Perhaps I’m way off (I haven’t had a lot of experience with machine learning), but mapping this sort of thing seems like something a neural network could learn to do extremely well. There may be hardware limitations or reasons to why this can’t be done, but in general optimising an engine’s performance looks like a problem with very measureable inputs and outputs, which would seem to lend well to a machine learning problem.
From a naive perspective, if engine manufacturers aren’t already investigating machine learning to optimise performance, it seems like something worth doing.