Basic Reliability Prediction Theory
7.1╇ Why Predict RAMS?
Reliability prediction (i.e. modeling) is the process of calculating the anticipated system
RAMS from assumed component failure rates. It provides a quantitative measure of how
close a proposed design comes to meeting the design objectives and allows comparisons to
be made between different design proposals. It has already been emphasized that reliability
prediction is an imprecise calculation, but it is nevertheless a valuable exercise for the following
reasons:
• It provides an early indication of a system’s potential to meet the design reliability
requirements.
• It enables an assessment of life-cycle costs to be carried out.
• It enables one to establish which components, or areas, in a design contribute to the major
portion of the unreliability.
• It enables trade-offs to be made as, for example, between reliability, maintainability and
proof-test intervals in achieving a given availability.
• Its use is increasingly called for in invitations to tender, contracts and in safety-integrity
standards.
It must be stressed that prediction is a design tool and not a precise measure of reliability. The
main value of a prediction is in showing the relative reliabilities of modules so that allocations
can
be made. Whatever the accuracy of the exercise, if one module is shown to have double the
MTBF of another then, when calculating values for modules in order to achieve the desired
system MTBF, the values allocated to the modules should be in the same ratio. Prediction also
permits a reliability comparison between different design solutions. Again, the comparison is
likely to be more accurate than the absolute values. The accuracy of the actual predicted value
will depend on:
1. Relevance of the failure rate data and the chosen environmental multiplication factors.
2. Accuracy of the mathematical model.
3. The absence of gross over-stressing in operation.
4. Tolerance of the design to component parametric drift.
Chapter 7
The greater the number of different component types involved, the more likely that individual
over- and under-estimates will cancel each other out.
It is important to state, at this point, that quantified reliability prediction (the subject of this
chapter and of Chapters 9 and 10) generally applies to random hardware failures only.
Systematic failures, which were introduced in Chapter 1 (and are dealt with in Chapter 17),
are NOT quantified by failure rates because they are not device specific and do not repeat
themselves once they have been designed out. They are therefore dealt with quite separately
by addressing the ‘rigor of life-cycle activities’.
7.2╇ Probability Theory
The following basic probability rules are sufficient for an understanding of the system
modeling involved in reliability prediction.
7.2.1╇ The Multiplication Rule
If two or more events can occur simultaneously, and their individual probabilities of occurring
are known, then the probability of simultaneous events is the product of the individual
probabilities provided, of course, that they are INDEPENDENT of each other. The shaded
area in Figure 7.1 represents the probability of events A and B occurring simultaneously.
Hence the probability of A and B occurring is:
pab = pa × pb
Generally
Pan = Pa × pb,…, × pn
7.2.2╇ The Addition Rule
It is also required to calculate the probability of either event A or event B or both occurring.
This is the area of the two circles in Figure 7.1. This probability is:
Figure 7.1
7.2.3╇ The Binomial Theorem
The above two rules are combined in the binomial theorem. Consider the following example
involving a pack of 52 playing cards. A card is removed at random, its suit noted, and then
replaced. A second card is then removed and its suit noted. The possible outcomes are:
two hearts
one heart and one other card
two other cards.
If p is the probability of drawing a heart then, from the multiplication rule, the outcomes of
the experiment can be calculated as follows:
probability of two hearts p2
probability of one heart 2pq
probability of no hearts q2.
Similar reasoning for an experiment involving three cards will yield:
probability of three hearts p3
probability of two hearts 3p2q
probability of one heart 3pq2
probability of no hearts q3.
The above probabilities are the terms of the expressions (p + q)2 and (p + q)3. This leads to the
general statement that if p is the probability of some random event, and if q = 1 – p then the
probabilities of 0, 1, 2, 3,…, outcomes of that event in n trials are given by the terms of the
expansion:
(p + q)nâ•›whicha•›equals
pn,â•›np(n － 1)q,a•‰a•￣
n(na•￣－a•￣1)p(na•￣－a•￣2)q2
a•￣a•￣___________a•￣
7.2.4╇ Bayes Theorem
The marginal probability of an event is its simple probability. Consider a box of seven
cubes and three spheres, in which case the marginal probability of drawing a cube is 0.7. To
introduce the concept of a conditional probability, assume that four of the cubes are black and
three white and that, of the spheres, two are black and one is white, as shown in Figure 7.2.
The probability of drawing a black article, given that it turns out to be a cube, is a conditional
probability of 4/7 and ignores the possibility of drawing a sphere. Similarly the probability
of drawing a black article, given that it turns out to be a sphere, is 2/3. On the other hand, the
probability of drawing a black sphere is a joint probability. It acknowledges the possibility of
drawing cubes and spheres and is therefore 2/10.
Comparing joint and conditional probabilities, the conditional probability of drawing a black
article given that it is a sphere is the joint probability of drawing a black sphere (2/10) divided
by the probability of drawing any sphere (3/10). The result is therefore:
pb/s = a•‰a•￣
_P_b_sa•￣a•￣ Ps
a•￣a•‰
given╛↜渀屮that: Pb/s is the conditional probability of drawing a black article given that
it is a
sphere; Ps is the simple or marginal probability of drawing a sphere; Pbs is the joint probability
of drawing an article which is both black and a sphere.
This is known as the Bayes theorem. It follows then that Pbs =
Pb/sPs╛↜渀屮or╛↜渀屮Ps/bPb⋅ Consider now
the probability of drawing a black sphere (Pbs)and the probability of drawing a white sphere
(Pws):
Ps = Pbs + Pws
Therefore
Ps
7.3╇ Reliability of Series Systems
Consider the two valves connected in series that were described in Figure 2.1 of Chapter 2.
One of the failure modes discussed was loss of supply, which occurs if either valve fails
closed. This situation, where any failure causes the system to fail, is known as series reliability.
This must not be confused with the series configuration of the valves shown in Figure 2.1.
It so happens that, for this loss of supply failure mode, the physical series and the reliability
series diagrams coincide. When we consider the over-pressure case in the next section it will
be seen that, although the valves are still in series, the reliability block diagram changes.
For loss of supply then, the reliability of the system is the probability that Valve A does not
fail and Valve B does not fail.