Vibration Analysis for Rotating Equipment: Reading the Warning Signs Before a Machine Fails

Rotating machines almost never fail without warning. They fail without anybody listening.

A pump that seizes on a Tuesday morning has usually been telling you for weeks. The bearing began shedding metal, a defect frequency appeared in the spectrum at an amplitude too small to feel by hand, then grew, then spread into harmonics, then raised the overall level past the point where a trained hand on the casing would have noticed. Somewhere in that sequence there was a cheap intervention. Nobody was measuring, so the first anyone knew was the noise and the smell.

Vibration analysis is the discipline of listening properly. It is the most information-dense diagnostic available for rotating equipment, it works while the machine is running under load, and it distinguishes between faults that look identical from the outside. This article covers what it measures, how ISO defines acceptable condition, what the common fault signatures look like in a spectrum, and how to build a programme that actually catches things.

What Vibration Actually Tells You

Every rotating machine vibrates. Perfect balance, perfect alignment and perfect bearings do not exist, so the question is never whether a machine vibrates but whether the vibration has changed and what its frequency content says about the cause.

Three measurement quantities matter, and choosing the wrong one hides faults.

Displacement, measured in microns, is how far the shaft or casing actually moves. It dominates at low frequencies and is the right quantity for slow-turning machines and for proximity probes watching shaft position inside a journal bearing.

Velocity, measured in millimetres per second, is the workhorse. Across the broad mid-band where most mechanical faults live, roughly 10 Hz to 1,000 Hz, velocity gives a reasonably flat response to fault severity. This is why almost every acceptance standard is written in mm/s and why an overall velocity reading is the standard health number for a machine.

Acceleration, measured in g or m/s², emphasises high frequencies. Rolling element bearing defects and gear mesh problems announce themselves in the kilohertz range long before they show up in a velocity reading, which is why acceleration and its derived envelope measurements are the early-warning channels.

An overall velocity number tells you the machine has a problem. The spectrum, which decomposes that vibration into its constituent frequencies, tells you which problem. A programme that records only overall levels is a smoke alarm. A programme that records spectra is a diagnosis.

The ISO 20816 Zones: What Counts as Acceptable

Engineers argue about acceptable vibration until someone produces the standard. ISO 20816-1 sets out the general framework, and the machine-specific parts define the numbers: ISO 20816-3 covers industrial machinery above 15 kW, and ISO 20816-2 covers land-based steam and gas turbines and generators above 40 MW.

The framework divides machine condition into four evaluation zones, measured as vibration velocity in mm/s RMS on the bearing housing.

Zone A is the condition of newly commissioned machines. Zone B is acceptable for unrestricted long-term operation. Zone C is unacceptable for continuous long-term running: the machine can be operated for a limited period while corrective work is scheduled. Zone D is severe enough to cause damage, and the machine should come off line.

The boundary values depend on machine size and on how the machine is mounted, which is the detail most people miss when quoting the standard. For medium machines in the 15 kW to 300 kW band on a rigid support, the boundaries fall at 1.4 mm/s between zones A and B, 2.8 mm/s between B and C, and 4.5 mm/s between C and D. The same machine on a flexible support is permitted higher readings, because a flexible foundation is expected to move. Larger machines above roughly 300 kW carry higher boundaries again.

A support counts as rigid when the lowest natural frequency of the combined machine and foundation sits above the machine's main exciting frequency. Otherwise it is flexible. Applying rigid-support limits to a flexibly mounted machine generates false alarms; applying flexible limits to a rigidly mounted one lets real faults run.

Two cautions on using zone tables at all. First, they are guideline values, and a manufacturer's own acceptance criteria for a specific machine override them. Second, and more importantly for anyone running a real plant, the trend matters more than the absolute number. A machine that has sat at 1.8 mm/s for three years and moves to 3.2 mm/s in a month is in trouble, even though 3.2 mm/s is inside Zone C rather than Zone D. A machine that has always run at 4.0 mm/s because of its foundation may be perfectly healthy. Baselines beat limits.

Want your rotating equipment assessed against the standard rather than against opinion? Our maintenance and asset management team establishes baselines and alarm bands per machine.

ISO 20816-3 evaluation zones for medium machines (15-300 kW) on rigid supports, vibration velocity in mm/s RMS. Boundaries differ for flexible supports and for larger machines. Source: ISO 20816-3:2022.

Reading the Spectrum: What Each Fault Looks Like

The diagnostic power of vibration analysis comes from the fact that different faults excite different frequencies, expressed as multiples of running speed. Running speed is written as 1×, twice running speed as 2×, and so on.

Unbalance produces a dominant peak at exactly 1×, mostly in the radial direction, with low axial content and a stable phase relationship. Amplitude rises with the square of speed, so an unbalance problem gets dramatically worse as the machine speeds up. It is the most common finding and usually the cheapest to fix, because balancing in situ is routine work.

