Complete Guide To Thermal Anomaly Detection (2026)
Averroes
Apr 07, 2026
Every hotspot has a story.
A junction box running 60°F above its neighbors, a heated substring that wasn’t there six months ago, a PID pattern creeping along string ends – each one is a specific fault, at a specific location, with a specific cost attached to it.
Find them early and they’re maintenance items. Miss them and they’re write-offs.
We’ll cover the full thermal anomaly detection workflow: types, causes, methods, interpretation, and response.
Key Notes
ΔT vs neighbors – not absolute temperature – is the correct diagnostic metric.
BOS components warrant urgent action even at moderate temperature differentials.
High-defect fleets degrade at ~2.6%/year vs ~1.2%/year in well-maintained plants.
Drone, fixed sensor, and handheld thermography serve different roles – most serious O&M programs run all three.
What Is A Thermal Anomaly? (& What It Isn’t)
A thermal anomaly is a consistent, localized temperature elevation above comparable components operating under the same conditions.
The key word is localized – a whole array running warm on a hot afternoon is not an anomaly. One module running 60°F hotter than its neighbors in the same string, under stable irradiance, absolutely is.
The Practical Metric Is ΔT (Delta-T)
The temperature difference between the suspect component and its direct peers.
Absolute temperature matters less than relative temperature, because irradiance, ambient heat, and wind all shift the baseline together.
What you’re looking for is what’s different, not what’s hot.
Working ΔT Bands For Operators:
ΔT vs Neighbors
Severity
Typical Action
~3–8°C
Low / monitor
Log, re-check on next inspection
~8–15°C
Medium / schedule repair
Plan maintenance, quantify yield impact
≥15–20°C
High / urgent
Prioritize; safety assessment required
One Important Caveat:
For balance-of-system components – junction boxes, connectors, DC cabling, combiner boxes – even moderate ΔT should be treated as urgent.
These components are where overheating transitions from a performance issue to a fire risk.
Types of Thermal Anomalies: What The Pattern Tells You
The pattern and location of a thermal anomaly in a solar array carries diagnostic information.
Here’s what each type typically signals:
Cell-Level Hotspots
One or a few cells are significantly hotter than the rest of the module.
Common causes include hard or soft shading, soiling, cracked cells, and cell mismatch.
Left unaddressed, these progress from localized performance loss to permanent damage: browning, glass cracking, and in severe cases, fire.
Multi-Hotspot / PID Patterns
Many small hot cells scattered across a module or along string ends, sometimes forming a speckled or banded pattern.
This is a classic signature of Potential Induced Degradation (PID), widespread microcracks, or long-term moisture ingress. The risk here is systemic – it’s rarely one module.
Heated Substring
One-third or two-thirds of a module is uniformly hotter than the rest, aligned with the internal cell string layout. This points to bypass diode activation from shading or mismatch, or a partially failed bypass diode.
A failed diode isn’t just a performance issue. It’s a genuine fire hazard.
Whole-Module Heating
An entire module runs clearly hotter than its string neighbors.
Typical causes: open-circuit faults, incorrect connections, severe internal degradation. That module is dissipating energy as heat rather than producing current.
Hot Junction Box
The junction box is significantly warmer than adjacent modules. Loose connections, corroded terminals, or failing bypass diodes inside the box.
Treat this as Priority 1 – it’s one of the highest fire-risk signatures in PV thermal inspection.
String-Level / Array-Pattern Anomalies
Repeating patterns like consistently hotter string ends, one side of a tracker row running warmer, or alternating hot/cool rows.
Causes range from PID at string ends to tracker misalignment to combiner or inverter faults.
The risk is systematic underperformance affecting many modules at once.
BOS Heating
Cables, connectors, and combiner boxes with localized hotspots. Even at moderate ΔT, overheating here is closely tied to arcing risk.
This is where you get your “acted on quickly vs. ignored” outcomes with the starkest difference in consequences.
Root Causes of Thermal Anomalies
Most thermal anomalies in solar arrays trace back to five root cause categories:
1. Shading & Soiling
Blocked light forces affected cells into reverse bias, dissipating current as heat instead of producing it.
Most reversible category – clean, re-inspect, anomaly typically gone
2. Installation & Handling Damage
Physical damage during transport or installation creates electrical weaknesses that surface thermally, sometimes months after commissioning.
