FDI and FTC operate after the fact. By the time they fire, something has already changed. Prognostics asks the opposite question — how much longer will this component keep working? — and pushes the diagnostic story forward in time. The economic stakes are enormous: a fleet that knows which of its assets needs maintenance next week has a fundamentally different cost structure than one that doesn’t.
This monograph is the third of a four-part diagnostics series:
- Fault Detection and Isolation
- Fault-Tolerant Control
- Prognostics and Health Management (this post)
- Frontiers in Diagnostics
What it covers
Six chapters plus references. About thirty-five minutes to read.
§1 — Forward in time. What prognostics actually predicts and why the question is harder than “estimate the time of failure.” The framing that separates this field from FDI.
§2 — Remaining Useful Life as a first-passage time. The mathematical object underneath all of prognostics: when does a stochastic degradation process first cross a failure threshold? Brownian motion, Wiener processes, the algebra that makes RUL estimation tractable.
§3 — Degradation modeling. The shape of getting old. Linear, exponential, gamma-process, Wiener with drift. Which model fits which physical mechanism (wear, fatigue, corrosion, dielectric breakdown).
§4 — Health indicators. Squeezing many sensors into one scalar. PCA, autoencoders, physics-derived indicators. Why a good HI matters more than a good RUL estimator.
§5 — Uncertainty — often more important than the estimate. A point estimate of RUL without a credibility interval is worse than useless — it’s actively dangerous. The uncertainty quantification techniques that make prognostics decision-grade.
§6 — Condition-based and predictive maintenance. Where all of this gets used. The scheduling problem on top of the prediction problem.
Read it
← Back to Autonomy