Rotating Equipment Reliability
Mechanical, electrical and production process reliability consulting for the rotating equipment industry: air separation, petrochemical and chemical, manufacturing, and power generation.
Mechanical, electrical and production process reliability consulting for the rotating equipment industry: air separation, petrochemical and chemical, manufacturing, and power generation.
RotaCore works with plants that depend on rotating machinery: air separation, refining and chemicals, manufacturing and power generation. We bring the three disciplines that decide whether a plant runs, mechanical reliability, electrical reliability and production process reliability, into one engagement, and we leave the site with a programme its own people can run.
Gabriel.Parra@r-dynamic.com · (512) 739-6518
Select a service for a full description of the work and what you receive. Every service is delivered against the machines and the process actually on site, not from a template.
Each sector has its own machinery, its own failure modes and its own operating constraints. The sector pages are technical guides: the equipment, the faults, how they show on the instruments, and the PdM programme that catches them. Select a sector.
Reliability is decided at every stage of an asset's life, and the cheapest place to fix a problem is always one stage earlier than where it is found. Select a stage.
RotaCore's programmes follow two established frameworks: the Certified Maintenance and Reliability Professional (CMRP) Body of Knowledge, which defines what a reliability organisation has to be good at, and Reliability Centered Maintenance (RCM), which defines how the maintenance strategy for each asset is decided. Select a pillar or a step.
Predictive maintenance is a set of measurements, each of which sees a different physical symptom at a different point in a failure. No single technology covers a plant. Select a technology for what it measures, what it catches, and where it belongs.
The matrix below shows which technology sees which failure mode. Select any cell.
The concepts that every programme is built on and every proposal is judged by. Each one is interactive.
A failure starts long before it stops the machine. The point where it first becomes detectable is P; the point of functional failure is F. Everything between them is the time available to plan. Each PdM technology finds P at a different place on the curve, and the inspection interval must be shorter than the P–F interval it is trying to use. Select a point.
Failure rate over an asset's life is not constant. Select a region for the failure mechanisms that dominate there and the maintenance strategy that fits. Then select a failure pattern: most rotating equipment does not follow the bathtub at all.
Inherent availability depends on two things: how long the machine runs between failures (MTBF) and how long it takes to get it back (MTTR). Move the sliders. Doubling MTBF and halving MTTR do the same thing to availability, and the second is usually the cheaper lever.
OEE multiplies availability, performance and quality into one number and exposes where production is being lost. Move the sliders; the waterfall shows where the time goes. World-class is often quoted at 85%; most plants measure in the 40s to 60s the first time they look.
Reactive maintenance is the most expensive kind; preventive is cheaper but wasteful; predictive and proactive cost least per avoided failure. Move the slider to shift a plant's mix from reactive toward predictive and watch total cost, emergency work and planned work change.
Anonymised engagements that show how the three disciplines come together. Select a case for the data, the diagnosis and the result.
Six dimensions that RotaCore assesses in the first week on site. Move the sliders to score your own plant; the radar shows the shape, the score shows the gap, and the notes say what closes it.
Unplanned downtime is paid for in lost production, overtime, expedited spares and collateral damage. Put in your own numbers to see what a reliability programme is worth, and what a typical programme recovers.
The decision RotaCore is most often asked to support. Move the sliders: the chart compares the cumulative cost of keeping an ageing machine against replacing it, and shows the year the lines cross.
Every recommendation is referenced to a recognised standard, so it can be audited and defended. Select a standard.
The bearing, the motor that drives it and the process that loads it are diagnosed by the same people. No hand-offs between a vibration contractor, an electrical contractor and a process engineer who have never met.
Every finding comes with the data, the test that would disprove it, and a confidence statement. Repairs are scoped from condition, and the machine is not handed back until the acceptance data says so.
Routes, limits, procedures and KPIs are built on the site\'s own machines and written down, then handed over. RotaCore stays as standing support for the reviews and the hard cases, not as a permanent dependency.
Integrally geared ASU compressors and turboexpanders, API process machinery, high-speed manufacturing drives, turbine-generator trains. The failure modes are different in each sector and so is the programme.
Specification review, FAT and SAT witness, commissioning baselines, the monitoring programme, and the root-cause analysis when something still fails. The whole life of the asset, not one slice of it.
API, ISO, IEEE and the SMRP Body of Knowledge behind every recommendation. Reports written so a plant manager, an insurer or an auditor can follow the reasoning.
Select a step. Most engagements start with an assessment and end with the site's own people running the programme.
For an assessment, a difficult diagnosis, or a reliability project in any of the four sectors, write or call.
The technical foundation behind every RotaCore programme, written for every sector we serve. It starts with the periodic table of rotating equipment faults, shows what those faults look like on a DCS trend, builds the analyst's skills from CAT I to CAT IV, and ends with the measuring equipment that produces the data. Each page carries an industry panel showing how the material applies on an air separation unit, in a refinery or chemical plant, in a manufacturing plant and in a power station.
The physics is the same everywhere; the machines, the sensors and the failure modes that dominate are not. Select a cell.
Cryogenic air separation runs on a small number of large, high-speed machines: integrally geared main air and booster compressors, nitrogen feed and recycle compressors, turboexpanders, and the motors, pumps, chillers and cooling tower drives around them. The machines are fixed-speed, tilt-pad, interstage water-cooled, and tightly coupled to a process that can hurt them in minutes. This section covers what is specific to the ASU: the motor analysis for its drives, the programme by machine, and the ASU panels and fault family in the Basics section.
What a complete programme on an air separation unit covers, machine by machine. Select a cell.
The sector with the widest range of rotating equipment and the longest list of standards governing it: API 610 pumps by the hundred, API 617 centrifugal and API 618 reciprocating compressors on the critical services, API 611 and 612 steam turbine drivers, fin-fan coolers, extruders and agitators, all protected to API 670 and most of it in a hazardous area. Pump reliability sets the maintenance budget; the compressors set whether the unit runs.
Select a machine type for what it is, how it fails, what it is instrumented with and where the reliability effort goes.
Most refinery pump failures are a seal or a bearing, and most seal and bearing failures start with where the pump is operated on its curve. Move the flow slider and watch the hydraulic forces, the vibration and the dominant failure mode change.
API 618 reciprocating compressors carry the hydrogen and the process gas, and they fail through valves, rings, packing and rod load, none of which a casing accelerometer sees well. The P–V card, from a cylinder pressure transducer against crank angle, is the diagnostic. Select a condition.
The API 617 process gas compressor and its API 612 steam turbine driver are the critical train of most process units. They run on tilt-pad bearings with proximity probes, keyphasors, axial position and bearing temperatures into an API 670 rack. The fault families are the same as the ASU compressors, with two additions: high-pressure gas forces that consume stability margin, and a driver that can overspeed. Select a case for the signature.
On a critical train the protection system is as important as the machine. API 670 defines the sensors, the monitor, the voting and the trip. Select each element for what it does and how it fails.
The generic rotating-equipment faults, with the ASU-only families removed. Select any cell for the signature, the checks that confirm it and the probable causes.
Vibration is one of six technologies a refinery or chemical plant runs. Select a cell for how each technology is used on each equipment class in this sector.
Pump MTBF is the reliability KPI the sector is judged by. The Pareto below is a typical distribution of failures across a 400-pump unit: a small number of services produce most of the work orders. Select a bar.
Industry surveys commonly quote refinery pump MTBF from under 12 months at poorly run sites to more than 48 months at the best. The gap is mostly operating point, alignment and seal selection, not pump design.