Cracked cells, bent frames, over-tightened clamps
Walking on modules is a more common microcrack cause than most teams expect
3. Manufacturing & Material Defects
Latent issues alter local current flow and heat dissipation – and they worsen gradually over years, not overnight.
Poor soldering, EVA browning, delamination, moisture ingress
Classic signatures: scattered hot cells, hot module edges, string-end warming
4. Electrical Connection & Wiring Faults
Loose or corroded connectors, poor terminations, and failing diodes show up as sharp, localized hot points – a very hot single spot rather than a diffuse warm area.
Strings or blocks run hotter than peers under identical conditions – the pattern is the tell
How These Causes Differ By System Type:
It’s also worth distinguishing reversible from structural issues:
Early hotspots driven by soiling or shade disappear after cleaning.
Persistent hotspots that remain under clean, unshaded conditions signal underlying cracks, corrosion, or lamination failure – and those don’t self-resolve.
Impact: Yield, Lifespan, Safety
Thermal anomaly detection isn’t about chasing perfect data, but about protecting revenue and equipment.
Yield Impact Is Immediate & Measurable
Silicon modules lose roughly 0.5–0.8% efficiency per °C above nominal operating temperature. A serious hotspot can cut an affected module’s output by up to 10–15%.
Aggregate that across a high-defect fleet and you’re looking at degradation rates of ~2.6%/year versus ~1.2%/year in low-defect plants – a gap that compounds every year the anomalies go unaddressed.
Modules with persistent hotspots reach replacement thresholds earlier, pulling forward capital costs that weren’t in anyone’s repowering budget.
Safety Risk Is Concentrated But Real
The absolute probability of a PV fire is still low. But most documented PV fires trace back to overheating components (hotspots, failing connectors, junction boxes) that were either missed or monitored and not acted on.
Regular thermal inspection is now specifically recommended as a fire-risk mitigation measure by major O&M guidelines, not just a yield optimization tool.
Detection Methods & Technologies
There are three primary approaches to thermal anomaly detection, each with a different sweet spot:
Manual / Handheld Thermography
A technician with a thermal camera – best for small-to-medium systems, commissioning checks, and targeted follow-up on faults identified via drone or SCADA.
Highest flexibility and close-range detail
Easy to correlate thermal findings with visible defects in real time
Not practical at scale – a 50 MW farm on foot isn’t a realistic option
Drone-Based Thermal Inspection
UAVs carrying radiometric IR and RGB cameras flying automated grids – the standard approach for utility-scale farms and large industrial rooftops.
What would take weeks on foot takes days in the air
Consistent altitude and angle enable reliable ΔT comparison across the entire site
Requires proper flight planning, pilot competence, and the right environmental conditions
Fixed Sensors & Continuous Monitoring
Permanently installed back-of-module sensors, reference modules, and SCADA-integrated monitoring – good for trending and catching systemic issues as they emerge.
Continuous data stream, no scheduled flight windows needed
Strong for early warning of inverter over-temperature or unusual string behavior
Low spatial resolution: tells you that a problem exists, not which cell it is
In Practice, Most Serious O&M Programs Combine All Three
Drones for periodic full-site coverage, fixed sensors for continuous early warning between flights, and handheld thermography for ground-truth verification of specific faults.
What The Data Quality Requirements Look Like:
IR sensor: Minimum 320×240, with 640×480 preferred for detailed classification
Ground sample distance (GSD): At least 5 thermal pixels per cell – with standard 150×150 mm cells, that means roughly 3 cm/pixel, which at a 13mm thermal lens requires flying at ≲24m AGL
Thermal sensitivity: NETD <0.08°C to resolve small ΔT between neighboring cells
Irradiance at time of inspection: ≥600–700 W/m², stable conditions, low wind, late morning to mid-afternoon
Get the conditions wrong (scanning at low irradiance, in strong wind, at sunrise with long shadows), and you’ll generate a dataset full of artifacts and missed anomalies.
How to Interpret Thermal Data Accurately
Having thermal images is one thing. Reading them correctly is another.
Real Anomaly vs False Positive
A real anomaly stays hotter than adjacent comparable components once irradiance and wind stabilize. It matches a known fault pattern and isn’t explained by visible dirt or shadow in the RGB frame.
