Infrared electrical inspection: severity classification by delta-T

You have the thermal camera on the switchgear, a screen full of warm joints and a report template asking for a severity class. The band table in circulation wants a delta-T — minor under 20, critical over 90 — but a joint your camera reads 20 °C above its neighbours may truly be running 100 °C hot. This guide gives the classification method that survives that problem: judge every hotspot by its difference from identical neighbours under the same load, correct the flagged few for emissivity, and add circuit criticality last. It also shows the two arithmetic checks that make the numbers defensible: the emissivity conversion and the pixel budget that fixes the camera standoff. The method fits panel surveys, busbar runs and switchyard inspections with LWIR infrared cameras.

Qualitative or quantitative: what each one answers

A qualitative inspection finds problems. It compares each component with identical neighbours carrying the same current and asks one question: is this one hotter than its peers? It never claims a true temperature and does not need one, because emissivity, reflections and cooling are shared across the whole comparison set.

A quantitative inspection reports actual temperatures. Every number it prints rests on three inputs the camera cannot measure for you: surface emissivity, reflected background temperature and distance. Get them right and the reading is defensible; get them wrong and the error can be several times the value you publish.

The decision is not either-or. Screen the whole asset qualitatively — that is how one pass covers 800 busbar joints — then go quantitative only at the exceptions, where a corrected temperature sets repair priority. The comparative pass finds; the quantitative pass ranks. The panel-level workflow sits in the thermal camera electrical inspection walkthrough; this guide stays on the classification question itself.

Why a fixed delta-T band misleads

Band tables grade severity by temperature difference alone. A scheme that circulates widely, for example, grades phase-to-phase difference as 1-20 °C minor, 21-45 °C important, 46-90 °C serious and 91 °C and above critical. The bands look authoritative and fail for a physical reason: the difference the camera shows is not the difference the joint has.

The relation holds when the camera emissivity stays at 1, both surfaces share a finish, the background is common, and the rise is small against the background level — the normal case in electrical surveys. It bites on bare metal: shiny aluminum and copper sit below 0.3 emissivity, so the camera suppresses the rise by that factor.

• Peers read 21 °C; the suspect joint reads 41 °C with the camera at e = 1. Apparent rise: 41 − 21 = 20 °C, which the table files under minor.
• Weathered bare aluminum, ε ≈ 0.2: real rise ≈ 20 / 0.2 = 100 °C, so the joint sits near 121 °C.
• The same 20 °C on a matte tape patch, ε ≈ 0.95: real rise ≈ 20 / 0.95 ≈ 21 °C, joint near 42 °C.

One camera reading, two realities a factor of five apart. The delta itself is emissivity-scaled, so no band built on raw apparent degrees can rank shiny hardware safely.

Compare like with like: the reference that replaces the band

The comparative method replaces absolute truth with internal consistency, and it only works when the comparison is honest. Same hardware, same material and finish, same current path, same ambient. A bolted aluminum splice compares with the identical splice on the neighbouring phase or one span away — never with a braided strap or a tinned lug, whose emissivity and cooling differ.

Phase-to-phase delta, the backbone of most survey reports, is valid when the load is balanced and the phases share a finish. Where phases carry different currents, compare each joint against its peers on the same phase instead.

This is where the method earns its keep: identical joints under identical load should sit within a degree or two of each other. Busbar practice treats a like-for-like rise of 5 °C as the recording threshold, because joint resistance only grows with oxidation cycles — a delta that exists is a delta that will grow. Record it, trend it against the last survey, and let the trend set urgency rather than the first reading alone.

Load multiplies what the camera sees

Heating at a resistive joint follows I²R, so the temperature rise tracks the square of the current:

The proportionality is an estimate — convective cooling rises roughly linearly with temperature over the tens-of-degrees range — but it is close enough to plan and to judge a survey.

• A joint that will run 80 °C above its peers at the rated 400 A, surveyed while the feeder carries 200 A: heat ratio (200 / 400)² = 0.25.
• Expected rise at survey time: 80 × 0.25 = 20 °C above peers — the defect is real, the camera sees a quarter of it.
• At 40% load a joint holds back 84% of its heat (0.40² ≈ 0.16), which is why light-load surveys come back clean while joints still fail.

Survey at the strongest normal load you can arrange, log amps per phase with every image, and re-survey flagged joints near full load before condemning or clearing them.

A severity scale you can defend

Rank from three signals in this order: the like-for-like delta, the corrected rise, and the consequence of failure. The tiers below are working defaults built on the bare-aluminum case, ε ≈ 0.2; tighten them with your own survey history once a few cycles accumulate.

Two refinements make the scale defensible in a report. Rate of change outranks magnitude: a joint that moved from 6 °C to 14 °C above peers in one year is a schedule item even though 14 °C sounds small, because oxidation accelerates. And weight by consequence: utilities put a hotspot at the source side or main incoming gear ahead of an identical temperature at the end of a feeder, because the outage it causes is bigger. Temperature starts the ranking; criticality finishes it.

