When I open the service lid of an electric vehicle battery pack and watch the cell voltages scroll down the diagnostic screen, the first thing I look for is not the absolute value of any single cell but the spread between the highest and the lowest one. That spread, a few tens of millivolts in a healthy pack and several hundred millivolts in a troubled one, tells me more about the remaining life of the electric vehicle battery pack than almost any other single number. In the pages that follow I want to set out, from my own practice, why that spread appears, how it quietly destroys usable capacity and driving range, what passive and active balancing can and cannot do about it, and how I set up a balancing session so that it actually finishes instead of running for a week and cooking a resistor board.
The central idea is simple. An electric vehicle battery pack is a series string of many individual cells that are supposed to behave as one. They never quite do. Manufacturing tolerances, ageing, thermal gradients, contact resistance and control errors all push the cells apart. Once they are apart, the weakest cell dictates when charging stops and when discharging stops, and the strong cells never get to give what they could give. Balancing is the process of pulling them back together, either by burning off energy from the strong ones or by moving energy from the strong ones into the weak ones.
1. Where Cell Inconsistency Comes From
I divide the root causes of imbalance in an electric vehicle battery pack into four families: birth defects, wear, environment, and electronics. Birth defects are the initial capacity, internal resistance and self-discharge spreads that leave the factory with the cells. Wear is the fact that no two cells age at exactly the same rate. Environment is temperature and mechanical stress. Electronics is everything the battery management system does, or fails to do.
Self-discharge is the quietest of all these mechanisms. Two cells with different self-discharge rates drift apart simply by sitting still, and the drift is strongly temperature dependent:
$$k_{sd} = A \exp\left(-\frac{E_a}{R T}\right)$$
where the pre-exponential factor and activation energy are cell-specific. A pack parked for months in a hot climate with a working but slow BMS can accumulate a surprisingly large spread with no driving at all. This is exactly the situation in which a passive balancing resistor, which only works while the pack is being charged, never gets enough time to catch up.
| Family | Mechanism | Typical observed effect | What I do about it |
|---|---|---|---|
| Manufacturing | Capacity, internal resistance and electrode coating tolerances | A few millivolts to tens of millivolts from new | Accept it, document the baseline spread, do not over-balance a new pack |
| Ageing | Different capacity fade and resistance growth per cell | Spread grows slowly and then accelerates | Track spread over time, replace outliers before they drag the pack |
| Self-discharge | Cell-specific leakage, strongly temperature dependent | Drift during storage, worst in hot climates | Store cool at mid state of charge, measure open-circuit voltage after rest |
| Thermal gradient | Coolant path and air flow give unequal cell temperatures | Cells near the hot end age faster and drift low | Verify thermal management, log temperature distribution |
| Connection resistance | Busbar and weld resistance differences | Voltage appears to sag under load on one cell only | Four-wire resistance check, re-torque or replace hardware |
| Battery management system | Balancing disabled or failed, sampling error, poor control strategy | Spread grows despite charging cycles | Functional test of the balancing path, calibration check of the sampling chain |
| Cell fault | Internal short, soft short, electrolyte dry-out | One cell falls persistently and cannot be lifted | Replace the cell or module, never keep balancing a damaged cell |
| Mixed population | New and old cells, or different chemistries and suppliers, in one string | Large, permanent, physically unavoidable offset | Match by capacity and resistance before assembly, not after |
Sampling error deserves a special mention because it is often mistaken for a real imbalance. The measurement chain of an electric vehicle battery pack includes a multiplexer, a reference, and an analog-to-digital converter whose resolution is set by the reference voltage and the bit depth:
$$LSB = \frac{V_{ref}}{2^n}$$
An eight-bit converter on a five-volt reference gives a step of nearly twenty millivolts. Instrumentation error is commonly specified as a percentage of reading plus a fixed offset:
$$\varepsilon = \pm\left(0.5\% \cdot V_{read} + 2\ \text{mV}\right)$$
When the numbers I read from the vehicle say that two neighbouring cells differ by twelve millivolts, I do not trust the vehicle. I cross-check with an independent measurement before I touch anything.
