Back
Manutenção Industrial

ISO 14224: how to structure reliability data

P
PM Run Team
August 23, 2026

ISO 14224 provides a basis for collecting and exchanging reliability and maintenance data in a standardized format. Its official scope covers petroleum, petrochemical and natural gas industries. The standard organizes a common language for operating experience and addresses equipment, failure and maintenance data categories as well as quality practices. It does not turn incomplete records into trustworthy analysis or replace company technical governance.

Practical application starts with a data project inspired by the public objective of ISO 14224 without reproducing protected tables, codes or normative content. To claim conformity, interpret a requirement or implement the standard in its scope, an organization must consult the licensed current edition, define applicability and involve competent professionals.

The problem standardization solves

A plant may hold thousands of notifications and remain unable to compare failures. One shift records a symptom, another records a component, and another types an action into the cause field. Pump, pump set and transfer system appear as if they were the same level. Downtime starts at notification opening in one area and at loss of function in another. The volume looks substantial while the population is incoherent.

Standardization creates three conditions. First, every record references a technical object and defined hierarchy level. Second, failure mode, mechanism, cause, consequence and action occupy distinct meanings. Third, time and exposure follow documented rules. The result is a dataset that can be reconciled, compared, and audited.

What the official ISO page supports

The ISO page identifies ISO 14224:2016 as the published third edition and shows it as confirmed in the review cycle. Its public abstract describes a basis for reliability and maintenance data during operational life. It identifies three broad categories: equipment data, failure data and maintenance data. It also names uses in reliability, availability, maintenance, safety and environment.

The abstract also limits scope. Direct costs, complete equipment data sheets, laboratory testing and complete analytical methods are not the declared central focus. That boundary matters. A failure vocabulary does not mean the standard alone provides an economic model, an RCFA or a maintenance strategy.

Minimum architecture for a data project

1. Governance

Define sponsor, data owner, catalog administrators, class specialists and a review routine. The system owner does not decide technical meaning alone. A change in a term, required field or time rule needs a version, justification and effective date.

2. Equipment boundary

State where each event accumulates. A hierarchy may include site, unit, system, equipment and maintainable item according to the company model. The purpose is to prevent the same type of event from being assigned sometimes to a system and sometimes to a component without a rule.

3. Failure dictionary

Separate required function, functional failure, observed failure mode, mechanism, cause and effect. Mode describes how function was lost or degraded. Mechanism describes the physical process. Cause explains the condition that initiated or allowed the mechanism when evidence exists. Effect records what was observed or produced. An unknown remains unknown.

4. Maintenance data

Record work type, action, replaced item, resource, start and end under selected milestones, and post-work condition. The order shows what was executed. It must not receive a confirmed cause merely because a part was replaced.

5. Quality and reconciliation

Measure completeness, validity, consistency, timeliness, uniqueness and traceability. A complete record can still be technically wrong. Sampling must return to notification, order, measurement and field evidence.

A ten-step implementation method

  1. Define the use. Select a concrete decision such as comparing recurrence by class or revising a plan.
  2. Bound the population. Fix classes, locations, period, status and exposure.
  3. Map current fields. Locate where equipment, failure, action, time and consequence are recorded.
  4. Diagnose quality. Count blanks, free text, duplicates, mixed levels and impossible times.
  5. Create an internal dictionary. Define terms, examples, exclusions and owners without copying protected content.
  6. Build a crosswalk. Relate current field, internal meaning and future process destination.
  7. Configure and test. Apply the design to a pilot class and preserve raw values for audit.
  8. Train through cases. Ask users to classify anonymized real records and explain their decisions.
  9. Review disagreement. Adjust catalog, interface or rule when specialists classify the same event differently.
  10. Scale with control. Publish a version, monitor quality and prevent ungoverned changes.

Worked case: compressor group K-301

Every number and record in this case is didactic. It is not a customer result, benchmark, ISO requirement or PM Run performance claim. A unit has five equivalent compressors. Engineering wanted to compare failures that removed capacity, but records mixed symptoms, components and actions.

