Accelerated rolling surface fatigue (L10) testing

The loss of material during a fatigue failure is in the form of pitting. The clearances shouldn't change as material loss is localized until it becomes severe.

I don't really understand your goal of pre-failing a bearing. Most vibration analysis starts with a baseline of a good machine and any deviation from the baseline indicates failure.

We make gear reducers and we are testing monitoring equipment. We have an endless supply of like-new machines to monitor but we don't have 20 years to wait for end-of-life.

Well if you have access to the bearings, then I guess you might be able to poke a tool between the rolling elements on either side of the cage to scratch either the inner or outer race (or rolling element themselves for that matter, fatigue spalling can occur on any of these locations).

As Tugboat mentioned, people in industrial plants monitor for rolling bearing defects using vibration quite well without needing any such experiments. Trending plays a role along with comparison to similar units. There's a lot of specizlized signal processing that can be applied towards detecting bearing defects. For example by trending CSI Peakvue Max Peak Wave you can be alerted to a defect long before it would affect the velocity overall. Further analyssis of peakvue and other spectra could help confirm the diagnosis and evaluate the severity. Sorry if I'm telling you what you already know. It's not clear what you're trying to accomplish.

=====================================
(2B)+(2B)' ?