Rank by repair cost and count, take the top 10%, assign each an owner, do the root cause (hydraulic, mechanical, operational or installation), fix it, and re-rank quarterly. Most sites halve their pump work orders in two years this way.
The hydraulic analysis of where the pump runs, the seal and bearing failure analysis, the alignment and installation standard, and the KPI reporting that keeps the programme honest.
NGL fractionation and gas plants are compressor-heavy: large reciprocating and centrifugal compressors, often engine- or gas-turbine-driven, in remote locations with lean staffing. Select a topic.
A manufacturing plant runs on hundreds of small and medium machines rather than a handful of large ones: fans, pumps, gearboxes, conveyors, mixers, extruders, presses, machine tool spindles, plant air compressors and the VSD-driven motors behind most of them. Almost all of it is rolling-element, most of it is route-based, and the reliability question is less "what is wrong with this machine" than "which of these four hundred machines matters and how do we cover them". Production losses are measured in OEE, and the maintenance programme is judged by it.
Select a machine type for how it fails, how it is monitored and where the effort goes.
With hundreds of machines, coverage is a budget decision. A criticality matrix ranks each machine by the consequence of its failure and the likelihood, and the strategy follows from the cell it lands in. Select a cell.
Consequence includes safety, environment, production loss and repair cost. Likelihood comes from failure history and operating context. The matrix is reviewed annually; machines move as the plant changes.
Most manufacturing drives are a motor, a belt or coupling, a gearbox and a driven machine. Each element has its own frequencies and its own faults. Select a fault for the spectrum and the test that confirms it.
VSDs are everywhere in manufacturing. They change the analysis (every frequency moves with the setpoint) and they create a failure mode of their own (bearing currents). Select a topic.
A spindle is a precision bearing system where a few microns of change is a scrapped part. The analysis lives at high frequency and small amplitude. Select a condition.
The generic rotating-equipment faults relevant to rolling-element machinery, belts and fans. Select any cell.
Industry surveys attribute more than half of rolling-element bearing failures to lubrication: too little, too much, the wrong grease, or contamination. Ultrasound-assisted greasing turns a calendar guess into a measurement. Select a stage of the greasing cycle.
Select a cell for how each technology is used on each equipment class in a manufacturing plant.
OEE is the manufacturing reliability KPI. The loss tree below shows where production time goes on a typical line; select a loss for what drives it and which discipline owns the fix. The interactive OEE calculator is on the home page.
A power plant is one very large train and a long list of auxiliaries that keep it fed. The turbine-generator runs on fluid-film bearings with a full turbine supervisory instrumentation set; the boiler feed pumps, draft fans, mills and conveyors are the balance of plant that trips the unit when they fail. Cogeneration adds the gas turbine and its heat recovery steam generator. The sector's KPI is availability, measured as equivalent forced outage rate and equivalent availability factor.
A coal-fired steam turbine-generator: HP, IP and LP turbine sections, the generator and its exciter, on eight or more fluid-film bearings. Select a bearing or a measurement for what it reads and what it catches. The supervisory set goes beyond vibration: it watches where the rotor is relative to the casing.
A large steam turbine passes through several critical speeds on every start, and the start is where most of the diagnostic information is. The Bode plot below shows a healthy run-up; select a condition to see how it changes.
The generator is the largest electrical machine on site and has fault families of its own. Select one for the signature and the test.
Boiler feed pumps, ID, FD and PA fans, pulverizers and mills, condensate and circulating water pumps, coal conveyors and crushers, cooling tower fans. Select a machine for how it fails and how it is monitored.
A combined-cycle or cogeneration plant adds a gas turbine, usually on fluid-film bearings with its own control and protection package, exhausting into a heat recovery steam generator that feeds a steam turbine or process steam. Select a topic.
The generic rotating-equipment faults, with the fluid-film and rotor families that matter on a turbine-generator. Select any cell.
Select a cell for how each technology is used on each equipment class in a power plant.
Power plants report availability in the NERC GADS terms: equivalent forced outage rate (the share of time the unit could not run when needed) and equivalent availability factor (the share of the period the unit was available at full load). Move the sliders.
72 fault "elements" in 11 families: the periodic table of rotating equipment faults. Each cell gives the signature, the checks that confirm it and the probable causes. Nine families apply in every industry RotaCore serves; the fluid-film family applies wherever tilt-pad and sleeve bearings run (ASU, petrochemical, power), and one family is specific to the cryogenic air separation process. Use the industry filter to see which apply to you, then the analyse-a-reading tool, the two plots, phase and the key concepts below.
Vibration data is presented in two forms, both from the same sensor. Select any part of either plot for an explanation of what it shows and how it is used.
Amplitude against frequency. Left to right is events per shaft revolution; height is amplitude in mils.
Shaft movement against time, before it is converted into a spectrum.
The two plots are the same reading. The waveform is what the sensor actually felt; the spectrum is the computer sorting that movement into "how much of it happens once per turn, twice per turn, and so on". Most faults are named by which bars are tall, but some, such as a knock from a cracked gear tooth, hide in the spectrum and only show up in the waveform. Look at both.
Amplitude says how much, frequency says how often, and phase says where the shaft is when the keyphasor mark passes. Comparing phase between two bearings, or between two directions, is what separates faults that have identical spectra. Select a fault below and watch the two bearing readings.
Phase is measured as the angle between the once-per-turn keyphasor pulse and the peak of the 1× waveform. Two readings that peak together are "in phase" (0°); two that peak half a turn apart are "180° out". The absolute number does not matter; the difference between two points does.
A spectrum shows amplitude at each frequency. The reference frequency is 1×, one event per shaft revolution; all other components are expressed as multiples of it. Faults are identified by which components are elevated, in which direction, and how they change with speed, load and time.
The table is the same in every plant; what changes is which families carry the work. Select an industry.
In order of how often each one turns out to be the answer on air separation plants. It tells you where to look first, not what you have. Start at the top. The dangerous ones sit low on the list because they are rare, not because they do not matter. Tap a row to open the fault.
Bars show share of findings (percent of diagnoses), not vibration amplitude. Amplitudes elsewhere on this page are in mils (shaft probes) or in/s (casing sensors). Select a bar to see the faults it contains, then select a fault to open it.
Bars show share of findings for that machine, not vibration amplitude; each machine adds to 100%. Select a machine, then select a bar to open the fault.
Be clear about the status of these numbers before quoting them. There is no published fault-frequency dataset specific to air separation units. These percentages are engineering judgement, structured so the ordering follows the patterns that published rotating-machinery statistics do support, and so that each machine's list reflects its duty. They are a starting prior for where to look, not measured data, and they should not be presented to a client as industry statistics.
What the published literature does support, and which shaped the ordering:
What is engineering judgement, not published data:
Replace both charts with the site's own CMMS and condition-monitoring history as soon as twelve months of records exist. Where the real order differs from this one, that difference is the useful finding: it tells you something about that plant's design, duty or maintenance practice. Until then, treat these as a reasoned place to start looking.
| Zone | Broad meaning |
|---|---|
| A | Newly commissioned condition. |
| B | Acceptable for unrestricted long-term running. |
| C | Unsatisfactory for long-term running. Plan an intervention. |
| D | Severe enough to cause damage. Investigate now. |
Zone boundaries depend on machine class, mounting and support stiffness — take them from ISO 20816 and the OEM manual for the specific machine, not from a single number. A rising trend against a machine's own established baseline is a stronger signal than any absolute limit.
The grey family at the end of the table covers measurement artefacts that look alarming and are not faults. Rule those out first.
