Usually, functional safety requires a deterministic system model: Any input given creates a defined, reproducible output. Error handling, test coverage and verification methods make sure that the entire system can be analyzed. Neural networks, on the other hand, make their decisions with approximated evaluations that are based on optimized training data.
In his article on elektroniknet, Eranyan Ravanan discusses how this black-box nature of Edge AI can be integrated into functional safety systems using a deterministic guardian layer. He gives an outlook on how the advantages of Edge AI can be leveraged in embedded development while ensuring safe monitoring during defined fault states.
The article is available on the elektroniknet website in German language.