Thermal imaging solar inspection is the most powerful diagnostic tool available for utility-scale PV plants. It reveals electrical faults that are entirely invisible to SCADA monitoring — panels that appear to perform normally on the production dashboard but are silently losing revenue through defective cells, failed bypass diodes, and developing hotspots.
This guide covers everything from the physics of why thermal imaging works, to IEC TS 62446-3 compliance requirements, to how AI is transforming fault classification at scale. Whether you’re commissioning your first drone solar inspection programme or evaluating inspection providers, this is the reference you need.
Key takeaway: Thermal imaging detects faults invisible to SCADA. IEC TS 62446-3 requires at least 600 W/m² irradiance on the module plane, low wind and clear sky, and every finding is graded by its temperature delta (ΔT) against healthy reference modules. AI drone inspection costs a fraction of a conventional drone survey or a manual inspection, and AI classification reaches up to 99% fault-detection accuracy.
The Physics of Solar Panel Thermal Signatures
A healthy solar cell converts approximately 20–22% of incident solar radiation into electricity. The remaining 78–80% is dissipated as heat. Under normal operation, this heat dissipation is uniform across a panel — all cells operate at similar temperatures.
When a cell is defective, its electrical behaviour changes. A cell experiencing reverse bias (forced into a non-generating state by shading or a bypass diode failure) dissipates more energy as heat than its neighbours, not less. This localised excess heat emission is the thermal signature that infrared cameras detect.
The key physics principle: Defective solar cells get hotter than healthy ones because they dissipate more energy as heat rather than converting it to electricity. The greater the electrical defect, the greater the temperature differential.
Minimum Detection Conditions
| Condition | Requirement | Why It Matters |
|---|---|---|
| Irradiance | ≥600 W/m² on the module plane (ideally ≥800 W/m²) | Below this, thermal contrast between defective and healthy cells is too low for reliable detection |
| Temperature differential | ΔT measured against a healthy reference module of the same type; a ΔT of around 10 K is the common reporting threshold | Separates real anomalies from normal module-to-module variation |
| Wind and cloud | Low wind (IEC TS 62446-3 sets a maximum of 28 km/h) and little or no cloud | Wind convects heat away and passing cloud changes irradiance mid-flight, both reducing thermal contrast |
| Time of day | Typically around solar noon | Maximum irradiance window |
IEC TS 62446-3: The Standard for Thermal Imaging Solar Inspections
IEC TS 62446-3 (Photovoltaic (PV) systems — Requirements for testing, documentation and maintenance — Part 3: Photovoltaic modules and plants — Outdoor infrared thermography) is the international technical specification governing how thermal inspections of solar plants should be conducted and documented.
What IEC TS 62446-3 Requires
| Requirement | Specification |
|---|---|
| Minimum irradiance | 600 W/m² on the module plane |
| Maximum wind speed | 28 km/h |
| Camera | Radiometric, calibrated, with thermal sensitivity (NETD) of 0.1 K or better |
| Image resolution | Enough pixels per cell to resolve single-cell anomalies, which sets the maximum flight altitude for a given lens |
| Documentation required | Irradiance, ambient temperature, wind speed, inspection date/time, camera details, inspector qualification |
| Anomaly classification | Class of Abnormality (CoA): CoA 1 no abnormality, CoA 2 thermal abnormality, CoA 3 safety-relevant thermal abnormality |
Why compliance matters: Lenders, insurers, and asset acquirers increasingly require IEC TS 62446-3 compliant inspection reports. Non-compliant reports may be rejected for insurance claims or financing reviews, regardless of the quality of the underlying inspection work. If you also report plant performance, see how IEC 62446 and IEC 61724 differ.
The IEC (International Electrotechnical Commission) publishes the full text of IEC TS 62446-3 in its webstore.
Fault Types Detected by Thermal Imaging
Bypass Diode Failures
What they are: Most crystalline modules contain 3 bypass diodes, each protecting one sub-string of cells. When a diode fails, or conducts continuously, its sub-string stops contributing — the module shows a heated third rather than a single hot cell.
Thermal signature: A rectangular warm zone covering one-third of the module (the sub-string protected by the affected diode). The pattern is distinct and geometric, and the diode junction box itself often shows as a hot point.