Full-size DCS trend samples across the industries RotaCore serves, grouped by machine, with the coast-downs first and an industries group at the end (refinery pumps and reciprocating compressors, plant fans and gearboxes, turbine starts, generators, draft fans and gas turbine combustion). Select a chart or its title to read what it shows. Use the filter to see one machine family at a time.
The DCS does not show you a spectrum. What reaches the operator screen from an API 670 machinery protection rack is a handful of scalar values — overall shaft vibration in mils peak-to-peak from each X and Y probe, axial position, speed, and alongside them the bearing metal temperatures and process tags. Diagnosis needs the condition monitoring system. But the shape of a one-second trend tells you a great deal about what kind of fault you are dealing with, and how fast you need to move.
Each sample below is a one-second trend of the kind an operator or manager can pull up without any specialist software. The charts are shown on their own so you can practise reading the shape first: select any chart or its title to reveal what it means, the checks, and the action. Learn the normal ones first.
Every compressor, pump, chiller and cooling tower fan on the plant is driven by an induction motor, and the largest air compressors may be driven by synchronous machines with brushless exciters. Motors add a family of faults that no other machine has: the electrical ones. They obey the grid, not the shaft, and that is how they are told apart. This section collects all of the motor and electrical material in one place. Select any lesson to open it.
The four electrical faults from the fault table, with the same detail panels. Select a cell.
The map below shows where each motor fault sits on a 2-pole, 60 Hz motor running at 3,580 rpm (59.7 Hz). Change the spectrum resolution and watch the electrical and mechanical peaks merge or separate. Select any peak.
Unbalance, and the carrier for pole-pass sidebands. Moves with load (slip). Disappears slowly on power-off.
Stator problems, air-gap eccentricity, phase imbalance, soft foot distorting the stator. Fixed by the grid. Disappears instantly on power-off.
Misalignment, looseness. 0.6 Hz below 2×LF on this motor: needs about 0.15 Hz resolution, or a zoom, to separate. At 400 lines over 0–1 kHz they are one peak.
A broken or cracked rotor bar means the rotor's magnetic field is weaker once per pole pass, which modulates the 1× vibration and the line current. The result is sidebands at ± pole-pass frequency (number of poles × slip frequency) around 1×, and around line frequency in the current. Move the severity slider.
If the rotor does not sit in the centre of the stator bore, the magnetic pull is uneven and the motor vibrates at twice line frequency. Whether the offset is fixed (static) or rotates with the shaft (dynamic) changes the signature. Select each and watch the air gap.
The rotor is offset but the offset does not turn: a bearing housing bored off-centre, a soft foot or a distorted end shield. 2×LF high, no sidebands. Fixed by re-centring the rotor or correcting the frame.
The rotor turns off its magnetic centre: a bent shaft, a bearing worn eccentric, a rotor not concentric with the journals. 2×LF with sidebands at ± 1×, and 1× with pole-pass sidebands. The minimum gap point rotates.
A motor bolted down with one foot unsupported distorts the stator when the bolts are torqued, which makes the air gap uneven. Reads as 2×LF that changes when a hold-down bolt is loosened. Check with a dial indicator on each foot.
The one test that settles electrical against mechanical for good. With the analyser recording, open the breaker. Magnetic forces stop in the same instant; mechanical forces coast down with the speed. Select a case and watch the trend.
Large main air compressors may be driven by synchronous motors: the rotor carries a DC field supplied by a brushless exciter through a rotating rectifier. There is no slip, so the induction-motor rules change, and the exciter adds its own faults. Select a condition.
The line current carries the same modulation as the vibration, and on a motor it is often the cleaner signal: it does not depend on a sensor location or a mounting. A clamp on one phase and a high-resolution spectrum in dB is the standard rotor bar test. Move the slider to see how the sideband depth is read.
| Sideband below line peak | Rotor condition | Action |
|---|---|---|
| > 60 dB | Excellent | None. |
| 54 to 60 dB | Good | Trend at the normal interval. |
| 48 to 54 dB | High-resistance joint or a bar developing a crack | Shorten the interval; take data at full load. |
| 42 to 48 dB | One or more broken bars likely | Plan the rotor inspection; watch for the bar lifting into the stator. |
| < 42 dB | Multiple broken bars | Remove from service at the first opportunity. |
Guide values in general use; confirm the interpretation against the motor's own history and load. Take current data at the highest steady load available.
A voltage on the shaft, from magnetic asymmetry in a large motor or from the common-mode voltage of a variable-speed drive, discharges through the bearing oil film. Each discharge pits the race; over time the pits line up into a washboard called fluting. Select a stage.
Motor bearings are rolling-element on everything except the very largest drivers, so the sensors are accelerometers on the bearing housings, horizontal and vertical at each end and axial at the drive end. The interactive motor diagram with its marker explanations is on the Condition Measuring Equipment page under Where Each Device Sits, and the motor DCS trends (breaker open, amps before vibration) are on the DCS Trends page under the Motors filter.
Both, at both ends. Electrical faults often read differently in the two directions because the magnetic pull follows the stator geometry; a large horizontal-to-vertical difference at 2×LF points at the frame or the feet.
Misalignment and a bent shaft show here; electrical faults mostly do not. High axial with 2×LF is worth a second look at the coupling before the motor.
Not a permanent sensor, but part of the motor route. One phase, full load, a high-resolution spectrum in dB. Ten minutes that answer the rotor bar question.
For operators, technicians and managers who need to collect readings, recognise the common patterns, and know when to escalate. Nothing here needs prior vibration training. Every section below opens on click, and every chart is interactive.
Every rotating machine shakes. A perfectly balanced rotor in perfect bearings would still move a little, because nothing is perfectly balanced and no bearing is perfectly stiff. What matters is not that a machine vibrates but how much, at what frequency, and whether that is changing.
Vibration is measured because it is the earliest practical warning most mechanical faults give. A bearing that will fail in three months, an impeller that has picked up a deposit, a coupling that has moved out of line: each of these changes the way the machine shakes long before it changes anything a person can see, hear or feel. The instrument sees it first. That is the entire justification for a vibration programme.
Vibration data comes in two forms and you will meet both. The waveform is the raw movement over time, the way the sensor actually felt it. The spectrum is the same data sorted by frequency: a row of bars, each one showing how much of the shaking happens at that rate. Most diagnosis is done on the spectrum, but some faults only show in the waveform.
Select any bar. The 1× bar is once per revolution and is the reference for everything else. 2× is twice per revolution. Bars to the left of 1× are slower than the shaft turns and are the ones to escalate on a compressor.
Select any part of the wave. The knock repeating once per turn is the reason to always look at the waveform: it can be almost invisible in the spectrum.
Thousandths of an inch. Used for proximity probes that watch the shaft directly. One mil is about the thickness of a sheet of paper. A healthy compressor pinion on this plant sits at 0.5 to 0.7 mils; alert is 1.5 and danger is 2.0.
Velocity, used for sensors bolted to the casing on motors, pumps and fans. Good for balance, alignment and looseness. A healthy motor bearing typically reads under 0.1 in/s.
Used for high-frequency content such as gear mesh and bearing defects. The number is small and means little on its own; the trend and the spectrum are what matter.
On an air separation unit, five patterns account for most findings. Learn to recognise these and you can screen the majority of alarms before a specialist is involved. Select each to open its full cell in the fault table.
