What Is Real-time Monitoring?
BMS needs to know what's going on inside the pack. Voltage per cell, total current, temperatures at multiple points. This data comes in continuously, not sampled once in a while. That's real-time monitoring.
Voltage Measurement
This is where most of the cost goes in a BMS design.
Each cell or parallel group needs its own voltage sense line back to an analog front end IC. For a 16S pack that's manageable. For a 100S+ EV pack you're looking at multiple AFE chips daisy-chained together, isolated communication between them, and a wiring harness that takes real effort to route cleanly. Noise pickup is a constant fight. Long sense wires act as antennas. Twisted pairs help. Keeping high-current paths away from sense lines helps more.
The AFE accuracy spec matters a lot more than people expect. NMC cells have maybe 1.4V of usable voltage range. At 4.2V you're full. At 4.25V you're plating lithium and killing the cell. That's 50mV of margin. If your AFE has ±15mV accuracy you've already burned through more than half your budget on measurement error alone. This is why high-voltage packs from any decent lithium battery pack supplier use the expensive AFEs-6815, 6813, that tier. The cheap ones work fine for 4S power tool packs. Not for traction.
LFP is more forgiving on the top end but the voltage curve is so flat in the middle that SOC estimation gets hard. You need good accuracy for a different reason.
Current
Hall effect sensors or shunts. Halls are electrically isolated which simplifies the design. Shunts are more accurate but they sit in the current path, so the BMS sensing circuit needs to handle common-mode voltage equal to pack voltage. Not trivial on a 400V system.
Shunts also dissipate power. A 100µΩ shunt at 500A drops 50mV and burns 25W. That's heat you have to manage. And shunt resistance drifts with temperature, so the current reading drifts too unless you compensate. Cheap BMS designs don't. Then SOC walks off over the course of a day and nobody knows why.
Temperature
Thermistors are cheap. Placement is the hard part.
A pack might have 200 cells but only 6-8 temperature sensors. Where do they go? The cells in the geometric center run hottest because they're surrounded by other heat sources. Cells near the casing lose heat to ambient. Cells near busbars pick up conducted heat from high-current connections. A lithium battery system manufacturer doing this properly runs CFD or at least a simplified thermal model before committing to sensor locations. The rest put one thermistor per module and hope for the best.
The sensor has to touch the cell. Not float in air near the cell. Air temperature inside an enclosure tells you almost nothing about cell surface temperature. We've seen 8-10°C differences between air and cell surface in packs that looked fine on paper.
Thermal interface material matters too. A dry contact between thermistor and cell can has high thermal resistance. The reading lags reality. By the time the sensor shows 45°C the cell might already be at 52°C and climbing.
What the BMS Does With the Data
SOC estimation is the main thing. Coulomb counting integrates current over time. OCV lookup correlates resting voltage to state of charge. Kalman filters or similar combine the two. None of these work perfectly. Coulomb counting drifts because current measurement isn't perfect and you can never know the true starting point. OCV lookup needs the pack to rest for a while which doesn't happen in continuous operation. The Kalman filter helps but it's only as good as the cell model it's built on, and cells age.
SOH estimation tracks degradation. Capacity fade, resistance growth. This usually means periodically running a controlled charge or discharge and comparing to baseline. Some systems try to estimate it online from operational data. Results vary.
Protection logic is simpler. Voltage too high, stop charging. Too low, stop discharging. Current too high, disconnect. Temperature too high, derate or disconnect. These are just threshold comparisons. Getting the thresholds right takes some thought-too tight and you false-trip constantly, too loose and you let the cells get damaged.
Balancing
Cells drift apart over time. Passive balancing burns off excess charge through resistors, typically at 50-100mA. It's slow. During a 4-hour charge cycle passive balancing might move 200-400mAh. If your cells are 2000mAh out of balance that's not going to cut it.
Active balancing transfers charge between cells using inductors or capacitors. Much faster, more efficient, more expensive, more complicated. For industrial lithium battery solutions where packs cycle hard daily, active balancing makes sense. For a pack that sits at 50% SOC most of the time with occasional use, passive is fine.
Communication
CAN bus for vehicles. Modbus for stationary. Both work. Pick whatever the rest of the system uses.
Cloud connectivity sounds good on paper. In practice half the sites have garbage cellular signal and the installer didn't budget for an external antenna. Local data logging with periodic upload works better for most commercial lithium battery provider deployments than assuming constant connectivity.
Standards
ISO 6469 and UN ECE R100 for automotive. UL 9540 for stationary storage. OSHA and local fire codes for industrial charging areas. A lithium battery OEM partner should know which ones apply to your target market. The isolation monitoring requirements in the automotive standards trip people up more than anything else in volume production.
Real-time monitoring isn't optional. The question is how much accuracy and sophistication you need, and that depends on the cells, the application, and the consequences of getting it wrong.
