “state” or “cycle.” The word “cell,” for example, can refer to a manufacturing cell, a battery cell or a spreadsheet cell. Its correct target language changes with the domain.
The representation can contain geometry, configuration, rules, recorded history, current data and simulated or inferred information. These are not interchangeable categories. A three-dimensional CAD model without relevant behaviour or data connections should not automatically be described as a continuously synchronized twin. A simulation using yesterday’s inputs should not be described as an observed condition today. Each claim requires evidence in the project description.
A productive intake interview asks six questions: What physical or operational element is referenced? Who supplies the data? How recent is the value? Was it measured, calculated or predicted? What decision may a reader make? What supporting record would verify that decision? If even one answer remains unknown, the target wording should avoid a stronger claim. A translation that preserves an uncomfortable qualification is more useful than a smooth sentence that invents certainty.
Distinguish observation, model output and instruction
Manufacturing systems communicate several kinds of statements: a sensor observed a value; a software model estimated an unobserved condition; a forecast predicted a future condition; an analytics rule raised an alert; a human operator approved work; or a control application issued a command. A translator who confuses these speech acts can change the authority of the message.
Take “the model recommends inspection.” This is advice from an analytical component, not proof that inspection is scheduled, authorized or completed. “The equipment was stopped” is an event claim requiring a record. “Stop the equipment” is an instruction that might be dangerous without operating context. The English forms are similar enough to be collapsed in some interfaces, yet the resulting safety implications are entirely different.
For every status field, preserve what generated the statement and what it permits. An explanatory label may add a brief qualifier such as “model estimate” when the underlying design provides that meaning; it must not add a new status invented by the translator. Where the system offers a fixed enumeration of values, confirm allowed values and their definitions with a software owner. Free-text naturalness is a poor substitute for respecting a coded state machine.
Protect time as rigorously as numbers
“Live” and “real-time” are frequently used as marketing shorthand, but an engineering reader needs to know whether a source is streamed continuously, polled periodically, buffered, synchronized at a defined interval or replayed from a historical record. A displayed value may be correct for its sampling instant and already outdated when the operator sees it. Translations must preserve terms such as sample time, receipt time, display time, event time and last synchronized time.
Keep time zones, daylight-saving context and timestamp formats unambiguous. An apparent time of 08:30 can mean the operator’s local time, site time or UTC, and a facility working across countries cannot safely guess. If the source stores UTC and renders local time, that relationship should be documented outside the translated label. Preserve the underlying machine timestamp in structured exports. Translate explanatory headings, not the data in a way that changes ordering.
When a dashboard says “last known good value,” it may not be a current measurement at all. Do not reduce this to “latest value.” The adjective “good” can indicate that a quality check passed at a past time; “known” warns that more recent updates could be absent. Readers need that distinction to avoid false confidence in a stale, although historically valid, reading.
Data lineage is a language problem
A value labelled simply “temperature” is often insufficient. It might originate from a thermocouple, an inferred process model, an imported laboratory report, a calculation using multiple points or a manually keyed observation. Each path has distinct quality and governance implications. Good localization makes the provenance accessible without forcing an operator to decode a technical database.
Build a term table with a source identifier, equipment tag, measured quantity, unit, origin, collection method, transformation, quality flag and approved target wording. Avoid silently renaming stable identifiers for aesthetic consistency. For example, tag MTR-07 may continue to be MTR-07 even when its surrounding label is translated. If a displayed number has been normalized or filtered, the translated description must not imply that it is an untouched raw sensor value.
Data lineage also includes version history. An engineering revision may supersede an old model, but its underlying historical observations do not magically become observations from the new version. Localized reports should distinguish “produced by model version 2.1,” “reviewed against version 2.1,” and “recalculated using version 2.1.” Every expression describes a different action, and a qualified reviewer should be able to trace which one occurred.
