Failure codes: make maintenance history useful without inventing causes

Topics: maintenance failure codes; equipment failure taxonomy; reliability data coding; work order failure history.

A reliability engineer pulls 12 months of corrective work orders for multistage centrifugal pumps from the enterprise asset management (EAM) system to justify a mechanical-seal piping flush upgrade. Out of 120 closed work orders, 68 carry the failure cause Normal Wear and Tear and 29 carry Other / Miscellaneous—the first and last entries in a 90-item alphabetical drop-down that the EAM requires before a technician can record labour hours and close a shift ticket. When the engineer reads the free-text technician narratives on those 97 tickets, 34 describe inboard seal leakage, 19 describe high bearing-housing vibration and 14 describe suction strainer blockage, none of which could be diagnosed to a root cause before bench teardown.

The core design question is how to structure maintenance failure codes so that work-order history distinguishes immediate field observations from confirmed failure causes without forcing frontline technicians to invent explanations at shift sign-off. A practical failure-coding workflow aligns its hierarchy with ISO 14224, scopes drop-down lists by equipment class in the EAM (IBM Maximo rotating and non-rotating asset hierarchies), allows technicians to close field work with a verified observation plus an explicit Cause Pending Review / Unconfirmed state, and assigns cause confirmation to a structured post-teardown or reliability review.

Separate observation, mechanism and confirmed cause by role and timing

Most EAM failure-coding breakdowns happen because the software treats failure reporting as a single timestamped gate at work-order completion. In real plant maintenance, the technician who isolates a pump, swaps a cartridge seal or replaces a Damaged coupling element at 02:00 can accurately report what was observed (such as external process-fluid leakage or abnormal vibration), which maintainable item was affected (the mechanical seal assembly or thrust bearing), and what immediate mechanism was visible on removal (scratched seal faces or discoloured lubricant). Establishing why that mechanism occurred—whether process cavitation during startup, flush-plan plugging, shaft misalignment or batch elastomer incompatibility—often requires workshop inspection, operating-trend review or metallurgical analysis days later.

International reliability-data standard ISO 14224 separates failure data into distinct taxonomic levels that map naturally to different moments and owners in the work-order lifecycle:

By allowing the field technician to record an accurate Failure Mode (VIB — High Vibration) and Visible Mechanism (BRG-SPALL — Bearing Raceway Spalling or INSP-PEND — Removed Assembly Sent to Workshop) while setting Failure Cause to UNCONF — Under Reliability Review, the EAM stops polluting the database with forced guesses just to release a timesheet.

Walk a 120-work-order centrifugal pump audit before and after coding reform

Illustrative example. A reliability team audits 120 closed corrective work orders on process centrifugal pumps across two quarters, comparing the legacy single-step mandatory drop-down against a two-stage reviewed coding workflow during a 60-day pilot:

• Under the legacy configuration, 68 tickets (56.7%) were coded Normal Wear and Tear, 29 tickets (24.2%) were coded Other / Miscellaneous (68 + 29 = 97 tickets, or 80.8% of all 120 work orders), and only 23 tickets (19.2%) carried a specific failure mode and cause that agreed with the technician's written narrative and workshop teardown sheet.
• During the structured pilot on the next 120 centrifugal pump corrective work orders, the EAM failure hierarchy is filtered to Centrifugal Pumps (ISO 14224 Class: PU) and split into Stage 1 (field sign-off of maintainable item + observed mode + visible mechanism) and Stage 2 (reliability/workshop review of cause for critical and repeat events).

Track the disposition of the 120 pilot work orders across the two-stage workflow:

1. Stage 1 (Field sign-off across all 120 work orders):
   • 114 work orders (95.0%) receive a specific, narrative-consistent Maintainable Item and Failure Mode (52 External Leakage — Mechanical Seal, 28 High Vibration — Bearing/Coupling, 19 Low Flow/Head — Impeller/Suction, 15 Instrumentation/Auxiliary Tubing Fault, where 52 + 28 + 19 + 15 = 114).
   • 6 work orders (5.0%) have incomplete or ambiguous field narratives returned to the shift supervisor within 48 hours (114 + 6 = 120).
2. Stage 2 (Cause confirmation across the 120 work orders):
   • 44 work orders (36.7%) have directly evidenced on-site causes verified by the maintenance supervisor at sign-off (such as Blocked Seal Flush Strainer verified on disassembly or Loosened Guard Fastener verified visually).
   • 38 work orders (31.7%) involve cartridge seals or bearing assemblies sent to the workshop with Stage 1 cause set to UNCONF — Pending Workshop Inspection. Within 14 days, workshop teardown confirms the cause on 31 of those 38 tickets (25.8% of total), while 7 tickets (5.8% of total) remain Unconfirmed — Evidence Inconclusive After Teardown because secondary damage obscured the primary initiation point.
   • 32 work orders (26.7%) are minor non-repeat auxiliary repairs below the asset's formal root-cause investigation threshold and are explicitly closed as Cause Not Investigated — Below Threshold (Mode & Mechanism Recorded).
   • 6 work orders (5.0%) are the narrative-mismatch tickets corrected via versioned supervisor review (44 + 31 + 7 + 32 + 6 = 120).

