Академический Документы
Профессиональный Документы
Культура Документы
“Appendix E”
Vibration: Fundamentals and Practice
Clarence W. de Silva
Boca Raton: CRC Press LLC, 2000
Appendix E
Reliability Considerations for
Multicomponent Units
In the practice of vibration (e.g., vibration monitoring, isolation, control, and testing), one depends
on the proper operation of complex and multicomponent equipment. Equipment that has several
components that are crucial to its operation, can have more than one mode of failure. Each failure
mode of the overall system will depend on some combination of failure of the components.
Component failure is governed by the laws of probability. Consider first some of the fundamentals
of probability theory that are useful in the reliability or failure analysis of multicomponent units.
E.1.2 UNRELIABILITY
The probability that the component will malfunction or fail during the time period t is called its
unreliability, or its probability of failure. It is denoted by F. Hence,
Because it is known as a certainty that the component will either survive or fail during the specified
time period t, one can write
The probability of survival of a component usually decreases with age. Consequently, the typical
R(t) is a monotonically decreasing function of t, as shown in Figure E.1. If it is known as a certainty
that the component is good in the beginning, then R(0) = 1. Because of manufacturing defects,
damage during shipping, etc., however, one usually has R(0) ≤ 1. For a satisfactory component,
R(t) should not drop appreciably during its design life Td. The drop is faster initially, however,
because of infant mortality (again due to manufacturing defects and the like), and later on, as the
component exceeds its design life because of old age (wear, fatigue, etc.).
It is clear from equation (E.3) that the unreliability curve is completely defined by the reliability
curve. As shown in Figure E.1, transforming one to the other is a simple matter of reversing the axis.
Example
Consider the rolling of a fair die. The set of total outcomes consists of six elements forming the space
S = {1, 2, 3, 4, 5, 6}
A = {Outcome is odd}
B = {Outcome is divisible by 3}
Then,
A = {1, 3, 5}
B = {3, 6}
Consequently,
A or B = {1, 3, 5, 6}
A and B = {3}
It follows that
℘( A) = 3 6 ; ℘( B) = 2 6 ; ℘( A or B) = 4 6 ; ℘( A and B) = 1 6
℘( A and B) = 0 (E.5)
℘( A and B) = ℘( A B)℘( B)
(E.6)
= ℘( B A)℘( A)
in which ℘(A/B) denotes the conditional probability that event A occurs, given the condition that
event B has occurred.
In the previous example of rolling a fair die, if it is known that event B has occurred, the
outcome must be either 3 or 6. Then, the probability that event A would occur is simply the
probability of picking 3 from the set {3, 6}. Hence, ℘(A/B) = 1/2. Similarly, ℘(B/A) = 1/3. It
should be noted that equation (E.6) holds for this example.
for independent events. Equation (E.8) is the product rule, which is applicable to independent events.
It should be emphasized that although independence implies that the product rule holds, the
converse is not necessarily true. In the example on rolling a fair die, ℘(A/B) = ℘(A) = 1/2. Suppose,
however, that it is not a fair die and that the probabilities of the outcomes {1, 2, 3, 4, 5, 6} are
{1/3, 1/6, 1/6, 0, 1/6, 1/6}. Then,
℘( A) = 1 3 + 1 6 + 1 6 = 2 3
whereas,
16
℘( A B) = =1 2
1 6 +1 6
This shows that A and B are not independent events in this sample.
Furthermore, ℘(B) = 1/6 and ℘(A and B) = 1/6. It is seen that Bayes’ theorem is satisfied by this
example.
dR(t )
r (t ) = (E.9)
dt
dF(t )
f (t ) = (E.10)
dt
where
In equation (E.10), f(t) is the probability-density function corresponding to the time to failure. It
follows that:
Also,
By substituting equations (E.11) through (E.13) into equation (E.6), one obtains
or
f (t ) f (t )
β(t ) = = (E.14)
R(t ) 1 − F(t )
Now suppose that there are N components. If they all have survived up to t, then, on the average,
Nβ(t)dt components will fail during the next dt. Consequently, Nβ(t) corresponds to the rate of
failure for the collection of components at time t. For a single component (N = 1), the rate of failure
is β(t). For obvious reasons, β(t) is sometimes termed conditional failure. Other names for this
function include intensity function and hazard function, but failure rate is the most common name.