A reading is only useful if it can be compared with the last one. That requires the same point, the same direction, the same sensor mounting and the same machine condition every time. Five rules:
You will rarely see a spectrum on the control room screen. What you see is the overall level from each probe, trended over time. That is enough to tell what kind of thing is happening, and how fast you need to move. The shapes to know:
Select each shape for what it means. These are simplified from the full trend library on the home page.
A CAT I analyst does not diagnose; a CAT I analyst recognises and escalates. Escalate immediately, before the next start, when you see any of these:
You do not need to design a sensor to use one well, but you do need to know what it is actually measuring, because the two sensor types on this plant measure different things. Select each to see it working.
A proximity probe watches the shaft from a fixed point and reports the gap. An accelerometer rides on the casing and reports how hard the casing is being shaken. On a fluid-film machine the casing barely moves, which is why the compressors carry probes. Full detail is on the Condition Measuring Equipment page.
The three numbers that describe any vibration, shown on the same rotating shaft. Drag the sliders and watch how each one changes what the sensor sees.
Amplitude is the size of the circle the shaft centre draws. Frequency is how fast it goes round. Phase is where the heavy spot is when the keyphasor mark passes the probe. Changing any one of them changes the waveform below in a different way.
Absolute limits exist and you need to know them, but on this plant the limits are simple and the same on every shaft probe: alert at 1.5 mils, danger at 2.0 mils. What a CAT I analyst must add to that is the trend. Select a scenario to see how the same reading means different things depending on where it came from.
Vibration analysts are certified to ISO 18436-2 in four categories. This guide follows the same ladder. Knowing what each level is expected to do is how you know what to escalate and to whom.
Collects data on a pre-defined route, recognises the common fault patterns, compares against limits and history, and escalates correctly. Does not set up measurements or diagnose.
Sets up the measurement, selects sensors and settings, performs single-channel diagnosis, balances in a single plane, and writes findings with a confidence statement.
Multi-channel and transient analysis, rotordynamic interpretation, root cause, alarm philosophy, and management of the programme. Signs off findings and repairs.
Modelling, torsional and structural analysis, the difficult and ambiguous cases, and the judgement calls. Trains and audits the levels below.
What the basic level looks like in each sector RotaCore serves: which sensors, which units, what the daily reading is compared against, and what gets escalated. Select an industry.
For analysts who set up the measurements, run the routes and make the first diagnosis. This level is about telling similar-looking faults apart, and about knowing which second measurement settles the question. Assumes CAT I.
Most false alarms on an ASU are setup errors, not machine faults. Before diagnosing anything, confirm the measurement was capable of seeing it.
Must cover the highest frequency of interest. Gear mesh on an integrally geared pinion runs to several kHz; a route set to 1 kHz will never see it. Blade pass on a fan needs the low end: set the range to reach down to fan speed or the fan's own 1× is missing.
Sidebands around gear mesh or around 1× on a motor are close together. 400 lines will smear them into one peak; 1600 or 3200 lines separates them. Resolution is what lets you tell a broken rotor bar from unbalance.
Four to eight averages with a Hanning window for routine spectra. A flat-top window for accurate amplitude on a known peak. No averaging when you are looking for a transient knock in the waveform.
The CAT II skill is discrimination. These pairs look the same on a single spectrum and are separated by a second measurement. Select each pair to see the deciding test.
Amplitude tells you how much. Phase tells you where in the turn, and comparing phase between two points is what separates unbalance from misalignment from a bent shaft. It requires a once-per-revolution reference from a keyphasor or a strobe.
Select a fault to see how the two ends of the rotor move relative to each other. In-phase at both bearings is unbalance. 180° apart is a couple or a bent shaft. 180° apart across a coupling, axially, is angular misalignment.
Motors, pumps and fans on this plant run on ball and roller bearings. Their faults produce frequencies that are not whole multiples of shaft speed, and that is how you recognise them. Each bearing has four:
Enter the shaft speed and the bearing geometry, or pick a common bearing, to see its defect frequencies. Faults are recognised because these fall between the shaft orders, not on them.
The progression is the same on every bearing: first a rise in high-frequency and envelope energy from lubrication distress, then a discrete defect frequency, then its harmonics, then sidebands, then a rising noise floor as the spall spreads. Each stage is weeks apart on a normal machine. Where you catch it decides how much notice you have.
Every compressor on this plant is integrally geared. That creates a specific analysis trap: each pinion runs at its own speed, set by the bull gear ratio, so 1× is a different number on every channel. Reading pinion 2's spectrum against pinion 1's orders produces nonsense.
Select a pinion to see its speed and where its orders fall. Gear mesh frequency is shared by every pinion because it is set by the bull gear, which is a useful cross-check.
The compressors and the expander run on tilt pad bearings and are watched by proximity probes. The data is different from a casing sensor in three ways that matter to a CAT II analyst.
The number is where the shaft is inside its clearance, in mils. The limit is physical: the rotor cannot move further than the bearing lets it. That is why alert and danger are in mils and why they are the same on every pinion.
The average gap voltage tells you where the shaft sits in the bearing. A dropping position is bearing wear before it is vibration. Trend it.
A bent shaft or a scored probe track reads as vibration at every speed. Measure runout at slow roll and subtract it, or every reading on that channel is biased.
A CAT II report states what was measured, what it is consistent with, and what would confirm it. It does not state a cause as fact. The template:
A CAT II analyst reads a spectrum as a set of families, not a set of peaks. Each family has a cause. Select a family below to see what generates it and what it means on our machines.
Harmonics are a series at integer multiples of one frequency. Sidebands are pairs spaced equally either side of a centre peak. A raised floor is broadband energy with no single frequency. A hump is a resonance being excited by whatever sits under it.
Generated by a non-sinusoidal once-per-turn event: an impact, a rub, a truncated motion, looseness. The more harmonics, the sharper the event. A pure unbalance gives 1× and almost nothing else.
Generated by modulation: something at the centre frequency being varied at the sideband spacing. Gear mesh with pinion-speed sidebands means the mesh force changes once per pinion turn, which is an eccentric or damaged pinion. Pole-pass sidebands on 1× mean rotor bar problems.
Random energy: turbulence, cavitation, flow separation, severe rolling-element wear, or simply a noisy cable. A floor that rises with load on one stage is process; a floor that rises across all channels on one date is instrumentation.
The spectrum averages away the very things that identify some faults: impacts, truncation, beating and modulation. The waveform keeps them. Select a pattern to see what produces it and how it appears in the spectrum.
CAT II covers single-plane balancing with a trial weight. The method is a vector calculation and the diagram below is the whole of it. Select each vector to follow the reasoning.
A resonance amplifies whatever forcing sits near it. It is the reason two identical pumps on identical bases can read completely differently, and the reason a small unbalance can produce a large reading. Select a region of the response curve.
The look-alike problems and the balancing and resonance work that each sector brings to the intermediate analyst. Select an industry.
For the analyst who diagnoses the difficult cases and the manager who owns the programme. This level covers rotordynamics on fluid-film machines, transient analysis, the process interactions specific to an ASU, and how to run a programme that produces findings rather than data. Assumes CAT II.
Every compressor pinion and the expander rotor floats on an oil film in a tilt pad bearing. The film has stiffness and damping, and those two properties decide whether the rotor is stable. A CAT III analyst has to understand what makes a rotor go unstable, because on these machines it is the fault with the least warning.
Select a condition to see its effect on the rotor's stability margin. The rotor is stable while the damping available exceeds the destabilising force. Anything that reduces damping (worn pads, low viscosity, lost preload) or increases the destabilising force (higher gas density, seal cross-coupling) eats the margin.
