Condition Monitoring: Vibration, Temperature & Beyond

What Condition Monitoring Actually Is (and Is Not)

Condition monitoring is the practice of measuring physical parameters of operating equipment to detect changes that indicate developing faults. That is it. It is not predictive maintenance, though it is a prerequisite for it. It is not IoT, though modern systems use networked sensors. And it is not a replacement for the experienced technician who can tell something is wrong by the sound a machine makes - it is a way to formalize and extend that same intuition with data.

The concept is straightforward: a healthy machine produces consistent vibration signatures, temperatures, pressures, and electrical characteristics. As components degrade, those parameters shift in predictable ways. A bearing with a developing inner race defect produces vibration energy at a specific frequency related to its geometry and shaft speed. A motor drawing progressively higher current is telling you something about mechanical loading or winding condition. Condition monitoring is the discipline of listening to what the equipment is already telling you.

This guide covers the practical side of condition monitoring: which parameters to measure on which assets, how to select sensor types, where to set alert thresholds, and how to build a monitoring program that maintenance technicians will actually use. We will skip the theoretical signal processing math and focus on what you need to know to instrument a production line and start getting actionable information out of it.

The Five Core Monitoring Parameters

There are dozens of measurable parameters, but five cover the vast majority of industrial failure modes. Master these before getting exotic with oil particle counts or ultrasonic leak detection.

Vibration monitoring is the workhorse. If you can only afford one parameter, this is it. A single triaxial vibration sensor on a motor-pump assembly can detect bearing defects, shaft misalignment, structural looseness, and impeller damage - often months before any of these conditions produce symptoms an operator would notice. Accelerometers in the $300-$600 range are reliable for most industrial applications. You do not need laboratory-grade sensors for production monitoring.

Temperature monitoring is the simplest to implement and the easiest for technicians to understand. A bearing running 15 degrees Celsius above its historical baseline is a clear signal that something has changed. Infrared temperature sensors can monitor without contact, which matters on moving equipment or in high-voltage environments. The limitation is that temperature is a lagging indicator - by the time a bearing is running hot, the defect is already well-developed. Temperature monitoring catches problems, but it does not give you as much lead time as vibration.

Current monitoring is underrated and often free if you already have variable frequency drives with built-in current measurement. Motor current signature analysis can detect broken rotor bars, eccentricity, and mechanical load changes. It requires no additional sensors on the motor itself, which makes it attractive for applications where mounting vibration sensors is difficult - submersible pumps, sealed motors, or equipment in hazardous areas.

Matching Parameters to Asset Types

The most common mistake in condition monitoring programs is applying the same monitoring approach to every asset. A centrifugal pump and a belt conveyor have completely different failure modes, and they need different monitoring strategies. Here is a practical mapping based on what works in production environments.

A few things to notice. Continuous monitoring does not mean you need someone watching dashboards 24/7. It means the sensors are sampling at regular intervals - typically every few seconds to every few minutes - and the system is running automated checks against alert thresholds. A technician reviews flagged alerts once or twice per shift. The system does the watching; the human does the thinking.

Oil analysis deserves a mention even though it is not a sensor-based technique. For gearboxes, hydraulic systems, and large reciprocating compressors, monthly oil samples analyzed for wear metals, contamination, and viscosity changes remain one of the most reliable condition indicators. A spike in iron and chromium particles in gearbox oil tells you about gear tooth wear long before vibration patterns change. Many plants run oil analysis in parallel with vibration monitoring on critical gearboxes, and the two methods catch different failure modes at different stages.

Setting Alert Thresholds That Actually Work

Threshold setting is where most condition monitoring programs either succeed or drown in false alarms. Set thresholds too tight and technicians get alert fatigue within the first month - they stop checking because every alert is a false positive. Set them too loose and you miss real problems. There is no universal right answer, but there is a structured approach.

The baseline-relative approach is almost always better than absolute thresholds. ISO 10816 gives you vibration severity categories for different machine classes - and they are useful as a starting point - but your pump running at 4.2 mm/s overall vibration might be perfectly healthy if it has always run at 4.0 mm/s. The same reading on an identical pump that normally runs at 1.8 mm/s indicates a serious problem. Context matters more than standards.