2. What a Voltage Difference Really Costs
The reason a millivolt spread matters is that it represents real energy. For a given cell, the energy gap implied by a voltage difference is the product of the difference and the capacity:
$$\Delta E = \Delta V \times Q$$
This single line is the most useful formula I carry in my head. A two-hundred ampere-hour cell with thirty millivolts of deviation is holding six watt-hours of mismatch. The same cell with three hundred millivolts of deviation is holding sixty watt-hours of mismatch. Table 2 does the arithmetic across the capacity range I meet most often.
| Cell capacity | 10 mV | 30 mV | 50 mV | 100 mV | 300 mV |
|---|---|---|---|---|---|
| 50 Ah | 0.5 | 1.5 | 2.5 | 5 | 15 |
| 100 Ah | 1.0 | 3.0 | 5.0 | 10 | 30 |
| 150 Ah | 1.5 | 4.5 | 7.5 | 15 | 45 |
| 200 Ah | 2.0 | 6.0 | 10.0 | 20 | 60 |
| 300 Ah | 3.0 | 9.0 | 15.0 | 30 | 90 |
On its own, that energy is small. Its significance is that it sets the boundary of the usable window. In a series string the pack can only charge until the highest cell reaches the upper cutoff, and can only discharge until the lowest cell reaches the lower cutoff. The usable capacity of the electric vehicle battery pack is therefore not the nominal capacity of a cell; it is the nominal capacity minus the offset that the weakest cell imposes:
$$Q_{usable} = Q_{nom} – \Delta Q_{offset} – Q_{reserve}$$
The consequences accumulate in a predictable order. First the electric vehicle battery pack loses usable capacity. Then it loses cycle life, because the strong cells are repeatedly pushed to the top of their voltage window while the weak cells are repeatedly pushed to the bottom. Then internal resistance rises, heat generation rises with the square of current, and the pack enters a spiral:
$$P_{loss} = I^2 R$$
| Stage | Observable symptom | Underlying mechanism | Reversibility |
|---|---|---|---|
| Early | Spread of tens of millivolts, no customer complaint | Normal dispersion and mild self-discharge | Fully reversible by balancing |
| Developing | Reduced range, charging terminates early | Highest cell hits cutoff before the others fill | Reversible by balancing plus strategy correction |
| Advanced | Power derating, discharge terminates early | Lowest cell hits cutoff while others still hold charge | Partially reversible, some capacity already lost |
| Severe | Continuous balancing, heat, resistor failure | One cell can never be lifted, balancing path runs permanently | Not reversible without cell replacement |
| Failure | Damaged balancing board, damaged sampling chip, cell venting risk | Thermal runaway of the discharge resistor or a shorted cell | Requires component replacement |
3. The Water-Cup Model
The model I use when explaining this to colleagues is a row of cups holding different amounts of water. Charging is pouring water in until the fullest cup reaches the brim line, at which point the pouring stops. Every cup that is not yet full stays not full. Discharging is pouring water out until the emptiest cup reaches the bottom line, at which point the pouring stops. Every cup that still holds water keeps holding it. The row of cups never empties and never fills, because the extremes set the limits, and the middle cups spend their whole life cycling in a narrow band that is neither full nor empty.
Translated to an electric vehicle battery pack, this is why a vehicle whose cells differ by a hundred millivolts cannot deliver the range printed on the specification sheet, even though every individual cell is perfectly healthy. It is also why inserting a few new cells into an aged electric vehicle battery pack guarantees an offset. A new cell and an old cell have different capacities, different internal resistances and different open-circuit voltage curves, so their voltages cannot be equal at every state of charge. Balancing can only equalise them at the instant of balancing.
