The assumption of constant failure rate is not essential to compute reliabilities and combine them as described in General Philosophy of Prediction. However, constant failure rate is necessary for the reliability distribution to be exponential.

The exponential distribution has an advantage over other statistical distributions in that it is described fully by the single parameter λ, or often by its reciprocal, MTBF, and the majority of standard failure rate data is presented as this single parameter. The added advantage of the exponential distribution is that reliability can be simply computed for series configurations using the sum of the individual item failure rates. (See equation (2.9).)

In practice of course, the individual items that make up a system may not all have constant failure rates. Some may have reasonably constant failure rates (e.g., some electronic components), and others may have failure rates that increase to varying degrees with time (e.g., wear in mechanical items). Further, some items may not have failure rates in the conventional sense. For example, one-shot explosive devices are usually said to be time-independent.

Consider now a system that is made up of items whose failure rates increase with time. Assume that the system is maintained so that:

The failure rate for new items is lower than for older ones. Thus, the failure rate of the system depends on the ages of the individual items. When all items are new, the system failure rate is low and increases as items age; but, whenever an item is replaced, it reduces the system failure rate. Thus, over a period of time, the system failure rate tends to oscillate, rising and falling in the form of a damped harmonic, and approaches a constant failure rate as illustrated in Figure 2-8.

From the above considerations, it can be seen that an assumption of constant failure rate is a reasonable basis for predicting the reliability of a complex repairable system in the long run even though all the individual items within the system may not exhibit constant failure rates. Clearly, however, the resulting prediction will be more approximate for systems comprising mainly items with time-related failure rates (e.g., mechanical systems) than for those with more constant failure rates (e.g., electronic systems).

It must also be recognized that the assumption of constant failure rate does not necessarily represent the same failure mechanisms in different types of items. For example, electronic component failures generally occur as a sudden breakdown whereas failures of mechanical parts occur through time-related failure mechanisms such as creep, corrosion, fatigue, wear, etc.. It is often the case that such failures may be foreseen and hence avoided.