What Is Remaining Useful Life (RUL)?
Remaining Useful Life (RUL) is the estimated time, or number of operating cycles, an asset can keep performing its function before it reaches a defined failure threshold. It is measured forward from the current moment, not from the day the asset was installed, and it is the central output of any predictive maintenance program.
RUL answers the one question behind every maintenance decision: how much longer can I run this machine before I have to stop and fix it? Get it right and you extract the full value of the asset while avoiding an unplanned failure. Get it wrong in either direction and you either scrap good life you already paid for or inherit a 2 a.m. breakdown.
Why Does RUL Matter?
Knowing that a bearing is degrading is useful. Knowing it has roughly three weeks left is what lets a planner order the part, book the technician, and schedule the repair during planned downtime. RUL converts a vague sense of risk into a date you can plan against.
The economics justify the effort. Siemens' True Cost of Downtime 2024 puts the cost of unplanned downtime at $1.4 trillion a year for the world's 500 largest companies, around 11% of revenue, with an idle automotive line running to $2.3 million an hour. Forward-looking condition data is what converts that exposure into planned, scheduled work.
How Is RUL Calculated?
The simplest formula uses design life and operating age:
RUL = Total Useful Life - Current Age
This is the age-based method, and it works for components that degrade predictably: timing belts, filter cartridges, calendar-limited seals.
Consider a pump bearing rated for 25,000 operating hours that has already run 18,400. Simple subtraction gives an RUL of 6,600 hours, about 13 months at a typical run rate. But subtraction assumes the bearing is wearing exactly as its designer intended. If the pump is running above rated speed, actual fatigue life is shorter, and only condition data will reveal it. Vibration monitoring might show RMS velocity has climbed from 2.1 mm/s at install to 6.8 mm/s, trending toward an 11.2 mm/s alarm threshold in roughly 3,800 hours, not 6,600. The condition-based estimate is far more honest than the age-based one.
Which Methods Estimate RUL?
Three methods dominate, trading off data requirements against accuracy.
1. Age-based (life-fraction). Uses rated service life and operating hours. No sensors required, only accurate runtime records. Good for low-variability consumables like belts, filters, and seal kits. Its weakness is that it ignores actual condition entirely.
2. Physics-of-failure (degradation models). Models the actual wear mechanism, such as crack propagation or bearing spall growth, then projects when the measured degradation will cross a failure threshold. Common inputs include vibration amplitude, temperature rise, oil particle counts, and motor current signatures. This method produces a confidence interval rather than a single number.
3. Data-driven (machine learning). Learns the relationship between sensor patterns and time-to-failure directly, without an explicit physical model. It handles complex multi-mode assets like gearboxes well, but has one hard requirement: enough historical run-to-failure examples from similar assets to learn from. That requirement is exactly what most sites cannot meet.
How Does RUL Differ From MTBF and MTTF?
RUL is often confused with reliability metrics that answer different questions.
MetricWhat it measuresScopeRULTime this specific asset has left before failureIndividual asset, real-timeMTBFAverage time between failures across a fleetPopulation, historicalMTTFAverage life before failure for non-repairable partsComponent populationP-F intervalTime from detectable degradation to functional failureFailure mode, design-time
The key distinction: MTBF is a fleet average that informs preventive schedules, while RUL is a per-asset estimate that informs a specific intervention. An asset can have an 8,000-hour fleet MTBF but only 400 hours of real RUL because it happens to be running hard in a hot, dusty corner of the plant.
How Does RUL Fit Into a Monitoring Pipeline?
RUL is the output of a chain that runs from sensor to work order: data acquisition (vibration, temperature, current, oil analysis sampled at the right frequency), feature extraction (turning raw signal into health indicators), degradation modeling (fitting the trajectory and projecting to threshold), and intervention triggering (an alert with enough lead time to procure parts and schedule the shutdown). A RUL estimate that lives in a dashboard but never triggers a work order delivers almost none of its potential value.
Where Is RUL Used?
RUL analysis shows up wherever failure is expensive enough to justify continuous monitoring. In aviation, engine health systems track thermal cycles and vibration to pull engines for inspection before the next flight cycle. In oil and gas, offshore operators coordinate compressor interventions into single maintenance campaigns to avoid repeat helicopter mobilizations. In manufacturing, asset monitoring on a stamping press lets planners bundle die changes and repairs into one downtime window. In wind energy, gearbox monitoring schedules expensive crane lifts weeks ahead. In mining, haul-truck drive monitoring routes trucks to the bay at shift change instead of failing in a remote pit.
Why Is RUL Data-Hungry, and What Changes With Physical AI?
Every accurate RUL method past simple subtraction depends on data, and conventional machine learning depends on a specific kind: long histories of run-to-failure examples for each asset class. Most sites cannot meet that requirement: they simply have not collected those histories, and thin, unlabeled data is the usual reason data-driven RUL stalls.
This is where a foundation model changes the equation. Newton, Archetype's Physical AI foundation model, is pre-trained broadly on physical signals, so it detects degradation and anomalies without years of labeled failure history for each machine. Physical AI platforms are the next step in predictive maintenance, which is why RUL work shifts from a per-asset modeling project into something that can begin closer to the moment sensors come online.
What Are the Limitations of RUL?
RUL is a probabilistic estimate, not a deadline. Confidence intervals widen the further out the prediction reaches, so "failure in 200 to 400 hours" is far more actionable than "failure in 3,000 to 8,000 hours." Estimates degrade when operating conditions shift, when sensor data is noisy or sparse, and when a failure mode is too rare to have training data. And the failure threshold itself is a judgment call: set it too conservatively and you waste life, too aggressively and you risk the failure you were trying to prevent.
Frequently Asked Questions
What Is Remaining Useful Life (RUL)?
RUL is the estimated time an asset can keep operating before reaching a failure threshold, measured from the present moment based on current condition rather than installation date.
How Is RUL Calculated?
The basic formula is Total Useful Life minus Age, but accurate RUL comes from degradation data. Physics-of-failure models project measured wear to a threshold; machine learning models learn degradation from run-to-failure data. Foundation models like Newton predict degradation without requiring years of labeled history per asset.
What Is the Difference Between RUL and MTBF?
MTBF is a fleet-wide average time between failures used to set preventive schedules. RUL is a per-asset, real-time estimate used to time an individual intervention. The same asset class can have a high MTBF and a low RUL depending on how one unit is being run.
Which Industries Use RUL Most?
Aviation, oil and gas, manufacturing, wind energy, and mining, wherever the cost of unplanned failure justifies continuous monitoring.
What Are the Main Limitations of RUL Prediction?
Uncertainty grows over longer horizons, estimates suffer when operating conditions change or sensor data is poor, and rare failure modes lack the training data machine learning needs. RUL should always be read as a range.
