botminds.aiRequest a demo
WorkflowAsset Management, Maintenance & Reliability

Asset Master Data Cleansing

Asset master data cleansing that fills each equipment class from the pages behind it, and leaves a value blank, with a reason, when no page shows it.

See the demoBook a demo
Walk the pump batch, 09:10 to 09:31 ↓
The problem

Asset master data that holds a tag, a description and little else

Most equipment records carry a tag and a short description. The impeller size, casing rating, seal plan and material class were never keyed, though they sit on datasheets, manuals and nameplates the plant already holds. One maker is spelled a dozen ways, and equipment long gone still raises planned work.

The trigger is usually a system move. The new system will not load a record with required fields empty, so the last weeks go to defaults, checked by whoever keyed them. After go-live those defaults look exactly like facts.

Built forMaster data leadsData reviewersEquipment stewardsMaintenance leadersInternal auditors

The Problem Today

Registers hold tags with little behind them, so system moves stall and defaults look like facts.

25 years
of one person keying today’s blanks by hand, at 2 minutes each our estimate

Keyed by hand today

  • Datasheets
  • Manuals
  • P&IDs
  • Nameplates
  • One maker, many spellingsNobody can count what they own or buy for it
  • Ghost recordsEquipment long gone still raises planned work

The Solution

  1. Read the pages behind each tagDrawings, datasheets, manuals, plates and repair records
  2. Fill by classRequired attributes from the page, or blank with a reason
  3. Read the sampleA second person reads values before seeing the proposal
  4. Fix the rule, rerunOne bad sample sends the whole batch back

Built-in controls

  • A reviewer never reads a batch they made
  • Safety values allow no wrong reading
  • Agents never write to the register

What You Get

The next system move loads records with a page behind them, and each blank carries its reason.

  • One name per makerSpellings merged once a steward agrees
  • Ghosts walked before retiringA sample walked, safety records in full
  • Blanks with a reasonA size no page shows stays empty
One batch, by the clock

09:10 to 09:31: the pump class, filled, sent back and filled again

A master data team fills the equipment register ahead of a system move, one class at a time. On this morning the pumps went through, and came back.

demoSample data: the plants, the people and the register.

  1. 01First thing

    86 classes as tanks, and the whole register sits at 49.1%

    Master data lead

    The tank farm shows 286,400 equipment records in 86 classes at three sites. A tank’s level is the share of its required attributes that are filled, against a dotted target of 85%. The agent suggests pumps next: 6,240 records at 37% with 15,998 pages on file. Isolation valves sit lower, but few of their pages exist.

  2. 0209:10

    Pumps at 37.0%, and one maker spelled 14 ways

    Master data lead

    The pump tank holds 149,760 required attributes: 24 for each pump. 55,411 are in the register with no page behind them yet, and 94,349 are blank. Impeller diameter sits at 20%, maximum impeller at 11%, seal drawing at 7%. Under the tank, one pump maker appears 14 ways. The lead starts the batch, and nothing reaches the register until a sample passes.

  3. 0309:13

    15,998 pages read, 76,378 values proposed, 17,971 left blank

    Master data lead

    Agents work through drawings, datasheets, manuals, nameplate photos and repair records, and 14 spellings become one name on 1,383 records. Of the 55,411 values already held, 41,204 match a page, 12,795 have no page and 1,412 disagree and go to the steward. The tank climbs to 88.0%, hatched, because it is proposed and nothing is written yet.

  4. 0409:16

    400 values to read, and the proposal hidden until each one is read

    Discipline data reviewer

    The rotating equipment reviewer gets the batch and a sample: 240 ordinary values with 2 wrong allowed, 120 safety values with none, and 40 blanks. Each value is read on its own page first, and the proposal shows only after. The top line says what a pass means: if 1 in 45 ordinary values were wrong, this sample would catch it 9 times in 10.

  5. 0509:17

    The plate says 11.75 inches; the agent read 13 from the model code

    Discipline data reviewer

    On a lean amine pump the nameplate carries its impeller on its own line: 11.75. The proposal is 13.00 in, taken from the last number of the model code. On older plates from this maker that number is the largest impeller the casing takes, and this one was trimmed. The reviewer marks the proposal wrong and picks the reason: read from the wrong place.