A false positive moves with the sun angle (reflection from glass), disappears after cleaning (soiling), or changes dramatically when you adjust the emissivity setting.
Common Misinterpretations That Cost Operators Time:
Confusing reflection with heating: Glass and metallic surfaces reflect IR. If apparent temperature changes sharply when you change viewing angle, you’re looking at reflection, not heat.
Over-trusting absolute temperatures: Without correct emissivity and distance settings, absolute readings can be off by 10°C or more. Always anchor your analysis in ΔT vs. neighbors.
Misreading auto-scaled palettes: Automatic color scaling can make a 2°C variation look dramatic. Always check what the scale actually represents before flagging something as a defect.
Ignoring conditions: A scan at 400 W/m² in 20 mph wind will understate or hide anomalies. If the data was collected outside recommended windows, weight your findings accordingly.
Correlating Thermal With Electrical Data Adds The Most Diagnostic Value
Thermal tells you where and how hot; electrical tells you how much it’s costing and often narrows the fault type.
Cross-check anomaly clusters against string or inverter current data:
If thermal shows problems but electrical data looks normal, you’re likely catching something early-stage.
If electrical underperformance exists without clear thermal features, look at PID, mismatch, or DC wiring issues that may be subtle in temperature but significant in output.
Thermal Anomaly Detection Software: How Averroes Automates Detection at Scale
At large sites, the bottleneck isn’t capturing thermal data, but processing it.
A single utility-scale inspection flight can generate tens of thousands of images. Manual review at that volume is slow, inconsistent, and expensive.
Averroes sits in the AI layer on top of your existing drone fleet and cameras – no new hardware required. The platform ingests your drone thermal and RGB imagery and runs purpose-built AI models to detect, classify, and report defects at scale.
What That Looks Like In Practice:
Processes up to ~80,000 images in under 20 hours, scalable to ~400,000 images/day
Outputs per-defect classification (e.g., “hotspot — severe”), ΔT severity, GPS/array location, and recommended action – not just a heat map
Detects hotspots, microcracks, soiling, vegetation encroachment, delamination, and thermal variations, including small-pixel defects that generic vision tools miss
95–98.5% defect detection accuracy with near-zero false positives
Models adapt to site-specific defect types with as few as 20–40 images per class (no need for thousands of labeled examples)
New or rare defects are flagged, reviewed, and folded back into the model so future inspections pick them up automatically
What’s Hiding In Your Thermal Inspection Data?
Get from raw drone imagery to actionable defect priorities.
Operational Response & Remediation
Detection without a response workflow is just data collection. The standard post-detection pattern is: detect → validate → classify → fix or monitor.
Three-Tier Triage:
Priority
Criteria
Action
Critical (P1)
Junction box, connector, cable with ΔT ≥15–20°C; module temperatures near 80–85°C; severe hotspots with matching electrical underperformance
Isolate if needed, dispatch immediately
Important (P2)
Cell/substring hotspots with ΔT ~8–15°C; PID patterns with measurable yield loss
Schedule repair in next maintenance window
Monitor (P3)
Low-ΔT hotspots (~3–8°C) linked to soiling, mild shading, or early degradation
Clean, re-inspect, track across future inspections
Typical Fixes By Issue Type:
Soiling-driven hotspots: Clean the modules, re-scan to confirm ΔT returns to normal. Many hotspots at large sites fall into this category and resolve completely.
Cracked cells, delamination, burn marks: Replace the module. Field repair isn’t practical or reliable for structural damage.
Connector and junction box issues: De-energize, inspect, re-terminate or replace. Treat as high priority given fire risk.
PID: Adjust grounding, modify inverter operating mode, or introduce anti-PID devices. Swap affected string-end modules as a short-term mitigation.
Repair vs Replace Decision…
Comes down to ΔT severity, measured yield impact, module age, and warranty status.
If a module has burn marks, visible delamination, or repeated severe hotspots after cleaning, replace it.
If the issue is clearly in the BOS and can be fixed by re-terminating a connector, fix it.
The Risk Of Inaction Compounds
Persistent hotspots accelerate degradation rates, void warranty coverage when documented defects are knowingly ignored, and in BOS components, increase the probability of insulation breakdown and DC arcing.
