Learn how thermal imaging inspection helps detect solar panel defects using drones, infrared cameras, AI analytics, and actionable inspection workflows.
Karthik Mekala
CMO
Published on
Thermal imaging inspection (sometimes called solar panel IR inspection, or run in the field as a thermal drone survey) uses infrared cameras to detect the heat signatures that solar module defects give off before they show up as measurable production loss. Drones now perform this inspection at utility scale, covering entire sites in hours instead of the weeks a ground crew would need with handheld thermal cameras.
This guide is written for asset owners, EPCs, and O&M teams who need to understand the process, the IEC 62446-3 standard that provides guidance for it, how AI analytics changes the economics, and what a compliant inspection report should actually contain.
Thermal imaging inspection is the use of infrared (IR) cameras to detect ΔT (temperature differential) on solar modules that indicate underlying electrical or mechanical defects. A module, cell, or connection running significantly hotter or cooler than neighbouring components under the same irradiance usually indicates an underlying defect.
Large utility-scale sites can contain hundreds of thousands of PV modules, which is why walking every row with a handheld thermal camera is not a viable inspection method and drone capture has become the practical alternative.
A defective cell or connection (cracked, delaminated, corroded, or poorly soldered) resists current flow. That resistance dissipates energy as heat rather than electricity, and the excess heat radiates outward as infrared energy an IR sensor can capture.
One caveat worth knowing: emissivity (how efficiently a surface radiates heat) varies slightly by material and viewing angle, which is why radiometric cameras and calibrated capture angles matter more than raw sensor resolution.
A compliant inspection follows a repeatable six-step process, not an ad hoc flyover:
Flight planning: Define altitude, image overlap, and capture window. Industry guidance based on IEC 62446-3 commonly uses plane-of-array irradiance of at least 600 W/m² and cloud cover no greater than 2 oktas, which in practice restricts capture to a mid-day window on clear or near-clear days.
Radiometric IR + visual capture: A dual-sensor drone flies a pre-programmed grid over the array, capturing radiometric thermal frames alongside visual imagery for cross-reference.
Radiometric data matters because it stores an actual temperature value in every pixel; a colorized thermal image only assigns colors based on relative brightness within that one frame, so it can't be compared reliably across images or expressed as a quantified ΔT.
Image processing: Individual radiometric tiles are stitched into a georeferenced thermal orthomosaic of the full site, preserving per-pixel temperature values.
AI-based defect detection: SenseHawk’s AI model detects and classifies anomalies using module type and thermal signatures, and automatically extracts the temperature differential (ΔT) for each detected anomaly.
Geospatial mapping: Every flagged anomaly is pinned to its exact module, string, and inverter/combiner, not just a rough GPS coordinate on a site map.
Reporting & remediation: The platform generates an IEC 62446-3-aligned report and dispatches geo-tagged field work orders, closing the loop between detection and repair.
At SenseHawk, this workflow runs through SenseHawk Therm, which layers AI-driven classification and IEC-aligned reporting on top of the raw radiometric capture. Across 200GW in 20 plus countries. Single operator can image above 100MWp per day
What this looks like in practice: a drone scans a 150 MW site in a single afternoon. The AI classifier flags 243 thermal anomalies across the capture. After validation against ΔT thresholds and image quality, 38 are confirmed as high-priority defects and dispatched as geo-tagged work orders, letting technicians repair the affected modules before the loss compounds into a measurable dent in monthly generation.
At that scale, a single flight over a 300 MW site can produce tens of thousands of individual thermal images, a volume that makes automated stitching and AI triage a requirement, not a convenience.
The majority of thermal defects rarely announce themselves through SCADA alone. A single underperforming string can hide inside portfolio-level generation data for months.
NREL research generally puts expected PV degradation in the range of roughly 0.5–1% per year, depending on technology and conditions. Thermal inspection helps operators identify localized defects and abnormal losses that can sit outside normal degradation trends.
