Condition-based maintenance (CBM) acts on evidence, not the calendar: monitor a machine's health and intervene only when a developing fault appears. It sits between run-to-failure and fixed-interval preventive maintenance, and it's the practical face of the predictive tasks in RCM.
It works because of the P-F curve: a failure becomes detectable (point P) well before it becomes functional failure (point F). The gap is the P-F interval — your warning. Your monitoring interval must be comfortably shorter than it.
For rotating machines, vibration analysis is the workhorse. The spectrum (FFT) turns vibration into peaks at multiples of running speed — and the pattern of peaks names the fault: 1× imbalance, 2× misalignment, harmonic trains for looseness, high-frequency non-synchronous peaks for bearings.
1 · What condition-based maintenance is
There are three ways to decide when to maintain a machine:
- Reactive (run-to-failure) — fix it when it breaks. Fine for cheap, non-critical, fail-safe items; catastrophic for anything else.
- Preventive (fixed interval) — overhaul every N hours regardless of condition. Simple, but you either replace healthy parts too early or miss faults that develop early — and intrusive overhauls introduce their own failures.
- Condition-based (predictive) — measure the actual health and act only when a fault is developing. You catch problems early and you don't disturb healthy machines.
CBM is the modern default for critical rotating equipment, and it is exactly what an RCM analysis selects when it chooses an "on-condition" task. But it only works if a failure gives warning — which is what the P-F curve describes.
2 · The P-F curve
As a failure develops, the machine's condition declines along a curve. At some point the fault becomes detectable by some technique — that's P, the potential failure. If nothing is done it continues to F, functional failure. The time between them is the P-F interval: the window in which you can catch it.
This gives the one rule that governs every monitoring programme:
The monitoring interval must be shorter than the P-F interval — by convention, at most half of it — or you risk inspecting on either side of the fault and missing it entirely. A fault with a three-month P-F interval needs monitoring at least monthly. This is the same P-F interval that drives task frequency in RCM.
3 · The condition-monitoring toolkit
"Condition monitoring" is a family of technologies, each detecting different failures at different points on the P-F curve. A good programme layers several:
| Technique | Detects | P-F warning |
|---|---|---|
| Vibration analysis | Imbalance, misalignment, looseness, bearing & gear defects, resonance | Early — the rotating-equipment workhorse |
| Ultrasound | Early bearing distress, leaks, electrical discharge, lubrication state | Earliest for some bearing/lube faults |
| Oil & wear-debris analysis | Wear metals, contamination, water, additive depletion | Early — sees inside the machine |
| Infrared thermography | Hot bearings, electrical connections, misalignment, blocked cooling | Mid — heat is a later symptom |
| Motor current signature (MCSA) | Rotor bar, eccentricity and load faults in motors | Early — non-intrusive, from the MCC |
| Performance monitoring | Efficiency / head / flow drift (e.g. off-BEP, wear, fouling) | Varies — process data you already have |
For rotating machines, vibration carries the most diagnostic information — so the rest of this guide is about reading it.
4 · Reading vibration: from waveform to spectrum
A vibration sensor records movement over time — the time waveform. That's hard to read directly, because every fault is superimposed. The breakthrough is the Fast Fourier Transform (FFT), which decomposes the waveform into its constituent frequencies — the spectrum. Now each fault appears as a peak at its own frequency.
The trick that makes diagnosis possible: express frequency in orders — multiples of the running speed (1× = once per revolution). A fault tied to rotation always lands at the same order regardless of machine speed, so the pattern of peaks is a fingerprint:
- 1× — once per revolution → imbalance (a heavy spot).
- 2× — twice per revolution → misalignment (often with strong axial vibration).
- Harmonic train (1×, 2×, 3×, 4×…), sometimes ½× → mechanical looseness.
- Non-synchronous high-frequency peaks (non-integer orders) with sidebands → bearing defects (BPFO, BPFI…).
- Number-of-vanes × running speed → vane-pass (hydraulic, in pumps).
The two most common of these — the 1× imbalance and 2× misalignment — have a direct cure in precision alignment & balancing: the spectrum tells you which fault you have, and that guide shows how to fix it at the source.
Interactive — Vibration spectrum diagnoser
Live modelPick a fault and read its signature in the spectrum. The x-axis is in orders of running speed; the dashed lines mark 1×, 2×, 3×. Then set the severity and watch the ISO 10816 zone change.
Vibration spectrum (FFT)
5 · How much is too much? ISO 10816 / 20816
The spectrum tells you what is wrong; the overall vibration level tells you how urgent it is. ISO 10816 (now 20816) defines broadband RMS velocity (mm/s) zones by machine class:
- Zone A — newly commissioned, good.
- Zone B — acceptable for long-term operation.
- Zone C — unsatisfactory; plan corrective action.
- Zone D — unacceptable; damage is likely — act now.
The zone boundaries shift with machine size and mounting, but the discipline is the same: trend the overall level over time, and let the spectrum diagnose any rise. A rising trend is more informative than any single reading.
6 · Turning it into a programme
Condition monitoring earns its keep only as a closed loop, and that loop is pure Bluestream toolbox:
- Prioritise by criticality — monitor the equipment whose failure matters (Criticality), not everything.
- Pick techniques by failure mode — let the FMECA and RCM say which failures need watching and which technique sees them on the P-F curve.
- Set intervals from the P-F interval — monitor at least twice within it.
- Trend, alarm, and act — and when something is found, run a proper root cause analysis so it doesn't come back.
Key takeaways
- CBM acts on evidence — between run-to-failure and fixed-interval PM, and the practical form of RCM's on-condition tasks.
- The P-F curve gives the warning window; monitor at least twice within the P-F interval.
- Layer techniques — vibration, ultrasound, oil, thermography, MCSA — each detects different faults at different points.
- The vibration spectrum names the fault — 1× imbalance, 2× misalignment, harmonic trains for looseness, non-synchronous peaks for bearings, vane-pass for hydraulics.
- ISO 10816/20816 grades severity; trend the overall level and diagnose with the spectrum.