Pilot assumptions

ElementInternal case definition
Pilot classFive equivalent centrifugal compressors
Historical windowJanuary 1 through June 30, 2026
Maximum calendar exposure21,720 hours, calculated as 5 compressors × 181 days × 24 h
Hours outside eligible state3,480 hours of planned outages, event downtime, and other periods excluded by the pilot rule
Accumulated exposure18,240 hours in eligible state
Included eventLoss or degradation of compression function beyond the internal limit
Excluded eventPlanned outage and order without loss of function
SourceReconciled notification, order, process history and measurement
Time unitHours, with time zone and milestone defined

The diagnosis found 27 related records. After reconciling duplicates and scope, seven were functional events. Nine were planned orders, six were duplicate notifications and five described anomalies without loss of function. No record was deleted. Every exclusion received a reason and a link to the original.

Exposure was reconciled from the physical ceiling of 21,720 calendar hours. Documented exclusion of 3,480 hours left 18,240 hours in eligible state. The seven events totaled 26.6 downtime hours, averaging 3.8 hours per event. The descriptive rate was 7 ÷ 18,240 × 1,000 = 0.384 event per 1,000 eligible hours. With only seven events and different mechanisms, that rate did not support causal comparison or a strategy change; the decision remained grounded in semantic separation and evidence for each event.

Before standardization

Raw textField usedProblem
High vibrationCauseSymptom recorded as cause
Replaced bearingFailure descriptionAction recorded as failure
TripLong textEffect without function and detailed object
Mechanical failureModeCategory too broad for comparison
NormalizedCauseWork outcome without a mechanism

Internal pilot dictionary

The team did not attempt to copy a normative taxonomy. It created internal definitions and recorded that formal implementation depended on the licensed standard. For the pilot, function was to compress gas within the defined range. Functional failure was inability to maintain flow or pressure. Mode was the observed form of loss. Mechanism was the physical process supported by evidence. Cause remained unconfirmed when no investigation existed.

Didactic IDObserved modeSupported mechanismCauseActionDowntime
E01Protective tripBearing degradationLubrication outside internal parameterReplace and revise task6.2 h
E02Flow below standardStage depositContaminant under investigationClean and sample4.5 h
E03External leakageSeal degradationUnconfirmedContain, replace and open RCFA3.8 h
E04Protective tripConfirmed instrumentation faultDegraded connectionRepair connection and test1.7 h
E05Unstable pressureMechanism unconfirmedUnconfirmedCollect additional data2.1 h
E06Start unavailableInterlock actuatedValid process conditionRestore condition0.9 h
E07Flow below standardConfirmed internal clearanceAccumulated wearRepair and revise limit7.4 h

The table is an original didactic model. It does not represent ISO 14224 codes, required structure or taxonomy. Its role is to show semantic separation. The team retained raw text beside the controlled term so a specialist could audit the transformation.

Decision produced

The pilot showed that protective trip was not a causal family. E01 and E04 shared a trip effect but had different mechanisms and actions. Combining them under one cause would create a false priority. The decision was to revise the lubrication plan for the bearing family, open investigation for the unconfirmed seal event and keep instrumentation failure in a separate workstream.

The team also decided that mode and effect would be required in the notification for this class. Mechanism and cause could remain unconfirmed. Later completion required evidence and an authorized role. That rule protects history from invented certainty.

Path into SAP PM

In the case architecture, equipment and functional location establish the address. The notification receives condition, malfunction, mode, effect, detection method and dates according to configuration. The order records operations, materials, time and executed action. Permitted measurements and attachments support evidence. Catalogs organize language while technical text preserves context that does not fit a code.

SAP official documentation describes failure data in notifications, including mode, effect and detection method in compatible scenarios. The organization must verify version, process and Customizing. The standard does not automatically determine how every field is implemented in a company's SAP environment.

PM Run can support mobility, planning, execution and return of data over SAP PM. It does not provide a ready-made ISO taxonomy, declare conformity or replace master-data governance. SAP remains the system of record.

Indicators and verification window

The pilot was evaluated for 180 days and at least 30 classifiable events or anomalies. The longer criterion applied. Indicators included:

  • completeness of pilot fields;
  • unclassified percentage, followed without artificial redistribution;
  • agreement between two specialists in a sample;
  • event duplication after reconciliation;
  • time from detection to formal notification;
  • percentage of records traceable to an order and evidence;
  • events per exposure only after validating the population.