Misalignment puts significant energy at 2×, frequently exceeding the 1× peak, and produces high axial vibration. That axial content is the giveaway: unbalance does not do it. Angular misalignment shows strongly axially, parallel offset more radially, and a phase difference of roughly 180 degrees across the coupling confirms it. Misalignment is the fault most often introduced by maintenance itself, which is why alignment should be verified after any coupling work and after the machine reaches operating temperature, not while it is cold.

Mechanical looseness generates a long series of harmonics: 1×, 2×, 3×, 4× and upward, sometimes with half-order components at 0.5× and 1.5× when there is genuine structural rattle. A spectrum that looks like a picket fence is usually loose hold-down bolts, a cracked baseplate, a worn bearing housing fit or degraded grouting.

Rolling element bearing defects are the most valuable diagnosis because they give the most warning. Each defect location generates its own non-integer frequency, calculated from bearing geometry: the ball pass frequency of the outer race, the ball pass frequency of the inner race, the ball spin frequency and the fundamental train frequency. These are not multiples of running speed, which is what makes them identifiable. In early stages the energy is at high frequency and low amplitude, invisible in a standard velocity spectrum, which is why envelope or demodulation techniques exist. A bearing typically progresses through months of detectable degradation before it becomes an overall-level problem, and by the time you can hear it, that window has closed.

Blade pass and vane pass faults in pumps, fans and compressors appear at the number of blades multiplied by running speed. A rise there points to flow disturbance, impeller damage, or too little clearance between impeller and cutwater.

Electrical faults in induction motors show at twice line frequency, meaning 100 Hz on a 50 Hz supply, and can be confirmed by the coast-down test: switch off the power and watch whether the peak disappears instantly. Mechanical vibration decays with speed. Electrically driven vibration vanishes the moment the supply does.

Cavitation in pumps does not produce a discrete peak at all. It raises a broad, random high-frequency hump, and it sounds like gravel. It is a hydraulic problem, meaning insufficient net positive suction head, and no amount of balancing or alignment will touch it.

Gear faults appear at gear mesh frequency, the number of teeth multiplied by shaft speed, usually with sidebands spaced at shaft running speed on either side. Growing sidebands mean the modulation is worsening, which points at a specific tooth or an eccentric gear.

From Data to Decision

Collecting spectra is not a programme. Plenty of plants own an analyser that produces beautiful plots nobody acts on. What turns measurement into avoided failure is a small number of unglamorous disciplines.

Fixed measurement points. Readings must be taken at the same physical locations, in the same three directions, with the same mounting method, every time. A magnet base on a painted surface and a stud-mounted accelerometer on a machined pad give different high-frequency responses. Mark the points permanently.

Consistent operating conditions. A pump measured at 60% flow and the same pump measured at full flow are two different machines as far as the spectrum is concerned. Record load, speed and process conditions alongside the reading, or the trend is noise.

Criticality-driven intervals. Not every machine deserves monthly attention. Route frequency should follow consequence of failure: critical, unspared machines earn continuous or monthly monitoring, important machines quarterly, and the general population annually or on condition. Trying to monitor everything at the same interval means monitoring nothing well.

Alarm bands set per machine. Blanket alarms based on a standard's zone table generate false positives on flexibly mounted machines and stay silent on rigidly mounted ones that are genuinely degrading. Set alarms from each machine's own established baseline.

A closed loop back to the work order. The finding has to become a job, the job has to be executed, and a post-repair measurement has to confirm the fault is gone. Without that last step nobody learns whether the diagnosis was right, and the programme's credibility erodes with every unverified call.

This is where vibration analysis fits into the broader shift from scheduled to condition-based work that we set out in preventive versus predictive maintenance. Vibration is the highest-value sensor in that toolkit for rotating machinery, but it is one input among several.

How each common rotating equipment fault appears in the vibration spectrum, expressed as multiples of running speed (1x). Source: MIMAH engineering diagnostic reference.

Where Vibration Analysis Stops and Other Techniques Start

An honest programme knows the limits of its primary tool.

Oil analysis sees things vibration cannot: contamination ingress, water, additive depletion, wear metals identifying which component is shedding material. On slow-turning and heavily loaded machines it frequently gives earlier warning than vibration does.

Infrared thermography finds electrical connection problems, cooling faults and insulation breakdown, none of which vibrate meaningfully until they are already failing.

Ultrasound catches the very earliest stages of bearing lubrication distress and is the correct tool for leak detection and for guiding regreasing, which is a common cause of bearing failure in both directions: too little grease and too much.

Motor current signature analysis detects rotor bar and air-gap problems that are hard to distinguish mechanically.