Resolve the joint: pixels decide what you can claim

A severity class is only as good as the pixels behind it. A spot reading wants the target spanning at least 3 × 3 pixels — the working rule behind every spot-size table — and a comparative screen still wants several pixels so one hot pixel cannot impersonate a joint. Pixel footprint at standoff distance D is 2 × D × tan(HFOV / 2), divided by the pixels across the frame.

For panel lines and bus runs you want the LWIR band, an Ethernet drop that reaches the panel, and a frame rate that keeps a walking survey live. The LS-IRC-0044 pairs a 640 × 512 LWIR array (8-14 μm, 50 Hz) with an 88.5° × 73.2° field of view, a C-mount, GigE and a 9-24 V DC supply.

Worked example with the wide fixed field: 2 × tan(44.25°) / 640 = 3.04 mrad per pixel. At 5 m that is 3.04 × 5 = 15.2 mm per pixel, so a 12 mm bolt head spans 0.8 pixels — invisible. To get 3 pixels on that 12 mm head the footprint must stay under 4 mm, which puts the standoff at 4 mm ÷ 3.04 mrad ≈ 1.3 m. At 1.3 m the head fills 3 pixels: tight, but enough to rank.

A narrower lens moves the standoff out. On a 1/1.2" sensor roughly 10.7 mm wide, a 17 mm lens gives 2 × atan(10.7 / 2 / 17) ≈ 35°, about 0.98 mrad per pixel: the same 3 pixels on a 12 mm bolt head from about 4 m. Focal length and sensor format trade against each other — how HFOV shifts when the same lens moves to another sensor — and the working distance and spot size method for safe scans works the standoff question end to end.

Reach comes from the glass in the thermal, SWIR and UV lens category. The LS-SWL-0006 is a 17 mm, F1.2 thermal (LWIR) lens for 1/1.2" sensors, with an 11-16 mm image circle. On that 1/1.2" format the 35° field moves the 3-pixel standoff from 1.3 m to about 4 m.

Field checks before you rank a hotspot

Six checks keep the numbers honest enough to rank:

1. Reflection check: polished metal mirrors LWIR. Step 30° sideways and re-frame; if the hotspot slides across the surface, it is a reflection from you or a neighbouring source, not the joint.
2. Emissivity discipline: screen at e = 1 so the whole frame shares one convention, then set a real emissivity only where a tape patch or dull target justifies it.
3. Load log: amps per phase beside every saved image — a delta without its current is uninterpretable next cycle.
4. Standoff log: same distance and angle every survey, so cycle-to-cycle deltas compare geometry as well as temperature.
5. Air movement: moving air cools a defect and its peers together, so deltas shrink in wind; weight calm-day readings accordingly.
6. Radiometric save: keep 16-bit counts rather than palette images — the radiometric jpg method for 16-bit temperature data shows the format, and raw thermal counts to temperature explains what the camera does inside.

When the comparative method struggles

• Everything shiny: the comparison survives, the correction does not. Phantom deltas from reflected sources are the main risk; the reflection check above and surveying while neighbouring processes idle keep them out.
• Closed panels: no line of sight means an infrared window in the door. The infrared window material guide covers the choice; fit identical windows per phase so every peer looks through the same optic.
• Mixed hardware in one frame: rank only against same-hardware peers. A braided strap beside a bolted splice is two populations, not one.
• Long lenses outdoors: a lens body crossing the outdoor-to-cabinet temperature span shifts focus, and soft focus reads cool. Athermalized LWIR optics hold focus across that swing; the infrared lens athermalization guide explains the mechanism.
• Unclear cause: when the image shows a hotspot but the mechanism is not obvious, work it through a structured infrared electrical hotspot troubleshooting routine before scheduling the repair.

One caution about over-correction: more precision on a one-pixel target is wasted effort. Move closer, fit a window, or resurvey under better load — the fix for a bad number is a better image, not a longer decimal.

The busbar survey, decided

• Screen qualitatively at e = 1, flag any joint 5 °C or more above identical peers, and log amps and standoff with every frame.
• Correct the flagged joints only — tape patch or entered emissivity — and rank by corrected rise, delta doubling since the last survey, and circuit criticality.
• Survey at the highest normal load available; treat a light-load clean bill as provisional.
• Fix geometry first: every joint spanning 3 pixels at your standoff decides the lens before anything else does.

Rigs built from components can take the sensor without the housing: the LS-IRC-0016 is a 640 × 512 LWIR module that streams over USB3.

For a fixed working distance — one bay, one frame, the same geometry every cycle — the MV-CI003-GL-N25 keeps an EFL 25 mm lens, an M34x0.75 mount and a GigE link in one 0.3 MP module.

Match the glass with the thermal, SWIR and UV lens finder, filter LWIR bodies by mount and interface in the camera finder, or let the system configurator turn the survey into an orderable build. More survey-side reading sits on the applications topic page, and we are glad to size the standoff for your joint list through /contact/.