Most battery problems give some warning first. The question is which layer of protection is watching.
Battery fires get the headlines, but the research on why battery energy storage systems fail tells a more useful story: most failures are not cell defects, most happen early in a system's life, the failure rate has fallen sharply as the industry learned, and many problems show signs before they become dangerous. This guide covers what actually causes BESS failures, how a cell fault develops into thermal runaway, the warning signals at each stage, and how layers of protection, from analytics to fire suppression, work together.
EPRI, with TWAICE and the Pacific Northwest National Laboratory, analysed the incidents in EPRI's global database of grid-scale BESS failures in 2024. Of the incidents with enough information to assign a root cause, integration, assembly and construction was the most common cause; only 11% were attributed directly to the cells. The balance of system and the controls accounted for most failed components. EPRI concluded that its data challenges the widespread assumption that the cell is the primary cause of failure.
Two more findings matter. Where a system's age was known, 72% of failures happened during construction, commissioning or the first two years of operation. And the failure rate per unit of deployed storage has fallen dramatically: EPRI's database page reports a 99% drop from 2018 to 2025. The sample of classified incidents is small, so the shares are indicative, but the direction is clear: monitoring has to cover controls, cooling, wiring and integration, not only cell chemistry.
Of the 2018 to 2019 incidents in EPRI's database, most occurred in South Korea, and EPRI notes that a significant fraction occurred when the state of charge was above 90%, higher than recommended limits. Many controls failures were related to restricting cell state of charge, voltage and current. Operating discipline is part of safety.
Thermal runaway is a chain reaction: a cell heats itself faster than it can shed the heat. A 2020 review in Joule by researchers at Tsinghua University describes three critical temperatures. At the first, abnormal heat generation begins, as the SEI film starts to break down. At the second, the trigger point, the temperature rise becomes violent, often when the separator collapses and causes an internal short circuit. The third is the peak. The review notes that fast charging and lithium plating can lower the onset temperature considerably.
Before any of that, a developing fault often shows up as subtle changes in how a cell behaves. A 2020 review of fault diagnosis in the journal Algorithms lists the usual signs: voltage drop, state-of-charge drop, temperature rise, rising internal resistance and swelling. Poor electrical connections are a separate, common cause: a 2020 review in IEEE Industrial Electronics Magazine notes that a poor connection raises resistance and generates abnormal heat.
Research on early detection gives a rough order in which signals appear.
No single system catches everything, so a well-designed BESS has layers, each backing up the one before. The earlier a problem is caught, the cheaper and safer the fix.
For failures related to operation and controls, EPRI recommends battery monitoring and analytics to augment BMS operation, generating trends and predictive analyses to identify potential failures early. For integration failures, it recommends training and quality checks during commissioning.
What the signals look like
One cell's resistance or self-discharge creeps up, it runs slightly warmer than its neighbours, or its voltage sags more under load. Or the fault is outside the cells: a loose connection heating up, a cooling unit losing capacity, a control setting drifting. Every reading is still inside the BMS limits.
What happens
Analytics compares each cell with its siblings and with similar cells across the fleet, watches cooling and connections too, flags the trend, and the team plans an inspection or a module swap.
UL's Fire Safety Research Institute investigated a 2019 explosion at a battery system in Surprise, Arizona, in which cascading thermal runaway filled the container with flammable gas that later ignited when the door was opened, injuring four firefighters. Its recommendations included remotely accessible gas monitoring and explosion prevention. EPRI's analysis of another incident notes that insulation alarms were not properly escalated to the operator before a fire. Both point to the same lesson: signals are only useful if someone, or something, acts on them.
Standards have moved on. NFPA 855 sets minimum installation requirements for stationary energy storage, and UL 9540A tests how thermal runaway propagates at cell, module, unit and installation level. Indian tenders reference UL 9540 or the IEC 62933-5 series at system level, and the Central Electricity Authority has published draft BESS safety rules that include cell-level BMS monitoring, gas detection and fire suppression requirements.
Analytics reduces risk; it does not remove it. In a 2023 Nature Communications study of fault detection on 347 electric vehicles, the best model cut expected accident and inspection costs substantially, but at its cost-optimal setting it still caught only about half of the faulty vehicles. Detection methods for internal short circuits remain an active research problem, with false alarms a real challenge.
That is the case for layers. Analytics catches many problems early and turns them into planned maintenance; the BMS, gas detection and fire protection catch what analytics misses.
Ingro Brain is the analytics layer: it compares every cell and module with its neighbours and with batteries of the same chemistry across every site Ingro runs, and watches cooling, connections and controls as well as cells. Drift that is still inside the BMS limits becomes an alert and a planned inspection for the Ingro team, not an emergency.
Ingro EMS supports the layers below it: it keeps the battery inside its state-of-charge and temperature limits, takes a faulted string out of service while the rest of the site keeps running, and records every alarm and action so nothing goes unescalated.
Battery health analytics and AI forecasting: state of health, early warnings before failure, and dispatch tuned to make the battery last.
4 questions
In EPRI's analysis of grid-scale incidents with a known root cause, integration, assembly and construction issues were the most common cause, and only 11% were attributed directly to cells. Balance-of-system components and controls, along with operation outside recommended limits, account for much of the rest.
Many can be caught early. Analytics that compares each cell with its peers and tracks trends can flag developing faults while readings are still inside the BMS limits. No method catches every fault, so analytics works alongside the BMS, gas detection and fire protection.
Early signs include a cell drifting from its neighbours in resistance, self-discharge, temperature or voltage, then impedance changes, then gas release when the cell vents. Research shows impedance and gas sensors can warn before temperature or smoke detectors respond.
Yes. EPRI reports that the global grid-scale BESS failure rate per unit of deployed storage dropped by 99% from 2018 to 2025, as designs, standards and practices improved.
Ingro Brain watches every cell and Ingro EMS runs the battery with its health in mind, on new projects and on batteries already in service.