SEAMIQ · Guides

Predictive maintenance for mining fleets, without the magic

Every vendor says predictive. Few can show a prediction with a confidence band, the evidence it was made from, and the record of what actually happened. This guide sets out what component life prediction has to deliver on a mine site, the questions that separate a model from a marketing slide, and how PredictIQ answers them.

Read the checklistSee PredictIQ
Remaining useful lifep10 / p50 / p90Fleet urgencyPrediction vs outcomeYour models, your IP
SEAMIQ · PredictIQ
PredictIQ by SEAMIQ
Definitions

Condition monitoring, prediction and the gap between them

Condition monitoring tells you the state of a component now: oil analysis wear metals, telematics fault codes, derates, temperatures. It is necessary and most mines have some of it. It does not, on its own, tell you when the component will reach end of life.

Prediction takes the condition evidence, the hours on the component, the utilisation rate and the application severity, and estimates the end-of-life date. A good prediction is a distribution, not a date: the earliest credible date, the median and the latest, with the band width stating how sure the model is. A single date without a band is false precision, and planning a rebuild on it is betting production on the median.

The mining problem

Why the OEM target is not a forecast

The manufacturer's target life for a component is arithmetic on a design assumption. Application severity, operator behaviour, ground conditions and maintenance quality move the real number by thousands of hours in either direction. The centreline estimate (hours to target at the current utilisation) is useful as a baseline. It is not a prediction.

Experienced reliability teams already know this and many have built their own predictive methods: regressions on utilisation, wear models on oil analysis, Monte Carlo simulations for the band. The method lives in a spreadsheet on one laptop, runs when someone has time, cannot be checked against outcomes at scale, and leaves when the person does.

The operational question is rarely about one component. It is which of thirty components across eight trucks the planner should act on first, and whether seven of them reaching end of life in the same half-year will overwhelm the workshop. Prediction has to rank the fleet, not just describe a part.

Ownership

Whose model is it?

When predictive capability is embedded in a vendor's platform, the methodology belongs to the vendor and the capability ends with the contract. A dealer or mine that has invested years in a predictive method should expect to run it on a platform that operationalises the method while leaving the model files, the feature engineering and the IP with the people who built them.

Where the cost hides

A wrong date costs money in both directions

Every component replacement is a bet on a date. Too late and the failure picks the time and place. Too early and the hours left in the component are thrown away. Neither shows up as a line called 'wrong date'.

01
Run past the band
A component past its credible life is a breakdown waiting for a time. When it comes, it is on a haul road, on night shift, with a crew that was planned elsewhere.
02
Changed out too early
A component pulled on the brochure interval with hours still in it. Across a fleet and a year, that is a rebuild programme you did not need to fund.
03
Secondary damage
A failure that was predictable damages the parts around it. The rebuild cost doubles and the downtime with it.
04
The rebuild shop in waves
Nothing for a month, then six at once. Rebuild capacity, exchange stock and the budget all take the shock, because nobody saw the cluster coming.
05
Long-lead parts bought on a guess
Major components ordered to the OEM interval rather than the fleet's own curve. Either the part is on a shelf for a year or it is not there when it is needed.
06
The prediction nobody can check
The reliability engineer's model and the fitter's instinct never meet on a record. When the component fails, there is no way to say whether the method was right.
A predictive method earns its keep only if it closes these leaks: a band rather than a date, a fleet-wide ranking rather than a single alert, and a record that lets the prediction be checked against what happened. The checklist below asks for each of those on a screen.
What mining needs

Four things a component prediction must carry

A band, not a date
p10, p50 and p90 end-of-life dates with the spread stated in days. Narrow is confident; wide is honest.
The evidence beside the number
Fault events, oil flags, downtime and PM compliance for the same component on the same screen.
A place in the fleet ranking
Every component ordered by urgency so the planner knows what to act on first.
A record against outcome
When the component completes its cycle, the gap between predicted and actual is shown and kept.
Buyer's checklist

How to judge a component prediction platform

Put these to any vendor, SEAMIQ included, and ask to see each one working on a screen with your own data.

Does the prediction come with a band or a date?
Ask for p10, p50 and p90 and how wide the band is on a component with little data.
Who owns the model?
Ask whether you can take your model files with you when the contract ends.
Can you see what the prediction was made from?
Hours, utilisation, target, model version, confidence, iterations, run date.
Does it compare prediction to outcome?
One component that completed its cycle, with predicted date, actual date and gap.
Is the fleet ranked, or just listed?
Which component the planner should act on first, and why.
Does the evidence sit next to the number?
A prediction without evidence is a guess with a decimal point.
Does it work across OEMs?
A mixed fleet on one urgency table with no manufacturer missing.
Can it go into a site meeting as paper?
If the mine manager cannot hold it, it will not change the plan.
How SEAMIQ approaches it

PredictIQ

PredictIQ is the execution layer for a predictive method. It ingests the operating evidence, runs your model (or the platform baseline) against live data, writes p10, p50 and p90 dates with confidence, ranks the fleet, puts the evidence on the same page as the number, and records every prediction against the outcome. The models and the methodology stay yours.

  • Deterministic centreline beside the probabilistic model output
  • Monte Carlo p10 / p50 / p90 bands with spread in days
  • Fleet urgency ranking and PCR schedule with the 180-day cluster window
  • Fault events, oil analysis, downtime and PM compliance per component
  • Prediction history, outcome recording and a printable hotsheet
Explore PredictIQ
SEAMIQ · PredictIQ
PredictIQ by SEAMIQ
Questions people ask

Questions people ask

Do we need a data scientist to use predictive maintenance?
Not to start. A centreline baseline and a platform model give a defensible position from day one. If you have a reliability team with a method, the platform should run their model rather than replace it.
What data do we need?
Component serials and install hours, machine hours, and whatever evidence you hold: telematics fault codes, oil analysis, maintenance events, recorded end-of-life outcomes. More evidence narrows the band.
Is a wide band a failure?
No. A wide band is the model being honest about thin data. The failure is a vendor hiding that width behind a single date.
How is accuracy proven?
By recording the actual end of life against the prediction, component by component, and showing the gap. Ask any vendor for that record.
Does this replace our OEM telematics portal?
No. It consumes the portal's data and adds the prediction, the ranking and the evidence record across any manufacturer's equipment.
PredictIQ
Let's talk

Bring us one component

Pick a component type you lose sleep over. Send the install hours, the machine hours and whatever evidence you hold. We put it through PredictIQ and show you the band, the ranking and the evidence page with your data.

Or email us directly: info@seamiq.io

No mailing lists. We reply personally within one working day.
The platform

Explore the other products

AssetIQ
Predictive asset management for mining fleets: work orders, components, downtime, lifecycle economics and spares in one platform.
GETIQ
Ground engaging tools intelligence for GET manufacturers: campaigns, wear data, consumption forecasting and commercial opportunity.
RebuildIQ
Structure, compliance and learning at every stage of the component rebuild, from strip through dispatch.
TrackIQ
Component tracking across every machine a component has ever lived on, with pull alerts and fleet health reporting.