Apps · Prognosis
RUL Predictor
Learn how bearings degrade from a fleet of run-to-failure histories, predict the remaining useful life of units still running, and find out what the predictions are worth against fixed-interval replacement.
Health indicator and forecastobservedpossible futuresfailure threshold
Training fleet
Every past unit stays flat, then degrades at its own rate. The spread of those rates is the model's prior.
Units in service90% interval check
What is a prediction worth?300 simulated units
Click Compare policies to simulate.
How the prediction works
- Health indicator. Each unit reports a number that stays near 0.05 while healthy and rises exponentially once a defect starts. Failure is defined as reaching 1.0.
- Learning from the fleet. For each past unit, a straight line is fitted to the logarithm of its rise. The spread of those slopes becomes the prior: what degradation rates are plausible.
- Detection. A unit is treated as degrading once its indicator stays above the healthy level plus four standard deviations for three cycles. Before that, RUL comes from the fleet's lifetimes alone.
- Bayesian update. After detection, the unit's own data are combined with the prior. Early on, the prior dominates and the interval is wide. As data accumulate, the unit's own slope takes over and the interval narrows.
- Policy. The predictive policy checks every 5 cycles and schedules a replacement when the 5th-percentile RUL falls within the lead time. It is compared with replacing every unit at a fixed age, and with running to failure.
The economics change with the cost ratio: when failures are cheap, run-to-failure wins; when they are expensive, fixed intervals become very conservative and predictions pay off. Read more in Prognosis: remaining useful life.