I think of RATE OF FAILURE. Could be other parameters to determine. You should "register" the time of functioning versus the speed of functioning, if I may say so.
This is one ideea.
Than, calculate the failure probability using Gumbel or Weibull distribution models. The data should be processed...