Revenue impact: The module loses roughly one-third of its output until the diode or module is replaced.
Repair urgency: HIGH. Unresolved bypass diode failures accelerate cell degradation in the affected zone and can progress to hotspots.
Hotspots
What they are: Localised overheating of a single cell, typically caused by partial shading, cell mismatch, soiling on a single cell, or a crack affecting one cell.
Thermal signature: A single cell or small cluster of cells at a much higher temperature than its neighbours, which can exceed 80°C under high irradiance.
Revenue impact: Moderate from the single cell, but high risk of panel damage and fire at temperatures above 80°C.
Repair urgency: CRITICAL above 80°C. Fire risk is the primary concern, which is why IEC TS 62446-3 treats safety-relevant thermal abnormalities as their own class.
Soiling and Shading
What it is: Bird droppings, dust accumulation, leaf debris, or pollen deposited on panel surfaces. Even partial soiling of a single cell can force bypass diode activation across that cell group.
Thermal signature: Irregular cool patches (soiled cells receive less irradiance, generate less heat) or warm patches (if soiling causes partial shading that forces reverse bias). Bird droppings show a distinctive pattern matching their physical shape.
Revenue impact: Losses scale with how much of the array is covered and for how long. A single bird dropping over one cell can push a whole sub-string onto its bypass diode.
Repair urgency: LOW-MEDIUM. Cleaning resolves the issue — no panel replacement required. But delayed cleaning means extended revenue loss.
Potential-Induced Degradation (PID)
What it is: Leakage current flowing through the module frame, driven by the high voltage differential between the module and the grounded frame/mounting structure. Over time, sodium ions migrate through the glass and affect cell performance.
Thermal signature: Affected modules typically sit at the negative end of a string and show a patchwork of warmer cells, often strongest near the frame, while overall output falls. The pattern repeats across modules in the same string position.
Detection challenge: PID has one of the subtlest thermal signatures of all fault types. It requires careful comparison against adjacent panels and may be easier to confirm through string-level production data or IV curve tracing than thermal imaging alone.
Revenue impact: Advanced PID can remove a large share of module power. Early-stage PID is partially reversible.
Affected panel types: p-type monocrystalline and multicrystalline panels are more susceptible than n-type. More common in high-voltage strings and in hot, high-humidity climates.
Delamination
What it is: Separation of the encapsulant layer (typically EVA) from the glass or backsheet. Creates air pockets that alter thermal conductivity and allow moisture ingress.
Thermal signature: Diffuse warm blotches, often appearing near panel edges or corners. The delaminated area has different thermal conductivity from intact laminate, creating an irregular warm pattern.
Revenue impact: Initially low, but delamination creates pathways for moisture ingress, which accelerates corrosion of cell contacts and leads to progressive degradation.
Repair urgency: MEDIUM. Not an emergency, but affected panels should be monitored and scheduled for replacement.
Cell Cracks (Micro-cracks)
What they are: Physical fractures in silicon cells caused by mechanical stress — transport damage, installation forces, hail impact, thermal cycling, or wind load.
Thermal signature: Subtle linear or irregular thermal anomalies following crack lines. However, many micro-cracks are electrically inactive (the crack does not break the electrical path) and show no thermal signature.
Detection limitation: Thermal imaging is unreliable for inactive micro-cracks. Electroluminescence (EL) imaging — which requires the panels to be energised at night and imaged with a specialised camera — is far more effective. Thermal imaging catches cracks only when they are large enough to affect cell electrical behaviour.
Revenue impact: Variable — inactive cracks may have minimal near-term impact. Active cracks that interrupt electrical paths cause production loss proportional to the affected cell area.
String and Connection Faults
What they are: Failed MC4 connectors, corroded junction box terminals, or open-circuit string connections.
Thermal signature: An entire string (row of panels) appears uniformly warmer than its neighbours on the thermal image — an open-circuited module converts none of its absorbed energy to electricity, so all of it becomes heat. This is the most visually dramatic thermal pattern, immediately obvious even on a low-resolution overview image. A failing connector also shows as a small, very hot point.
Revenue impact: Complete loss of the string’s generation — the output of every module in it.
Repair urgency: CRITICAL. Complete string loss is the highest-revenue-impact fault type and should be investigated within days, not weeks.