Keep units and thresholds coupled
Factories cross national conventions. Temperature can be expressed in °C or °F; pressure in Pa, kPa or bar; velocity in m/s or mm/s; energy in kWh; vibration in a context-specific quantity. Conversion changes numbers and may affect the significance of rounding, limits and tolerances. A translator must not treat unit conversion as a harmless copy-edit when operating thresholds are involved.
Translate the unit label, if needed, but ask the engineering owner whether the measurement is to remain in its original unit or be converted under a controlled localization specification. A threshold of 80 °C is not 80 °F. A change from gauge pressure to absolute pressure is not a mere change of notation. Even when arithmetic is easy, reference conditions and measuring conventions can make a direct conversion misleading.
For threshold messages, retain the operator, comparison and trigger basis: above, at or above, below, below or equal to, exceeded once, exceeded for ten consecutive seconds, or within a moving average. “Peak pressure exceeded threshold” is not the same as “average pressure exceeded threshold.” The decision can depend on which statistic is involved. Translation review should test numerical examples directly against the implemented logic.
Versions, revisions and approval
A twin’s geometry, behaviour, configuration, software package, training data and production recipe may all have separate versions. Saying “current model” without indicating which object was revised is often inadequate. Documented workflows should distinguish draft, under review, validated, released, deprecated and archived statuses where those states exist. They are not universal ISO categories to impose on every system.
A translation should preserve the difference between verification and validation. In common engineering practice, verification asks whether specified requirements are met; validation considers fitness for intended use in a defined context. Neither word should be substituted casually for “tested” or “approved.” A model may pass unit tests without being approved to support maintenance decisions. Localized status text must not close that gap by tone alone.
Release notes require particular care. A patch may fix display formatting but not change the numerical prediction. A model update may change a prediction even if the software interface looks identical. Separate what was changed, what was checked and what remains unknown. Where original documentation is ambiguous, open a question with the product owner instead of resolving a governance decision in translation.
A clear translation workflow for the manufacturing team
Begin with an inventory. Collect screen strings, tooltips, alarm catalogues, engineering drawings, data dictionaries, workflow diagrams, release notes, sample reports and training materials. Identify the reader for each component: operator, maintenance engineer, integrator, auditor, manager or external customer. A text that is appropriate for a dashboard may not supply enough context for a contract deliverable.
Next, map every label to a field definition. Protect immutable identifiers and placeholders. Confirm plurals, gender, line wrapping, non-Latin characters and character limits in actual UI builds, not just spreadsheet previews. Then compile a bilingual glossary of equipment names, data-quality states, model outputs and control verbs. Assign a responsible specialist to each high-risk term. A translator should not be left to arbitrate the difference between “model confidence” and “instrument accuracy.”
Review the translation by reconstructing the source meaning: give a bilingual engineer only the target-language screen and ask them which asset, time, status and action they infer. Compare their answer with the system design. Run this exercise with both normal and fault states. Finally, test exported PDFs, mobile layouts, touchscreen truncation and accessible text. The best translation in a document may still fail once the UI clips the word “not.”
Worked examples: how an apparently small change becomes an operational error
The examples below are invented learning scenarios. They do not reproduce proprietary standards clauses or recommend changing live control logic. For each case, the issue is not merely which dictionary entry sounds most natural. The question is what a reader would believe, what evidence supports that belief and whether the translated text permits an action that the original text did not.
Consider every case as a three-stage exercise. First identify the factual claim and authority. Second draft a clear phrase in the destination language that maintains its bounds. Third ask another reader to state what happened or should happen. If their answer differs from the original system state, the translation needs more work. Where a source lacks critical context, mark a query for the system owner rather than pretending to know.
