Why equipment vibration monitoring does not catch an early defect
Formally, equipment vibration monitoring is implemented: sensors are installed, routes are done, setpoints are displayed on the control panel. Yet the unit still seizes and stops the line.
The reason is not the hardware — the signal about an incipient defect is present in the data, but the acquisition and processing scheme does not let it through. It is lost in three places.
First — a one-off measurement. A route once per shift gives a point, not a picture: between two points the defect passes through several stages of development. A roller that has just begun to spall reaches failure, and instead of dynamics the operator gets a completed fact.
Second — a static threshold by absolute value. The setpoint is taken with a margin to the nameplate level to avoid false alarms, and it trips already at an irreversible stage.
Plus different units have different backgrounds: the same level for a new and a worn unit means opposite things.
Third — no trend and no binding to operating mode. Vibration depends on load, speed, and temperature, so absolute values between shifts are not comparable.
Early wear of a bearing unit is weakly manifested in the overall level — it is visible in the high-frequency part of the spectrum and in the signal envelope, where no one looks during a one-off measurement.
The result is a sudden stop that was predictable weeks before it. Below we analyze how this is closed: continuous acquisition, normalization to operating mode, trends by characteristic frequencies of a specific unit.
Early detection of defects by vibration and temperature
A bearing defect, misalignment, or loose mounting manifests in the vibration spectrum long before the unit starts rattling.
We decompose accelerometer signals into harmonics: a rise in amplitude at the rolling frequency and its sidebands indicates incipient spalling, while an increase in the rotational component indicates a centering violation.
Condition monitoring of units is continuous, not a route once per shift, so a developing defect is recorded in the first hours, not at the stage of destruction.
Temperature works as an independent confirmation channel. Heating of the bearing housing, gearbox, or motor winding cuts off false alarms based on vibration alone: only the coincidence of two signs triggers, and the on-duty shift does not receive unnecessary alarms.
For the hardware part, we use industrial accelerometers and vibration analyzers SKF — this provides a stable signal where consumer-grade sensors drown in shop-floor noise.
The history accumulated for each unit provides a wear trend: the growth rate of harmonics is extrapolated to the threshold level, and we obtain an estimate of remaining useful life.
Repair is planned in advance — in the window of a planned line shutdown, with a ready bearing and an available maintenance technician — not at night after a failure.
In practice, the share of sudden failures is reduced severalfold, and the time for unit diagnostics drops from several hours to reviewing a single card with graphs and a verdict.
Data collection modes: continuous, periodic, hybrid
Measurement mode is determined not by platform capabilities but by the physics of the fleet: how quickly a defect develops, whether the unit has power and stable communication, and how critical downtime is.
A bearing in a high-speed pump degrades in hours — it must be seen continuously, whereas wear of a low-speed conveyor gearbox is read by the slope of a trend over weeks.
An error in either direction costs the same: infrequent polling leaves a blind zone, too frequent pollutes the channel and storage with noise in which the useful sign is lost.
| Mode | When suitable | Limitations |
|---|---|---|
| Continuous measurements | Critical units: compressors, turbines, pumps with a high cost of failure. Spectrum and early warning are needed | Requires power and a stable channel for each acquisition channel; the volume of raw data grows linearly with the number of points |
| Periodic routes | Standard pumps, fans, areas with a low cost of downtime | Blind zone between routes: a defect can go from detection to failure within a single interval |
| Hybrid scheme | Mixed fleet: critical units under continuous monitoring, the rest on schedule and triggers | A unified tag and time model is needed, otherwise route measurements are not comparable with continuous trends |
A separate layer is wireless sensors: they remove the cable routing limitation and cover units in hard-to-reach areas — overhead racks, shafts, enclosures, where a route technician rarely gets.
The price for this is battery life and a scheduled replacement procedure, which must be built into the same scheme, otherwise part of the fleet quietly drops out of monitoring.
Therefore we assemble the hybrid scheme by classes of units and their cost of failure, not by default: first we decide where continuity is mandatory and where a route is sufficient.
Implementation stages: from fleet survey to system handover
The order of work is structured so that each subsequent stage relies on the data from the previous one.
First we record which failures need to be prevented, and only then select measurement points and thresholds — otherwise the system collects a stream of data but does not answer the question of what to do with it.