  6. 0609:19

    Seven wrong sizes, and the whole pump batch goes back

    Discipline data reviewer

    The agent points to the other values in the sample read by the same rule. The reviewer reads each plate: 7 are wrong, and 2 are right because those impellers are full size. That is over the limit, so the batch goes back with its reason. Nothing is written, and the tank falls back to 37.0%.

  7. 0709:25

    The impeller rule is fixed by maker, then tested on the nine plates

    Equipment class steward

    The lead proposes a new rule: take the size from an impeller line or a later shop record, use the model code only where that maker puts the fitted size in it, and otherwise leave it blank. The steward tests it before countersigning: on the 9 plates, 7 sizes correct to the impeller line and 2 stay, and 40 more plates show no new wrong value.

  8. 0809:31

    The rerun lands lower, at 87.6%, with 574 sizes left blank

    Master data lead

    The rerun reads 1,698 impeller sizes again. 1,124 now come from a real impeller line or a shop record, and 866 of those changed. 574 have only a model code behind them, so they stay blank with that reason. The tank stops at 87.6%, below the first run, and the screen says why. A fresh draw of 400 values is ready for the reviewer.

  9. 0909:31

    One of the 574: a crude pump with no impeller on its plate

    Master data lead

    The lead opens one of them, an atmospheric bottoms pump. Its datasheet states no fitted impeller, and the model code gives only the casing maximum of 15.00 in. So impeller diameter stays blank, with the reason only a model code gives a size, until the pump is next opened.

149,760 pump attributes, and where each one stands after the rerun

The pump class after the fixed rule, as the class screen shows it. None of it is in the register until a sample passes.

Proposed from a page75,804
Already held, and a page agrees41,204
Already held, no page yetRegister values still waiting for a page12,795
Two pages disagreeWith the steward1,412
Blank, with a reason574 of them impeller sizes with only a model code18,545
Required pump attributes149,760

demoSample data, from the working solution.

Four things a quick fill would have written into the register

Each one looks like a clean value, or a clean record, until someone opens the page behind it.

A model code read as a size

On older plates the last number is the biggest impeller the casing takes. The sample caught 7 trimmed pumps read that way.

One maker under 14 names

1,383 pump records spell one maker 14 ways, so nobody could count its pumps. A look-alike seal company stays apart.

Two pages that disagree

Driver power, seal plan, rated head and one relief valve’s set pressure each carry two values. The steward picks one, with a reason.

Records with nothing behind them

4,836 ghost records across three sites, some still raising planned work, listed with their evidence before anything is retired.

demoSample data, from the working solution.

Around the tank farm

Ghost assets, spellings, disagreements and the field list

What the stewards, the field technician and the auditor open between batches.

A value and its pageEach of 24 attributes shows its state and proof. The 11.75 in impeller comes from the plate’s own line, and 13.00 in sits as maximum impeller.
One maker, 14 spellings6,214 spellings for 1,138 real companies. The lead agrees each merge and the steward countersigns; a different company on a similar name is kept apart.
Where two pages disagree3,206 disagreements, each with both values and a due date. The steward rules with a reason; the reviewer countersigns or sends it back.
Ghost assets, with evidence4,836 records with nothing behind them. Each row shows its drawing, datasheet and last work, and the planned work it would stop.
The field listNameplates to photograph on today’s route, and the ghost walkdown: found it, or not there.
Proof and sampling settingsChanges wait for the steward’s countersignature. Safety values allow no wrong reading, and prove it or leave it blank cannot be turned off.
The morning on the audit trailEach step and signature from 09:10 to 09:31, with the seven wrong sizes listed value by value in the rejection line.

demoSample data, from the working solution.

Who makes, reads and settles an asset data cleansing batch

The person who builds a batch is never the person who reads its sample.