Frequently Asked Questions
What is the difference between active and passive thermography for solar panels?
Active thermography applies an external heat source to stimulate the panel before imaging, while passive thermography captures heat the panel generates naturally under normal operating conditions. For solar PV inspection, passive thermography is standard – the sun does the work, and you inspect under live load. Active thermography is used in controlled lab or manufacturing settings, not field O&M.
Can thermal anomaly detection be used at night?
Thermal anomaly detection on solar panels requires the array to be under load, which means daytime operation with sufficient irradiance. Night-time thermography can detect residual heat in BOS components like inverters, combiner boxes, and cabling – but it won’t reveal cell-level or module-level defects. For meaningful PV inspection, irradiance of ≥600–700 W/m² is the baseline requirement.
How accurate is drone thermal inspection compared to ground-based thermal inspection?
Drone thermal inspection matches or exceeds ground-based accuracy for array-wide defect detection when flown at correct altitude, GSD, and under suitable conditions. Ground-based handheld inspection has an edge in close-range resolution and immediate RGB correlation for individual fault verification. In practice, drone inspection finds the anomalies; handheld inspection confirms and classifies them.
What thermal camera specifications do I need for solar panel inspection?
For reliable solar panel thermal anomaly detection, the minimum is a 320×240 IR sensor with thermal sensitivity below 0.08°C NETD – though 640×480 is preferred for higher altitudes or detailed cell-level classification. Ground sample distance should hit at least 5 thermal pixels per cell, which at standard cell sizes means flying at roughly 24m AGL or below with a 13mm lens.
Conclusion
Thermal anomaly detection is, at its core, a prioritization problem.
You have a site generating thousands of data points per inspection flight, and somewhere in that data are the junction boxes on their way to arcing, the PID patterns quietly compounding across string ends, and the soiling-driven hotspots that disappear after one cleaning cycle.
The difference between operators who manage this well and those who don’t usually comes down to whether they have a system (for detection, interpretation, triage, and response) or whether they’re reacting to alarms after the damage is already done.
The tools exist to run that system properly. If you want to see what AI-powered detection looks like applied to your inspection data specifically – your module types, your fault patterns, your scale – Averroes is worth an hour of your time. Book your free demo now.
Every hotspot has a story.
A junction box running 60°F above its neighbors, a heated substring that wasn’t there six months ago, a PID pattern creeping along string ends – each one is a specific fault, at a specific location, with a specific cost attached to it.
Find them early and they’re maintenance items.
Miss them and they’re write-offs.
We’ll cover the full thermal anomaly detection workflow: types, causes, methods, interpretation, and response.
Key Notes
What Is A Thermal Anomaly? (& What It Isn’t)
A thermal anomaly is a consistent, localized temperature elevation above comparable components operating under the same conditions.
The key word is localized – a whole array running warm on a hot afternoon is not an anomaly. One module running 60°F hotter than its neighbors in the same string, under stable irradiance, absolutely is.
The Practical Metric Is ΔT (Delta-T)
The temperature difference between the suspect component and its direct peers.
Absolute temperature matters less than relative temperature, because irradiance, ambient heat, and wind all shift the baseline together.
What you’re looking for is what’s different, not what’s hot.
Working ΔT Bands For Operators:
One Important Caveat:
For balance-of-system components – junction boxes, connectors, DC cabling, combiner boxes – even moderate ΔT should be treated as urgent.
These components are where overheating transitions from a performance issue to a fire risk.
Types of Thermal Anomalies: What The Pattern Tells You
The pattern and location of a thermal anomaly in a solar array carries diagnostic information.
Here’s what each type typically signals:
Cell-Level Hotspots
One or a few cells are significantly hotter than the rest of the module.
Common causes include hard or soft shading, soiling, cracked cells, and cell mismatch.
Left unaddressed, these progress from localized performance loss to permanent damage: browning, glass cracking, and in severe cases, fire.
Multi-Hotspot / PID Patterns
Many small hot cells scattered across a module or along string ends, sometimes forming a speckled or banded pattern.
This is a classic signature of Potential Induced Degradation (PID), widespread microcracks, or long-term moisture ingress. The risk here is systemic – it’s rarely one module.
