Condition Monitoring Glossary

Every term twice — once precisely, once in the words you would actually use in a room with a budget holder.

Updated · 2026-08-09

Fault detection is the easy half. The half that decides whether anything gets funded is translation — turning a spectrum into a maintenance date, a bearing frequency into a work order, a leakage trend into a budget line.

So each term here is given twice: the precise definition first, then → the plain-room version where a useful one exists. If you can only say a thing one way, you can only say it to one audience.

The working rule that goes with it: define on first use in a mixed room, then use the term freely. Explaining every term every time is its own kind of condescension.

Sensors and the measurement chain

• Accelerometer — sensor converting casing acceleration into an electrical signal. → "the vibration sensor."
• Piezoelectric (piezo) sensor — crystal plus seismic mass; acceleration squeezes the crystal, producing charge proportional to acceleration. Wide flat response, a few Hz to 30 kHz. Wired. → "the high-end wired sensor type."
• Charge (in the piezo sense) — the crystal's electrical output, proportional to acceleration. Not battery charging. Piezo sensors never "charge"; they are powered through the cable.
• IEPE / ICP — the standard for powering piezo sensors through their own signal cable using constant current, running continuously for decades. → "powered off the same cable that carries the signal."
• MEMS accelerometer — micro-machined silicon measuring deflection capacitively, usually with the ADC on-chip. Cheap, microamp power, battery-friendly. Modern industrial units reach 8–10 kHz and beyond; cheap ones stop around 3 kHz. → "the chip-based sensor that makes wireless affordable."
• DAQ (data acquisition) — the electronics that filter and digitise the analogue signal: anti-aliasing filter plus ADC. An external box in the wired world, built into the node in wireless — where the sampling specification is fixed by silicon and cannot be upgraded.
• Triaxial sensor — three sensing elements at right angles: one housing, three directions.
• Mounting resonance — the spring-mass resonance of the sensor and its attachment. It sets the usable frequency ceiling: stud around 10 kHz and above, then adhesive or epoxy pad, then magnet at roughly 2–5 kHz, then handheld at about 1 kHz. Content near it reads inflated; content above it is invisible. → "how the sensor is attached decides which faults it can physically see."
• Stud mount / epoxy pad / magnet mount — attachment methods in descending order of stiffness and usable bandwidth.
• g-range and dynamic range — the maximum measurable acceleration before clipping. Saturation is non-linear, and sprays false harmonics across the spectrum. Worth asking any vendor for both the g-range and the overload behaviour.
• Sampling rate — how many times per second the ADC measures. Nyquist requires more than twice the highest frequency of interest; industry practice is 2.56×.
• Fmax — the highest frequency a measurement captures, equal to sampling rate ÷ 2.56.
• Anti-aliasing filter — a low-pass filter before the ADC, preventing above-Nyquist content from masquerading as false low frequencies.
• Burst sampling — a wireless node wakes, samples at high rate for a second or two, transmits, and sleeps. Good battery economics, but snapshots rather than continuity. Ask for burst length, interval, and what is computed on the node.
• Scalar versus spectral transmission — budget nodes transmit only summary numbers such as RMS and band energies, so the diagnostic detail never leaves the sensor. Ask whether you get full spectrum and waveform, or scalars only.
• Transmission path — the metal between the fault and the sensor. Every joint and gasket attenuates, high frequencies worst. Mount on the bearing housing, in the load zone, on the stiffest shortest path — never on fan covers, shrouds or fins.
• Proximity probe — a non-contact sensor watching shaft position directly, used on journal-bearing turbomachinery. An X-Y pair at 90° produces an orbit plot of shaft centreline motion.

