A single vibration spectrum captures a machine at one steady speed, but many faults only reveal themselves while the speed is changing. A cascade plot, also called a waterfall plot, solves this by stacking many spectra one behind another against machine speed or time as the machine runs up or coasts down. Reading the plot as a three-dimensional map shows how each frequency component grows, shrinks, or moves as speed changes. That evolving picture is how analysts catch resonances, subsynchronous instabilities, and rubs that a steady-state snapshot never shows.
Cascade spectrum plot in one line: A cascade or waterfall spectrum plot is a stack of vibration frequency spectra recorded across a range of machine speeds, usually during a startup or coastdown, displayed as a three-dimensional map of amplitude versus frequency versus speed. It lets an analyst see how order-related and subsynchronous components change with speed, which exposes critical-speed resonances, oil whirl, and rubs that a single fixed-speed spectrum cannot reveal.
A cascade plot is built by capturing a fresh spectrum at frequent intervals while the machine sweeps through a range of speeds, then drawing each spectrum offset behind the last so the collection forms a surface. One axis is frequency, a second is amplitude, and the third is the machine speed at which each spectrum was taken. Because the plot is usually gathered during a transient event such as a run-up or a coastdown, it turns a few seconds or minutes of changing operation into a single readable map rather than a stack of disconnected readings.
The power of the format comes from how different components behave as you scan up the speed axis. Components that are tied to running speed, the so-called order lines at one, two, or three times turning speed, appear as ridges that march outward in frequency as speed rises, because their frequency is a fixed multiple of a rising speed. A component fixed in frequency regardless of speed, such as a structural resonance, appears as a vertical ridge that stays at the same frequency while the machine accelerates past it. Reading which ridges slope and which stay put is the core skill the plot rewards.
This separation is exactly what a single spectrum cannot give you. At one steady speed, a resonance and a running-speed harmonic can land on top of each other and look like one peak, and there is no way to tell them apart. On a cascade plot the two behave differently as speed changes, so the analyst sees them diverge and can assign each peak a cause. The plot trades the simplicity of one reading for the diagnostic depth of watching the whole vibration signature move.
Critical speeds are the clearest example. As a rotor accelerates through a resonance, the amplitude of the running-speed component swells and then falls back as the machine passes through and out of the resonant band. On a cascade plot this shows up as a bulge in the one-times order ridge at a particular speed, pinpointing where the critical speed sits and how much the machine amplifies at it. A single spectrum taken above or below that speed would show a modest peak and give no hint that a resonance is lurking a few hundred rpm away.
Subsynchronous instabilities are the other headline use. Oil whirl in a journal bearing appears as a ridge sitting below the running-speed line, typically near a fraction of turning speed, and it tracks with speed like an order line. Oil whip is the dangerous escalation where that subsynchronous component locks onto a rotor critical speed and stops tracking, appearing on the cascade plot as a ridge that follows the fraction until a threshold speed and then goes vertical, holding at a fixed frequency while the machine keeps accelerating. That transition from a sloping ridge to a vertical one is a signature that is almost impossible to read from steady-state data.
Rubs and looseness also declare themselves on the plot. A rub tends to generate a spray of harmonics and sometimes fractional components that come and go at particular speeds, producing a busy, intermittent pattern across the speed axis rather than a clean set of order lines. Because the cascade plot shows the whole run-up rather than one operating point, an intermittent event that touches only a narrow speed band is captured instead of being missed by a snapshot that happened to be taken at a quiet moment.
Building a good cascade plot depends on catching the machine while its speed is actually changing, which is a coordination problem as much as an analysis one. The data collector has to be armed and recording before the operator hits start or trips the machine offline, and it needs a speed reference so each captured spectrum can be tagged with the rpm it belongs to. Machines that start and stop unpredictably, or that live at a remote unmanned site, are exactly the ones where a portable analyst standing by is impractical, and where continuous monitoring earns its place.
A monitoring platform that streams field data to the cloud changes the economics of transient capture. When vibration channels and a tachometer signal are logged continuously, a run-up or coastdown does not have to be scheduled around a visiting technician; the event is already recorded and a cascade plot can be reconstructed after the fact from the historized data. Merobix pulls the underlying vibration and speed signals into a browser-accessible history, so an analyst can pick a past startup, build the waterfall from the stored spectra, and study a resonance or an instability without having been on site when it happened.
The same continuous record turns cascade analysis from a rare commissioning exercise into an ongoing check. Comparing the waterfall from today's coastdown against one captured months ago shows whether a critical speed has shifted, whether a subsynchronous component has appeared that was not there before, or whether a rub has developed. Because the transient data lives in one place and is tied to speed, that comparison becomes a routine part of watching a fleet of remote turbomachines rather than a special trip that only happens when something already sounds wrong.
In most vibration software the two terms mean the same thing: a stack of spectra displayed against speed or time. Some tools reserve waterfall for spectra stacked against time and cascade for spectra stacked against machine speed, but the distinction is not universal. In practice analysts use the words interchangeably to describe the same three-dimensional map of amplitude, frequency, and a changing operating condition.
Use a cascade plot whenever the machine's condition depends on speed or when you suspect a resonance or an instability. Startups and coastdowns are the natural times to capture one, because the machine sweeps through its whole speed range and past any critical speeds. A normal steady-state spectrum is fine for tracking imbalance or a bearing fault at operating speed, but it cannot show how a component behaves as speed changes.
You can stack spectra against time without a speed signal, which still shows components appearing and disappearing during a transient. However, without a tachometer you cannot label the speed axis or reliably separate order lines from fixed-frequency resonances, since both simply move with time. A tachometer or a clean once-per-revolution reference is what lets the plot distinguish speed-related components from structural ones, so it is strongly preferred for diagnostic work.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.