Just spam solves. Only when you get to PLL, if you don’t recognize the case from just how you are holding the cube when you get there, stop, don’t do anymore turning, and sit there and try to deduce what case you have to a small group of possible cases. Then as you do this you’ll be able to notice more specific things and narrow it down faster and more accurately. That what I did. You technically never have to actively learn two sided PLL recognition if you just do solves and stop and try to recognize a case from the natural angle you’re at when you get to PLL. Plus once you are comfortable with how your OLL affects cp by just experience with it, it makes prediction your PLL way easier since you can already know what kind of PLL group it can fall into and then recognize faster.