Case 1: Equipment identity
Source situation. A dashboard labels an asset “Press 04 — cell B,” but its export includes only “Press 04.” This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Retain the cell context where necessary because another plant may have a Press 04. Confirm the unique asset identifier and naming convention before shortening labels. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. The reviewer must distinguish a readable display name from the primary equipment key; an appealing short phrase must not silently merge two machines. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 2: Sensor versus model
Source situation. A table displays “measured bearing temperature” beside “estimated bearing temperature.” This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Keep measured and estimated visibly different, including on tiny mobile cards and in legends. A model-derived estimate is not a direct reading. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Test by asking the operator which column corresponds to a physical measurement and which depends on assumptions or a model. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 3: Forecast horizon
Source situation. A plot title says “predicted vibration, next 30 minutes.” This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Preserve both the predictive character and the time horizon; translating it as “vibration trend” may hide the future-looking claim. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Check the x-axis and reference time, because a 30-minute forecast starting at an older timestamp is not necessarily the next 30 minutes from now. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 4: Data quality
Source situation. A sensor reading carries a “suspect” quality flag after a communication dropout. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Do not translate the value as verified, normal or safe. The doubtful status must travel with the number. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. If the UI displays the reading without the quality flag in another language, this is a product defect to escalate, not stylistic smoothing. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 5: Stale value
Source situation. A tile shows “last valid reading at 14:02” after a 14:18 network outage. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Retain the historical timestamp and do not render the tile as a current condition simply because it is the latest displayed value. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. A bilingual acceptance test should disconnect the data source and verify that both locales communicate staleness equally clearly. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 6: Simulated trial
Source situation. An engineer records “simulation indicates lower cycle time” before a physical trial. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Do not rewrite the statement as “cycle time reduced.” The calculation is not an observed production outcome. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Request the modelling assumptions and baseline. An improvement measured in a simulated scenario has different evidence from a verified shop-floor change. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 7: Digital command
Source situation. The interface shows “stop command queued,” then “stop command executed.” This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Maintain distinct queued and executed states. Queue acceptance does not prove that equipment stopped. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. A reviewer should trace acknowledgements from the actual control system rather than allowing a generic “stopped” translation to span all states. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 8: Maintenance advice
Source situation. A condition-monitoring module recommends inspection at the next planned shutdown. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Retain “recommendation” and the planned timing. The source is not an immediate shutdown instruction. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Ask the maintenance owner whether the suggestion is advisory, mandatory under a site rule or a system-enforced interlock; the translator cannot decide. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 9: Model revision
Source situation. A forecast was produced under model v1.8, while the current dashboard runs v2.0. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Keep the provenance v1.8 in historical exports. Avoid replacing it with the presently installed version in translated captions. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Historical evidence should remain reproducible, and the receiving engineer must know which software assumptions generated each result. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 10: Geometry revision
Source situation. A work instruction refers to CAD revision C and a digital assembly representation revision D. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Preserve each revision label with the correct artefact, rather than describing both as the “latest model.” The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Check whether a revision denotes geometry, simulation code or manufacturing recipe; labels can resemble one another while referring to separate objects. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 11: Confidence estimate
Source situation. A model outputs “confidence interval” for a predicted parameter. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Do not transform it into “accuracy guaranteed.” The meaning of the statistical interval depends on the model and method. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Consult the data scientist for the interval’s definition and population assumptions before writing friendly explanatory text. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 12: Error code
Source situation. Machine state 0x0A is “communication lost,” not “machine fault.” This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Keep the stable code and translate its controlled definition. Loss of communication does not establish a physical failure. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Check the fault dictionary and any nested cause code. Mistaking communication loss for mechanical damage can trigger unnecessary maintenance. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 13: Threshold boundary
Source situation. An alarm fires when temperature is greater than 85 °C, not at 85 °C. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Preserve the strict comparison. “At least 85 °C” changes the boundary and may misdescribe the logic. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Test 84.9, 85.0 and 85.1 in an offline specification example; the target label should match the intended outcomes. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 14: Averaging period