- Fleet survey. Inventory of units, modes, and operating conditions. Priority is units that have already failed and whose shutdown is most expensive, not blanket coverage.
- Sensor placement scheme. We select points by unit kinematics: bearing supports, axis directions, access without disassembly. For each point we record which defects it catches.
- Installation and connection. Mounting, routing, checking channels for interference from neighboring machines. Here it is visible where the signal is clean and where indirect signs are needed.
- Threshold setup. Baselines are based on mode statistics, not a single measurement. Startup, run-up, and load changes are analyzed separately; warning and alarm levels are oriented to ISO 10816.
- Training the model on failure history. Repair logs are matched with the signal archive; some periods are held out from training.
- Validation on the failure archive. A retrospective run shows how many known failures the system would have seen and how long before shutdown; the threshold and features are adjusted based on the result.
- System handover. Procedures for reacting to a signal, training of the operations service, review of the first alerts together with the customer's engineers.
Scope of supply: diagrams, documentation, maintenance procedures
A vibration and temperature monitoring system lives for years: personnel change, maintenance contractors change, and the equipment fleet itself changes.
So that operation does not depend on the developer, the documentation package is assembled so that each loop can be deployed, checked, and restored on your own.
- Sensor placement diagram for each unit: installation points, orientation, places where installation is unacceptable due to heat or flow.
- Acquisition configurations: polling rate, averaging windows, warning and alarm thresholds separately for each equipment group.
- Description of models and features: which spectrum components and temperature trends are considered normal, and which indicate bearing wear, imbalance, or lubricant degradation.
- Maintenance procedure: frequency of channel verification, procedure for replacing sensors, actions when communication with a measurement point is lost and in case of a false alarm.
- Operations manual: what the operator does when a signal triggers, to whom and within what time it is passed, how confirmation and taking equipment out for repair are recorded.
- Response rules: which alarm level raises the on-duty operator, which stops the unit, who has the right to remove an interlock.
- Integration scheme: how data goes to the existing dispatch system and where measurement history is stored for incident review.
The package is versioned together with the configuration: when a new unit is commissioned, both the model description and the thresholds are updated.
Handover includes training — the customer's engineers run installation, setup, and alarm review on live equipment, not from a presentation.
Case study: how the share of sudden stops decreased in a pump area
On the pilot area with pump units, condition was taken by a route once per shift: the mechanic recorded vibration and temperature in a log, and the trend for the unit was reconstructed after the fact.
A bearing defect or growing axial displacement manifested as a sudden stop — after the pump had already gone into catastrophic failure.
Sudden failures accounted for about a third of area downtime, and the time from the first sign to reaction was measured in days: between routes the unit managed to reach a stage where the rotor, not the bearing, is replaced.
We equipped the pump units with stationary vibration and temperature sensors, set reference levels by machine class according to ISO 10816, and configured two boundaries — warning and alarm.
The system maintains a trend and analyzes the spectrum, so a bearing defect, misalignment, and cavitation are distinguished from each other rather than reduced to a single "threshold exceeded."
Metrics before and after
The share of sudden stops at the pilot decreased by about threefold, and the time from a warning signal to taking the unit out for repair shortened from days to several hours.
False alarms were filtered out by tuning thresholds for each pump type — the operator stopped going out to inspect for every peak.
The main shift is the transition to planned repair: bearing replacement, alignment, and seal replacement fall within the area's shutdown window, not at the moment of maximum load.
For the pilot units, the share of work closed as planned grew from half to the majority, and the area stopped keeping an emergency pump reserve.
What to do if sensors on the units are already installed?
Installed sensors are not an obstacle but a ready measurement loop to which read access remains to be obtained.
The question is almost never about removing everything and installing your own, but about where to get the raw signal of the required bandwidth and sampling rate instead of the averaged value already being sent to the control system.
Therefore we start with an inventory: sensing element, interface, whose loop it is, and whether parallel acquisition is technically possible.
On a mixed fleet of units, one unit provides a digital stream over a bus, the second an analog signal through an existing controller, and the third is represented only by a relay contact.
