[garbageman]

While it would be neat to build batteries that handle manufacturing defects akin to a CPU with 4 cores but 3 useable, I suspect the additional complexity, wiring, and circuitry is limiting. Batteries today are almost (or actually?) a commodity. Detecting and simply replacing the entire battery is probably cheaper and easier at grid scale.

[londons_explore]

The wiring and complexity isn't very much - you simply need a contactor that can 'short' the offending set of cells. It only needs to close once, so can simply be made from a spring and meltable material. And you need ~100 of these per 400 volt battery.

Beyond that, all the complexity is in software. Software needs to monitor cell voltages and currents to detect a self-heating cell. Software then needs to stop balancing that cell up (ie. let it discharge). At the moment the cell voltage hits zero, software needs to close the 'short' across the cell, permanently taking it out of the circuit.

This design might occasionally prevent charging the entire battery for a few minutes during this process. Specifically, when a cell is midway through being taken out of circuit, it can only be discharged, and would be dangerous to recharge.

But a few minutes of downtime per year seems acceptable to me.