A loosening terminal in a substation — what happens between two thermal surveys?

A cable lug connection in a medium-voltage cell starts working loose three years after installation. The cause is rarely dramatic: thermal expansion and contraction from daily load cycles, creep in aluminium conductors, or vibration from a switching operation nearby.

The connection still works. Measure it and current flows; the voltage drop looks negligible. But the contact area has shrunk, and resistance at that point has begun to rise.

The physics of the chain

As contact resistance rises, I²R loss at that point increases. The connection heats. That heat puts the insulation around it — PVC, polyamide, epoxy — under thermal stress.

Here is the critical part: those polymers begin to degrade well below their ignition temperature, and they release a measurable outgassing signature. Smoke and flame are the last link in this chain, not the first.

The interval in between is measured in weeks, sometimes months. The problem is not that there is no indication during that period — it is that nobody is watching.

What a quarterly survey misses

Thermal imaging is a sound method and it does find a loose connection. Its limit is not in the technique but in its frequency.

A survey every three months takes four snapshots a year of a process that unfolds over weeks. For those snapshots to work, three things have to be true at the same moment:

• Load must be high enough. Heating scales with the square of current. A survey taken at low load may not show a temperature difference that is obvious at full load.
• The connection must be in line of sight. A point behind a cover, beyond a barrier, or inside a cable duct is invisible to a thermal camera. Without an IR window, the panel has to be opened.
• The defect must already be mature. Early on, the difference is a few degrees and can be mistaken for normal load variation.

When all three hold, the method does its job. When they do not, there is no data at all about that connection until the next survey three months later.

What can be measured in between

ArcFire Sense (ARF-005) is mounted inside the panel and watches outgassing across several channels: volatile organic compounds (VOC and TVOC), carbon monoxide (CO), hydrogen (H₂), nitrogen dioxide (NO₂) and equivalent carbon dioxide (eCO₂). The same probe reads ambient and surface temperature (±0.2 °C), relative humidity and pressure.

The gas channel has two advantages over a thermal camera. First, it needs no line of sight — gas produced by a connection behind a cover disperses through the panel’s internal volume. Second, it measures continuously; the problem does not have to reveal itself during a moment of high load, only at some moment.

This does not replace the thermal survey. It fills the gap between surveys.

Heating is not always a fault

A simple temperature threshold does not work in this application. A cell running at full load on a summer afternoon gets hot — that is expected behaviour, not a fault. A system watching an absolute threshold either produces constant false alarms, or, once the threshold is raised far enough, misses the real event.

ArcFire Sense scores risk against a physics-informed model: thermal rise is evaluated together with load current. An expected temperature rise under increasing load is not classified as a fault. What raises the alarm is not absolute temperature but deviation from the equipment’s own reference signature — and that signature is learned for that specific panel during commissioning.

Per-channel Kalman filtering also keeps switching transients, cover-opening events and ventilation cycles out of the risk function.

Reporting from an unstaffed site

A substation has no corporate network and is usually remote. The communication layer is chosen to match the site: across distribution networks where many buildings are spread over a wide geography, the LoRaWAN Module (CM-L300) covers kilometre-scale areas from a single gateway; at critical points needing full-resolution telemetry, the Cellular GSM Module (CM-G200) is used.

But the substantive point is this: the decision is made on site. Risk scoring and alarm generation run on the IQ Panel at the edge. A communication outage does not stop alarm generation, and a dry-contact relay output can drive a warning circuit with no connection at all.

During a coverage gap, outgoing data is held in a local buffer and transmitted in order once the link returns — no hole appears in the trend history.

In short

A quarterly thermal survey is a method that looks four times a year. The gas signature of a loosening terminal forms in the weeks in between. The two are not rivals: one gives periodic verification, the other continuity.

On an unstaffed site, the absence of the second means the problem is noticed only once it has become a fault.