Source situation. A report presents a five-minute rolling average of motor current. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Keep the window length and the fact that it is an average. Replacing it with “instant current” misstates the measurement. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Technical reviewers can check the sample definition and whether the averaging window is trailing, centered or another calculation. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 15: Site time zone
Source situation. A global production dashboard labels 07:30 UTC and 15:30 Singapore time. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Keep the zone designators with both values. Do not simply translate “morning shift” across plants with different shift calendars. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Reconcile the displayed time with the underlying event timestamp, including daylight-saving changes elsewhere. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 16: Scenario comparison
Source situation. A model compares “baseline scenario” and “proposed configuration.” This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Do not call the proposed configuration an installed upgrade. It is a hypothetical or planned alternative until evidence says otherwise. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Ask whether the baseline is a historical measured dataset or a modelled reference; the label must not erase that distinction. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 17: Equipment availability
Source situation. A chart shows “availability 92%” for a defined production interval. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Do not turn this into a guarantee of 92% uptime or safety without the denominator and exclusion rules. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Review which stoppages were included, what observation interval was used and whether planned maintenance was excluded. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 18: Manual override
Source situation. The status says an operator overrode an automatic recommendation. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Protect the direction of authority: the human made the change; the model did not autonomously actuate it. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. An audit should verify who initiated the override, when it occurred and whether the displayed state refers to requested or completed execution. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 19: Uncalibrated input
Source situation. A measurement is accepted into the model but its sensor calibration has expired. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Do not translate accepted-for-processing as valid-for-quality-assurance. Different systems can allow processing of doubtful input. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. The correct label should reveal the status without implying a certification or conformity decision. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 20: Digital shadow
Source situation. A vendor brochure uses “digital shadow” for one-way data flow from machine to model. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Avoid silently replacing it with a two-way synchronized digital twin if the project’s terminology depends on direction of flow. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Ask the architect what integration actually exists and document that answer in the termbase instead of correcting the product’s marketing in translation. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 21: Missing data
Source situation. A model imputes missing production observations using a defined rule. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Label imputed or estimated values rather than translating them as collected readings. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. A reviewer needs to know the fraction of data filled and what method did so, because that can affect interpretability. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 22: Process completion
Source situation. A workflow shows “inspection report uploaded” but not “inspection approved.” This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Keep document receipt apart from authorization. Uploading a file proves neither technical sufficiency nor a sign-off. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Test the UI when upload succeeds but review fails; both language versions must still distinguish those outcomes. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 23: Remote asset
Source situation. A diagnostic session names a remote machine and local operator station. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Protect machine location and operator location separately. “Local” can refer to the workstation, plant or network segment. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Ask whose perspective defines “local” and whether the translation will be displayed at multiple sites. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 24: Alert severity
Source situation. An analytics package labels a forecast “warning,” while the safety system labels “critical alarm.” This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Keep the source subsystem and configured severity. A dashboard prediction is not automatically an equipment safety interlock. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Do not invent a harmonized color or severity term until the engineering governance owner approves its semantics. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 25: Supplier handoff
Source situation. A vendor exports a digital twin package marked “for evaluation,” not “approved for production.” This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Preserve the intended-use restriction even if the receiving team is eager to deploy. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Acceptance of a file package does not establish validation, verification, configuration control or permission to run its outputs in production. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 26: Model retirement
Source situation. A report refers to a retired algorithm maintained only for historical reproducibility. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Translate retired and read-only accurately, without suggesting that past results were retroactively invalidated. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Readers should know whether retirement means unsupported future use, revoked acceptance or merely replacement by a better model. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Case 27: Security restriction
Source situation. The API documentation labels a field “restricted—maintenance personnel,” not publicly shareable. This detail matters because the receiving reader may not see the original system interface or supporting diagrams when a text is forwarded, exported or republished.
Translation decision. Keep the access restriction with the correct role and scope across languages. The translated sentence should allow a technically literate person to identify the same object, same event, same evidence level and same practical limitation as in the source.
