Reliability: Analysis and Measure
isn't amenable to treatment with the ubiquitous normal distribution, a fact that should
catch the attention of any practicing engineer with only that bullet in his gun.
Reliability data has other distinguishing features. The data are usually censored,
which means the exact failure times are not known so the observations can only provide
bounds on the actual failure times. Inferences and predictions usually require
extrapolations, making engineering and physics-based modeling an important adjunct to
statistical methods. Whereas many *statistical* problems focus on parameter
estimation (e.g.: mean, standard deviation), these are not of primary interest to
engineers who need specific measures of product reliability (e.g.: failure probabilities,
life distribution quintiles, failure rates).
Meeker and Escobar, Statistical Methods for
Reliability Data, Wiley, 1998. This is an excellent book. If
your bookseller is out of stock, you can get a copy quickly from amazon.com.