Demand for that capability is scaling fast. According to Verified Market Research, the drone solar inspection market was valued at roughly $1.2 billion in 2024 and is projected to reach $6.8 billion by 2032, a reflection of how central drone thermal imaging has become to utility-scale O&M budgets.
Comparing outcomes with and without a thermal inspection program makes the gap concrete. The resolution rate, generation loss, and payback figures below are drawn from SenseHawk's own analysis of 14 GW of thermography across 17 countries:
Without Thermal Inspection | With Thermal Inspection | |
|---|---|---|
Defect visibility | Detected only after measurable yield loss or SCADA alarms | Detected at the cell/module level before yield loss compounds |
Detection speed | Weeks of manual walkdowns for a large site | Full utility-scale site scanned in hours; 100+ MWp/day per operator |
Traceability | Approximate location, often re-walked to confirm | Anomaly geo-tagged to exact module, string, and inverter |
Warranty leverage | Limited evidence for manufacturer claims | Radiometric, timestamped data suitable for warranty and bankability audits |
Remediation rate | Findings often sit in spreadsheets or PDFs | Map-based work orders drive faster closure; resolution rates rising from 46% to 91% |
Financial exposure | Losses accumulate silently; median 1.4% generation loss (range 0.1%–20%) from undetected thermal defects | Losses caught and remediated within a 3–5 month payback window on utility-scale sites |
Each defect type produces a distinct thermal signature, and that signature is what tells a reviewer how urgently it needs a work order. Scan the summary below, then read the detail on any row that applies to your site:
Defect | Thermal Pattern | Typical Priority |
|---|---|---|
Hotspot | Bright, localized spot on a single cell | Medium to High |
Bypass diode failure | Uniform heating across a full sub-string | Medium to High |
String/combiner failure | Entire string reads hot or cold vs. neighbors | High |
PID | Pattern-based heating across a table or row | Medium to High |
Junction box fault | Localized heat at the rear connection point | Medium to High |
Soiling/shading hotspot | Diffuse spot that shifts position across scans | Low |
Module cracking | Irregular pattern following a crack line | Medium to High |
The hotspot detection solar operators run most frequently targets a single overheating cell rather than an entire string or module.
Thermal appearance: Small, sharply localized bright spot on a single cell.
Root cause: Cell cracks, delamination, or shading-induced reverse bias.
Business impact: Accelerates cell degradation and, in severe cases, creates a fire risk.
Corrective action: Isolate and inspect the module visually; replace under warranty if cell-level defect is confirmed.
Thermal appearance: A full sub-string (typically one-third of the module) heats uniformly rather than a single point.
Root cause: Diode failed open or short, no longer routing current around a shaded/faulty sub-string.
Business impact: Consistent partial power loss across every affected module; can mimic soiling if misdiagnosed.
Corrective action: Confirm with electrical testing; replace the junction box or diode assembly.
Thermal appearance: An entire string of modules reads uniformly hot or cold relative to adjacent strings.
Root cause: Loose connector, blown fuse, or combiner-box fault interrupting current flow.
Business impact: Full string offline, one of the largest single-event production losses an O&M team will see.
Corrective action: Prioritize as an urgent work order; dispatch electrical technician same-day.
Thermal appearance: Pattern-based heating following a predictable spatial layout across a table or row (often edge or center modules).
Root cause: Voltage stress between cell and frame drives ion migration, degrading cell performance over time.
Business impact: Power loss that varies widely by severity and exposure time, with field studies documenting cases well into double digits and worsening over the asset's life if untreated.
Corrective action: Confirm with I-V curve testing; evaluate PID-recovery devices or negative grounding retrofits.
Thermal appearance: Localized heating at the module's rear junction box rather than on the cell surface.
Root cause: Corroded connections, water ingress, or poor factory soldering.
Business impact: Risk of connector melt or fire in advanced cases; often precedes full module failure.
Corrective action: Replace junction box assembly; flag manufacturer batch for broader QA review.