Signals and spectra

• Time waveform — the raw signal, amplitude against time, everything superimposed. Shape carries meaning: sinusoidal means a smooth fault, spikes mean impacts, one-sided truncation means rub or looseness, distortion means harmonics are coming.
• FFT / spectrum — the waveform decomposed into amplitude against frequency. Each fault type has a spectral address. → "the chart that untangles the mess into labelled peaks."
• Running speed (1×) — shaft rotation frequency, RPM ÷ 60. The reference for everything else. Not to be confused with natural frequency.
• Natural frequency — what a structure rings at when struck; a property of mass and stiffness, independent of RPM. A forced frequency landing on one is resonance, and is dangerous.
• Orders (2×, 3× …) — integer multiples of running speed. Order tracking normalises the spectrum by measured speed, so faults stay at fixed orders even as RPM drifts.
• Harmonics — the integer-multiple family produced by distorted, non-sinusoidal periodic motion. Fourier's rule: distortion breeds harmonics, and the more clipped or rattly the waveform, the taller the family.
• Sub-synchronous — content below 1×: belt frequencies, cage frequency, oil whirl.
• Fractional harmonics (½×, 1½× …) — a severe looseness signature.
• Sidebands — pairs flanking a carrier peak, spaced at the modulation rate. The spacing names the modulator: ±1× on BPFI is the load-zone sweep; ±FTF on ball-spin frequency is the cage carrying the bad ball; ±pinion-speed on gear mesh is a pinion-carried fault. → "the satellite peaks that tell us which part is driving the problem."
• Amplitude modulation — one signal's amplitude varying at the rate of another; the time-domain cause of sidebands.
• Beating — two close frequencies drifting in and out of phase, the envelope swelling at their difference frequency. Its existence proves both components are present, since nothing beats against itself. Period is the reciprocal of the beat frequency.
• Phase — the timing relationship between two measurements. It splits look-alike faults: ends in phase is static unbalance, ends opposed is couple unbalance, a roughly 180° break across a coupling is misalignment. Frequency says what, amplitude says where, phase says which.
• RMS — overall vibration energy. A late-stage indicator that climbs steeply near failure.
• Crest factor — peak divided by RMS. High values mean impulsive content.
• Kurtosis — statistical spikiness. Conventionally expected to rise during early impulsive damage while RMS is still flat — though not on every channel of every bearing.
• Band alarm — an alert on energy within a chosen frequency window. Cheap and computable on-node, but it only finds what you thought to look for, windows miss when speed drifts, broadband faults escape, and same-frequency noise fools it.
• Time-synchronous averaging — tacho-locked averaging over many revolutions, so shaft-synchronous content reinforces and everything else averages toward zero. It filters by synchronicity rather than by frequency.
• High-pass and band-pass filtering — discarding low content, or selecting a window. Filters choose; they do not resurrect what the hardware chain never captured.

Faults and their signatures

• Unbalance — a heavy spot, producing a smooth dominant radial 1× that grows with its cause: dirt, lost weight, erosion. Fans are the classic case, since collecting dirt asymmetrically is what a fan does for a living.
• Misalignment — shaft centrelines that do not meet at the coupling. Offset or parallel misalignment gives radial 2×; angular misalignment gives strong axial 1× and 2×. The distorted waveform brings a harmonic march. It is the leading killer of mechanical seals and the second of bearings. → "the shafts are fighting each other."
• Looseness — mechanical gaps in bolts, fits or structure. A harmonic picket fence, fractional harmonics when severe, a one-sided truncated waveform, and directional behaviour. Self-aggravating: hammering loosens things further.
• Soft foot — a machine frame not sitting flat. It produces the looseness signature and distorts the stator into air-gap eccentricity, which also pumps twice-line-frequency vibration — so it feeds both the mechanical and electrical sides and produces an unusually loud beat. The fix is a torque check.
• Bent shaft and eccentricity — 1× family look-alikes of unbalance; phase separates them.
• Resonance — an operating force landing on a natural frequency, with dangerous amplification. It changes with structure, not with load.
• BPFO / BPFI / BSF / FTF — bearing defect frequencies for outer race, inner race, ball spin and cage. Geometry multiplied by shaft speed, and importantly non-integer multiples of 1× — anything off the integer grid should make you suspect a bearing. Rules of thumb: BPFO ≈ 0.4·n·f, BPFI ≈ 0.6·n·f, and the two sum to n·f, which is a conservation law rather than a coincidence. → "each bearing part has its own tell-tale frequency."
• Spall — a surface fatigue crater on a race or rolling element; the thing the rollers strike.
• Bearing failure stages — high-frequency ringing only, then discrete fault frequencies, then harmonics and sideband families, then broadband low-frequency chaos as RMS climbs. Energy marches down the spectrum as damage matures.
• Gear mesh frequency — teeth multiplied by shaft speed; a normal resident of the spectrum. Its sidebands are the fault, and their spacing says which shaft carries it. Asymmetric sideband families indicate a maturing fault.
• Cavitation — pump vapour bubbles collapsing, producing broadband high-frequency noise. A process problem, not a bearing problem.
• Rub — a rotating part contacting a stationary one; truncated waveforms and odd harmonics.

