Reading the paper by Tye and Cundall in the June issue of the journal, I was started on a train of thought on the same lines.

I agree with the paper but feel that the argument has not been taken far enough. The main conclusion is:–

So long as a random failure is no more expensive and troublesome than a controlled replacement, it is always cheaper to wait until an item fails before replacing it. This is quite independent of the law of failure with time.