Thermal appearance: Diffuse or partial hotspot pattern that shifts position across scans taken at different times of day.
Root cause: Dust, bird droppings, vegetation, or structure shadow raising localized cell resistance.
Business impact: Recoverable loss with no physical damage, but consistent yield drag if uncleaned.
Corrective action: Schedule targeted cleaning or vegetation management rather than warranty replacement.
Thermal appearance: Irregular, jagged thermal pattern that visually follows a crack line rather than a clean circular hotspot.
Root cause: Mechanical stress from transport, installation, hail, or thermal cycling.
Business impact: Progressive degradation as moisture ingress worsens the crack over time.
Corrective action: Document with visual + thermal overlay for warranty claim; monitor or replace depending on severity.
IEC TS 62446-3 is the key technical specification for outdoor infrared thermography of PV modules and balance-of-system components on operating plants. It defines the conditions under which a thermal inspection is considered valid and repeatable, rather than a best-effort scan. The specifics below reflect how the standard is consistently summarized across industry guidance; teams working toward a compliance audit should confirm exact clause language against the current edition they've licensed.
Key requirements commonly cited from the standard include:
Irradiance threshold: Plane-of-array irradiance at or above 600 W/m² (or roughly 30% of rated load) at the time of capture, since lower irradiance produces ΔT values too small to reliably attribute to a defect.
Sky conditions: Cloud cover capped at 2 oktas to avoid transient shading skewing readings.
ΔT reporting format: Findings reported with quantified temperature differentials and geolocation, not just a marked-up photo.
Image resolution: Sufficient spatial resolution to resolve individual cells and identify cell-level thermal anomalies, with the required resolution depending on camera specifications, flight altitude, and inspection objectives.
A peer-reviewed study in EPJ Photovoltaics confirms that both irradiance and drone flight altitude materially affect the reliability of IR thermography readings, reinforcing why these variables are treated as compliance-relevant rather than left to operator judgment.
Compliance matters beyond audit-readiness: warranty claims to module manufacturers and bankability reviews by lenders or insurers increasingly require IEC 62446-3-aligned evidence before they'll act on a finding. Non-compliant scans (off-angle, low-irradiance, uncalibrated) are routinely rejected as unusable data.
Challenge | Mitigation |
|---|---|
Weather dependency | Schedule capture windows around forecasted irradiance and cloud cover (≥600 W/m², ≤2 oktas) rather than fixed calendar dates; build slack into inspection schedules for weather slippage. |
Large datasets | Utility-scale sites generate tens of thousands of radiometric images per flight; automated stitching and cloud processing pipelines are necessary; manual review doesn't scale past a few MW. |
False positives | AI classification models trained on labeled defect libraries reduce false flags from reflections, vegetation, or transient shading; human QA review on a sampled subset catches remaining edge cases. |
Flight permissions | Coordinate BVLOS/waiver requirements and site-specific airspace restrictions well ahead of the inspection window; utility-scale sites near airports or restricted zones need lead time measured in weeks. |
Battery limitations | Plan flight legs against realistic per-battery coverage and stage swap-in batteries on-site to avoid gaps in capture during the valid irradiance window. |
Data processing time | Edge-processing or same-day cloud pipelines shrink the gap between flight and actionable report, which matters since raw data has no value sitting unprocessed for weeks. |
Inconsistent reporting | Standardize on IEC 62446-3-aligned templates so findings are comparable across inspection cycles, vendors, and site portfolios instead of locked in one-off PDF formats. |
Not every drone service that offers “thermal scans” delivers inspection-grade data. Most vendors in this space still stop at the flight: they hand over a folder of thermal images, or at best a PDF with a few anomalies circled, and leave classification, prioritization, and dispatch as someone else's problem. That gap is where the value actually lives.
A genuine inspection platform doesn't just capture radiometric imagery; it runs AI classification on top of it, maps every anomaly to a specific module and string, and pushes the confirmed defects into a work-order pipeline your field team already uses. The practical difference is what happens after the scan: inspection-grade platforms turn thermal data into classified, geo-tagged findings that can move directly into remediation workflows.
