Thermal cameras for bearing failure detection: how early is early

You already own, or are about to buy, a thermal camera, and bearings look like the obvious first job: friction makes heat and heat kills bearings. The honest answer from plants that run both programs is that thermography finds bearings that are already failing, vibration analysis finds them while grease or a planned swap can still save them, and ultrasound hears the problem before either. This guide gives the temperature rise you should act on, the spot-size arithmetic that decides whether you can measure it at all, and a rule for pairing the three methods.

The decision, up front

Use the thermal camera as the screening pass and vibration analysis as the confirming instrument. Act on a bearing when its housing runs about 10 °C above an identical sibling under the same load, read on matched high-emissivity targets. Below that level, emissivity mismatch, load swings and ambient drift can produce the same reading on a healthy machine. Vibration analysis picks up the same defect weeks to months before the housing reaches that level, and ultrasound flags lubrication problems before any heat appears.

Thermal still earns its place on the route. One image covers a row of machines in seconds, and it is the strongest tool you have for electrical hotspots. What it usually cannot do is save the bearing itself: by the time a cap is 10 °C hot, the rolling surfaces are generally already damaged, and the win is avoiding seizure, shaft damage and the unplanned stop.

How a bearing dies, and what each sensor sees

Bearing failures progress in a dependable order. Lubrication degrades first: the oil film thins, metal-to-metal contact begins and friction rises. Ultrasound hears this immediately - a bearing in lube distress reads clearly louder, in decibels, than its own baseline. As surfaces pit and spall, defect frequencies appear in the vibration spectrum; this is the stage vibration analysis is built for, and in practice it arrives weeks to months before failure. Only once the damage is established does friction heat the housing enough to see at a distance, and the run from visible heat to failure is short.

Ultrasound also gives you a lubrication procedure the other two cannot: grease while listening, stop the moment the level stops falling and begins to rise again. Over-greasing churns the bearing and heats the housing on its own, so the listening method protects you from the cure.

Researchers have pushed thermography earlier with image processing on recorded thermograms - separating fault signatures with transforms such as the two-dimensional discrete wavelet transform. That is offline analysis software, though, not something you get by walking past with a camera.

Why the action level is 10 °C, not 2 °C

A thermal camera does not read temperature; it reads radiance and converts it with the emissivity you enter. What it collects is a mix:

Painted and oxidized surfaces sit near ε = 0.9, so the reflected term stays small. A bare machined cap can sit below ε = 0.3, and then the camera mostly sees the room. Enter 0.95 against a shiny cap whose true emissivity is 0.2 and a 70 °C housing can display near 40 °C - a 30 °C error that turns a critical bearing green.

Load adds its own spread. Two identical motors at different loads, or the same motor after a duty change, differ by a few degrees with nothing wrong. Add ambient drift between the start and end of a walk-around, and a 1-2 °C difference means nothing. Ten degrees above a same-load sibling on matched targets stands out of that noise for every operator, every shift.

Between 3 °C and 5 °C you can still act, but only with the noise removed: taped or painted targets, identical load, same camera position - or a permanently mounted camera that trends against itself. Sensor sensitivity is not the constraint. A modern microbolometer's NETD sits well under 0.1 °C, far below these margins, so put the effort into emissivity and placement; the NETD and lens choice guide covers how far that number takes you. For the physics behind the sensor itself, see camera sensor technologies by spectral range.

Worked example: screening a motor row at 3 m

Inputs:

• Camera: the LS-IRC-0044, whose data prints an 88.5° × 73.2° field of view over 640 × 512 pixels.
• Standoff: 3.0 m, chosen to keep you off the coupling guards.
• Target: a 40 mm wide bearing cap on a mid-size motor.
• Decision level: 10 °C above the same-load sibling, from this guide's rule.

The arithmetic:

1. 88.5° ÷ 640 = 0.138° per pixel = 2.4 mrad.
2. 3,000 mm × 2.4 mrad = 7.2 mm of cap per pixel.
3. 40 mm ÷ 7.2 mm = 5.5 pixels across the cap - above the 3-pixel minimum for a trustworthy reading.
4. Frame coverage: 2 × 3.0 m × tan(44.25°) = 5.8 m wide, and 2 × 3.0 m × tan(36.6°) = 4.5 m high.