Heated Substring
One-third or two-thirds of a module is uniformly hotter than the rest, aligned with the internal cell string layout. This points to bypass diode activation from shading or mismatch, or a partially failed bypass diode.
A failed diode isn’t just a performance issue. It’s a genuine fire hazard.
Whole-Module Heating
An entire module runs clearly hotter than its string neighbors.
Typical causes: open-circuit faults, incorrect connections, severe internal degradation. That module is dissipating energy as heat rather than producing current.
Hot Junction Box
The junction box is significantly warmer than adjacent modules. Loose connections, corroded terminals, or failing bypass diodes inside the box.
Treat this as Priority 1 – it’s one of the highest fire-risk signatures in PV thermal inspection.
String-Level / Array-Pattern Anomalies
Repeating patterns like consistently hotter string ends, one side of a tracker row running warmer, or alternating hot/cool rows.
Causes range from PID at string ends to tracker misalignment to combiner or inverter faults.
The risk is systematic underperformance affecting many modules at once.
BOS Heating
Cables, connectors, and combiner boxes with localized hotspots. Even at moderate ΔT, overheating here is closely tied to arcing risk.
This is where you get your “acted on quickly vs. ignored” outcomes with the starkest difference in consequences.
Root Causes of Thermal Anomalies
Most thermal anomalies in solar arrays trace back to five root cause categories:
1. Shading & Soiling
Blocked light forces affected cells into reverse bias, dissipating current as heat instead of producing it.
2. Installation & Handling Damage
Physical damage during transport or installation creates electrical weaknesses that surface thermally, sometimes months after commissioning.
3. Manufacturing & Material Defects
Latent issues alter local current flow and heat dissipation – and they worsen gradually over years, not overnight.
4. Electrical Connection & Wiring Faults
Loose or corroded connectors, poor terminations, and failing diodes show up as sharp, localized hot points – a very hot single spot rather than a diffuse warm area.
5. System-Level Design & Grounding Issues
Poor grounding schemes and specific inverter operating modes drive PID. The anomalies appear systemic rather than random.
How These Causes Differ By System Type:
It’s also worth distinguishing reversible from structural issues:
Impact: Yield, Lifespan, Safety
Thermal anomaly detection isn’t about chasing perfect data, but about protecting revenue and equipment.
Yield Impact Is Immediate & Measurable
Silicon modules lose roughly 0.5–0.8% efficiency per °C above nominal operating temperature. A serious hotspot can cut an affected module’s output by up to 10–15%.
Aggregate that across a high-defect fleet and you’re looking at degradation rates of ~2.6%/year versus ~1.2%/year in low-defect plants – a gap that compounds every year the anomalies go unaddressed.
Equipment Lifespan Takes The Slower Hit
Sustained localized overheating accelerates encapsulant browning, solder fatigue, interconnect corrosion, and backsheet damage.
Modules with persistent hotspots reach replacement thresholds earlier, pulling forward capital costs that weren’t in anyone’s repowering budget.
Safety Risk Is Concentrated But Real
The absolute probability of a PV fire is still low. But most documented PV fires trace back to overheating components (hotspots, failing connectors, junction boxes) that were either missed or monitored and not acted on.
Regular thermal inspection is now specifically recommended as a fire-risk mitigation measure by major O&M guidelines, not just a yield optimization tool.
Detection Methods & Technologies
There are three primary approaches to thermal anomaly detection, each with a different sweet spot:
Manual / Handheld Thermography
A technician with a thermal camera – best for small-to-medium systems, commissioning checks, and targeted follow-up on faults identified via drone or SCADA.
Drone-Based Thermal Inspection
UAVs carrying radiometric IR and RGB cameras flying automated grids – the standard approach for utility-scale farms and large industrial rooftops.
Fixed Sensors & Continuous Monitoring
Permanently installed back-of-module sensors, reference modules, and SCADA-integrated monitoring – good for trending and catching systemic issues as they emerge.
In Practice, Most Serious O&M Programs Combine All Three
Drones for periodic full-site coverage, fixed sensors for continuous early warning between flights, and handheld thermography for ground-truth verification of specific faults.
What The Data Quality Requirements Look Like:
Get the conditions wrong (scanning at low irradiance, in strong wind, at sunrise with long shadows), and you’ll generate a dataset full of artifacts and missed anomalies.