4. Translating Millivolts into Kilometres
The relationship between the voltage spread and the loss of driving range comes from the width of the working voltage window. Take a nickel-manganese-cobalt electric vehicle battery pack whose cells operate between 3.5 V and 3.9 V. The window is 0.4 V wide. A spread of 0.1 V therefore occupies a quarter of the window. If the vehicle covers 500 km on a good day, the simple proportional estimate is:
$$\Delta D = D_{full} \times \frac{\Delta V}{V_{max} – V_{min}} = 500 \times \frac{0.100}{0.400} = 125\ \text{km}$$
In practice I treat this as an upper bound rather than a precise prediction, because the relationship between open-circuit voltage and state of charge is not linear, and because the vehicle does not stop exactly at the cutoff. It is nevertheless the right order of magnitude, and it is the number that convinces a customer that a hundred millivolts is not a rounding error.
| Voltage spread | Share of working window | Loss on a 500 km pack | Loss on a 400 km pack | Loss on a 300 km pack |
|---|---|---|---|---|
| 10 mV | 2.5% | 12.5 km | 10 km | 7.5 km |
| 30 mV | 7.5% | 37.5 km | 30 km | 22.5 km |
| 50 mV | 12.5% | 62.5 km | 50 km | 37.5 km |
| 100 mV | 25.0% | 125 km | 100 km | 75 km |
| 200 mV | 50.0% | 250 km | 200 km | 150 km |
For lithium iron phosphate the linear estimate is far less reliable, because the flat plateau means that a large change in stored charge produces a very small change in voltage:
$$\frac{dV_{OCV}}{dSoC} \approx 0 \quad \text{on the plateau}$$
For that chemistry I always convert voltage to charge rather than reasoning directly from millivolts, using coulomb counting over a known interval:
$$SoC(t) = SoC(t_0) – \frac{1}{Q_{nom}} \int_{t_0}^{t} \eta\, I(\tau)\, d\tau$$
5. Passive Balancing
Passive balancing is the dissipative approach. A switch and a resistor are placed across every cell in the electric vehicle battery pack, and whenever a cell is above the target the switch closes and the resistor turns the excess into heat. It is cheap, it is everywhere, and it is perfectly adequate for the slow drift caused by self-discharge. It is also slow, and it is the source of most of the burned balancing boards I have replaced.
The behaviour is fully described by three relationships. The discharge current is set by Ohm’s law, the dissipated power by the product of voltage and current, and the time to remove a given amount of energy by the ratio between them:
$$I = \frac{U}{R}$$
$$P = U \times I = \frac{U^2}{R}$$
$$t_{bal} = \frac{E_{diss}}{P} = \frac{\Delta V \cdot Q \cdot R}{U^2}$$
Substituting the numbers for a typical resistor board makes the problem obvious. With a thirty-three ohm resistor across a cell sitting at 3.3 V, the current is only about 0.1 A and the power only about 0.33 W. Removing the six watt-hours represented by thirty millivolts of deviation on a two-hundred ampere-hour cell takes:
$$t_{bal} = \frac{0.030 \times 200 \times 33}{3.3^2} \approx 18.2\ \text{h}$$
Six watt-hours takes most of a day. Sixty watt-hours, the three-hundred-millivolt case, takes roughly 182 hours, which is more than seven and a half days of continuous balancing.
| Resistor | Current | Power | Time for 6 Wh | Time for 60 Wh |
|---|---|---|---|---|
| 10 ohm | 0.330 A | 1.089 W | 5.5 h | 55.1 h |
| 22 ohm | 0.150 A | 0.495 W | 12.1 h | 121.2 h |
| 33 ohm | 0.100 A | 0.330 W | 18.2 h | 181.8 h |
| 47 ohm | 0.070 A | 0.232 W | 25.9 h | 258.9 h |
| 100 ohm | 0.033 A | 0.109 W | 55.1 h | 550.5 h |
The thermal consequence is what actually ends the life of the hardware. Inside a real battery management system I see a dense row of balancing resistors next to the sampling and management chips. When one cell can never be brought into line, its resistor never stops working. The junction temperature climbs, the solder softens, and eventually the resistor cracks or opens. When it opens, the neighbouring components pay the price, and I have seen the management chip beside a failed balancing resistor destroyed along with it.