Inside the bearing. Choose the bearing type and the condition, and watch the shaft, the oil wedge and the effects on film, temperature and stability.
A plain sleeve bearing generates its own cross-coupled stiffness from the oil wedge, which is what drives oil whirl at just under half speed. Tilt pads let each pad rotate to face the load, which removes most of that cross-coupling. That is why classic whirl is rare on these machines and why a sub-synchronous component points elsewhere.
At high discharge pressure, the gas in impeller eyes and labyrinth seals produces a tangential force on the rotor. API 617 calls this aerodynamic cross-coupling and requires it to be analysed. It scales with gas density and pressure, which is why it appears on the high-pressure stages and follows load rather than speed.
The bearing's damping is what holds the margin. Worn pivots, lost preload, a pad fitted wrongly, oil too hot or too thin: each reduces damping. A rotor with healthy pads can tolerate a cross-coupling that the same rotor with worn pads cannot. This is why a bearing job is the most likely time to find an instability.
On a compressor, the single most useful CAT III discrimination is whether a change follows load or speed. They point at completely different faults and completely different fixes.
Select a case. Amplitude that scales with the square of speed and returns to the same value at the same speed is unbalance. Amplitude that follows discharge pressure or flow at constant speed is instability or a process effect. On a fixed-speed compressor, only the second case can happen while running.
Steady-state data tells you what the machine is doing now. Transient data tells you what the rotor-bearing system is. A coast-down, plotted as amplitude and phase against speed, is the single most information-rich measurement a CAT III analyst can take.
A Bode plot: amplitude and phase against speed. Select each feature. The critical speed is where amplitude peaks and phase passes through 90°. The height of the peak is set by damping: a sharp, tall peak means low damping. A critical that has moved since commissioning means the support stiffness has changed.
On an air separation unit, a large share of vibration findings are process events with a vibration symptom. A CAT III analyst reads the process trends with the vibration, not after it. The interactions that matter:
Every fault cell on this site lists candidate causes. None of them is a finding until it has evidence from the machine in question. Before a repair is authorised, a CAT III analyst gets at least two of the following to agree:
The exact moment the reading changed and what the plant was doing in that minute. A cause that cannot explain the timing is not the cause.
Bearing metal temperature, axial position, oil analysis, motor current, process tags. Vibration alone rarely proves anything.
Change the load, change the oil temperature, open the breaker, overlay an old coast-down. If the fault does not respond as the hypothesis predicts, cross it off.
A photograph, a borescope image, a worn part, a lab result. This is what turns a likely cause into a proven one.
Has this machine done this before? What was done, and did it hold? A cause that keeps coming back was never the cause.
Write "consistent with" and "most likely" when that is what the evidence supports. Keep "caused by" for when you can show it. An overconfident cause gets the wrong part replaced and closes the job.
A condition monitoring programme is judged on findings acted upon, not on data collected. The programme manager's job is to make the second lead to the first. The elements:
Select each element for what it involves on an ASU.
Alert and danger are different instruments. Alert is a call to investigate and can be set by the programme. Danger is a protective trip with a voted time delay and is set by the machinery protection philosophy. The manager owns the first and defends the second.
Set from the machine's own baseline, not from a table. A sensible starting rule is two to three times the healthy running level, capped below danger. On this plant: 1.5 mils on the shaft probes.
Set from bearing clearance and OEM data, with a short voted delay. 2.0 mils here. Never raised to keep a machine running; a bypass goes through management of change with a time limit on it.
The alarm most programmes lack. A level that has doubled in a week is a finding even if it is still below alert. Trend-based alerting catches the fouling ramp and the slow bearing wear that fixed limits miss.
| Configuration | What to expect |
|---|---|
| Independent FNC | Takes warm gaseous nitrogen from the cold box and raises it to liquefier or pipeline pressure. Clean, dry gas, so fouling is rare; suction conditions swing with the cold box, and it sees every column upset first. |
| Independent RNC | Circulates nitrogen around the liquefier refrigeration loop at high pressure ratio. Usually the highest-pressure machine on site, with the hardest-worked pinions, but the bearings and the oil still lead the findings. A liquefier trip removes its load in seconds, so surge and load tracking instability come next. A liquefier trip removes its load in seconds. |
| Combined FNC/RNC | Feed and recycle stages on one bull gear, one casing set, one lube system. Anything that happens on one service reaches the other through the gear and the thrust balance. Diagnose it as one machine, never as two. |
On all three the drive is large and often variable speed. The ASU-family cells for icing, adsorber carryover and liquid ingestion do not apply, because nitrogen service is dry and clean. Bearings and lubrication lead, then surge and load tracking instability, then impeller rub and water from the interstage coolers. Gear mesh and thrust follow.
Every fault cell lists likely causes. None of them is a finding until you have evidence from your own machine. Before a repair is authorised or a report is issued, get at least two of these to agree:
Be honest in the report about which it is. Write "consistent with" and "most likely" when that is what the evidence supports, and keep "caused by" for when you can show it. An overconfident cause gets the wrong part replaced, closes the job, and the fault comes back with the credibility of the whole programme attached to it.
Every compressor on this plant is interstage water cooled through shell-and-tube exchangers. That puts cooling water on the other side of a tube wall at every stage of the MAC, the BAC and the nitrogen machines, so water ingestion is a credible fault on all of them rather than a rare one. It is worth treating as a standing possibility whenever a compressor changes suddenly:
Liquid through a compressor erodes impellers, can cause a rub, and can bend a rotor. Prove which cooler before restarting: the machine damage is the consequence, and the exchanger is the cause.
Every compressor pinion, bull gear and expander here runs on tilt pad bearings. That changes what to look for:
Shaft vibration limits for these machines come from the OEM and the site protection philosophy. Where a general benchmark is needed, the usual starting point for fluid-film machines is the ISO 20816 shaft vibration guidance, which scales with running speed — high-speed pinions and expanders have far tighter limits in mils than a 1800 rpm motor.
Two probes 90° apart on a compressor pinion or the expander rotor give more than two amplitudes. They give the path the shaft centre draws each turn (the orbit) and the average position of the shaft within its clearance (the centreline). Both are diagnostic in ways a single amplitude is not. Select an orbit.
Every compressor, gearbox and expander on the plant runs on a shared oil system, and the oil passes over every bearing and gear before it goes back to the reservoir. That makes it a sampling point for the whole machine. Wear that has not yet changed the vibration has usually already changed the oil. Select a stage of the timeline.
On a fluid-film machine, oil analysis and vibration are the two halves of one picture. Oil sees material leaving the surfaces; vibration sees the geometry that the missing material leaves behind. Oil generally moves first.