• Collect 2-4 weeks of baseline data on each asset before setting any thresholds
• Set the Advisory threshold at 2x the baseline standard deviation above the mean - this catches genuine trends while filtering out normal variation
• Set the Alert threshold at 3.5-4x the standard deviation, or at 150-200% of the baseline mean, whichever is more conservative
• Set the Alarm threshold at the equipment protection limit (bearing temperature rating, pump cavitation point) or at 250-300% of baseline
• Review and adjust thresholds at 30, 60, and 90 days. If a sensor generates more than 3 false alerts per month, the threshold is too tight
• Seasonal adjustments matter: a cooling water pump will run differently in July than in January due to ambient temperature and demand changes

A practical target: each sensor point should generate fewer than one false alert per month in steady-state operation. If your 50-sensor deployment is producing 200 alerts per month and only 15 of them lead to actual work orders, you have a 92.5% false positive rate and your technicians have already learned to ignore the system. Better to miss an occasional real alert with conservative thresholds than to bury genuine signals in noise.

Sensor Selection, Placement, and Installation

Sensor technology has gotten dramatically better and cheaper in the last five years. Wireless vibration sensors that cost $2,500 in 2018 now cost $300-$600 and last 3-5 years on a single battery. That changes the economics of condition monitoring for mid-market manufacturers. But cheaper sensors do not help if they are installed wrong.

For most plants starting a condition monitoring program, wireless sensors on 80-90% of monitored points with wired sensors on a handful of truly critical assets is the right split. The cost difference adds up fast: instrumenting 50 points with wired sensors might cost $75K-$100K installed, while the same coverage with wireless is $20K-$40K.

Placement matters more than sensor quality. A $1,500 precision accelerometer mounted on a motor shroud with a magnet will give you worse data than a $300 wireless sensor properly stud-mounted to the bearing housing. The vibration signal attenuates and distorts as it travels through structural boundaries. You want the sensor as close to the bearing or component of interest as possible, on a flat machined surface, with rigid mechanical coupling.

• Mount vibration sensors on bearing housings, not on sheet metal covers or structural frames
• Use stud mounting or high-quality industrial adhesive for permanent installations - magnetic mounts are for route-based spot checks only
• Orient the sensor axis correctly: radial (horizontal and vertical) for most bearing faults, axial for thrust bearing and misalignment detection
• Temperature sensors belong on the bearing housing or in a thermowell, not on external surfaces where ambient air distorts readings
• For motor current monitoring, install CTs (current transformers) on individual motor feeds, not on branch circuits serving multiple loads
• Label everything. A sensor generating great data is useless if nobody knows which asset it is attached to. Use asset tags that match your CMMS

Building a Monitoring Route vs. Continuous Online Systems

Not every plant needs - or can afford - continuous online monitoring from day one. Route-based monitoring, where a technician walks a defined path collecting readings with a handheld device on a weekly or monthly schedule, has been the backbone of condition monitoring for 30 years and still has a place.

The hybrid approach is the most practical starting point for mid-market manufacturers. Put continuous wireless sensors on your 10-15 most critical assets - the ones where an unplanned failure costs you more than $20K in downtime and repair. Run a monthly handheld route on the next 30-50 assets. Review the economics annually and shift assets from route-based to continuous as the data justifies it.

One advantage of route-based monitoring that gets overlooked: it puts a trained technician physically in front of the equipment on a regular schedule. They see oil leaks, feel unusual heat, hear changes in sound, and notice things that no sensor can detect. The best programs combine continuous sensor data for early anomaly detection with periodic human inspection for the things that sensors miss. Neither approach alone is sufficient.

Common Mistakes and How to Avoid Them

After working with hundreds of plants implementing condition monitoring, the same mistakes come up repeatedly. Most of them are organizational, not technical.

The biggest mistake of all is not acting on the data. A condition monitoring system that identifies a bearing defect six weeks before failure has done its job. If the maintenance planning process cannot get a bearing change scheduled and executed within that six-week window, the monitoring investment is wasted. Before buying sensors, make sure your work planning and scheduling process can consume the information they produce. The technology is the easy part. The workflow change is where programs succeed or fail.

Getting Started: A 30-Day Quick Start Plan

You do not need a massive project plan to get started with condition monitoring. Here is a 30-day quick start that gets sensors on equipment and data flowing with minimal overhead.

After 30 days you will have a functioning condition monitoring system on your most critical assets, generating real data and actionable alerts. It will not be perfect - you will need to tune thresholds over the next 60-90 days as you learn what normal variation looks like for each asset. But you will be monitoring, and you will be building the data foundation for a broader program.