Drone thermal imaging is only one input into a broader O&M stack; for a wider view of what that stack should cover, see SenseHawk's rundown of the Features of Solar Project Management Software.
Evaluate providers and platforms against this checklist:
Sensor resolution: Confirm the drone captures true radiometric data (per-pixel temperature values), not a visual-IR blend that only looks like a thermal image.
AI defect classification accuracy: Ask what anomaly types the model distinguishes (hotspot vs. diode failure vs. string fault) and how classification accuracy is validated against ground-truth data.
IEC 62446-3-aligned reporting: Verify reports include irradiance conditions, ΔT values, and resolution metadata, not just annotated images.
Integration with O&M/work-order systems: A defect list in a PDF is a dead end. Look for direct integration into your work-order or CMMS pipeline, not a separate report to manually re-enter.
Geo-tagged traceability: Every anomaly should map to a specific, serialized module, which matters for both warranty claims and repeat-inspection trend tracking over the asset's life.
Thermal imaging inspection has shifted from a periodic, manual diagnostic exercise to a drone-first, AI-assisted discipline that's now essential to protecting yield at utility scale. The combination of radiometric capture, IEC 62446-3-aligned methodology, and AI-driven classification turns a single flight into a prioritized, geo-tagged action list rather than a folder of thermal photos.
Thermal data only creates value when it leads to repairs. Platforms like SenseHawk Therm connect drone inspections directly to geo-tagged work orders, helping utility-scale teams move from defect detection to remediation faster, instead of leaving findings to sit in a report nobody actions. For sites that already route SCADA alarms into field dispatch, see how SenseHawk closes that same loop between solar alarms and autonomous drones.
What is thermal imaging inspection?
Thermal imaging inspection uses infrared cameras to detect ΔT on solar modules, connections, and balance-of-system components that indicate defects such as cell damage, diode failure, or string faults, often before those defects cause measurable production loss.
How accurate are drone thermal inspections?
Accuracy depends on meeting appropriate IEC 62446-3 inspection conditions, including suitable irradiance, sky conditions, image resolution, and flight parameters. Radiometric drone thermography combined with AI classification can then help distinguish different types of thermal anomalies.
Can thermal imaging inspection detect problems that visual drone inspection misses?
Yes. Visual (RGB) inspection catches surface-visible issues like cracked glass, soiling, or vegetation overgrowth, but it cannot detect internal electrical faults such as bypass diode failure, PID, or early-stage hotspots, which have no visible signature but show up clearly as a temperature anomaly.
What weather conditions are needed for accurate thermal imaging inspection?
Thermal inspection guidance based on IEC 62446-3 commonly uses plane-of-array irradiance of at least 600 W/m² and cloud cover no greater than 2 oktas at the time of capture. Wind speed also matters indirectly: strong wind cools module glass and can suppress the ΔT signal a defect would otherwise produce.
Is drone thermal imaging inspection accurate enough for warranty claims?
It can be, provided the inspection follows IEC 62446-3 methodology and the report documents irradiance conditions, ΔT values, and image resolution alongside geolocation. Manufacturers, lenders, and insurers may require this level of documentation before acting on a claim or assessment; ad hoc scans without standardized conditions are frequently rejected as insufficient evidence.
How much does drone-based thermal imaging inspection cost per MW?
Cost per MW varies by site size, geography, and whether AI analytics and reporting are bundled with the flight service, since larger sites benefit from economies of scale in flight planning and processing. Rather than quoting a flat figure, most operators evaluate cost against payback period. SenseHawk's own field data across utility-scale sites shows thermal inspection programs typically pay for themselves within 3–5 months through recovered generation.
The practical difference is what happens after the scan: inspection-grade platforms turn thermal data into classified, geo-tagged findings that can move directly into remediation workflows.
Karthik Mekala
CMO