| Test | What it measures | What it tells the vibration analyst |
|---|---|---|
| Viscosity at 40 °C | Whether the oil is still the grade the bearings were designed for. | Low viscosity thins the film; the first vibration symptom is a slow 1× rise with bearing metal temperature rising, then instability. A film problem that shows in the oil first. |
| Water (Karl Fischer, ppm) | Dissolved and free water in the oil. | Our compressors are interstage water-cooled with shell-and-tube exchangers. Water in the oil says a cooler or seal is leaking. Water destroys the film, strips additives and rusts babbitt. Expect bearing distress on the affected machine to follow. |
| ISO 4406 particle count | Cleanliness: particles per ml in three size bands, reported as three code numbers. | A rising count with no change in wear metals is contamination entering, not wear. A rising count with rising Fe or Cu is wear generating debris. Tilt pads with a scored journal often begin here. |
| Spectrometric wear metals (ppm) | Fe, Cu, Pb, Sn, Cr, Al and others, particles below about 5 µm. | Tin, lead and copper together are babbitt: a pad is wiping. Iron alone with the gearbox on the same system is gear or journal steel. Aluminium points at a labyrinth seal. This is the test that names the component. |
| Ferrography / analytical ferrography | Larger ferrous particles, their shape and count. | Cutting wear (long curls) is abrasion from contamination. Fatigue platelets are rolling contact damage on a gear tooth. Spherical particles are early fatigue. Shape tells the mechanism. |
| TAN and oxidation (FTIR) | Acid number and oxidation by-products. | Oxidised oil varnishes. Varnish on tilt pads and in servo valves changes clearance and damping. On a machine that has become marginally unstable with no other change, look at the oil's condition, not just its level. |
| RPVOT / RULER | Remaining antioxidant additive. | Warns that the oil is about to oxidise before TAN moves. A charge-life question, not a machine question, but it sets the interval for everything above. |
Vibration alarm on one pinion with no oil change: mechanical or process, local to that stage. Oil alarm (babbitt metals or water) with vibration still flat: film distress beginning; watch bearing RTDs, shorten the vibration interval, plan the outage. Both moving together: the failure is underway and the timing is now a shutdown decision.
Same point, same method, same interval. Sample from a live line upstream of the filter, machine at normal operating temperature. A sample from the reservoir drain is sediment, not oil. Trend against the previous samples from the same machine; a single result is nearly useless.
Water in oil is the number one contaminant on these machines because of the interstage coolers. Cold ambient oil on start-up thickens the film and raises bearing metal temperature transiently. The turbo expander oil and the compressor oil are usually separate systems; do not compare their trends.
Every mechanical fault dissipates energy, and almost all of that energy ends as heat. A thermal camera shows where the heat is. It cannot see inside a compressor casing, but on motors, couplings, pump and fan bearings, gearboxes and electrical gear it often confirms or rules out a vibration finding in one image. Select a hot spot on the image.
Airborne and structure-borne ultrasound instruments listen in the 20 to 100 kHz band, well above hearing and above most of the vibration spectrum. Friction, impacting and turbulence all generate energy there long before they generate enough force to move a housing. On rolling-element bearings this is the earliest warning available; on fluid-film machines it is a leak and electrical tool. Select a band.
Instruments report dB above a baseline taken on the same bearing when healthy. Widely used field guide: +8 dB lack of lubrication; +16 dB early failure; +35 dB and above, advanced failure with audible content. Vibration envelope data typically begins to show the defect around the +16 dB point. Ultrasound therefore extends the warning by weeks to months on slow or lightly loaded bearings.
The single most valuable use on motors, pumps and fans. Add grease while listening: the level drops as the film re-forms, then rises again when the bearing is full. Stop there. Over-greasing is as common a cause of motor bearing failure as under-greasing, and a schedule cannot tell the difference.
Compressed air, instrument air, nitrogen and vacuum leaks are turbulent and loud in the ultrasound band. Corona, tracking and arcing in switchgear generate ultrasound before they generate heat. Both are routes for the same instrument on the same day as the vibration route.
Each technology sees a failure at a different point in its development. The P–F curve places them in order from the first detectable sign (P) to functional failure (F). The programme manager's job is to have the right technology watching at the right interval so that P is found with enough time to plan. Select a point on the curve.
The advanced analysis and the programme management that each sector demands, and the technologies beside vibration that carry it. Select an industry.
For the specialist who is called when the standard methods have not settled it: rotordynamic modelling, torsional analysis, structural resonance, and the judgement calls that carry real risk. This level is less about new techniques and more about knowing the limits of every technique below it. Assumes CAT III.
A lateral rotordynamic model represents the rotor as a chain of beam elements with mass and stiffness, supported on bearings represented by stiffness and damping coefficients that vary with speed and load. From it you get the undamped critical speed map, the damped natural frequencies, the mode shapes, the unbalance response and the stability (log decrement) at each operating condition.
Select a mode. The first bending mode of a pinion puts maximum motion near mid-span and least at the bearings, which is why probes inboard of the bearings see more than probes at them. The mode shape also tells you where a balance correction is effective and where it is not.
Predicting where the criticals are and how far they sit from running speed; predicting the stability margin under a stated cross-coupling; predicting the response to a stated unbalance; deciding whether a proposed bearing or seal change helps. It converts "we think" into "the model says", which is a different conversation with an OEM.
Telling you what is wrong with a real machine. The bearing coefficients in the model assume a bearing built to drawing with the specified oil at the specified temperature. A worn pivot, a lost preload, a wrong clearance or a rub are not in the model, and a machine with any of them will not behave as the model predicts. Divergence between model and measurement is itself the finding.
Level I is a screening calculation using an empirical cross-coupling estimate; if the predicted log decrement is above a threshold and the machine sits in the low-risk region of the flexibility-ratio chart, no further analysis is required. Level II requires a full seal and impeller force model. The nitrogen recycle and booster machines on this plant are the ones most likely to sit near the Level II boundary.
A shaft train twists as well as bends. Torsional natural frequencies are set by the inertias of the motor rotor, coupling, gear and pinions and the torsional stiffnesses between them. They are excited by motor torque pulsations, VSD harmonics, gear mesh and process pulsation. The vibration is angular, not lateral, so casing accelerometers and proximity probes largely do not see it. The evidence arrives as broken coupling elements, sheared keys and fatigue-cracked shafts, usually with little warning.
A torsional Campbell diagram: natural frequencies against speed, with excitation orders drawn as lines from the origin. An intersection inside the operating range is an interference and needs a separation margin or a damping justification. Select an order line to highlight its intersections.
The machine is not the only thing that can resonate. Baseplates, pedestals, fan decks, pipework and bearing housings all have natural frequencies, and any of them coinciding with an excitation amplifies a small force into a large vibration that no amount of balancing will remove. The expert's tools are the bump test, the operating deflection shape and, where needed, a full modal survey.
An operating deflection shape of a fan deck at fan speed. Select points to see their relative amplitude and phase. The shape shows the deck bending as a plate about its long axis, which tells you where a stiffener would be effective and where it would do nothing.
Strike the stationary structure with an instrumented hammer or a dead-blow mallet and capture the ring-down. The frequencies that ring are the natural frequencies. Compare with the forcing frequencies the machine will produce at speed. A natural frequency within about 15% of a forcing frequency is a resonance risk.
Measure amplitude and phase at many points on the running structure at the frequency of interest, all referenced to one fixed sensor. Animate the result. The ODS shows how the structure is actually moving, which is what tells you whether to stiffen, add mass, or move the excitation.
First move the excitation away (change speed, change blade count, fix the unbalance driving it). Second change the natural frequency (stiffen or add mass at the antinode). Damping alone is rarely enough for a structural resonance. Detuning by trial is a last resort.
The cases that reach a CAT IV analyst are the ones where the standard method has given an ambiguous answer. These are the ones specific to this plant, with the discriminator that settles each.
Single-plane trim balancing is a CAT II skill. The expert cases are multi-plane balancing of flexible rotors, balancing through a critical, and knowing when balancing is the wrong answer.
Influence coefficient balancing: the response vector at each probe for a trial weight at each plane. Select a plane to see its influence at both bearings. A plane with strong influence at one bearing and weak at the other cannot correct a couple; you need both planes.
At this level the technical answer is often clear and the hard part is the decision. These are the calls that come to a CAT IV analyst on an ASU, with the reasoning that should go into each.
