Thermal cable hotspot diagnosis: when the heat is not at the fault

A survey image that shows a feeder cable warm along its run, with terminals that look clean, rarely means the cable is at fault. The usual source is a connection hidden behind the panel: its heat conducts along the copper conductor and warms the sheath tens of centimetres from the joint. What follows gives the conduction distances behind that pattern, the emissivity traps that hide a real hotspot behind perspex and porcelain, the load scaling that keeps readings comparable between surveys, and the micro-ohmmeter test that confirms the suspect connection before anyone condemns a healthy cable.

The camera reads the sheath, not the fault

A thermal camera maps the temperature of the first solid surface in its view. On a feeder that surface is the sheath; the conductor carrying the fault current runs inside it, and copper moves heat along its length almost as readily as it moves current. The image is a heat map whose source sits displaced along the conductor — sometimes tens of centimetres, sometimes a panel wall away.

Two checks sort most surveys on the spot. Compare phases: one cable clearly warmer than its two neighbours points to that phase's connections, at the gland and at the far termination. All three equally warm points to load, grouped-rating derating, panel ambient or blocked ventilation — a capacity question, not a connection fault.

For the wider panel routine — busbars, breakers, fuse ways and survey intervals — see thermal camera electrical inspection: finding hotspots in panels. This page covers the case where the heat sits on the cable itself.

How fault heat travels: the one-decay-length rule

Heat made at a joint flows two ways: along the conductor, and out through insulation and sheath to the air. Along the conductor it decays exponentially, because every metre of copper leaks heat through its surface. That produces the signature pattern: warmest near the source, fading with distance in one direction only.

Here ΔR is the joint's extra resistance, k is copper's 390 W/(m·K), A the core cross-section, p its circumference (π × d), and h the surface heat loss — about 10 W/(m²·K) for still indoor air through insulation. h is the model's weakest input: doubling it shrinks λ by 30%, halving it grows λ by 41%.

For copper cores in still air the decay length lands near a third of a metre: 0.33 m for 95 mm², 0.39 m for 185 mm² and 0.44 m for 300 mm². The working rule follows: a joint's footprint stays visible for about one decay length, fades past two, and the gradient always points back at the source.

Worked example: a 150 μΩ joint on a 150 A feeder

Set-up: a 6.6 kV three-core XLPE feeder with 95 mm² copper cores, carrying 150 A. The gland-end termination on one phase has degraded from a healthy 25 μΩ to 150 μΩ, and the panel wall hides it from the camera.

Extra fault power: ΔR = 150 − 25 = 125 μΩ, so P = (150 A)² × 125×10⁻⁶ Ω = 2.8 W. Small by motor standards, but concentrated at a single lug.

Conductor geometry: d = 2·sqrt(A/π) = 2·sqrt(95×10⁻⁶ / π) = 11.0 mm, so p = 0.0346 m. Joint rise, treating the core as a fin: ΔT_joint ≈ P / sqrt(h·p·k·A) = 2.8 / sqrt(10 × 0.0346 × 390 × 95×10⁻⁶) = 2.8 / 0.113 ≈ 25 K above ambient. Decay length: λ = sqrt(390 × 95×10⁻⁶ / (10 × 0.0346)) = 0.33 m.

A camera that cannot see the lug still sees the cable: a 9 K bump at 33 cm from the gland, on a matte XLPE sheath, is a clear and reportable hotspot. Treat the table as order-of-magnitude — heat shared between the three cores through the fillers stretches the warm zone somewhat.

That 2.8 W is also a continuous bill the feeder pays; the same arithmetic turns survey frames into thermal imaging energy analysis when losses are costed.

Reading the gradient

The gradient decides where the outage goes:

• Warmest at the gland, cooling outward: the fault is at that cable's termination inside the panel. Plan the outage at the panel end, not along the cable.
• A peak part-way along the run: suspect that point — a splice on a repaired cable, or water in the core. Here the cable genuinely is the fault.
• All three phases warm with a flat gradient: load above the grouping-corrected rating, hot panel ambient or lost ventilation. Check clamps and air paths before touching a joint.
• One phase warm over its whole run: grade both of its terminations; the micro-ohmmeter test below says which one carries the fault.

Image both ends of every suspect feeder. The far termination at a transformer bushing is often the more accessible of the two.

Emissivity traps: perspex, porcelain and shiny metal

A LWIR camera sees only the 8–14 μm band, and glass and most plastics are opaque there. A perspex sheet in front of the terminations therefore ends the optical story: the camera records the sheet's front face — perspex emissivity ≈ 0.94 — not the joint behind it. Adjusting the emissivity setting will not look through it.

Porcelain adds a second trap: it is a good thermal insulator, so its surface lags the conductor by seconds to minutes and reflects anything hot nearby. Read the metal cap or flange — thermally close to the joint, emissivity honest.

The documented way to watch live joints behind a barrier is an infrared inspection window: a germanium or similar crystal pane transmitting 8–14 μm, fitted over the termination. Where no window exists, the micro-ohmmeter test below does the confirming at the next outage.

