00:00Ever wondered how electric vehicles know how much charge is left in their battery?
00:04The state of charge which tells you how charged the car is relies on complicated calculations
00:10of current or voltage, sometimes adjusted for temperature. If it's too low, the vehicle might
00:17shut down unexpectedly. If it's too high, the battery could overcharge, leading to overheating
00:23and in rare cases, fire. This risk is why the EV battery industry hasn't actively adopted
00:30artificial intelligence yet, despite the potential of increased efficiency.
00:34You have problems with the proving causality and liability with an AI component because
00:41it's black box. You can't look inside and say what went wrong. So if an AI component fails,
00:50it's usually something to do with either the hardware or the training data.
00:55Experts say integrating AI into the battery estimation system could make EVs go farther
01:01and last longer. If you can use the sensors, one temperature sensor, one voltage sensor and one
01:08current sensor directly into an AI, that AI is more competent to estimate the charge and will adapt
01:19more quickly to new releases of battery packs and stuff like that, because the old way is very
01:25cumbersome and slow. So there are benefits for AI.
01:29But it's also a task where mistakes can literally ignite into serious trouble.
01:34Skogloon's team intentionally fed the AI faulty data to see how it acted and saw that the output was
01:41way off. It's important to do this kind of test because we proved that it was really
01:50not that robust. As a solution, the team suggests a system that keeps AI in check and shuts it down
01:57when it acts out, which they call a safety cage. This isn't the only way to bring AI into EV batteries,
02:04but researchers say this is one of the most realistic solutions.
Comments