The question is never "is it above the limit" but "what is the failure mode, how fast does it progress, and what does failure cost". A slow fouling ramp above alert can run to the next planned stop. A sub-synchronous component below alert cannot. Write the reasoning down before the decision, not after.
A request to raise a danger setpoint to keep running is a request to remove a protective function. The answer is a management-of-change review with a time limit, a stated basis, and a named owner. The analyst's job is to state the failure mode the trip protects against and what the margin actually is.
The OEM has the model, the drawings and the fleet history. The analyst has the measurement. When they disagree, the measurement wins on what the machine is doing and the OEM wins on what it was designed to do; the gap between the two is the finding. Bring the transient data, not just the steady-state.
A bearing job on a tilt-pad machine is signed off against the pre-outage coast-down, not against the alarm limits. Same critical speed, same peak height, same slow-roll value, same resting position: then it is the same machine. Anything else is a question.
The expert's most valuable sentence. A confident wrong diagnosis at this level replaces the wrong part, closes the job, and costs the programme its credibility when the fault returns. State what the data supports, what it rules out, and what measurement would settle it.
Every difficult case is a teaching case. Write it up with the data, the reasoning, the wrong turns and the confirmation. The CAT II analyst who reads it is the one who will catch the next one early.
A Campbell diagram plots natural frequencies against speed with the excitation orders drawn as lines from the origin. Where a line crosses a natural frequency there is a potential resonance; whether it matters depends on damping and on whether anything actually excites that order. Select a crossing.
Oil whirl and oil whip look alike in a single spectrum and behave completely differently in a cascade. Telling them apart decides whether the fix is a bearing change or a rotor change. Select a region of the cascade.
Sub-synchronous, typically 0.42 to 0.48×, tracking speed. The fluid film itself is driving the rotor around the clearance. Cured by the bearing: preload, geometry, clearance, load direction, tilt pads instead of plain sleeves. Our machines have tilt pads precisely to prevent it, so whirl on a tilt-pad machine means a pad problem.
The whirl frequency reaches the first rotor natural frequency and locks there while speed continues to rise. Amplitude grows to the clearance. The rotor is now resonating with the film feeding it. A bearing change alone often fails; the rotor's first mode and its damping have to change.
The pattern seen on our N2 compressors and boosters: a sub-synchronous component that appears with load, not speed, on a fixed-speed machine. Cross-coupled aerodynamic forces at the impeller or seals, or a bearing that loses its film with load. Distinguished by the process trend, which is why the load-tracking lesson comes first.
A rotor is stable when its damping exceeds the destabilising forces acting on it. Log decrement is the number that expresses that margin: positive is stable, zero is the threshold, negative is unstable. API 617 asks for a stability analysis on every centrifugal compressor and sets minimum values. Move the cross-coupling slider to see the margin consumed.
With two orthogonal probes, the spectrum can be computed for the orbit rather than for each probe separately. The result has a positive side (forward, with rotation) and a negative side (backward, against rotation). The split tells you what kind of force is acting. Select a component.
The expert-level studies and judgement calls by sector. Select an industry.
Every sensor on the plant, what it measures, how it works, where it belongs and where it is the wrong choice. The machine diagrams at the bottom show where each device sits on the real equipment. Select any device or any marker.
Four sensor types cover everything on an air separation unit. Each one measures a different physical quantity and is right for a different kind of machine. Select a device to see how it works and its cut-away.
The choice follows from the bearing type and the frequency of the fault you are trying to see. The rule is short: fluid-film bearings get proximity probes, rolling element bearings get accelerometers. The reasons are what a CAT II analyst has to be able to explain.
Select a fault to see which sensor sees it and why. The frequency axis is the deciding factor: displacement is strongest at low orders, acceleration is strongest at high frequency, and the two cross over around blade pass.
The shaft floats on an oil film and the casing barely moves even when the rotor moves a lot. An accelerometer outside is reading through that film and the housing and misses most of it. The limit on a fluid-film machine is clearance, not force, and only a probe measures where the rotor actually is.
Rolling contact and almost no clearance. Force passes straight into the housing where a casing sensor reads it clearly. A probe would have almost nothing to measure. The faults that matter are high-frequency and non-synchronous, which is acceleration territory.
Gear mesh runs to several kHz, far above what a probe can follow, so mesh condition needs an accelerometer on the housing. But the pinions inside run on tilt pads, so their rotor behaviour needs probes. Different faults, different physics, different sensor.
A reading is only as good as the weakest link between the shaft and the screen. Select each stage to see what can go wrong there and how it shows in the data.
Most measurement errors are mounting errors. The sensor can be perfect and the data worthless. Select a mounting method to see its usable frequency range and what it costs you.
A magnet mount resonates far below a stud mount and hides everything above its resonance. Gear mesh and bearing defect frequencies are exactly the content a magnet mount loses.
Select a machine, then select a flashing marker. Each marker explains what is measured at that point, why that sensor, and what it catches. Diagrams are generic schematics of the standard machine types on an ASU, not any manufacturer's drawing.
Vibration travels. A fault in the rotor, the coupling, the base or the alignment loads the whole machine and shows on several probes at once. When a change appears at one stage, or at one end of a rotor, and the neighbouring channels do not move at all, the most likely explanation is that the source is local to that measuring point and there is nothing to carry it anywhere else. Most often it is the bearing at that stage: a worn pad, a lost preload, a damaged pivot, a bearing that has lost its oil film or its fit. The other common explanation is the impeller or the process at that stage: fouling or damage on that one wheel, or flow that has gone rough, such as turbulence, recirculation or stall local to that stage. Check both before concluding anything about the rotor as a whole.
Rolling-element faults live in acceleration and envelope data. Balance, alignment and looseness live in velocity. Fluid-film and rotordynamic behaviour lives in shaft displacement, which is why the compressors and the expander carry proximity probes and the motors and pumps do not.
Which devices carry the programme in each sector, and the installation rules that go with them. Select an industry.
Industry studies attribute between a third and a half of all rolling-element bearing failures, and most tilt-pad and gear failures, to lubrication: the wrong lubricant, the wrong amount, contamination, or oil that has stopped protecting. A lubrication programme is the cheapest reliability programme a plant can run and the one most often missing. This section covers what a good programme looks like, how to read oil analysis, the ISO cleanliness code, water, wear metals, varnish, grease, and how it ties to vibration.
Six elements, all of which have to exist for the programme to work. A plant that buys the right oil but stores it in an open drum outdoors has no programme. Select an element.
Viscosity decides film thickness, and film thickness decides whether the surfaces touch. Oil thins with temperature, so the grade is chosen for the operating temperature, not the ambient. Move the sliders: the chart shows the viscosity of each ISO VG grade at your operating temperature against what the bearing needs.
Cleanliness is reported as three numbers: the particle counts per millilitre larger than 4, 6 and 14 µm, each converted to a code where every step doubles the count. Move the sliders to build a code, then compare with the target for the machine type.
Servo hydraulics 15/13/10. Turbine and compressor circulating systems 16/14/11. Gearboxes 17/15/12. Bearing lubrication systems 16/14/11. New oil from the drum is often 19/17/14 or worse: filter it in.
Bearing fatigue life is sensitive to particles in the size of the film thickness: a few microns. Going from 19/17/14 to 16/14/11 can double or triple rolling-element bearing life. Tilt pads are scored by the same particles.
Filter new oil in through a 3 µm filter; seal reservoirs and fit desiccant breathers; keep transfer containers clean and dedicated; filter continuously on circulating systems; sample monthly and trend the code.
Select a cleanliness level below for the bearing life effect.