Decision: 3 m is the right distance for a motor row. One frame takes in roughly 5.8 × 4.5 m of plant, every 40 mm cap inside it resolves with more than 5 pixels, and the 10 °C sibling comparison is valid at that resolution. Shrink the target to a 20 mm cap and it covers only 2.8 pixels - fail. Halve the distance to 2.0 m and the footprint drops to 4.8 mm per pixel, putting 4.1 pixels on the same cap - pass. The rule: under 3 pixels across the cap, close the distance or fit a narrower lens.

For fixed installations that trend one point over time, temperature vs position line profiles from radiometric pixel data shows how to turn those recorded images into a trend you can act on. And if you have been using a single-point gun so far, the sibling in the same frame is the whole method - a spot pyrometer measures one cap at a time and cannot compare, which is why a frame saves the walk-around.

Thermal, vibration and ultrasound compared

The decision per case:

• One technician, hundreds of assets: monthly thermal walk-around, vibration on anything it flags.
• Critical machines - gearboxes, main drives, fans without spares: permanent vibration monitoring, thermal as the weekly sanity check.
• Chronic grease problems: ultrasound at the grease station, thermal only to confirm the fix.

Where the thermal camera is the strongest tool

Open the same walk to electrical assets and thermography stops being the latecomer. Terminations, contactors, drives and fuse holders produce heat long before they fail and vibrate not at all, so there is nothing for the other two methods to find. Finding hotspots in panels covers the routine; size the camera by IFOV for electrical inspection because terminal bars are small targets at panel distance; and severity classification by delta-T turns what you measure into a repair priority.

One rule belongs here: never open an energized panel just to point a camera through it. Follow your site's arc-flash procedure, and where panels must stay closed, choosing an infrared window material lets the camera look through the door safely.

Choosing the camera and lens

The band is settled for you: machinery at 20-100 °C lives in LWIR, so the choice is resolution, field of view and how the camera mounts. The LS-IRC-0044 is the walk-around shape of this job - a 640 × 512 LWIR camera over GigE, with an 88.5° × 73.2° field of view, 50 Hz frame rate, 8-14 μm band and a 9-24 V DC supply.

For a station that stays on a machine, a fixed camera with a narrower lens beats walking: the MV-CI003-GL-N25 puts a 25 mm EFL on an M34x0.75 mount for points that need reach, and the LS-IRC-0016 is a USB3 module at 640 × 512 for tight enclosures.

Values as printed in each product's data sheet; open a part for its drawing and full specifications.

When the lens changes the pixels-on-target answer, shop thermal, SWIR and UV lenses rather than cameras. The LS-SWL-0006 is a 17 mm F1.2 LWIR lens with an 11-16 mm image circle for small thermal sensors. Outdoors, remember that infrared lens athermalization decides whether focus survives the seasons.

Browse the full infrared cameras range, filter it in the camera finder, or start sizing a lens in the configurator.

When the usual answer is wrong

• A cool bearing is not a healthy bearing. Friction power scales with speed, so slow-turning or well-ventilated bearings stay near ambient well into damage; the absence of heat says nothing, and slow machines belong on the vibration route.
• Siblings under different loads break the 10 °C comparison. Compare drive-end against non-drive-end on the same machine, or trend the same point over time, instead of comparing across machines.
• Guards that hide the cap push you onto the adjacent frame. That point still trends usefully, but treat it as a relative measure, not an absolute level.
• Sun-loaded outdoor machines carry a moving heat pattern of their own. Scan at the same hour every round, or after dark.
• A bearing that got hotter right after greasing may simply be over-greased. Purge the excess and re-scan before ordering a replacement.

Checklist for the scan day

• Matte-black tape or paint dots on every cap you will compare, ε 0.95 set on the camera.
• Same load, same speed, machines warmed up, for anything you compare.
• Three pixels minimum across the cap - confirm at your first stop and hold that distance all day.
• Log every image with asset ID and running hours; the trend outweighs any single reading.
• 10 °C on a sibling: raise a work order. 3-5 °C on taped targets: schedule vibration analysis within the week. Anything audible: ultrasound now.
• Panels and drives on the same route - one pass serves both worlds.

Start from the infrared cameras category, size a complete fixed station in the system configurator, or send us the distances and cap sizes and we will run the spot-size numbers against your layout before you order.