Smart Array firmware polls every physical drive for its SMART or SAS log pages and compares them with HPE's thresholds. A drive that crosses one is flagged Predictive failure, and HPE's guidance is to replace it under warranty; on a logical drive that is otherwise Optimal, that is sound, because the rebuild onto the replacement reads survivors that are healthy. On a logical drive that is already degraded, the same rebuild reads a predictive-failure survivor from end to end, and the error that crossed the threshold is the error that stops the rebuild.
Logical drive failed is the end of that road: the controller has lost more members than the RAID level tolerates, and the volume is offline. The physical drives it dropped are usually still readable on a bench prepared to wait, and the one dropped first still holds every stripe written before it dropped. Every member is imaged and the array reassembled in software from the images, with the controller's metadata on the drives giving the layout; that is the array site's work, done on this bench.
The clocks are the HPE-specific case. Bulletin a00092491, from November 2019, named SAS SSDs on firmware before HPD8 that fail at 32,768 power-on hours; a00097382, from March 2020, named another line failing at 40,000 hours before HPD7. HPE's own words were that after the failure neither the SSD nor the data can be recovered, and that drives put into service together will fail nearly simultaneously. Some of those drives answer at controller level on the bench and some do not; the first look is honest about which.