There is a second, subtler penalty that customers notice before hardware failure: the energy burned during balancing is energy removed from the electric vehicle battery pack. Every watt-hour dissipated in a balancing resistor is a watt-hour that will not move the vehicle.
6. Active Balancing
Active balancing moves energy rather than destroying it. An inductor, a capacitor, a transformer or a bidirectional converter is used to shuttle charge from the higher cells to the lower cells in the electric vehicle battery pack until the voltages converge. Because the energy is conserved rather than discarded, the efficiency is high, the hardware can be compact, and the result is usually better than what passive balancing achieves on a stubborn pack.
The trade-off is complexity and cost. Energy can generally only be moved between a limited number of nodes at a time, so the balancing current for a whole pack is modest, and the control electronics and magnetics add cost and failure modes of their own. I describe the transfer efficiency as:
$$\eta = \frac{E_{delivered}}{E_{extracted}}$$
and the time to move useful energy as the ratio of that energy to the transfer power scaled by efficiency:
$$t_{active} = \frac{E_{diss}}{\eta \cdot P_{transfer}}$$
| Topology | Energy path | Strengths | Limitations |
|---|---|---|---|
| Switched capacitor | Cell to neighbouring cell | Simple, no magnetics, moderate cost | Slow for long strings, large inrush currents, limited voltage difference handling |
| Buck-boost inductor | Adjacent cells, bidirectional | Good efficiency, well understood | Neighbour-only transfer means many hops to cross a long string |
| Flyback or transformer | Whole pack to any cell | Direct transfer, good for long strings | Transformer cost, leakage, isolation design effort |
| Bidirectional DC to DC | Pack bus to string nodes | High current capability, fast | Cost and control complexity, more failure modes |
| Attribute | Passive | Active |
|---|---|---|
| Energy handling | Dissipated as heat | Transferred between cells |
| Typical balancing current | 0.03 A to 0.33 A | 0.5 A to 5 A |
| Efficiency of the balancing act | Zero by definition | Typically high, losses only in conversion |
| Cost | Low | High |
| Best suited to | Self-discharge drift on healthy packs | Long strings, large offsets, repeated balancing demands |
| Main failure mode | Overheated or cracked resistors, damaged neighbouring chips | Converter or transformer failure, control faults |
| Heat generated inside the electric vehicle battery pack | Significant for large spreads | Low |
7. How the Chemistry Changes the Strategy
The two chemistries I work with most behave so differently that the balancing strategy has to change with them. For lithium iron phosphate the nominal cell voltage is 3.2 V, the working range is roughly 3.0 V to 3.4 V, the overcharge guard is 3.65 V and the over-discharge guard is 2.2 V. The discharge curve falls quickly from the top, then settles onto a long plateau near 3.2 V, then drops steeply at the end. Charging is quick and quiet between thirty and eighty percent state of charge because the voltage barely moves.
For nickel-manganese-cobalt the nominal cell voltage is 3.7 V, the working range is roughly 3.5 V to 3.9 V, the overcharge guard is 4.2 V and the over-discharge guard is 2.7 V. The discharge curve is more sloped overall. It descends slowly at first, holds a plateau around 3.6 V, then falls away faster. Charging is quick between twenty and ninety percent state of charge, and the correspondence between voltage and remaining capacity is much clearer.