Master data lead
Sees
The tank farm, each batch and the reasons values stay blank
Signs
Runs and sends batches, proposes rule fixes, merges, retirements and settings
Discipline data reviewer
Sees
Each sampled value on its own page, before the proposal
Signs
Signs a batch whose sample passes, or sends the whole batch back
Equipment class steward
Sees
Disagreements between pages, and the required attributes for each class
Signs
Countersigns rule fixes, merges and settings, and settles disagreements
Field technician
Sees
Nameplates to photograph and the ghost walkdown on today’s route
Signs
Records found it or not there for each ghost walked
Internal auditor
Sees
Any value traced to its page, and the audit trail
Signs
No signature; traces values and takes the evidence file away
The agents

Seven agents fill the register’s blanks, and none writes to it

Where two pages disagree or a call is needed, the agent stops and a person takes it.

Source reader

Reads drawings, datasheets, manuals and nameplate photos, and ties each page to a tag.

  • Never changes a source
  • Never uses a superseded revision as proof
Name matcher

Makes one clean name from many spellings of a maker.

  • Never merges two companies on a similar name
Attribute finder

Fills each class’s required attributes by its reading rules.

  • Never infers a likely value
Proof checker

Checks that each value has its page and leaves the rest blank with a reason.

  • Never picks a side
  • Never writes to the register
Sampler

Draws the sample across pages: ordinary values, safety values and blanks.

  • Never shows the reader the proposal first
Ghost finder

Finds records with nothing behind them, and tagged items missing from the register.

  • Never retires anything
  • Checks the stores list and test records first
Readiness checker

Tests each record against the new system’s load rules.

  • Never loads anything

Signing a batch, picking a side and retiring a record stay with people

The agents read, fill and point. These calls belong to named roles.

No likely valuesA value with no page stays blank. The attribute finder may not fill in what a pump that size usually has.
Agents never sign a batchA batch reaches the register only when the reviewer reads the sample and it passes. One bad sample sends the whole batch back.
Two pages that disagree go to the stewardThe proof checker sends both values and their pages; the steward decides, and the reviewer countersigns.
Nothing is retired by an agentThe ghost finder lists records with their evidence. A walkdown comes first, safety records in full, and people sign.
Older revisions are never proofA superseded page stays on file but cannot back a value, and this cannot be turned off.

What master data leads ask about asset master data cleansing

What is asset master data cleansing?

Asset master data cleansing fills and corrects the equipment records in a maintenance system or ERP: the attributes each class needs, one name per maker, and records for equipment that is gone. Here agents read the pages behind each tag, fill what a page shows and leave the rest blank with its reason. Before anything is written, a second person reads a sample.

Where does the equipment master data come from?

From documents the plant already holds: datasheets, vendor manuals, nameplate photos, repair records and drawings. A drawing proves only a tag and its service. Later records win over earlier ones, so a shop record of a trimmed impeller beats the plate.

How is EAM data quality proven before a load?

By a sample read from the pages, per batch. A second person reads ordinary values, safety values and blanks, each on its own page, before seeing the proposal. Safety values such as design pressure, material class and relief set pressure allow no wrong reading. The screen states what a pass means and what it does not.

What happens to values no page can prove?

They stay blank, each blank counting as a result with its reason: no page shows it, a nameplate photo is needed, only a model code gives a size, or two pages disagree. Each reason routes to the person who can close it: the steward, the field technician or the owner of a file.

How are ghost assets handled?

The ghost finder lists records with no drawing, datasheet or recent work, checks the stores list first and shows the planned work each would stop. The field technician walks a sample, and every safety record in full. Retirement then needs the lead, the steward and the reviewer, in that order.

Does this help an asset data migration to a new system?

Yes. The readiness screen tests records against the new system’s load rules by site and shows what blocks the load, with an owner on each blocker. Agents never load anything.

Can the sampling rules be relaxed to get a batch through?

The lead proposes a change and the steward countersigns it. Sampling and walkdown limits cannot change while a sample is open, even with a signature, and the zero allowance on safety values cannot be raised.

Asset Management, Maintenance & Reliability

More for maintenance and equipment data teams

All 10 solutions →
CMMS Data Build

Every tag, plan and spare ready on day one, with its reason attached.

1-min demoWork Order Failure Coding

Vague closeout notes traced to the real failure cause, each code tied to its note.

1-min demoRoot Cause Failure Analysis

The real cause signed, and the sister pump caught still running.

1-min demo

Start with one class
from your own register.

We’ll walk you through one class of your equipment, its pages, its sample and the blanks that stay blank.

Book a demoBrowse solutions