In 2018, a parked Tesla Model S caught fire in a Shanghai parking garage. No crash. No external ignition source. The battery pack went into thermal runaway on its own, and the fire burned for hours. Firefighters pumped thousands of litres of water into the vehicle before it stopped reigniting.
Incidents like that one, and there have been others across multiple manufacturers, tend to get reported as battery failures. They’re not, exactly. The cells didn’t just decide to catch fire. Something upstream went wrong: a reading missed, a threshold crossed without a response, a cascade that moved faster than the system could intervene.
Understanding what a battery management system actually does is the starting point for understanding why that chain happens, and how modern systems are designed to break it.
What the BMS Is Actually Managing
Take a 75 kWh battery pack apart and you won’t find one battery. You’ll find thousands of individual cells wired in series and parallel combinations. Each cell is slightly different from its neighbours. Different internal resistance, different capacity, different thermal behaviour under load. The pack number on the spec sheet is the sum of all of them working together.
Left unmanaged, those differences compound. A cell with slightly higher internal resistance generates more heat under the same current load. It ages faster. Its capacity drifts further from its neighbours. In a series string, the weakest cell limits the whole string’s usable range. In a parallel group, current distribution becomes uneven. Fifty cycles in, the difference is negligible. Five hundred cycles in, you have a genuinely weak cell sitting inside a string that treats it like it’s still normal.
The battery management system watches all of it: cell voltages one by one, temperatures across the pack, current in both directions. It’s also running estimation algorithms in the background, calculating state of charge and state of health on every measurement cycle. Call it a monitor and you’re underselling it. It’s making control decisions constantly.
What Happens When a Cell Gets Hot
Normal lithium-ion cells operate safely between roughly 15°C and 45°C. Push above that range and the chemistry starts misbehaving. Electrolyte decomposition accelerates. The SEI layer, the solid electrolyte interphase, is a thin protective film on the anode. It breaks down under excess heat. Internal resistance climbs. Heat generation increases. Which raises temperature further.
That feedback loop is thermal runaway. Once it starts in earnest, it’s self-sustaining. The cell doesn’t need external heat input anymore because it’s generating its own. Cell temperatures in runaway can exceed 800°C. The gases released are flammable. In a tightly packed module, the heat propagates to adjacent cells. One cell becomes a module. A module becomes a pack.
Temperature creeping up on one cell? Charge current drops. Pack temperature rising broadly? The cooling loop, including pump, chiller, and coolant circuit, ramps up its output. A cell voltage that sags faster than the model predicts under load suggests elevated resistance, possibly lithium plating on the anode. The BMS flags it, pulls back its discharge contribution, and writes the event to the diagnostic log. None of this requires driver input.
Cell Balancing: The Slow Problem That Becomes a Fast One
Thermal events get the headlines. Cell imbalance is the quieter problem that sets them up.
Active balancing moves charge between cells using small DC-DC converters, where higher cells feed lower ones during operation. Passive balancing takes the simpler route: burn the excess off as heat through a resistor. Active wins on efficiency. Passive wins on cost and simplicity. Neither is free because passive adds heat to a pack that’s already working to stay within temperature limits, and that’s its own problem.
Either way, the goal is keeping cells within a tight voltage window, typically a few millivolts. A pack where all cells are balanced ages uniformly. A pack with persistent imbalance develops weak points. Under high discharge rates such as hard acceleration, those weak cells see disproportionate stress. Over time, they become the cells most likely to trigger a thermal event.
State of Charge Is an Estimate, Not a Reading
One of the less obvious facts about battery management is that state of charge, the percentage figure on the dashboard, is never directly measured. There’s no fuel gauge in a lithium cell. Voltage correlates loosely with charge state, but the relationship shifts with temperature, age, and discharge rate.
BMS algorithms estimate state of charge using a combination of voltage, current integration via coulomb counting, and model-based correction filters like extended Kalman filters. The accuracy of that estimate determines whether the system charges to a safe ceiling, discharges to a safe floor, and correctly predicts remaining range. Get the estimate wrong and the pack either gets underused or pushed past its safe limits.
According to a review published in the journal Energies, state of charge estimation accuracy remains one of the most active research areas in BMS development, directly affecting both battery longevity and safety margins in real-world EV operation.
Why BMS Design Is the Real Safety Frontier
Cell chemistry gets most of the research attention. Solid-state cells, silicon anodes, sodium-ion chemistry: the research pipeline is full. Every few months there’s a new announcement about a breakthrough cell technology that will make today’s lithium-ion packs obsolete. Maybe. But none of that changes the fact that the cars on the road right now are running on existing cells, and those cells are only as safe as the software and hardware watching over them.
A well-designed battery management system doesn’t prevent thermal runaway by making cells that don’t fail. It prevents it by catching the early signals. A cell running 3°C hotter than its neighbours. A voltage that drops 20 mV faster than the model predicts. Acting on those signals before the feedback loop starts is not a passive safety feature. It’s continuous, real-time engineering running silently in the background every time the car moves.