| Parameter | Lithium iron phosphate | Nickel manganese cobalt |
|---|---|---|
| Nominal cell voltage | 3.2 V | 3.7 V |
| Working voltage range | 3.0 V to 3.4 V | 3.5 V to 3.9 V |
| Overcharge guard | 3.65 V | 4.2 V |
| Over-discharge guard | 2.2 V | 2.7 V |
| Plateau behaviour | Very flat, voltage insensitive to charge | Sloped, voltage informative |
| Region of rapid charging | 30% to 80% state of charge | 20% to 90% state of charge |
| Best balancing point | Full charge, at the top of the curve | Around nominal voltage |
| Reason | The plateau carries no information, so balancing there is meaningless | Balancing at full charge accelerates lithium plating |
The practical rule I follow is blunt. For lithium iron phosphate, balancing in the middle of the plateau is a waste of time, because two cells at the same voltage can hold very different amounts of charge. I balance the entire electric vehicle battery pack, not a single module, either at full charge or at the cutoff, and I push every string to the top of the curve, where the voltage finally becomes steep enough to make the result meaningful:
$$SoC\ \text{error} \le 3\% \quad \text{when balancing at full charge on the steep region}$$
For nickel manganese cobalt I balance near the nominal 3.7 V and avoid balancing at the 4.2 V ceiling, because holding cells at full charge accelerates the growth of lithium dendrites and shortens cycle life.
8. When Balancing Is Actually Required
Not every spread needs intervention. I use a small set of triggers, and I insist on a rest period before I measure anything, because polarisation and surface charge make a freshly cycled pack look far worse than it is.
| Condition | Nickel manganese cobalt | Lithium iron phosphate |
|---|---|---|
| Static spread after 2 to 4 hours of rest | Greater than 30 mV | Greater than 50 mV |
| Dynamic spread during charge or discharge | Greater than 50 mV | Greater than 80 mV |
| Capacity below expectation at full charge | Investigate | Investigate |
| One cell reaching cutoff before the others | Investigate | Investigate |
| Cycle count above 300 | Schedule preventive balancing | Schedule preventive balancing |
| Cell or module replacement performed | Mandatory balancing | Mandatory balancing |
Two statistical measures help me decide whether a pack is drifting as a whole or whether a single cell is misbehaving:
$$\Delta V = V_{max} – V_{min}$$
$$\sigma_V = \sqrt{\frac{1}{n}\sum_{i=1}^{n}\left(V_i – \bar{V}\right)^2}$$
If the range grows but the standard deviation stays flat, the whole pack is spreading. If the standard deviation grows while the range is stable, one or two outliers are running away, and those outliers are candidates for replacement rather than balancing.
9. The Diagnostic Workflow I Use
Measuring every cell with a handheld multimeter across a large electric vehicle battery pack is slow and error-prone, so I use a dedicated diagnostic application with a controller area network interface. The application runs on a laptop. One end of the interface connects to the laptop by universal serial bus. The other end has a high line, normally red, and a low line, normally black, which I connect to pins six and fourteen of the vehicle diagnostic connector respectively.
With the wiring verified I open the application, select the host mode, enter vehicle diagnostics, choose the model, confirm, and connect. From there I read the total pack voltage, the highest and lowest cell voltages, current, state of charge, insulation resistance, and the live voltage of every individual cell. Being able to see all cells at once is what turns a two-hour guessing exercise into a ten-minute diagnosis. The screen shows immediately whether the problem is a general drift or a single rogue cell.
My sequence is as follows. First, confirm the reported spread with an independent instrument. Second, look at the spread under load as well as at rest, because a cell with high internal resistance looks fine at rest and collapses under current. Third, examine the temperature distribution across the pack, since a hot region ages the cells inside it faster. Fourth, check the balancing history in the battery management system logs, because a pack that balances every single cycle is telling me something is structurally wrong. Fifth, decide whether the answer is balancing, repair, or replacement.
10. Screening Before Balancing
Balancing a pack whose cells differ in internal resistance is like levelling a floor with a broken spirit level. Before I connect a balancing instrument I measure the internal resistance and the capacity of every group. If the resistance spread exceeds five percent, balancing the voltage alone will not hold, and the offending cells must be replaced.
$$\frac{\left|R_i – \bar{R}\right|}{\bar{R}} \times 100\% \le 5\%$$
| Chemistry | Typical cell capacity | Expected cell resistance | Action threshold |
|---|---|---|---|
| Lithium iron phosphate | 20 Ah class | Below 10 milliohm | Spread above 5% or absolute value above 10 milliohm means replace |
| Nickel manganese cobalt | 50 Ah class | Below 8 milliohm | Spread above 5% or absolute value above 8 milliohm means replace |
The same logic applies to capacity. A cell that has lost a fifth of its capacity cannot be balanced into agreement with its neighbours for more than one cycle, because the moment the pack is cycled the difference reappears in the state of charge rather than in the voltage.
