Good article, you talk extensively about longer term degradation which is fine, but maybe you should consider incipient failure detection (diagnostics) for some items that can start to fail at any time in the operational life and deteriorate relatively rapidly over days or a few weeks. A bearing failure is a good example. Also you might consider machine state detection so you can seperate sensor time series data into steady state, where the machine is running steadily (Its easier to spot anomalies in this state) and transient state such as machine statrt up. Transient state diagnostics may be a bit more difficult, but necessary. An example is measuring the stiction and hysteresis of powered valve cycling, or monitoring the speed torque characteristic of an induction motor on start up to check the health of an induction motors squirrel cage (or rotor cage). Hope to hear more on this fascinating subject.