I am trying to use ML for maintenance optimization aka predictive maintenance. One of the interesting challenge is, using tensor flow to predict possible Alarms in the next 72 hours by analysing the plant history of, say, past 3 yrs. Decision makers get excited about this possibility. They get hooked 
I haven’t picked up the tips and tricks of ML yet but certain correlation results i found in my experiment were quite shocking. Very good learning material is available on Youtube.
One thing for sure. ML and AI are the buzz word to make money with old data lying unused in the corner of a control room