Notice the analytical difference:

• In the legacy dataset, a query for mechanical-seal flush problems returned almost nothing because technicians selected Normal Wear and Tear to close the screen.
• In the structured dataset, all 52 mechanical-seal leakage events are immediately countable at the Maintainable Item + Failure Mode level regardless of whether a root-cause teardown was performed, and the 31 workshop-confirmed teardowns provide an uncorrupted denominator showing that 22 of the 31 inspected cartridge failures resulted from plugged API Plan 11/21 flush orifices rather than seal-face material selection. That separation of backlog status and verified history also strengthens maintenance backlog triage.

Govern taxonomy updates and AI coding suggestions without rewriting history

Two governance rules protect the long-term integrity of maintenance failure records:

1. Version taxonomy changes instead of remapping historical codes in place: If reliability engineering splits an older failure mechanism code (SEAL-LEAK) into two new codes (SEAL-DYNAMIC-FACE and SEAL-STATIC-ORING) on 1 November, apply the new codes prospectively from Revision 2 of the taxonomy. Never run a bulk SQL update that arbitrarily assigns historical SEAL-LEAK work orders to one of the new codes without physical evidence.
2. Keep AI coding assistants strictly advisory with cited narrative spans: Natural-language tools can read a technician's free-text remark ("Found inboard cartridge seal weeping; flush line cold to touch and strainer plugged with scale") and suggest Maintainable Item: Mechanical Seal, Failure Mode: External Leakage, Mechanism: Blocked Flush / Thermal Distress, together with the highlighted text span. However, the technician or reliability reviewer must explicitly accept or reject the suggestion, and the AI assistant must default to Unconfirmed — Under Reliability Review whenever the narrative describes only a symptom ("Pump tripping on high vibration; replaced coupling and aligned") without evidence of the underlying cause.

Ownership and observable acceptance measures

Assign clear roles across the coding lifecycle:

• Executing technician / craft lead: Records the affected maintainable item, observed failure mode, visible physical mechanism and a concise factual narrative of what was found and what action was taken.
• Maintenance supervisor: Checks at shift closeout that the selected maintainable item and failure mode match the work performed and that unverified causes are marked Pending Workshop/Reliability Review or Below Threshold.
• Reliability engineer / workshop inspector: Reviews assemblies routed for teardown or repeat-failure analysis, appends the confirmed failure cause with a link to the inspection report, and curates the class-specific failure hierarchy.

Measure the quality of your failure-coding workflow using four observable indicators:

1. Code-to-narrative agreement rate: Percentage of sampled monthly work orders where an independent reliability reviewer agrees that the recorded maintainable item, failure mode and mechanism match the free-text narrative and parts issued.
2. Catch-all bucket share: Combined percentage of closed corrective work orders coded as Other, Miscellaneous or unverified Normal Wear and Tear (target: near zero once explicit Unconfirmed and Below Threshold states are available).
3. Pending-cause closure rate and age: Percentage and median turnaround time of work orders marked Pending Workshop/Reliability Review that receive a documented teardown outcome (Confirmed Cause or Inconclusive After Teardown).
4. Provenance retention: Audit confirming that post-closeout reliability updates store the reviewer ID, date and evidence reference alongside the original technician entry rather than overwriting it.

Summary and next actions

Maintenance failure codes become trustworthy when software stops forcing technicians to invent root causes before an assembly has even reached the workbench. Separating observed symptoms and visible mechanisms from confirmed causes—and providing honest codes for pending teardowns and minor repairs—gives reliability engineers clean component-failure frequencies today and verified causal evidence where inspections actually occurred.

1. Audit 100 recent corrective work orders for one priority equipment class (such as centrifugal pumps, compressors or control valves) to measure catch-all usage and code-to-narrative agreement.
2. Restructure your EAM failure picklists by equipment class so technicians see only class-relevant maintainable items, 8–12 failure modes and clear physical mechanisms.
3. Add explicit Unconfirmed — Pending Workshop/Reliability Review and Not Investigated — Below Threshold options to the failure-cause field so field sign-off never requires guessing.
4. Establish a weekly 30-minute reliability and workshop review queue to close pending teardown causes with documented inspection references.

To scope a failure-history quality review for one asset family, bring your current EAM failure-code hierarchy and an anonymised export of 100 closed corrective work orders with their free-text narratives and spare-parts issues. The assessment will benchmark code-to-narrative agreement and design a class-filtered, two-stage coding workflow for your EAM.