How to Interpret Thermal Data Accurately
Having thermal images is one thing.
Reading them correctly is another.
Real Anomaly vs False Positive
A real anomaly stays hotter than adjacent comparable components once irradiance and wind stabilize. It matches a known fault pattern and isn’t explained by visible dirt or shadow in the RGB frame.
A false positive moves with the sun angle (reflection from glass), disappears after cleaning (soiling), or changes dramatically when you adjust the emissivity setting.
Common Misinterpretations That Cost Operators Time:
Correlating Thermal With Electrical Data Adds The Most Diagnostic Value
Thermal tells you where and how hot; electrical tells you how much it’s costing and often narrows the fault type.
Cross-check anomaly clusters against string or inverter current data:
Thermal Anomaly Detection Software: How Averroes Automates Detection at Scale
At large sites, the bottleneck isn’t capturing thermal data, but processing it.
A single utility-scale inspection flight can generate tens of thousands of images. Manual review at that volume is slow, inconsistent, and expensive.
Averroes sits in the AI layer on top of your existing drone fleet and cameras – no new hardware required. The platform ingests your drone thermal and RGB imagery and runs purpose-built AI models to detect, classify, and report defects at scale.
What That Looks Like In Practice:
New or rare defects are flagged, reviewed, and folded back into the model so future inspections pick them up automatically
What’s Hiding In Your Thermal Inspection Data?
Get from raw drone imagery to actionable defect priorities.
Operational Response & Remediation
Detection without a response workflow is just data collection. The standard post-detection pattern is: detect → validate → classify → fix or monitor.
Three-Tier Triage:
Typical Fixes By Issue Type:
Repair vs Replace Decision…
Comes down to ΔT severity, measured yield impact, module age, and warranty status.
The Risk Of Inaction Compounds
Persistent hotspots accelerate degradation rates, void warranty coverage when documented defects are knowingly ignored, and in BOS components, increase the probability of insulation breakdown and DC arcing.
Frequently Asked Questions
What is the difference between active and passive thermography for solar panels?
Active thermography applies an external heat source to stimulate the panel before imaging, while passive thermography captures heat the panel generates naturally under normal operating conditions. For solar PV inspection, passive thermography is standard – the sun does the work, and you inspect under live load. Active thermography is used in controlled lab or manufacturing settings, not field O&M.
Can thermal anomaly detection be used at night?
Thermal anomaly detection on solar panels requires the array to be under load, which means daytime operation with sufficient irradiance. Night-time thermography can detect residual heat in BOS components like inverters, combiner boxes, and cabling – but it won’t reveal cell-level or module-level defects. For meaningful PV inspection, irradiance of ≥600–700 W/m² is the baseline requirement.
How accurate is drone thermal inspection compared to ground-based thermal inspection?
Drone thermal inspection matches or exceeds ground-based accuracy for array-wide defect detection when flown at correct altitude, GSD, and under suitable conditions. Ground-based handheld inspection has an edge in close-range resolution and immediate RGB correlation for individual fault verification. In practice, drone inspection finds the anomalies; handheld inspection confirms and classifies them.
What thermal camera specifications do I need for solar panel inspection?
For reliable solar panel thermal anomaly detection, the minimum is a 320×240 IR sensor with thermal sensitivity below 0.08°C NETD – though 640×480 is preferred for higher altitudes or detailed cell-level classification. Ground sample distance should hit at least 5 thermal pixels per cell, which at standard cell sizes means flying at roughly 24m AGL or below with a 13mm lens.
Conclusion
Thermal anomaly detection is, at its core, a prioritization problem.
You have a site generating thousands of data points per inspection flight, and somewhere in that data are the junction boxes on their way to arcing, the PID patterns quietly compounding across string ends, and the soiling-driven hotspots that disappear after one cleaning cycle.
The difference between operators who manage this well and those who don’t usually comes down to whether they have a system (for detection, interpretation, triage, and response) or whether they’re reacting to alarms after the damage is already done.
The tools exist to run that system properly. If you want to see what AI-powered detection looks like applied to your inspection data specifically – your module types, your fault patterns, your scale – Averroes is worth an hour of your time. Book your free demo now.