What Is Voltage Monitoring?
Last year a German customer reached out because three forklifts in their warehouse suddenly refused to work. Turned out the batteries were fine-the BMS had drifted about 40mV on voltage readings and decided the packs were overcharged. Locked everything down. Three machines dead for a full day. They bill by the hour over there.
Pretty typical story, actually. Voltage monitoring sounds like the most basic thing a BMS does. And yet this "basic function" causes more field problems than almost anything else.
What Voltage Tells You
Anyone working with batteries knows terminal voltage is basically a real-time mirror of cell state. SOC estimation depends on it. Overcharge and overdischarge protection depends on it. Cell consistency checks depend on it. IEC 62619:2022 puts it bluntly: voltage is more critical than current or temperature, and the BMS must cut charging before or as soon as cell voltage hits the safety ceiling.
Why so definitive? Because voltage changes first. Rising internal resistance, capacity fade, lithium plating risk-these show up in voltage behavior weeks or months before temperature anomalies. By the time you get a thermal alarm, things have already gone sideways.
Toyota's AHR10W hybrid battery system offers a useful reference. 168 cells at 1.2V each, series-connected to 201.6V total. The ECU treats every two modules as one monitoring unit, tracking 14 units total. This grouping lets the system pinpoint which unit contains a problem cell. Same logic applies to industrial packs-you can't give every single cell its own communication channel, but you need enough granularity to trace issues back to module level at minimum.
Four Detection Methods
Battery pack voltage detection has settled into four main approaches. Most buyers talking to a lithium battery pack supplier won't ask about this detail, but this detail sets the reliability floor for the entire BMS.
Relay-and-Capacitor
First is relay-and-capacitor isolation sampling. Simple concept: capacitor samples voltage from the cell, then you measure the capacitor. Problems are obvious too-slow sampling, poor accuracy, relay wear. Some early energy storage projects used this. Mostly obsolete now.
Floating-Ground
Second is floating-ground detection. A window comparator checks whether current ground potential works for A/D conversion; if not, D/A adjusts it. Tests fine in the lab. Falls apart in the field. Forklifts, AGVs-motor drive interference is too heavy. Ground potential won't stay put.
Common-Mode
Third is common-mode detection. All cells measured against one reference point, precision resistor dividers scale everything down, then you subtract to get individual cell voltages. Simple circuit. But resistor errors stack up. Works okay under 8S. Past that, accuracy gets questionable. This stacking problem can't be fully fixed with calibration-it's baked into the approach.
Differential-Mode
Fourth is differential-mode detection. Op-amps reject common-mode voltage and measure the differential across each cell directly. Accuracy beats the other three by a solid margin. Trade-off is circuit complexity and cost. Packs over 12S usually split into multiple detection modules, each handling a segment, communicating results over a bus. Most lithium battery system manufacturers doing industrial work have moved this direction.
The Leakage Current Trap
Easy to overlook this one.
Voltage detection circuits draw current from the cells. Tiny amounts-microamps to milliamps-but continuous. And here's the catch: in a series pack, cells closer to the negative terminal carry more leakage current. Take a 16S pack. Cell 1 at the positive end only sees leakage from its own detection circuit. Cell 16 at the negative end sees cumulative leakage from all 16 detection circuits, plus the BMS controller, plus whatever else references pack negative.
Over hundreds of cycles, negative-end cells discharge deeper and age faster. Consistency drifts. This isn't a cell quality problem. It's a system design problem.
Fixes exist: raise detection circuit input impedance to cut absolute leakage; add switches to disconnect sampling paths when not measuring; or just spec slightly higher capacity cells for negative-end positions and accept the asymmetry. If a LiFePO4 battery wholesale supplier can't answer questions about this, their packs will probably develop accelerated imbalance in the field. Good filter question when you're vetting vendors.
Accuracy and Why It Matters More Than Specs Suggest
IEC 62619 says voltage protection must act before or as cell voltage reaches the safety threshold. Sounds like there's margin. In practice, there isn't much.
Take LFP. Full charge around 3.65V. Danger zone starts around 3.70V. That's a 50mV window. If detection accuracy is ±30mV-common in cheap BMS designs-measurement uncertainty alone eats more than half your safety margin. BMS shows 3.65V, real voltage might already be 3.68V. We run ±5mV internally. Needs better ADCs, tighter voltage references, more careful PCB layout. Costs more. But compared to what goes wrong when accuracy slips, not worth mentioning.
Had a case: customer's pack ran two years, detection accuracy drifted from ±8mV at delivery to ±35mV. No fault codes-drift was gradual. One day during charging, a cell actually hit 3.72V while the system thought it was 3.65V and kept going. Got lucky that cell held up. Otherwise that's an incident report.
This kind of progressive failure hides well. Annual calibration checks are the minimum.
Balancing, Temperature, and Everything Else
Active or passive, balancing only works if you know which cells are high and which are low. Bad detection means blind balancing. Passive balancing bleeds through resistors. Typical 50mA balancing current takes hours to pull one cell down 0.1V. If voltage measurement errors flip the direction even occasionally, you're making imbalance worse while burning energy. When evaluating an industrial lithium battery vendor, ask about balancing strategy and detection accuracy together. "Has active balancing" means nothing if the underlying data is garbage.
Temperature compensation is another thing that rarely shows up in spec sheets. Same cell at 3.30V reads differently at 25°C versus 0°C. Internal resistance shifts, measurement circuit itself drifts with temperature. Without compensation, SOC estimates vary seasonally. Low-temperature charging gets trickier-cells at -5°C don't behave like cells at room temperature. BMS that only watches voltage without cross-referencing temperature might allow charging when it shouldn't, or block charging when it could. You have to ask the supplier's technical team directly about this stuff.
What Failure Looks Like
Sudden Fault
Voltage monitoring problems show up a few ways. Most obvious is sudden fault. One cell voltage reads 0V or full pack voltage. System throws a code and locks out. Annoying, but at least the system knows something broke.
Gradual Drift
Worse is gradual drift. Measured values slowly diverge from reality. No fault codes. Protection thresholds effectively disabled. Might run two years before anyone notices-usually because something bad happened.
Intermittent Contact
Then there's intermittent contact issues. Vibration loosens sense wire connectors. Voltage readings come and go. Sometimes reproducible, sometimes not. Nightmare to troubleshoot.
Voltage monitoring goes deep if you want to dig, but the core point is simple: this is the data foundation for everything else the BMS does. Foundation isn't solid, nothing built on top matters.
When you're sourcing battery packs, don't just look at capacity, cycle life, price. BMS detection accuracy, detection architecture, calibration protocols-these internals matter just as much. Questions welcome. We've stepped on enough landmines over the years to have opinions.