In the exercise, baseline completeness was 46%. Internal criteria were at least 95% completeness, unclassified below 5% without forced classification, duplicates below 2%, and 100% sample traceability. These are didactic targets, not ISO requirements or market benchmarks.

Select the discipline that answers the question

ISO 14224 addresses data structure and quality within its scope. Reliability engineering integrates decisions and strategy. FMEA anticipates modes and effects. RCFA investigates an observed event. MTBF summarizes a defined population. Each discipline answers a different question and may use the same governed history.

Limits

  • The official ISO 14224 scope is sector-specific and must be respected.
  • Use in another sector can inform design but does not create automatic conformity.
  • The public abstract does not replace access to the licensed standard.
  • A taxonomy without review governance only standardizes error.
  • A required field can increase completion and reduce quality when definitions are weak.
  • Plant comparison requires equivalent populations, exposure and rules.

Quality control before analysis

The K-301 pilot established four checks before any indicator was calculated. The first confirmed that equipment, position, and operating period formed a coherent population. The second searched for duplicate records of one event using time proximity, technical object, and description. The third checked the sequence from detection and shutdown through work start and return to service. The fourth compared the controlled term with original text and attached evidence.

Rejected records were neither deleted nor forced into a convenient class. They received a quality state, reason, and correction owner. This queue separated master-data gaps, semantic disagreement, and genuine absence of information. Completeness therefore measured usable fields rather than fields containing any value.

Agreement sample

Two specialists independently classified ten pilot records. Eight had full agreement, one differed between effect and mode, and one remained inconclusive because evidence was insufficient. Joint review corrected the internal definition, preserved the inconclusive state, and added a training example. An agreement target would be enforced only after calibration.

This control reduces a common risk: consistent charts built from inconsistent meanings. Before comparing units, a company should demonstrate that people recognize the same event, apply the same time boundary, and preserve unknown status when evidence does not support classification.

Change control for taxonomy and mapping

The pilot assigned an owner to every definition, catalog entry, and SAP mapping. A proposed change required reason, affected records, effective date, migration rule, and approval. Existing records were not silently rewritten because that would alter historical series without an audit trail.

Versioning also covered extraction logic. If one report treated a trip as an event while another merged it with a preceding derate, rates could not be compared. The data dictionary therefore recorded inclusion rules, exposure source, units, null treatment, and query version. A release note identified any series break.

For K-301, the team froze the pilot dictionary for 90 days, reviewed exceptions every two weeks, and scheduled a formal decision at the end of the 180-day window. Only approved definitions would move to the wider compressor population. This sequence protected analysis from uncontrolled catalog growth.

Official references

Frequently asked questions

Does ISO 14224 apply to every industry?

The official scope is petroleum, petrochemical and natural gas industries. Other operations may study its structuring principles but should not claim automatic conformity. Applicability, licensing, internal requirements and the standards of the relevant sector need review.

Does the standard provide every SAP PM code?

The standard and the system have different roles. An organization must design the mapping among its licensed taxonomy, process and SAP configuration. Copying terms without defining use, level and governance does not create comparable data.

Must every notification contain a cause?

A cause must not be invented to fill a field. Symptom, mode, effect, mechanism and cause have different meanings. When cause is unconfirmed, the unknown state must remain and can trigger investigation according to criticality.

To connect execution, planning and field feedback to SAP PM, learn about the PM Run operational layer. Taxonomy, conformity and engineering decisions remain the responsibility of the company.

ISO 14224
Reliability data
Failure taxonomy
SAP PM
Reliability engineering
Industrial maintenance
PM Run

Built for SAP.Not just adapted. Native.

PM Run connects planning, field execution, and supervision through native SAP integration, with no parallel spreadsheets, no re-entry at end of shift, and no data loss.

Used by leading operations in their sectors

Logo Volkswagen
Logo Eurofarma
Logo Saint-Gobain
Logo Marcopolo
Logo Moura
Logo Alpargatas

Back to blog