Fault Severity Classification
IEC TS 62446-3 grades findings by Class of Abnormality (CoA 1–3). Most O&M teams add a ΔT-based severity band on top so that repair crews can prioritise. The bands below are common industry practice rather than limits taken from the standard.
| Fault Type | Typical ΔT | Severity Band | Production Impact | Urgency |
|---|---|---|---|---|
| Hotspot (critical) | >40 K | 3 (safety-relevant) | Single cell, but fire and module-damage risk | CRITICAL — fire risk |
| String fault | Whole string warmer than neighbours | N/A | Entire string offline | CRITICAL |
| Bypass diode failure | 20–40 K | 2 | About one-third of module output | HIGH |
| Bypass diode failure (minor) | 10–20 K | 1 | Partial sub-string loss | MEDIUM |
| Soiling/shading | Variable | 1–2 | Varies with coverage and duration | MEDIUM |
| Delamination | 10–25 K | 1–2 | Low at first, rising with moisture ingress | MEDIUM |
| PID | Subtle/diffuse | Requires interpretation | Can be large in advanced cases | MEDIUM |
| Micro-cracks (active) | 10–20 K | 1 | Proportional to inactive cell area | LOW-MEDIUM |
Thermal Camera Specifications for Solar Inspection
Minimum Requirements for IEC TS 62446-3 Compliant Inspection
| Specification | Minimum | Recommended |
|---|---|---|
| Thermal resolution | 320×240 | 640×512 |
| NETD (noise equiv. temp. diff.) | ≤0.1 K (100 mK) | ≤50 mK |
| Temperature accuracy | ±2°C or ±2% | ±1°C or ±1% |
| Spectral range | 7.5–14 μm (LWIR) | 8–14 μm |
| Data type | Radiometric | Radiometric with paired RGB |
Common Cameras Used in Professional Solar Inspection
- FLIR Vue Pro R (640×512, drone-mounted, radiometric)
- DJI Zenmuse H20T (dual thermal+RGB, integrated with Matrice 300)
- Teledyne FLIR A50/A70 (fixed mount for continuous monitoring)
- FLIR T-Series (handheld, for manual inspection)
Radiometric vs non-radiometric: For IEC TS 62446-3 compliance and AI analysis, radiometric thermal data (every pixel has a calibrated temperature value) is required. Non-radiometric (visual-only thermal images) cannot be used for compliance documentation or accurate fault severity assessment. Emissivity also has to be set correctly: module glass has a high emissivity when viewed close to perpendicular, but reflections from the sun and sky increase at shallow viewing angles.
Drone-Based vs Ground-Based Thermal Inspection
| Dimension | Drone Inspection | Manual Handheld |
|---|---|---|
| Coverage | Many MW per day | A small fraction of a MW per day |
| Relative cost | A fraction of manual with AI analysis (a conventional drone survey sits in between) | Highest |
| Camera resolution (at panel) | 640×512 from altitude | 640×480 at a few metres |
| Ground sample distance | Set by flight altitude and lens | Millimetre-level at close range |
| IEC TS 62446-3 compliance | ✅ Yes (with radiometric camera) | ✅ Yes |
| GPS accuracy (fault location) | Centimetre-level with RTK | Depends on manual logging |
| Structural fault detection | ✅ Yes (aerial RGB) | ❌ Not typically |
| Annual inspection of a large plant | Days | Weeks to months |
The altitude trade-off: Drone inspection trades close-range resolution for coverage speed. Flight altitude has to be planned so that the ground sample distance still meets the IEC TS 62446-3 resolution requirement for the camera and lens in use; at that resolution, every anomaly above the reporting ΔT is visible. For micro-crack detection, close-range EL imaging is required regardless of inspection method. Our thermal drone inspection guide covers flight planning, altitude and emissivity settings in more depth.
How AI Classifies Thermal Anomalies Automatically
Traditional drone inspection required a human analyst to review every thermal image and manually identify anomalies. Every flight produces thousands of radiometric images, so this was a significant bottleneck.
