What is remaining useful life (RUL) and how is it estimated?

What RUL measures

RUL is the remaining time before an asset stops meeting its requirement. In practice, the time is defined through a health indicator, a measurable quantity that trends as the asset degrades. The moment the indicator reaches the failure threshold is the moment RUL reaches zero. Two things must therefore be defined before any estimate: a health indicator that trends, and a threshold based on loss of function rather than a guess. Without a threshold, the question 'when is it failed?' has no answer. Estimation methods fall into three families: life-data models, such as Weibull, that use age; extrapolation of the degradation trend; and similarity or machine-learning models trained on run-to-failure histories.

Worked example: trend extrapolation (hypothetical)

Suppose a compressor's weekly health indicator reads 0.40, 0.44, 0.48 and 0.52 in weeks 1 to 4, and the failure threshold is 1.0.

• Slope: (0.52 - 0.40) / 3 = 0.04 per week.
• Distance left to threshold: 1.0 - 0.52 = 0.48.
• RUL: 0.48 / 0.04 = 12 weeks, counted from week 4.

The result assumes a straight-line trend. If the reading accelerates, 12 weeks may be optimistic. A slope fitted to only four points also carries uncertainty, which is why a range is needed.

Limits and common mistakes

A trend-based estimate is only as good as its threshold and its stability assumption. Maintenance resets the health indicator, so the old trend no longer applies after a repair. Changes in load, speed or product mix can change the degradation rate, and a model fitted under one regime can mislead under another. Few failures in the history mean few examples to test against. A common mistake is to extrapolate a noisy indicator without smoothing it. Another is to report one number without a band that widens with the horizon. Every estimate should carry its time stamp, its inputs and the date of the last maintenance event.

Follow-up questions