What to Measure, and How Often: Battery Monitoring That Earns Its Keep

A badly monitored battery can be worse than an unmonitored one. Bad telemetry produces numbers that look like knowledge, and operators make decisions on them. The pattern is familiar across the industry: a pack reporting a comfortable state of charge that cannot carry its load for five minutes, or a temperature alarm wired to a sensor sitting on the enclosure wall rather than on a cell. Good monitoring is not about collecting more data. It is about collecting the few measurements that genuinely predict failure, at a rate fast enough to catch it, and presenting them so that somebody acts.

Start from the failure you are trying to catch

Before choosing a monitoring architecture, write down how your application actually fails and how that failure announces itself. A standby string in a telecom cabinet typically degrades through slow capacity fade and one cell drifting to high internal resistance — a process that takes months and stays invisible until the mains drop. A cycling storage system or a mobile medical cart fails differently: thermal excursion under load, resistance building at a terminal, imbalance widening with every cycle. The first case needs accurate periodic per-cell voltage and resistance measurement. The second needs continuous current and temperature at a far higher sample rate. Specifying a generic monitoring system without making that distinction is how sites end up paying for channels nobody ever reads.

The four measurements that carry the most information

Per-cell or per-block voltage is the backbone. Pack voltage alone conceals almost everything, because a weak cell and a strong cell in series average into a healthy-looking total. Temperature belongs at the cell, not on the cabinet wall — a sensor mounted on sheet metal reports the room. Current should be measured with a shunt or Hall sensor sized for the real peak rather than the nominal draw, because it is the input to every state estimate that follows. Finally, connection integrity: terminal and interconnect resistance, checked on a schedule, is one of the most common causes of a hot spot in an otherwise healthy pack, and it is the cheapest thing on this list to check.

State of charge is an estimate; state of health is a trend

Coulomb counting drifts. Open-circuit-voltage lookup is unreliable on chemistries with a flat discharge plateau — LiFePO4 sits close to 3.2–3.3 V per cell across most of its usable range, so a voltage-only state of charge on an LFP pack is little better than a guess through the middle of the curve. Practical systems blend both methods and re-anchor the estimate at the knees, where voltage moves quickly and tells you something. State of health is a different kind of quantity: nearly meaningless as a single reading, and valuable as a trend. Log capacity actually delivered on real discharges, internal resistance measured at a consistent temperature and state of charge, and the imbalance spread over time. Three carefully repeated measurements a year beat continuous data that is never compared against itself.

Getting the data out of the cabinet

The protocol matters less than committing to one your site already speaks. Modbus RTU and Modbus TCP remain the workhorses in industrial and storage installations. CAN bus is the standard where a pack has to talk to an inverter or a vehicle controller. SNMP traps are still the practical route into telecom and IT monitoring stacks. Dry-contact relays are the correct answer where a single alarm needs to reach a legacy panel with no software in the path. Whichever you choose, insist on documented registers and a published data map before purchase. A monitored pack whose data map is proprietary and undocumented is a pack you will not be able to integrate later, and that constraint tends to surface at the worst possible moment.

Alarms people will actually act on

Every alarm should have an owner, a threshold with a stated reason behind it, and a defined response. Thresholds copied straight from a datasheet's absolute maximum ratings produce alarms that fire only once it is too late to prevent anything. Thresholds set too tight produce noise, and noise gets muted — after which the system is decorative. A two-tier structure works well in practice: a warning band that adds an inspection to the next scheduled visit, and a trip band that acts immediately and automatically. Then review the alarm log quarterly and retune. An alarm that has fired forty times without anyone changing anything is telling you about your thresholds, not about your battery.

Monitoring is worth specifying at the same time as the pack itself, not bolted on afterwards. Sensor placement, shunt sizing and communication interfaces are all cheaper decisions on the drawing than they are in the field.

If you are specifying a monitored battery pack, a BMS or a storage installation and want a second engineering opinion on what to instrument, write to us at info@lmelectro.com. We reply within 24 hours.