Drone Thermal + AI: How We Caught a BESS Hotspot 6 Weeks Early
Six weeks. That is how much warning we got. And six weeks was exactly enough to make the difference between a controlled maintenance intervention and a much more expensive emergency response.
I am telling this carefully because it shows something important about what the combination of AI agents and autonomous drones actually catches. It also shows why internal sensors alone leave a gap that is bigger than most people think.
The setup
The system was a 50 MWh LFP BESS co-located with a 35 MW solar farm in southern Spain. It was four years into operation. The most recent annual SoH test showed 89 percent. The BMS dashboard looked normal. HVAC parameters looked normal. No alarms were active.
From a standard monitoring perspective, everything seemed fine.
ClearSpot’s AI Solar Performance Monitoring for 20MW+ Portfolios had been running on the site for about eight months. During that time, it had built baseline internal resistance profiles for all 12 racks across three containers.clearspot
In week three of July, during peak summer heat, the agent detected something. Not an alarm. Not a threshold violation. A trajectory change.
Rack 7 in Container 2 showed a 0.8 percent increase in internal resistance over the previous three weeks. In isolation, that was small and still within normal variation. But the agent also flagged that Container 2’s HVAC unit had shown a 4 percent reduction in cooling capacity over the previous six weeks.
Each trend by itself was unremarkable. Together, they pointed to thermal stress in a system already operating near its thermal ceiling.
The agent did what it was designed to do: it triggered an autonomous drone thermal survey of Container 2’s exterior.
What the drone found
The drone flew the next morning at 06:40, before the day’s peak heat. That mattered because ambient thermal noise was low.
The thermal camera, with 640×512 resolution and 0.05°C sensitivity, showed a hotspot on the eastern face of Container 2, about 1.2 meters from the northern corner. The delta T versus the adjacent container wall was +8.7°C.
That is not minoris not angle-related solar gain. That is heat being generated internally and conducted through the container skin from the area directly associated with Rack 7.
The internal BMS sensor for Rack 7 showed 28°C, which still looked normal. But the exterior wall at that location showed 47°C on a 34°C ambient morning. The problem was not visible to the internal sensor because the heat was concentrated in a section of the rack where airflow was not reaching properly.
Internal sensors showed no alarm. The drone thermal scan showed a clear anomaly.
ClearSpot’s Technology page describes this closed-loop model: SCADA and internal data identify the trend, drones confirm the physical condition, and agents decide the next action.clearspot
The intervention
The Battery Health Surveillance Agent classified the condition as a potential blocked cooling channel with elevated thermal-stress risk on Rack 7. It recommended a physical inspection of the cooling channel and HVAC servicing for Container 2.
That work was completed within four days.
The maintenance team found a partially blocked cooling channel caused by debris accumulation from a degraded gasket. The blockage covered about 35 percent of the channel cross-section. It was not enough to trigger an existing alarm, but it was enough to reduce effective cooling on Rack 7 in summer heat.
The HVAC inspection confirmed the cooling-capacity reduction the agent had already flagged. A worn fan blade assembly was causing roughly 12 percent reduced airflow in the upper section of the container.
Both issues were repaired in a single visit. Total cost: about €2,800.
The technician who completed the work made the point better than I can: if it had run through August, the site could have been looking at accelerated capacity fade on Rack 7, and possibly cell-level thermal stress.
The math of what did not happen
If the hotspot had continued through August, the likely outcome would have been a combination of avoidable loss and avoidable risk.
A six-week period of elevated thermal stress on Rack 7 could have led to roughly 1.5 to 2.5 percent accelerated SoH loss on that rack and early refurbishment costs of about €85,000.
In a worse case, the asset could have been curtailed pending investigation. That would have meant at least two weeks of revenue loss at roughly €18,000 per week, plus curtailment of the co-located solar farm and possible insurance consequences. The combined downside could have exceeded €78,000 before any long-term portfolio impact.
What we actually spent was €2,800.
Why the closed loop matters
This is why the combination of agents and drones matters so much. The agent alone had a statistical signal, but no visual confirmation. The drone alone had no reason to fly. Together, they formed a closed loop that turned a weak signal into an actionable maintenance decision.
That is the value of ClearSpot’s AI Solar Performance Monitoring for 20MW+ Portfolios and Technology stack working together. It is also why the broader Solar Asset Management Software layer matters: it gives operators a portfolio-level view of issues before they become failures.clearspot+2
What this changes
The BESS safety conversation is usually focused on fire suppression, thermal runaway, and internal sensor coverage. Those are all important. But they are also reactive.
What caught this issue was something different: continuous AI-powered internal signal analysis combined with external drone thermal imaging. Not a hard threshold. Not a BMS alarm. A pattern that only became obvious when two small signals were connected.
That is why external thermal monitoring should not be treated as optional. It is part of how you see what internal sensors miss.
FAQs
What is the main benefit of drone thermal monitoring for BESS?
It can detect exterior hotspots, HVAC issues, and container-level thermal anomalies that internal sensors may miss.
Why did the internal BMS not detect the issue?
The problem was localized, and airflow was not reaching the affected area well enough for the internal sensor to register a critical alarm.
How early was the issue caught?
The combination of agent analysis and drone inspection provided about six weeks of lead time.
Why is a closed loop important?
Because the agent creates the hypothesis and the drone confirms the physical condition. That prevents unnecessary dispatch and improves accuracy.
Which ClearSpot pages are most relevant here?
The most relevant pages are AI Solar Performance Monitoring for 20MW+ Portfolios, Technology, and Solar Asset Management Software.clearspot+2