AI-based thermal analysis works as follows:
- Pre-processing: Raw radiometric thermal images are corrected for vignetting, atmospheric absorption, and emissivity assumptions
- Panel detection: Computer vision identifies individual panels and their boundaries in the image
- Anomaly detection: Statistical comparison of each panel’s temperature distribution against adjacent panels and against a site-wide baseline
- Classification: Machine learning model classifies the anomaly type based on spatial pattern, temperature differential, and location within the panel
- Severity scoring: Assigns a severity band and IEC TS 62446-3 class of abnormality based on the ΔT measurement and fault pattern
- Report generation: Automated fault list with GPS coordinates, panel ID, fault type, severity, and recommended action
Detection accuracy: Well-trained AI classification models reach up to 99% fault-detection accuracy on bypass diode failures and hotspots (the most common and financially significant fault types). Accuracy is lower for subtle faults like early-stage PID and delamination, where human review of AI-flagged anomalies is still recommended.
Reading a Thermal Inspection Report
A standard IEC TS 62446-3 compliant thermal inspection report should contain:
| Section | What It Includes |
|---|---|
| Executive summary | Total anomalies by class, estimated revenue impact, priority action list |
| Inspection conditions | Date, time, irradiance levels, wind speed, temperature — confirms IEC compliance |
| Equipment details | Camera model, drone type, pilot certification, software version |
| Site overview image | Thermal orthomosaic of entire site — shows distribution of anomalies spatially |
| Fault register | Every anomaly with: GPS coordinates, panel ID, fault type, IEC class, ΔT, photo, recommended action |
| Priority repair list | Faults ranked by revenue impact (not just severity class) |
From Detection to Action: Prioritising Repairs
A thermal inspection of a utility-scale site typically returns hundreds of anomalies. Not all warrant immediate repair. The prioritisation framework below feeds straight into a drone-led solar panel maintenance plan:
Immediate Action (Within 7 Days)
- Safety-relevant hotspots (ΔT >40 K or above 80°C) — fire risk
- String open-circuit faults — complete string loss
Planned Action (Within 30 Days)
- Band 2 bypass diode failures (ΔT 20–40 K)
- Any fault whose estimated annual revenue loss exceeds the cost of the repair visit
Next Scheduled Maintenance
- Band 1 anomalies (ΔT 10–20 K)
- Soiling events (schedule cleaning crew)
- Early-stage delamination (monitor, schedule for replacement)
Conclusion
Thermal imaging is not a supplementary tool for solar inspection — it is the primary method for detecting the fault types that matter most financially. SCADA monitoring tells you what your inverters see. Thermal imaging tells you what your panels are actually doing, and feeding those results into solar panel monitoring software closes the gap between the two.
For utility-scale portfolios, AI drone thermal inspection delivers IEC TS 62446-3 compliant documentation, same-day fault reporting, and detection accuracy equivalent to or better than manual inspection — at a fraction of the cost and in a fraction of the time.
See ClearSpot’s IEC TS 62446-3 compliant drone inspection in action. Book a demo to see how the platform handles inspection planning, AI analysis, and compliance reporting for your portfolio.
Frequently Asked Questions
It uses radiometric infrared cameras, usually drone-mounted, to measure the surface temperature of every PV module and flag cells, sub-strings or strings that run hotter than healthy reference modules. Those temperature deltas (ΔT) reveal faults that SCADA data cannot see.
A hotspot is a cell, or small group of cells, running far hotter than its neighbours because shading, soiling, cracks or mismatch force it into reverse bias. Hotspots above about 80°C are a fire and module-damage risk and should be repaired first.
Hotspots, bypass diode failures, open strings and failing connectors, soiling and shading, delamination, potential-induced degradation and electrically active cell cracks. Inactive micro-cracks usually need electroluminescence (EL) imaging instead.
At least 600 W/m² irradiance on the module plane, low wind, little cloud and a calibrated radiometric camera. Irradiance, ambient temperature, wind speed and camera details must be recorded in the report so the findings can be audited.
A radiometric camera stores a calibrated temperature for every pixel, so ΔT can be measured and each anomaly graded. Non-radiometric images only show relative colour and cannot support IEC TS 62446-3 documentation or accurate severity scoring.
Most operators inspect at least once a year, many twice a year, and again after hail, storms or other severe weather. Sites with recurring faults or open warranty claims justify more frequent flights.
AI detects each module, compares its temperature distribution with its neighbours and the site baseline, classifies the anomaly pattern and assigns a severity, producing a geolocated fault register within hours instead of days of manual image review.