Essentially, reliability studies provide predictions. They predict the future behavior of a device or system, based on past information and experience. Since predictions cannot be made with certainty, they are inherently probabilistic.
This is equally true for most engineering design techniques which also involve predictions of future performance. However, very rarely are probabilistic approaches used in these methods - they are mostly based on deterministic techniques. This also applies to the older methods used in reliability studies. While more complex and difficult, probabilistic thinking is gradually establishing itself in many areas of engineering (Endrenyi, 2000).
In this approach, system design and operating policies are based on pre-selected tests: failure criteria are defined so that certain combinations of system and load conditions must not represent immediate system breakdown or even excessive component stress. To make sure that these criteria are met, “worst-case conditions” are analyzed and the calculated stresses and strengths for the case are set apart by a “safety factor”.
Variability in input data is ignored (data provide spectra, not fixed numbers).
Selection of “worst-case” conditions is arbitrary: important conditions may be omitted, unlikely conditions included.
The assumption of no failure risk in designs satisfying traditional criteria is misleading; in fact, the approach provides no idea how safe the design actually is.
The effort to stay on the safe side often results in over design.