@corlin Can confirm. I see one impedance factor for now is the high cost of setting up machine learning systems to work on the 20% (Pareto) of cases they _might_ be useful. Usual analysis works fine on 80+% of issues. Sensor anomalies/motor health blips/weird stuff are rare; machine learning approaches need data so are weak on rare problems.
Someone'll do it anyway and chip away on the cost and accumulate more data.