Industrial AI
Remaining useful life: reading the estimate
Learn what remaining useful life predicts, which data support the estimate, and how to use uncertainty when planning equipment maintenance.
Remaining useful life, usually shortened to RUL, estimates how much operation remains before a defined maintenance or failure condition. The unit might be hours, days or operating cycles.
The estimate depends on the equipment's condition, the model and its assumptions about future use. It is not a guaranteed expiry date.
MathWorks' published explanation shows an engine example with an estimated 65 cycles remaining and a bearing example with roughly 9.5 days to a defined threshold. Those are separate model demonstrations. Their numbers are not interchangeable maintenance intervals.
Match the method to the evidence
| Available evidence | Type of approach | What the comparison needs |
|---|---|---|
| Lifetime records for comparable components | Survival analysis | Comparable populations, recorded failures and treatment of units still operating |
| Sensor histories that continue to failure | Similarity-based estimation | A current degradation path and relevant historical paths |
| A condition indicator and a defined threshold | Degradation modelling | A credible trend and a threshold with an operating meaning |
The equipment history limits the choice. If preventive work replaces components before failure, the records contain incomplete lifetimes. Treating each replacement as a failure would change the event the model predicts.
The maintenance comparison explains where this forecast fits beside scheduled and condition-based work.
Read the uncertainty with the estimate
A point estimate gives one value. The maintenance decision also needs information about the uncertainty around it and the conditions under which the forecast remains valid.
Check how the model defines its interval. A confidence interval for an estimated quantity and a prediction interval for a future outcome answer different questions. The label alone is insufficient; the method must explain what is covered.
Turn the estimate into a maintenance decision
Compare the forecast with inspection time, spare-part lead time and the next authorised maintenance window. Keep the consequence of acting too early beside the consequence of waiting too long.
A reported RUL cannot settle those trade-offs without an agreed decision rule. Nor does it establish that the model can safely extend an existing maintenance requirement.
Record the estimate when it is issued, then retain later observations and the eventual inspection or replacement finding. Evaluating only the last prediction before intervention would conceal how the forecast behaved earlier.
Keep RUL separate from MTBF
Mean time between failures summarises operating time relative to counted failures over a defined history. It does not mean a particular bearing will last that long from today.
RUL makes a conditional statement about the remaining operation under its model. Neither measure can replace the other simply because both use hours.
When reviewing a remaining-life display, ask for the predicted event, unit, issue time, uncertainty definition and assumed operating conditions. Without those fields, a precise-looking number is difficult to use responsibly.