Performance monitoring, meaning flow, head, pressure differential and efficiency, catches degradation that produces no vibration signature at all: a fouled heat exchanger, a worn impeller ring, a blocked strainer.

The strongest programmes correlate. A rising vibration trend at bearing defect frequencies, plus rising iron content in the oil, plus a warmer bearing housing on the thermal survey, is not three findings. It is one finding with three confirmations, and it is the kind of evidence that gets a shutdown approved.

Running critical rotating assets without a monitoring baseline? Our industrial engineering services cover condition monitoring, root cause analysis and rotating equipment diagnostics across turbines, pumps, compressors and generators.

Building a Programme Without Buying Everything

The usual failure of a new condition monitoring programme is that it starts too large. Somebody buys an analyser, builds a route covering four hundred machines, collects data for six months, generates no actionable findings because there are no baselines yet, and the budget is questioned before the programme has had time to prove itself.

A better sequence starts small and narrow. Rank the machine population by consequence of failure, not by how easy it is to reach. Take the top tier, typically a few dozen assets whose failure stops production or endangers people, and establish proper baselines on those. Set alarm bands from those baselines rather than from a textbook. Prove the value on that tier by catching two or three real faults and documenting what the intervention saved against what an unplanned failure would have cost. Then expand.

Route the difficult and dangerous points to permanent sensors rather than to a technician with a probe. Anything requiring a ladder, a confined space entry or a guard removal will eventually be skipped or recorded from an approximate location, and both outcomes corrupt the trend.

Above all, resist the temptation to buy a system before you have decided who analyses the data. The instrument is the cheap part. The competence to look at a spectrum and say "that is a fluid-film bearing instability, not unbalance, and balancing it will make things worse" is what the programme actually runs on.

Which vibration quantity to measure, and the frequency band each is suited to. Choosing the wrong quantity hides the fault. Source: MIMAH engineering diagnostic reference.

Frequently Asked Questions

How often should vibration readings be taken? It depends on consequence of failure rather than machine type. Critical unspared machines justify continuous monitoring or monthly routes. Important but spared machines suit quarterly intervals. The general population is usually adequately served annually or on condition. The interval must be short enough that a developing fault cannot progress from detectable to catastrophic between two readings.

Can vibration analysis predict exactly when a machine will fail? No, and any vendor promising a failure date should be treated carefully. What it provides is a severity assessment and a rate of change, which together support a decision about how long you can safely run and when to schedule the repair. That is genuinely valuable. A precise remaining-life figure is not what the technique delivers.

What is the difference between overall vibration and spectral analysis? An overall reading is a single number summarising energy across a frequency band. It tells you something has changed but not what. A spectrum separates that energy by frequency, which is what allows unbalance, misalignment, looseness and bearing defects to be told apart. Overall readings are useful for screening; spectra are required for diagnosis.

Our machine reads inside the ISO limits but keeps failing. Why? Almost certainly because the standard's zone tables were written for broad machine classes and cannot know your specific machine, its foundation or its duty. A machine can sit comfortably inside Zone B while a bearing defect frequency climbs steadily in the high-frequency band, because that energy contributes little to the overall velocity number. This is exactly why baselines, trends and envelope measurements matter more than compliance with a table.

Is portable data collection enough, or do we need permanent sensors? For most of a plant's machine population, a well-run portable route with fixed measurement points is entirely adequate and far cheaper. Permanent instrumentation earns its cost on machines that are critical, that run at variable conditions, that are unsafe or impractical to reach, or where the failure develops faster than the route interval.

Does vibration analysis work on slow-speed machinery? It works, but the technique has to change. Below roughly 100 rpm, velocity measurements become unreliable and displacement or specialised low-frequency accelerometers with long averaging times are required. Ultrasound and oil analysis often prove more informative on very slow machines.

The Cheapest Repair Is the One You Schedule

The economics of rotating equipment maintenance are not subtle. A bearing replaced during a planned outage costs the part, a few hours of labour and no production. The same bearing left until it seizes costs the shaft, possibly the housing, the coupling, the downtime, the expedited freight on parts nobody stocked, and the overtime of a crew working through a night they did not plan for. The ratio between those two numbers is routinely an order of magnitude.

Vibration analysis is how you get to choose which of those two you pay. It does not prevent wear. It converts an unpredictable failure into a scheduled task, which is the only thing that ever actually reduces maintenance cost.

Across four decades of turbine, generator and rotating equipment work, from steam turbine overhauls in Nigeria to root cause analysis at White Nile Sugar in Sudan, the finding is consistent: the machines that surprise people are the machines nobody was measuring.

Have a machine that vibrates, runs hot, or failed once already without explanation? Talk to our engineering team. We will measure it properly, tell you what the spectrum says, and give you a repair scope based on the fault rather than on guesswork.