11. Setting Up a Balancing Session
For lithium iron phosphate I balance the whole electric vehicle battery pack at full charge or at the cutoff. I aim to bring every string to 3.65 V at the top of the steep region, and I accept a state of charge error of up to three percent because the curve is finally informative there.
For nickel manganese cobalt I balance near 3.7 V. If I am using a charger-balancer in a parallel charging mode, where all strings are charged toward the same target simultaneously, I set the target below the over-voltage guard and above the under-voltage guard with a deliberate margin.
| Parameter | Setting | Reason |
|---|---|---|
| Working mode | Charge balancing, all cells toward one target | Brings the whole string up together rather than chasing a moving reference |
| Line type and current limit | 5 A when the sense harness is unfused; 0.8 A when it is a printed board harness with fuses | Prevents overload of the sensing and balancing wiring |
| Chemistry selection | Nickel manganese cobalt | Selects the correct voltage profile and protection windows |
| Under-voltage protection | 2.5 V per cell | Minimum safe cell voltage for this chemistry |
| Over-voltage protection | 4.25 V per cell | Maximum safe cell voltage for this chemistry |
| Voltage cutoff | 4.18 V per cell, that is the guard minus 50 mV | Keeps the target 50 mV below the over-voltage guard in parallel charging mode |
The margin is not decorative. In parallel charging mode, the instrument stops when the string reaches the cutoff, and any overshoot from cabling resistance or measurement lag must still land below the guard. Fifty millivolts is the smallest margin I am comfortable with on this chemistry. For lithium iron phosphate the equivalent session targets 3.65 V per string with the guard set above it and the cutoff set just below it.
12. Balancing After a Cell Replacement
Replacing a cell inside an electric vehicle battery pack leaves two populations with different behaviour. The new cell is tight, low in resistance, and slightly higher in capacity. The old cells are looser and lower. Connecting a balancing instrument and waiting is the slow route.
The quick route I use is a pairing trick. When one string has been replaced, I take the new string and one un-replaced original string, discharge both to empty together, then charge the capacity released from the original string back into the two strings in parallel. Because both strings start from the same empty reference, the parallel charge equalises their state of charge directly rather than fighting voltage offsets at an arbitrary point on the curve. What would otherwise take days of resistor dissipation takes a fraction of the time.
13. A Worked Case
A vehicle arrives with a nickel manganese cobalt electric vehicle battery pack. The nominal cell voltage is 3.7 V. During the diagnostic session the highest string reads 3.700 V and the lowest reads 3.600 V, giving a spread of 0.100 V. The cells are otherwise healthy, and the internal resistance spread is within three percent, so balancing rather than replacement is the right answer.
| Quantity | Value | Source |
|---|---|---|
| Nominal cell voltage | 3.700 V | Chemistry specification |
| Highest cell voltage | 3.700 V | Diagnostic reading |
| Lowest cell voltage | 3.600 V | Diagnostic reading |
| Voltage spread | 0.100 V | Difference |
| Working window | 0.400 V | 3.9 V minus 3.5 V |
| Share of window occupied by the spread | 25% | 0.100 divided by 0.400 |
| Full-charge range | 500 km | Specification |
| Estimated range loss | 125 km | 500 multiplied by 25% |
| Internal resistance spread | 3% | Measured, within limit |
Before the session I verify the internal resistance and capacity of every group. Because the resistance spread is only three percent, well inside the five percent screening limit, balancing is worth performing. Because the chemistry is nickel manganese cobalt, I balance near nominal voltage and keep the target clear of the upper guard. After the session the spread falls to a few millivolts, and the customer recovers the majority of the missing range, subject to the genuine capacity loss that no amount of balancing can return.
