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:

  1. Fault Detection and Isolation
  2. Fault-Tolerant Control
  3. Prognostics and Health Management (this post)
  4. 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

Open Monograph No. 3 →

← Back to Autonomy