BMS Data Already on Site Can Flag Battery Faults Weeks Early

Claudius Jehle, co-founder and managing director of Volytica Diagnostics, argued on 5 August 2026 that diagnostic software can detect subtle changes in cell voltages, internal resistance, balancing times and thermal behaviour days or weeks before conventional alarms trip.

He cites an 80MWh North American storage project where developing faults progressed over several weeks without triggering either the battery management system or the SCADA alarms, and were later shown to have been identifiable days in advance. In one case abnormal thermal patterns in a module were detectable several days before the condition became critical.

The argument he makes against the industry's usual response is the interesting part. The default reaction to a detection failure is to add sensors, and his position is that this is misplaced: a battery management system already gathers thousands of measurements per site. The analysis works by comparing relationships — between parameters, between individual cells, and against the site's own historical operating patterns — rather than by watching any single value against a threshold. A cell whose internal resistance is rising relative to its neighbours is informative long before its absolute temperature is alarming.

That makes this a data plumbing problem, which is squarely integrator territory. The obstacle at most sites is not sensing but access. Per-cell voltage, resistance and balancing data exists on the BMS CAN or Modbus bus, and site SCADA typically logs only aggregate string-level values because that is what the operator interface needed. Getting the cell-level data off the bus at usable resolution, timestamping it consistently, and shipping it to an edge or cloud analytics layer is work most storage sites have simply not done.

The value case is straightforward and unusually easy to state: a fault caught weeks early instead of a container lost. Whether the analytics deliver on that at any given site is a separate question, but the data they would need is already being produced and mostly discarded.