What a crew needs from a thermal finding

An inspection report can list hundreds of anomalies. A crew can work a handful in a day. Between the two sits a translation, usually done by hand, from a hotspot in an image to a job a technician can drive to, confirm and fix. The finding has to carry six parts for that translation to work.

01Where, to the module

Standard drone GPS is accurate to two to five metres, which can send a crew to the wrong row; RTK correction narrows that to one or two centimetres ^[1]. Better still is a component identity: block, row, string and module, in the same names the as-built drawings and the SCADA tags use.

02Under what conditions

A temperature difference means little without the conditions it was measured under. IEC TS 62446-3 sets a minimum of 600 W/m² of plane-of-array irradiance for valid thermal data, and wind above about 5 m/s cools hotspots enough to compress their signature ^[2]. A finding should carry irradiance, wind, ambient temperature and time of capture, so that a smaller ΔT on a cloudier day is not mistaken for an improvement.

03How severe, on a stated scale

There is no single standard scale. The IEC standard does not prescribe categories, and many operators use three or four tiers by temperature difference, from monitoring below 10 °C to urgent action above 20 °C ^[1]. Whatever the scale, it belongs on the finding, next to the measured value, so two reports can be compared.

04What it is costing

Severity and cost are separate measures. A string fault and a cell hotspot can share a severity class and differ by orders of magnitude in lost energy. String and combiner faults accounted for 26.89% and 21.51% of observed power loss across 373 GW of US solar in 2025 ^[3]. The energy at stake should set the order of a crew's work.

05The evidence

A crew should see what the analyst saw: the thermal crop, the visual crop and the reason the finding was made. In a 2026 study, a model that wrote out its diagnostic reasoning across thermal, visual and electroluminescence images reached 93% accuracy on 1,927 real modules, and a blind concordance study found its explanations closely aligned with a certified inspector's assessments ^[4]. With the reasoning attached, a crew can check the call on site.

06Its history

A finding that carries its own history tells a crew whether they are looking at something new, something worsening or a repair that did not hold. That history only exists if every finding is tied to the same component, flight after flight.

Aperture's findings carry five of the six today: the module, assigned on the surveyed layout and flagged exact, conflict or approximate; the conditions at capture; ΔT measured on the °C raster as the maximum minus the mean over the module; one of nine causes and one of eight next steps, set by an expert; and a follow-up that the next campaign reads. History is kept at string level until module identity clears its gate. Energy at stake waits on site energy data. Until a site supplies it, a labelled baseline stands in, and no loss figure is quoted as a measurement.