Survey under load: rise scales with current squared

Survey at 40% of rated current or more. Heating goes as the square of the current, and 0.40² = 0.16: at 40% load a joint dissipates under a sixth of its rated-load power, which is why a survey on a lightly loaded morning misses a fault that trips out in summer.

Record the amperes with every frame, and scale before comparing surveys taken at different loads:

The 9 K bump from the worked example was measured at 150 A. Against a 200 A reference survey it corresponds to 9 × (200/150)² = 16 K — (200/150)² = 1.78, and 9 × 1.78 = 16 K, which moves it up a severity band on its own.

Severity, always referenced to a similar joint under the same load, follows the bands most electrical testing schedules publish: under 4 K, watch; 4–15 K, repair at the next planned outage; over 15 K, act at once.

Confirm the joint electrically: the micro-ohmmeter reading

A torque check proves the bolt was turned; it says nothing about the contact faces. Oxidation, lost plating and fretting leave a joint tight and still resistive — hundreds of micro-ohms where a sound face reads in the tens. Re-torquing does not move that reading; only a resistance measurement does.

The test: isolate and prove dead, then inject DC across the joint — typical instruments supply 10–200 A — read the millivolts and divide by the current to get micro-ohms. Compare the three phases; the outlier is the fault.

Camera choice and pixel size for panel work

Resolve before you grade. Survey practice puts at least 3 × 3 pixels on a graded target, so a 20 mm lug needs pixels of about 7 mm or finer; work backwards from that budget to working distance.

The LS-IRC-0044 is a GigE thermal camera for the 8–14 μm band with an 88.5° × 73.2° field of view, a C-mount and a 9–24 V DC supply. Its full specification, including the resolution, is on the card below.

At a 2 m working distance the 88.5° axis spans 2 × 2 m × tan(44.25°) ≈ 3.9 m, so each of the 640 pixels covers about 6.1 mm — enough to flag a 50 mm gland area with margin, and just meeting the 3 × 3 rule on a 20 mm lug. For confident grading, step closer or fit a narrower thermal-band C-mount lens.

Check your own distance and pixel budget with the focal length calculator:

For a permanent watch on one termination, the MV-CI003-GL-N25 ships with a fixed 25 mm lens, streams 0.3 MP over GigE and mounts on M34×0.75 — a narrow view that suits a single bushing or gland plate seen through an inspection window.

The table lines up the three LWIR choices used in this guide:

All 3 share: Pixel size 17 μm. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

Browse the camera finder by interface, resolution and spectral band, or open the infrared cameras category for the LWIR and MWIR range.

Record, scale and trend

Save radiometric frames, not viewer screenshots: a frame that keeps per-pixel temperature can be re-measured, re-scaled and re-graded months later. The export path is in export high-bitdepth grayscale images from thermal cameras. Log three values beside every frame: load amperes, emissivity setting and working distance.

Trend the same joints survey to survey after scaling to a common current. A joint drifting 3 K, then 9 K, across a year is a different work order than a stable 9 K.

Two edges of the method: when one point is all you need, a spot pyrometer can replace the camera — the trade-off is in thermal imaging camera vs spot pyrometer: when one sensor sees all. When the target is tens of metres away or several hundred degrees, a cooled MWIR camera earns its keep; the selection logic is in cameras for continuous steel casting monitoring: heat and distance. More survey-facing reading sits on the applications topic page.

When the usual answer is wrong

• Warm air from below: a cable rising past hot equipment picks up convected heat evenly along its run. Look at what sits under the cable in the same frame before blaming a joint.
• Reflected backgrounds: midday sun on a south-facing gland, a polished gland plate or a hot busbar fakes a hotspot on low-emissivity surfaces. Shift your viewing angle; a fake hotspot moves, a real one stays.
• Sheath circulating currents on single-core feeders warm the armour evenly end to end — a real loss, but not a connection fault.
• Even warmth on all three phases at full loading: capacity, grouping and ventilation, as above. The joints may be innocent.
• A genuine mid-run fault: an old repair splice, or water ingress, peaks in the middle of the run with a clean gradient on both sides.

Before the next survey: a working checklist

Before:

• Confirm the feeder runs at 40% of rated current or more, and log the amperes with the frames — every later scaling depends on them.
• Map the barriers: perspex, glass and unwindowed covers decide which joints you can see and which you must confirm electrically.
• Set emissivity per surface: about 0.93 on XLPE sheaths, and a tape patch on anything shiny you must read.

During:

• Keep 3 × 3 pixels on every graded target; step closer or change lenses where the pixel budget fails.
• Image both ends of each suspect feeder, and save radiometric frames.

After:

• Scale every rise to a common current, then grade on the 4 K and 15 K bands.
• Put the suspect joint on the next outage list with a micro-ohmmeter test specified — the reading closes the case, not the torque wrench.

Take the gradient, the load and one micro-ohmmeter reading, and the image becomes a work order. If a survey frame does not resolve, send it through the contact page with the load current and working distance noted — or start a build at the camera step of the configurator.