This is a normal situation, and it is solved not by a single scheme but by a connection map for each type.
| What is on the unit | How we connect | Limitations |
|---|---|---|
| Sensor with an open protocol on a bus | Directly, reading over the existing bus | Polling schedule must be coordinated with the automation owner |
| Analog sensor on a controller | Parallel acquisition through an isolated gateway; we do not break the circuit | Polling rate is limited by the settings of the existing system |
| Closed controller without bus access | Through a gateway at the system output, export of averaged values | No source waveform — fine bearing diagnostics are unavailable |
| Sensor outside the operating range or band | Replacement with a new one with a separate channel | Installation and cable routing, but a diagnostic sign appears |
| Relay, limit switch, annunciator | We leave it in the existing loop, do not touch it | It has no value for assessing the technical condition of the unit |
Everything that cannot be acquired directly is routed through a data acquisition gateway: it reads the bus in read-only mode, buffers values during communication loss, and provides them with timestamps, while normalizing different data transfer formats into a unified tag model with a unit of measurement. The control loop remains untouched — interlocks and protections work as they worked.
Replacement is justified only where the signal physically does not contain a sign of malfunction.
If a sensor provides an averaged value once per minute, the rise in bearing vibration at an early stage is not visible through it — the data is formally there, but no decisions are made from it.
Engineers' concerns: false alarms, installation, operation
False alarms are almost always born not in the sensor but in the model: startup and shutdown, load changes, seasonal drift of bearing temperature, a neighboring unit on a common frame.
Therefore the reference base of "healthy" condition is built separately for each mode, not as one curve for everything.
An alarm is confirmed by an observation window and return hysteresis, plus the growth of the envelope at bearing frequencies is monitored — the overall vibration level by itself is too noisy a sign.
We perform installation without stopping the unit: during commissioning, points are placed on temporary mounting pads, then transition to stationary ones — a point takes from one to two hours in a planned downtime window.
As for mode changes, labels for startup, load, and repair come from the process control system or are marked manually, and the transient is not interpreted as an alarm.
How do you filter out false alarms without losing a real defect?
Thresholds are adaptive: they are calculated from the spread of a specific point in a specific mode, not from nameplate values. An event is considered confirmed if the sign persists for several polling cycles and repeats on the next entry into the mode.
Misses are monitored on a labeled archive — the share of false negative results remains within the limits fixed at acceptance.
Is equipment shutdown mandatory for installation?
No. The main points are installed on a running unit; stopping is needed only for stationary installation and cable routing — and this fits within planned downtime. Some points can temporarily be run on temporary pads to capture a baseline profile before the nearest repair.
What is within the responsibility of the operations service?
Response according to the procedure: confirm or reject the event, record the cause, mark the repair and unit replacement. These labels return to the model as markup and refine the thresholds.
Routes and manual measurements are not required — only a person who will close the alarm and write down how it ended.
What happens after repair or a change in operating mode?
The model does not transfer the old norm to a new unit: after bearing replacement or a load change, profile retraining is launched on fresh data. Until statistics accumulate, a conservative threshold operates so that an early failure does not remain unnoticed.
Let's discuss the task: what data is needed for assessment and plan
The first layer of information about the facility is enough — we clarify the rest ourselves on a call. From this set we build the layout of monitoring points, the data model, and the order of work, so instead of general promises you get an assessment tied to your fleet.
- Fleet composition and operating conditions: types of units, quantity, power and criticality of each, operating mode (continuous, shift, seasonal), environment and load. This is enough to understand where vibration and temperature carry a diagnostic load and where measurement will add nothing.
- Access to the accounting system: which sensors and controllers are already installed, where archives are written, how many months of history exist, whether export is available, and how network segments are arranged. If there is no accounting, write that — it changes the acquisition scheme, not cancels it.
- What already hurts: which units failed, how it manifested — vibration growth, overheating of the bearing unit, sudden failure without precursors. Using these cases, we check whether the system would have seen them on the available data.
- Monitoring points and polling rate: for which units continuous monitoring is needed, where periodic acquisition is sufficient, whether real-time detection is required or archive analysis once per shift.
- Requirements for actions: to whom and how the system reports exceeding a threshold, whether integration with existing systems is needed and in what form you want to see reports.
- Work approval procedure: who accepts the result, whether phased acceptance is needed, whether there are downtime windows for installation and rules for contractor access to the site.
Next we return with a sensor layout, a staged work plan, and an assessment. Describe the task in free form — some points will be closed in correspondence, and we will clarify the rest on a call.