Had the spread been three hundred millivolts, the same cells would have represented sixty watt-hours of mismatch on a two-hundred ampere-hour pack, and if the pack had been balanced passively with a thirty-three ohm resistor the process would have run for more than a week of continuous dissipation, generating heat the whole time. That is precisely the scenario in which a balancing resistor fails and takes its neighbouring chip with it.
14. Why Balancing Is Not a Permanent Cure
I never let a customer believe that a balancing session fixes the electric vehicle battery pack forever. Every cell changes with use, and the rates of change are not identical:
$$Q_{loss} = B \exp\left(-\frac{E_a}{R T}\right) t^{z}$$
Cells that live in the warm end of the pack lose capacity faster than cells in the cool end. Cells with slightly higher self-discharge drift lower every time the vehicle rests. If one string in the electric vehicle battery pack has an internal resistance or capacity that is dramatically different from the rest, it will reach a different state of charge after every drive and after every charge, and the battery management system will have to start balancing again as soon as the next cycle begins. The balancing resistor on that string works continuously, dissipating energy that the customer paid to store, and generating heat inside the pack.
The second reason balancing is not a cure is that not every balancing fault originates in the cells. I have chased balancing anomalies to failed resistors, to open sense lines, to a corrupted sampling channel, to incorrect thresholds left behind by a previous repair, and to control strategies that close the balancing switches at the wrong point on the voltage curve. A pack that balances constantly and never converges is more likely to have an electronics problem than a cell problem, and I always test the balancing path and the measurement chain before I condemn a cell.
| Symptom | First suspect | Check I perform |
|---|---|---|
| One cell never reaches the target | Cell capacity or resistance mismatch | Capacity test and resistance comparison against the string average |
| Balancing never terminates | Wrong threshold or wrong point on the voltage curve | Review the balancing strategy against the chemistry plateau |
| Balancing resistor very hot | Continuous dissipation on a cell that cannot be lifted | Confirm the cell condition, replace if necessary |
| Voltage readings noisy or inconsistent | Sampling chain fault | Calibration check, harness inspection, comparison with an independent meter |
| Pack balances then drifts within one cycle | Distribution or self-discharge mismatch | Standing open-circuit voltage test over several days |
| Spread grows only under load | Connection or internal resistance | Four-wire resistance measurement of busbars and welds |
15. Conclusions
The voltage spread in an electric vehicle battery pack is not a cosmetic number. It is the difference between the energy a pack holds and the energy a pack can actually deliver. The conversion is direct and worth committing to memory, because it turns a millivolt reading into watt-hours:
$$\Delta E = \Delta V \times Q$$
From there, the logic unfolds without surprises. Passive balancing is adequate for the slow drift of healthy cells, and it is the right default for its cost, but its currents of a tenth of an ampere and its powers of a third of a watt mean that large spreads take days and produce heat that eventually destroys hardware. Active balancing moves energy instead of burning it, balances faster and cooler, and costs more to build and to repair. The chemistry decides where balancing should happen: at the top of the curve for lithium iron phosphate, where the plateau finally gives way, and near nominal voltage for nickel manganese cobalt, away from the ceiling that accelerates ageing.
My working sequence is always the same. Screen the electric vehicle battery pack for internal resistance and capacity spread first, because balancing cannot repair a mismatched population. Verify the measurement chain, because an instrumentation error and a real imbalance look identical on the screen. Select the correct point on the voltage curve for the chemistry in front of me, set the protection and cutoff values with a deliberate margin, and only then start the session. Afterwards, confirm the spread has collapsed and explain to the owner that balancing equalises the cells at one instant and that time, temperature and use will pull them apart again. A well-managed electric vehicle battery pack is not one that never develops a spread. It is one whose spread is measured, understood, and corrected before it turns into lost range, lost life and heat.