In view of equation (E.10), one can write equation (E.14) as a first-order linear, ordinary
differential equation with variable parameters:
dF(t )
+ β(t ) F(t ) = β(t ) (E.15)
dt
F(0) = 0 (E.16)
t
0
∫
F(t ) = 1 − exp − β( τ)dτ
(E.17)
t
0
∫
R(t ) = exp − β( τ)dτ
(E.18)
It is observed from equation (E.18) that the reliability curve can be determined from the failure-
rate curve, and the reverse.
A typical failure-rate curve for an engineering component is shown in Figure (E.3). It has a
characteristic “bathtub” shape, which can be divided into three phases, as in the figure. These phases
might not be so distinct in a real situation. The initial burn-in period is characterized by a sharp drop
in the failure rate. Because of such reasons as poor workmanship, material defects, and poor handling
during transportation, a high degree of failure can occur during a short initial period of design life.
Following that, the failures typically will be due to random causes. The failure rate is approximately
constant in this region. Once the design life is exceeded (third phase), rapid failure can occur because
of wearout, fatigue, and other types of cumulative damage, and eventual collapse will result.
It is frequently assumed that the failure rate is constant during the design life of a component.
In this case, equation (E.18) gives the exponential reliability function:
This situation is represented in Figure E.4. This curve is not comparable to the general reliability
curve shown in Figure E.1. As a result, the constant failure rate should not be used for relatively
large durations of time (i.e., for a large segment of the design life) unless it has been verified by
tests. For short durations, however, this approximation is normally used and it results in considerable
analytical simplicity.
R(t ) = R1 (t ) R2 (t ) K RN (t ) (E.20)
where Ri(t) is the reliability of the ith component. If there is no component redundancy, which is
assumed in equation (E.20), none of the components should fail (i.e., Ri(t) ≠ 0 for i = 1, 2, …, N)
for the object to operate properly (i.e., R(t) ≠ 0). This follows from equation (E.20).
In vibration testing, a primary objective is to maximize the risk of component failure when
subjected to the test environment (so that the probability of failure is less in the actual in-service
environment). One way of achieving this is by maximizing the test-strength-measure function
given by
TS = ∑ F (T )Φ
i =1
i i (E.21)
in which Fi(T) is the probability of failure (unreliability) of the ith component for the test duration T,
and Φi is a dynamic-response measure at the location of the ith component. The parameters of
optimization may be the input direction and the frequency of excitation for a given input intensity.
Regarding component redundancy, consider the simple situation of ri identical subcomponents
connected in parallel (ri th-order redundancy) to form the ith component. The component failure
requires the failure of all ri subcomponents. The failure of one subcomponent is assumed to be
independent of the failure state of other subcomponents. Then, the unreliability of the ith component
can be expressed as
Fi = ( F0 i )
ri
(E.22)
in which F0i is the unreliability of each subcomponent in the ith component. This simple model
for redundancy might not be valid in some situations.
There are two basic types of redundancy: active redundancy and standby redundancy. In active
redundancy, all redundant elements are permanently connected and active during the operation of
the equipment. In standby redundancy, only one of the components in a redundant group is active
during equipment operation. If that component fails, an identical second component will be auto-
matically connected.
For standby redundancy, some form of switching mechanism is needed, which means that the
reliability of the switching mechanism itself must be accounted for. Component aging is relatively
less, however, and the failure of components within the redundant group is mutually independent.
In active redundancy, there is no need for a switching mechanism; but the failure of one component
in the redundant group can overload the rest, thereby increasing their probability of failure (unreli-
ability). Consequently, component failure within the redundant group is not mutually independent
in this case. Also, component aging is relatively high because the components are continuously active.