SystemAsset Management, Maintenance & Reliability

Work Order Failure Coding

Work order failure coding that reads what technicians actually wrote, codes each note for what it says, and leaves blank what it cannot support.

Follow 2,318 bearing notes to the top of the list ↓
The problem

Brg fail in the note, the top code on the list, eight years running

The note is written at the end of a shift, and the code is whatever sits near the top of the pick list. One pump failure opens several work orders, and each one counts. The teardown and the oil sample that say what really happened sit as attachments nobody opens.

So the bad-actor list ranks what was written down. A problem spread thinly across a unit’s pumps never reaches the top, and the reliability engineer who suspects it would need years to read the notes by hand.

Built forReliability engineersSenior reliability engineersProcess safety engineersMaintenance supervisorsMaintenance data analysts

The Problem Today

Technicians write brg fail and pick any code, so failure history cannot say what really broke.

13,756
hours to read 412,680 sample notes by hand, at 2 minutes each our estimate

Pieced together from notes today

  • Notes
  • Teardowns
  • Oil samples
  • Old codes
  • One failure, many ordersFirst call, oil change and rebuild all count
  • Proof sits elsewhereTeardowns and oil samples never reach the order

The Solution

  1. Notes and reports readShorthand, teardowns and oil samples read; routine work set aside
  2. Coded by equipment classPart, mode, mechanism and cause, each tied to its words
  3. Cannot code, with the reasonWhen a note holds nothing, the agent says so instead of guessing
  4. Confirmed on a sampleAn engineer reads a random sample; a senior engineer checks others

Built-in controls

  • Safety equipment goes to a named person
  • The technician’s note stays as written
  • Any record can be reopened, any wave reversed

What You Get

Reliability teams see what really fails, and the bad-actor list stops ranking what was written down.

  • Bad actors counted by failureThe list changes only after two people agree
  • New closeouts coded on the shiftA short question reaches the crew while they remember
  • Better notes, crew by crewPlain tips by crew, not by person
One family of notes, one morning

Work order failure coding on 2,318 coker pump notes that all say bearing

A refinery has eight years of work orders across fourteen units. Its delayed coker has 78 pumps, and their notes keep saying bearing. Here is the morning a reliability engineer and a senior engineer settle what those notes really mean.

demoThe refinery, its pumps, the people and every count are sample data.

  1. 01Morning

    412,680 work orders, and the words on the left lead to codes on the right

    Reliability engineer

    The sorting line draws what technicians wrote against the failure mode, mechanism and cause the agents propose. 168,240 orders were routine, with no failure. Of the 244,440 failure orders, 194,350 have a proposed code, 4,262 are held for a person on safety equipment, 21,386 are not sure yet and 24,442 cannot be coded.

  2. 02Morning

    5,674 pump notes that say only fixed, and none of them guessed

    Reliability engineer

    Narrowed to pumps, the bottom ribbon is the honest one. Notes that say fixed, ok, or same as last time go to cannot code, with the reason, and each row opens the order as it was written. The agent does not guess from an empty note. Those under a year old can still be asked about.

  3. 03Morning

    A gas detector note says false alarm, and the agent refuses to code it

    Process safety engineer

    Relief valves, shutdown valves and detectors go one at a time. A hydrogen gas detector note says false alarm, sensor poisoned. A poisoned sensor can read normal and miss gas, so the agent will not call it a false alarm. It asks whether the detector was bump tested and responded, and leaves impact, severity and detection for her.

  4. 04Unit 31

    2,318 bearing notes, old codes split three ways, and the trace behind them

    Reliability engineer

    At the coker pumps the old codes on 2,318 bearing notes are bearing fail, other and mechanical failure, as they stood. The trace beside them shows what the agents read: the notes, 1,104 teardown reports and 312 oil samples. They joined 1,940 orders into 418 failures, then left 137 for a person because they cannot be told apart.

  5. 05The family

    1,286 orders where the proof sits in the note or the attachment

    Reliability engineer

    The family holds the notes with proof: seal, flush, grit, black oil. Each note keeps the mode its own words give. Noise is noise, hot is overheating, a re-align is a follow-on job. The first failure of each is on the seal, external leakage, with the bearing as the dependent failure. Twenty-four of the 25 sampled notes are already read.

  6. 06The family

    Vanadium at 38 ppm in the bearing oil, at a level no bearing makes

    Reliability engineer

    The one note left to read says brg fail, noise and hot, oil in the housing black. The agent cites four things: the teardown, the work order on the seal leak, the flush flow sitting at 1.4 gal/min against a 4 gal/min minimum, and an oil sample. Opened, the oil sample shows vanadium 38 ppm and nickel 14 ppm, coke metals in a bearing housing.

  7. 07The family

    25 of 25 read and agreed, so the family can be confirmed

    Reliability engineer

    With the last row agreed, the engineer confirms the family. The dialog says plainly what this is: 1,286 work orders confirmed on a sample of 25, and nothing moves until another engineer has checked. He picks his reason from the list and can add a note for the audit trail.