Electrical and current signature analysis

• ESA / MCSA — reading fault-induced structure in motor current. Signature is not draw: draw is the amps consumed, signature is which frequencies ride on the current and at what spacing. A soft-foot motor draws near-normal current while its sidebands tell the whole story. → "the motor's electricity carries a fingerprint of its mechanical health."
• Slip — the rotor's permanent lag behind the rotating stator field. Synchronous speed is twice line frequency divided by pole count; the nameplate figure is full-load speed, and slip shrinks as load falls, so a lightly loaded motor turns faster than its nameplate. Rotor-bar faults appear as sidebands at ±2× slip around line frequency.
• Twice-line-frequency vibration — 100 or 120 Hz magnetic-pull vibration. Every induction motor has a little; air-gap problems produce a lot. It collides with the mechanical pole-count harmonic and produces beating. The power-cut test separates them: the electrical component vanishes at the instant of trip, while the mechanical component persists through coast-down.
• Air-gap eccentricity — an uneven rotor-to-stator gap from soft foot, poor fits or a bowed rotor. It pumps twice-line-frequency vibration and current sidebands, and is the home turf of current signature analysis.
• VFD switching noise — variable-frequency drives spray 2–16 kHz content, both electrical and magnetostrictive. An innocent high-frequency resident that regularly fools band-energy alarms.

Monitoring strategy and maintenance

• Condition monitoring — measuring machine health indicators, continuously or periodically.
• Predictive maintenance — acting on measured condition to schedule work before failure.
• Proactive or precision maintenance — eliminating root causes such as misalignment, unbalance and poor lubrication so that faults stop occurring. The ladder runs run-to-failure → preventive → predictive → proactive.
• Preventive maintenance — fixed-interval work. Worth remembering that intrusive maintenance introduces failures through contamination, mis-torque and infant mortality.
• Reliability-centred maintenance — a methodology born from airline studies showing most failure modes are not age-related, which makes calendar overhauls indefensible for much equipment.
• P-F curve — the timeline from potential failure, the first detectable point, to functional failure. Techniques are placed along it by how early they detect. The map is here.
• P-F interval — the warning window between detection and failure. This is the number every "how early can you detect" conversation is really about.
• FMEA — failure modes and effects analysis; an asset-by-asset account of how things fail.
• Criticality analysis — ranking assets by consequence of failure, which is what drives tiered monitoring.
• Run-to-failure — deliberately not monitoring cheap redundant assets. Entirely legitimate when the spares are real and the failure is contained.
• Enveloping / demodulation — high-passing around a structural resonance and extracting the impact repetition rate, for early bearing detection. It needs one genuine resonance beneath the sensor's ceiling; envelope an empty band and you demodulate noise.
• Ultrasound, thermography, oil analysis — sister techniques covering, respectively, early bearing wear and leaks and arcing; heat patterns and electrical connections; and wear debris and contamination. Each owns failure modes the others miss.
• Sensor fusion — independent modalities agreeing, such as vibration and current, which buys confidence, fewer false alarms and mutual blind-spot coverage.
• Time-based versus condition-based — calendar against evidence. A plant that "never breaks down under monthly maintenance" is buying that reliability at enormous recurring cost, and monitoring offers the evidence to safely stretch the intervals.
• CMMS — the computerised maintenance management system holding work orders and history. The data is usually dirty and archaeology should be expected; repeat offenders in it are where the business case lives.
• Historian — the plant time-series database for process data.
• Baseline — a machine's healthy signature, against which deviations are flagged. Insufficient baseline is the standard cause of missed detections early in a deployment — and a baseline learned from a machine that was already degrading is worse than none, because it looks like one.
• Shelf-ware — bought, ignored, auto-renewed. Prevented by workflow design — the alert becoming a ticket in the customer's own CMMS, with a named owner and an escalation path — rather than by better sensors.

Protocols and OT

• Modbus — a legacy master/slave polling protocol using registers and coils, in serial (RTU) and Ethernet (TCP) forms. No authentication and no encryption, so it lives behind segmentation.
• MQTT — a lightweight publish/subscribe protocol with a broker in the middle, topics, and three quality-of-service levels. Its critical property in an industrial deployment: the client initiates an outbound connection, so no inbound firewall port is ever required.
• OPC-UA — the modern industrial protocol, with a real security model built in.
• Gateway — the site device aggregating sensors and connecting outward. The thing the IT department will interrogate.
• OT (operational technology) — plant-floor networks and systems, as distinct from IT. An egress rule is firewall permission for one specific outbound path, and is worth offering in writing before it is demanded.
• TLS — transport encryption for that outbound connection.