Inside every hard drive is a stack of platters spinning at five or seven thousand revolutions a minute and an arm carrying the read heads, which fly over the platters on a cushion of air thinner than a hair. When the drive starts, the heads read the firmware's service area on the platter and then the data. If a head has failed, or has been damaged in a fall, or is too weak to read the surface, the arm cannot find its place. The firmware moves it, waits, gives up, and parks it with a click. Then it tries again.
Every attempt is a failed head dragged across a platter it is meant to float above. On a drive dropped while running, the heads may already have touched the surface; on a drive that has simply worn out, a head that cannot read can still scratch. A single click is a warning. Twenty clicks over an evening of trying is a platter with a ring scored into it, and a ring on a platter is data nobody will read again.
The repair is mechanical and it is done once. The failed heads are replaced with a matched set from a donor of the same family, in clean air, and the drive is imaged head by head, weakest surface last, with the reads timed to what the platters can stand. On the drives where the firmware is what sent the heads astray, the service area is corrected from the terminal first. Either way the drive is never made to work again; it is made to read once.