  8. 08Second check

    The senior engineer answers three rows of her own, outside his 25

    Senior reliability engineer

    The second check is not a glance at his work. The senior engineer gets three rows the first engineer did not read, opens each one and agrees or disagrees, with the note, the codes and what the agent read beside it. Only then can she countersign or send it back.

  9. 09Second check

    Countersigned, and 1,286 codes move to the Unit 31 wave

    Senior reliability engineer

    She countersigns. The banner says who confirmed, who answered three rows and what moved: 1,286 codes to the Unit 31 wave for loading. A reopen button sits beside it, because any record can be reopened and any wave reversed.

  10. 10After

    Seal flush at the coker goes from nowhere to first, at $9.4M

    Senior reliability engineer

    The bad-actor list is recounted only after two people agreed. The coker seal flush takes first place across 52 pumps, and FCC slurry pumps fall to second at $8.9M. Per pump it is far smaller than an FCC slurry pump, and the screen shows that too. A further $4.8M waits on the timing family.

Where 244,440 failure work orders landed

Every failure order gets one of four outcomes. None of them is a guess.

Proposed code, with citations194,350
Held for a person: safety equipment4,262
Not sure yet, under the confidence line21,386
Cannot code, with the reason24,442
Failure work orders244,440

demoSample data, from the working solution.

What the bearing notes were hiding at the coker pumps

The bearings were the end of a chain

A blocked flush filter starved the seal, hot oil leaked along the shaft, steam carried it past the isolator and the bearing failed. The technician wrote brg fail, and he was not wrong.

About 4.6 work orders to each failure

First call, oil change, bearing job, shop rebuild, re-align. 1,940 orders turned out to be 418 failures, about one per pump per year.

A problem spread too thin to rank

No single pump was near the top, because the cost was spread across 52 tags. Counted as one problem, it leads the list.

demoSample data, from the working solution.

Behind the codes

Failures, families and crew notes behind each code

Six screens the reliability team and the supervisors work from.

418 failures, each with its ordersOrders joined by pump and date into failures, the first failure on the seal, and an undo on each join.
What the words were hidingFour patterns, each a chain with its citations and what does not fit. The seal flush chain is confirmed by two people; the others wait.
Every recode in the familyAll 1,286 orders, old code to new, page by page. Open any one to correct it or reopen it, or export the lot.
Families waiting to confirmLargest families first, each with its sample size and clock. 435 families, 21,386 not sure yet, 3 waiting for a second check.
Good notes, by crewHow many notes are good enough to code, crew by crew, and a tip drafted for the supervisor to send, by crew only.
Audit trailEach read, confirmation, countersignature and agent step on its own line, from the first read of the notes to the recounted list.

demoSample data, from the working solution.

Who confirms a family of failure codes, and who checks the safety items

Reliability engineer
Sees
The family’s proposal, its random sample and what the agent read for each row
Signs
Confirms the family after reading the whole sample
Senior reliability engineer
Sees
Rows of her own, outside the first engineer’s sample
Signs
Countersigns or sends back
Process safety engineer
Sees
Each safety equipment order, one at a time, with the question that decides it
Signs
Decides impact, severity and detection herself
Maintenance supervisor
Sees
Crew trends and questions for the shift
Signs
Sends every question and tip to the crews
Maintenance data analyst
Sees
Load waves, rejects and the reversal file
Signs
Prepares the write-back, which needs two signatures

Calls the agents leave to the engineers and the crews

Coding safety equipmentRelief valves, shutdown valves, detectors and trip transmitters go to the process safety engineer, one at a time.
Guessing on an empty noteA note that says fixed is marked cannot code, with the reason, and the crew can be asked.
Declaring a root causeThe pattern is handed to a root cause analysis. The agent counts what fits and what does not.
Changing the noteWhat the technician wrote stays as written. Codes sit beside it.
Writing to the maintenance system aloneA load wave needs two signatures and comes with a reversal file.

What reliability engineers ask about work order failure coding

What is work order failure coding?

Work order failure coding means giving each failure work order a part, failure mode, mechanism and cause that its note supports. Here agents read the notes, teardowns and oil samples, propose codes from your equipment class lists with citations, and say when a note cannot be coded. Engineers confirm families on a sample and a second engineer checks.

Which failure codes does it use?

Your own licensed code list, loaded by class. The proposal stays inside the class list, and a value the note does not support stays blank, with the reason.

How does it improve maintenance work order data quality going forward?

New closeouts are coded as they land. When a note is too thin, a short question goes to the crew through the supervisor while they still remember, and crews get plain tips on what a good note holds.

Does free text work order classification count one failure many times?

No. Orders for the same failure, a first call, an oil change, the bearing job, the shop rebuild, are joined by pump and date into one failure, and any join can be undone.

How does bad actor analysis change?

The list is recounted only after a family is confirmed and countersigned. You can switch between the list as first coded and as it stands now, with failures, orders, cost and cost per tag on each row.

Can the codes be loaded back to the maintenance system?

Yes, in waves. Each wave is test loaded, its rejects held, and it needs two signatures. A reversal file goes with it.

Bring one unit’s work orders
and the notes nobody reads.

We’ll read them with you, code a family, and show what the old codes were hiding.