Rotating equipment rarely fails without warning. Bearings wear, shafts drift out of alignment, mountings loosen, and each of these changes the way a machine vibrates long before anything stops. Vibration monitoring is one of the oldest forms of condition monitoring, and industrial IoT has made it cheaper to start. The hard part is no longer the sensor. It is deciding what to measure, what normal looks like, and who does something when it changes.

Choose Machines Worth Watching

Resist the urge to instrument the whole plant. Vibration data is most useful on assets where failure is costly, spares have long lead times, and the failure mode shows up as a change in vibration. Motors, pumps, fans, gearboxes, and compressors fit well. A simple conveyor idler that costs little and is swapped in minutes does not.

A useful way to shortlist: ask maintenance staff which machines caused the most unplanned stops or the worst nights. Their memory of past failures is better than any spreadsheet at pointing to the first five assets.

Understand What the Signal Is Telling You

You do not need to be a vibration analyst to start, but you should know the basic ideas. Overall vibration level is a single summary number that rises as a machine deteriorates; it is a good first alarm. Frequency analysis breaks the signal into components, and different faults tend to appear at different frequencies related to shaft speed. Imbalance, misalignment, looseness, and bearing defects each leave a recognizable pattern, though interpreting them takes training or a vendor’s analysis tooling.

Sensor placement matters more than people expect. Mount sensors rigidly, as close to the bearing housing as practical, on a clean, flat surface, in a direction recommended by the machine builder or sensor vendor. A sensor glued to a flimsy guard measures the guard. Check the sensor manufacturer’s current documentation for mounting and range guidance.

Collect a Baseline Before You Alarm Anything

The biggest mistake in early projects is setting alarm limits on day one from generic charts. Every machine has its own normal. Run the sensors for a few weeks across the operating conditions you actually see: different loads, speeds, and product runs. Record the typical range for each state, because a pump at full load and a pump at half load do not vibrate alike.

Also note maintenance events. When a bearing is replaced or a coupling realigned, vibration should change, and that is a free test that your sensor can see real differences. If it cannot, you have a placement or sensitivity problem to fix now.

Set Alerts People Can Act On

An alert is only useful if someone knows what to do with it. Start with two levels. A watch level means look at the trend and schedule an inspection. An action level means plan intervention soon. Tie each level to a named role and a standard response, not just an email to a shared inbox.

Decision	Questions to settle	Typical starting choice
Which assets	Failure cost, spares lead time, suitable failure mode	Five to ten critical rotating machines
Measurement	Overall level only, or frequency data too	Overall level first, add frequency later
Baseline period	Enough to cover load and speed changes	Several weeks across normal operation
Alert levels	Watch and action, with owners	Two levels, reviewed monthly
Response	Inspect, lubricate, realign, replace	Written playbook per asset type

Example: A Cooling Water Pump

Suppose a plant has a cooling pump that has failed twice during summer peaks. A sensor is mounted on the motor-side bearing housing and another on the pump side. After a month of baselining, the team learns the overall level sits in a steady band at normal load and rises modestly when flow is throttled. Weeks later, the pump-side reading climbs slowly while load stays constant. The watch level triggers, a technician checks lubrication and finds it low, and the trend flattens after servicing. Nobody celebrated, because nothing broke, which is exactly the point.

Avoid Common Pitfalls

Do not treat a rising number as proof of failure; confirm with a second signal such as temperature or a hand-held reading. Do not ignore the operating context, since a product change can alter vibration without any fault. Do not let data pile up unreviewed: schedule a short weekly check of trends, even if nothing is alarming.

Network and data decisions deserve thought too. Wireless battery sensors are easy to deploy but sample less often; wired sensors can capture richer data but cost more to install. Match the choice to how fast the fault you fear can develop. A bearing that degrades over weeks does not need high-rate streaming.

Safety comes first. Mounting sensors on rotating equipment must follow your site’s lockout and isolation procedures, and work in hazardous areas may require certified equipment. Consult your safety team and the equipment documentation.

Measure Whether It Is Working

After six months, ask plain questions. Did the system flag any real issues early? Were there false alarms, and why? Did maintenance planning change? If the answer to the last question is no, the data is decoration. Adjust thresholds, retrain on the playbook, or drop assets that never show useful signals.

Frequently Asked Questions

Do I need a vibration analyst on staff? Not to start. Overall-level trending is approachable for maintenance technicians. Deeper frequency diagnosis benefits from training or vendor support, and you can add it for the assets that justify it.

How many sensors per machine? Often one per bearing housing on the critical assets is the starting point, but follow the equipment and sensor documentation for your specific machine.

Can I use vibration data alone? It is powerful, but pairing it with temperature, current draw, or operating state gives better context and fewer false alarms.

How long until I see value? Usually after the baseline is stable and the first genuine drift appears. That timing depends on how often your assets actually degrade.

Is wireless good enough? For slow-developing faults, often yes. For fast or high-frequency fault detection, check whether the sampling capability suits your needs.

Next Step

Pick one critical pump or motor, ask maintenance for its last three failures, and write down which signal might have warned you earlier. That note becomes the justification for your first sensor and the first line of your playbook.