Water enters through breathers, seals, coolers and condensation, and leaves as rust, hydrogen embrittlement, additive depletion and a collapsed film. Below saturation it is dissolved and invisible; above it, free water settles and damages whatever it touches. Move the sliders.
A routine oil sample carries a dozen tests. Each fault condition moves a particular set of them, and the pattern names the problem. Select a condition for the tests that move, the example trend, and the action.
| Test | What it reports | Alarm thinking |
|---|---|---|
| Viscosity @ 40 / 100 °C | Grade and shear or dilution loss | ±10% from new oil: investigate. ±20%: change. |
| Water (KF) | Dissolved + free water, ppm | 200 ppm bearings, 500 ppm gears; any free water is an action. |
| ISO 4406 code | Particle contamination | Two codes above target at any size: action. |
| Wear metals (ICP) | Fe, Cu, Pb, Sn, Cr, Al, Ni <5 µm | Rate of change, not absolute value; trend per machine. |
| Contaminant metals | Si, Na, K, B, Ca, Mg | Si = dirt; Na/K/B = coolant or additive; Ca/Mg = wrong oil mixed in. |
| Ferrous density / PQ | Large ferrous debris | Catches particles ICP misses; a rise with flat Fe is big debris. |
| Analytical ferrography | Particle shape and size | Cutting, fatigue, sliding, spheres: names the mechanism. |
| TAN / TBN | Acidity (oxidation) / reserve alkalinity | TAN rising 0.3–0.5 above new oil: oxidation under way. |
| FTIR | Oxidation, nitration, water, soot, additive bands | Fast screening; confirm with the chemical tests. |
| RPVOT / RULER | Remaining oxidation resistance / antioxidant | Below 25% of new: plan the change before varnish starts. |
| MPC (varnish potential) | Colour of insoluble degradation products | Under 15 good; 15–30 monitor; 30–40 abnormal; over 40 critical. |
| Foam / air release / demulsibility | Surface-active properties | Degraded in oxidised or contaminated oil; servo and bearing problems follow. |
Varnish is the thin, hard, insoluble film that forms on bearings, servo valves, coolers and reservoirs when an oil's degradation products exceed its ability to hold them in solution. Group II base oils, hot spots, electrostatic discharge and air entrainment all drive it. It sticks servo valves, raises bearing temperatures, plugs coolers and filters, and causes trips that look like instrumentation problems. Select a stage of its formation.
Move the slider to read the MPC (membrane patch colorimetry) scale, the standard varnish-potential test.
Sticking inlet guide vane and governor servo valves on compressors and turbines (unexplained trips, slow response); bearing metal temperatures rising with vibration flat; cooler approach temperatures rising; filters loading faster.
Thermal and oxidative stress (hot bearings, hot spots), electrostatic spark discharge in fine filters on dry Group II oils, air entrainment and micro-dieseling, and the antioxidant package running out. RPVOT and RULER see the antioxidant going before the varnish appears.
Monitor MPC and RPVOT quarterly on every turbine and compressor oil. Above MPC 30: fit a varnish-removal system (resin or electrostatic), review bearing and cooler temperatures, and plan the change-out with a flush. Choose an oil with proven varnish resistance when the charge is replaced.
Most motor, pump and fan bearings are grease-lubricated, and both over-greasing and under-greasing kill them. The interval depends on bearing size, speed, temperature and the grease; the amount depends on the bearing dimensions. Move the sliders for a first-pass interval and quantity, then confirm each bearing with ultrasound.
Select a sampling point on the circulating system for what a sample from there tells you and what it hides.
Select an industry.
Ultrasound instruments listen in the band above hearing, 20 to 100 kHz, where friction, impacts, turbulence and electrical discharge produce energy long before they produce vibration or heat. On rolling-element bearings it is the earliest warning available and the only reliable way to grease; on plant air it finds the leaks that cost more than any single machine; in switchgear it hears corona before thermography sees heat. This section covers the physics, the instruments, bearings, lubrication, leaks, steam traps, valves, electrical and slow-speed applications, and the programme.
High-frequency sound is directional and does not travel far, which is what makes ultrasound good at locating a source. The instrument shifts the band down into the audible range (heterodyning) so the operator hears it in headphones, and reports a level in decibels. Select a part of the chain.
Two ways to listen. Airborne, with a scanning module or a parabolic dish, for leaks, electrical discharge and anything that radiates into the air. Structure-borne, with a contact probe or a magnet-mounted sensor, for bearings, gears, valves and steam traps. Select a technique.
A bearing's ultrasound level is read against its own baseline, taken when it was known to be healthy at the same point with the same instrument. Move the slider through the ladder; the sound sample and the matching vibration spectrum change with it.
The one application that pays for the instrument on its own. Add grease while listening: the level falls as the film re-forms and stops falling when the bearing is full. Select a stage.
A leak is turbulent and loud in the ultrasound band, and compressed air is the most expensive utility in most plants. Move the sliders to estimate what a leak costs and what a survey of the whole plant is worth.
A steam trap discharges condensate and holds steam; a failed one either blows steam continuously or holds condensate and goes cold. Ultrasound on the trap body, with a temperature reading, tells which. Valves that should be shut but pass make the same turbulent sound at the seat. Select a condition.
Electrical discharge ionises air and produces ultrasound before it produces heat or a visible mark. An airborne scan of switchgear, bus ducts, insulators and transformer bushings from a safe distance finds it with the enclosure closed. Select a discharge type.
Below about 100 rpm a damaged bearing generates almost no vibration energy and the defect frequencies sit below what a route accelerometer resolves. Conveyor pulleys, cooling tower fan shafts, kiln and mill trunnions and agitators are the common cases. Ultrasound hears the impacts regardless of speed. Move the speed slider to see the two technologies swap places.
What an ultrasound programme contains, by sector. Select an industry.
Every mechanical, electrical and process fault dissipates energy, and almost all of it ends as heat. A thermal camera sees that heat as an image, which makes thermography the fastest survey technology on the plant and the usual confirmation for a vibration or ultrasound finding. It is also the easiest to get wrong: emissivity, reflections, distance and load all change what the camera reports. This section covers the physics, the camera, electrical, mechanical, motor and process applications, severity criteria, and the programme.
A camera measures radiated energy, not temperature. A dull painted surface radiates most of its heat (high emissivity); bare shiny metal radiates little and reflects the surroundings instead. Move the sliders to see how a wrong emissivity setting changes the reported temperature of the same object.
Resolution, spot size and distance decide whether a small hot component can be measured at all. Move the distance slider: the spot the camera can resolve grows with distance, and a connection smaller than the spot reads cooler than it is.
Loose, corroded or undersized connections, overloaded conductors, failing breakers and phase imbalance all show as heat. Severity is judged by the temperature rise above a similar component under the same load. Move the slider through the delta-T criteria.
Select a component on the switchgear image for the typical finding.
Mechanical heat is friction. A bearing running hot against its twin, a coupling hotter than the shafts, a belt hotter than its sheave, a gearbox with one hot corner. Select a finding for the image, the vibration cross-check and the cause.
A motor's thermal image shows the stator frame, the bearings, the cooling path and the terminal box, each with its own story. Select a region.
Thermography reaches well beyond rotating equipment: insulation and refractory loss, heat exchanger fouling, tank and vessel levels, steam traps, boiler and HRSG tubes, fired heater tubes and flue gas leaks. Select an application.
A thermography report is an image pair (thermal and visible), the measured temperatures and the reference, the load, the ambient, the severity class and the recommended action. Select a priority level.
Select an industry.