The public discourse around electric two-wheeler battery fires in India tends to focus on the event itself: the temperature spike, the venting, the flame. What is less often discussed is the period before the event, when the cell is sending signals that something has changed from normal operation. Those pre-event signals are detectable with standard BMS instrumentation in many cases, and understanding them is the basis for early warning systems that can interrupt a charge or discharge before thermal runaway becomes irreversible.
This is not a claim that thermal runaway is always preventable. There are failure modes (mechanical damage, severe internal short circuit, extreme overcharge) that can progress too quickly for any software intervention to matter. But a large fraction of thermal runaway events in field batteries result from slow-developing degradation: cells that have accumulated internal damage over many cycles, that have been cycling at elevated temperatures for extended periods, or that have experienced gradual electrolyte decomposition that has lowered the onset temperature for exothermic reactions. These events develop over a longer timescale and present measurable precursor signals before the point of no return.
The thermal runaway initiation sequence
Understanding what precursor signals to look for requires understanding the sequence of events that leads to thermal runaway in a lithium-ion cell. The initiation sequence varies somewhat by failure mode (overcharge, external heat, internal short, lithium plating), but the general progression for a cell degrading under normal use conditions follows a pattern.
As a cell accumulates SEI growth and lithium plating over many charge cycles, the anode's ability to accept lithium during charge is gradually compromised. The SEI layer thickens and becomes more resistive. Lithium deposits on the graphite surface reduce active surface area. The internal resistance of the anode-side interface increases. Under the same charge protocol, the cell requires a higher overpotential to maintain the charge current, and the local temperature at the anode-electrolyte interface rises more than in a fresh cell at the same C-rate and surface temperature.
If the cell continues to be charged aggressively under these degraded conditions, the elevated anode temperature begins to accelerate the very reactions causing the degradation: faster SEI growth, faster plating. This is a slow positive feedback that over time drives the cell toward a condition where the internal exothermic reactions can outpace the cell's heat rejection capacity. At some point, the cell's internal temperature exceeds the threshold for decomposition of the electrolyte or the cathode material (for NMC cells, cathode decomposition begins around 180 to 200 degrees Celsius under adiabatic conditions), and thermal runaway initiates.
Pre-event temperature signals
The most direct precursor signal is anomalous temperature behavior during charge. Specifically: the temperature rise per unit charge delivered should be roughly predictable from a thermal model of the cell at its current state of health. A cell that is heating faster than expected per unit charge, at a given C-rate and ambient temperature, is generating more heat internally than a healthy cell would. This can indicate elevated internal resistance (from SEI growth or plating), electrolyte decomposition, or early-stage separator damage.
The diagnostic signature is a temperature rate (dT/dt) that is higher than the model predicts for the current charge state. A BMS that tracks a simple thermal model of the cell and flags when the measured temperature rate exceeds the model prediction by more than a threshold (for example, 1.5 degrees Celsius per minute above predicted at a given C-rate and SoC) is detecting a change in the cell's heat generation behavior. This does not always mean imminent thermal runaway, but it means the cell is behaving differently than a healthy cell and warrants a reduction in charge rate or a diagnostic check.
A more specific precursor is temperature behavior after charge termination. In a healthy cell, temperature after the end of charge drops relatively quickly as the resistive heat generation stops and the cell dissipates to ambient. In a cell with significant ongoing exothermic reactions (from electrolyte decomposition products reacting, or from ongoing SEI growth driven by a reactive lithium surface), the temperature may continue to rise briefly after current is removed, or it may drop more slowly than expected. A post-charge temperature that does not follow the expected cooling curve is a signal that the cell has a heat source that is not current-dependent.
Impedance-based precursors
Electrochemical impedance changes preceding thermal runaway have been documented in controlled laboratory studies, though translating these findings to field-deployed BMS is a challenge. The primary impedance signature is a gradual decrease in the real part of the impedance at low frequencies, sometimes accompanied by a change in the shape of the charge transfer semicircle in the Nyquist plot.
The physical interpretation: as the SEI layer becomes thicker and more mechanically unstable from repeated temperature cycling, it begins to develop micro-cracks and localized thin spots. These thin spots lower the effective SEI resistance at those locations, producing a net decrease in the low-frequency impedance even as the average SEI thickness is growing. The impedance decrease is a sign of SEI non-uniformity, which is associated with localized plating and localized heat generation.
For a field BMS without full EIS capability, the practical equivalent is tracking the high-frequency resistance (from 1 kHz impedance measurement or from the voltage response to a current step at charge start). A sudden decrease in high-frequency resistance, particularly one that is not explained by temperature increase, can indicate SEI cracking or internal short circuit onset. The challenge is that high-frequency resistance also decreases with increasing temperature, so the signature requires temperature-corrected tracking to be interpretable.
Voltage behavior as a precursor
Voltage-based precursors are the most accessible to standard BMS firmware because they require only the voltage measurement that is already present for SoC estimation and overvoltage protection. The relevant signatures are more subtle than simple overvoltage detection.
The first voltage precursor is a shift in the charge curve shape: the SoC at which the constant-current phase transitions to constant-voltage changes as the cell degrades. In a healthy cell, the CCCV transition occurs at a predictable voltage that correlates well with a specific SoC from the OCV curve. In a cell with significant plating, the CCCV transition may occur at a lower voltage or higher SoC than expected, because plated lithium on the anode surface contributes to the effective electrode potential in ways that shift the apparent equilibrium.
The second voltage precursor is voltage noise or micro-oscillations during rest periods at high SoC. These are small, rapid fluctuations in open-circuit voltage that can indicate micro-short circuit activity: lithium dendrites that have grown to the point of making intermittent electrical contact across the separator. Each contact event causes a tiny local discharge, which appears as a small negative voltage step followed by recovery. These events are detectable with voltage sampling rates of 1 Hz or better and require filtering to distinguish from measurement noise, but they are a documented precursor to internal short circuit progression.
How to act on these signals
The value of precursor signals depends entirely on what the BMS does when it detects them. A BMS that monitors for these patterns but has no response logic is not a safety system. The appropriate response structure is graduated: early precursors trigger a reduction in maximum charge current; stronger or multiple coincident precursors trigger a reduction in charge cutoff voltage (charging only to 90% or 80% SoC until the cell is inspected or replaced); the most severe signals trigger a charge termination and an alert to the user or fleet operator.
We should be clear that none of these signals are perfectly specific to thermal runaway precursors. A temperature rate anomaly can indicate a bad thermistor connection, not a degrading cell. A voltage noise signature can come from connector vibration, not dendrites. The diagnostic value comes from combinations of signals: temperature anomaly plus impedance trend plus voltage behavior shift together are a much stronger indicator than any single signal in isolation. The BMS logic should require at least two independent signal channels to agree before triggering any response that affects charge behavior, to minimize false positives that reduce user trust in the system.
The engineering work here is not glamorous. It is building and tuning a set of model-based anomaly detectors, validating them against accelerated aging cells, and calibrating the thresholds to balance sensitivity against false positive rate. But it is the difference between a BMS that detects that a cell has entered a precursor state while there is still time to act, and one that waits for the overvoltage or overtemperature condition that arrives only at the point of no return.