Review and recovery. Localization exports can leak protected operational identifiers; a release reviewer should check both text and access metadata. Record any unanswered technical question beside the string, appoint the responsible engineer and retain the approved answer with the revision. This is preferable to converting a question about system behaviour into an unsupported editorial assumption.
Frequently asked questions
Does ISO 23247-1:2021 guarantee digital twin prediction accuracy?
No. The public catalogue describes framework principles and requirements, not a universal performance guarantee for every algorithm. Prediction accuracy depends on the real implementation, training and validation evidence, data quality and conditions of intended use. A translator should preserve what the specific vendor or test record actually says.
Is every 3D model a manufacturing digital twin?
Not automatically. A geometry file can be a valuable model, but an integrated digital twin project may contain data connections and behaviour or services absent from a static CAD representation. Ask what project-specific meaning is intended before using a stronger term.
How do I translate “live” when data arrives every few minutes?
Ask for the system’s update policy and latency expectations. “Most recently updated” or a label with the actual refresh time may be more informative, but the approved wording must match the implemented source definition.
Should codes and machine IDs be translated?
Generally keep stable machine identifiers, engineering tags, schema keys and error codes unchanged while translating adjacent explanations. Any localized aliases should map deterministically to the same underlying identifier.
Can AI translation tools handle these records?
They can assist with drafts and terminology discovery, but generated text must be checked against approved technical definitions, source metadata, version context and UI behaviour. Never upload confidential plant diagrams or live operational records without authorization and suitable data controls.
What is the single best check before release?
Ask a bilingual domain expert to use only the translated artefact to identify what was measured, inferred, approved and commanded. If the reviewer draws a different conclusion from the source specification, the text has not passed.
A four-week practical study plan
During week one, build an asset and data-flow diagram from a fictional workstation: one physical press, two sensors, a model, a dashboard and an operator. Label each observation and model-generated claim, then translate the labels into your target language. Require another reader to reconstruct the data flow from the translation alone.
During week two, create a terminology sheet with thirty entries. Separate machine identity, process states, timestamps, simulation outputs, control verbs, uncertainty terms, alerts and version labels. Attach a short usage note and one prohibited mistranslation to every high-risk entry. Revise based on actual counterexamples rather than merely choosing shorter words.
During week three, localize a miniature screen flow containing normal, degraded and fault states. Check that “not connected,” “data stale,” “model unavailable,” “manual override” and “command awaiting acknowledgment” remain distinguishable when displayed in a narrow panel. Ask the engineering reviewer to identify a safe next action for each state, then record any ambiguity.
During week four, assemble a sample release pack with bilingual screen captures, a versioned glossary, decision log, outstanding queries and evidence of a second-reader test. Demonstrate how the same process would handle a corrected sensor label or a changed prediction algorithm without rewriting past records as if they belonged to the new release.
Conclusion
Manufacturing digital twin translation succeeds when an engineer in another language can tell what is real, what is modelled, how fresh the information is and who has authority to act. Correct terminology matters, but the larger accomplishment is preserving the route from data to conclusion without inventing a measurement, endorsement or command. Clear phrasing helps operators use the system; documented qualifications prevent clarity from becoming false certainty.
Official source and connected learning
Official ISO catalogue: https://www.iso.org/standard/75066.html (consulted 5 November 2026). This article is independent educational commentary based on the public scope and original examples, not a reproduction of the protected standard or an assurance of conformity. Organizations should consult authorized technical specifications and their accountable experts before operational release.
General translation architecture: https://edukatesg.com/2026/09/18/master-art-of-translation-complete-system-moving-meaning-between-languages/ (EDKSG-TRANS-MASTER-WORLD-000).
Vocabulary Learning Hub: https://edukatesg.com/vocabulary-learning-hub/ . How English Works | Translation and Meaning Across Languages: https://edukatesg.com/how-english-works-translation-and-meaning-across-languages/ . These resources supply the wider learning foundation; this article remains a focused manufacturing application.