The Future of Machine Learning

The battery life extension problem or the more, that is very interesting why because it was of high value. But you can take the same problem to the buildings or maybe even a train. Yeah, though we need to make use and inject domain know-how into the ML model somehow. So, what are the ways to add explicit or at least somehow explicit knowledge in the world of machine learning? And I'm thinking in the direction of maybe simulation or graphs or even synthetic data or something like that.