Why is RUL estimated as a range instead of a single number?

Where the uncertainty comes from

A remaining-life estimate extends an observed trend into the future, and three kinds of uncertainty enter that step. The first is measurement noise: each health indicator value carries sensor and environmental effects. The second is parameter uncertainty: the true slope cannot be known exactly from a short history. The third is future uncertainty: if load, speed or the maintenance plan changes, the degradation rate changes too. A single point estimate merges these into one figure and hides their effect. A range keeps them visible, so the planner can see how early or how late the asset might fail.

Worked example: how the range is computed (hypothetical)

Suppose a pump's health indicator is 0.6 today, the failure threshold is 1.0, and the trend rises 0.2 per week. The point estimate is (1.0 - 0.6) / 0.2 = 2 weeks. Now suppose the rate is only known to lie between 0.15 and 0.30 per week. The fastest case gives 0.4 / 0.30 = about 1.33 weeks, and the slowest gives 0.4 / 0.15 = about 2.67 weeks. A planner who needs a safe window can act on the early end. If the band is too wide for the planning window, more measurements may narrow the slope estimate. That narrowing does not remove future uncertainty about operating conditions.

Common mistakes and limits

The most frequent mistake is treating the band as a guarantee. P10 to P90 describes the model's spread, not a promise about the machine. The second mistake is using a fixed band width for every horizon, since uncertainty grows as the forecast reaches further. The third is comparing estimates after changing the threshold without recomputing the band. A band is also only as honest as the model behind it. After a repair resets the indicator, or when operating conditions change, old bands no longer apply. Each estimate should be recalculated with each new reading and logged with its inputs.

Follow-up questions