Applied R&M Manual for Defence Systems

Part D – Supporting Theory

CHAPTER 9

AVAILABILITY DEMONSTRATION PLANS

CONTENTS

1 INTRODUCTION 2

2 CONCEPTS AND TERMINOLOGY 2

3 STATISTICAL TEST PLANNING 4

4 DEMONSTRATION TEST PLANS 4

Issue 1
Chapter 9
Availability Demonstration Plans

1 INTRODUCTION
1.1 The purpose of an Availability Demonstration is to establish that Availability
performance of the equipment is better (or not) than a specified level, with some degree of
confidence. It is assumed in this Chapter that the demonstration is carried out on a sample of
a population. The sample is the equipment under test, which is representative of the total
population of equipment produced, or intended for production.

1.2 The test equipment is considered as being either in an up or down state. The test is
started with the equipment in the up state. After a period a failure occurs, causing the item to
enter the down state, and a new cycle begins. PtCCh24 gives guidance to what down time
should be considered relevant, depending on the Availability measure (i.e. Operational,
Intrinsic or Effective Availability) which is specified in the requirement.

2 CONCEPTS AND TERMINOLOGY

2.1 General
2.1.1 Availability demonstrations are hypothesis tests (see PtDCh7). The essential first step
is the definition of the hypothesis about the parameter to be tested.

2.1.2 There are three standard forms of test which can be used:
a) Fixed number of failures terminated
b) Fixed time terminated
c) Sequential.

2.2 Consumer’s Risk

2.2.1 In any test of a sample there is a risk of accepting ‘bad’ equipment. Whatever the
minimum standard the consumer sets, there is a finite probability that the sample will be such
that a population worse than this standard will be accepted. This risk is quantified by the
following definitions:
a) Minimum Acceptable Availability (A β ) is a value selected such that the risk
associated of accepting the equipment of the value is tolerable.
b) Consumer’s Risk ( ) is the probability of accepting the equipment with a true
Availability equal to the Minimum Acceptable Availability (A β ). Typically, has
a value of between 5% and 20%.

2.3 Producer’s Risk

2.3.1 The reverse of Section 2.2.1 is also possible, in that there is also a risk of rejecting
‘good’ equipment. To ensure that the producer has a chance of passing the demonstration, he
must design for a target Availability which is significantly greater than the Minimum

Page 2
 Applied R&M Manual for Defence Systems
Part D – Supporting Theory
Acceptable Availability. The producer’s target Availability (A α ) is dependent on the
Consumer’s and Producer’s Risk, and the duration of the test.

2.3.2 The Producer’s Risk ( i s the probability of rejecting the equipment with a true
Availability equal to the target Availability. In other words, the probability of rejecting
equipment with a true Availability greater than A α will be less than Typically, has a
value between 5% and 20%.

2.4 Discrimination Ratio

2.4.1 The Discrimination Ratio (D) is a parameter which measures the ‘power’ of the test.
The smaller the D, for a given and  the steeper is the Operating Characteristic (OC) curve
(see PtDCh7) and the better is the test. For an Availability Demonstration, the discrimination
ratio is the ratio of the consumer’s minimum acceptable unavailability (1 − A α ) the
producer’s target unavailability (1 − A β )

1 − Aβ
D=
1 − Aα

2.4.2 The selection of the producer’s and consumer’s risks will be a compromise between
test accuracy and cost. The nearer the discrimination ratio is to one, the more expensive the
test will be, but the better the test is at discriminating between good and bad.

2.4.3 Typical values for the Discrimination Ratio range between 1.5 and 3 with 2 being the
most common. This is the same as other statistical demonstrations.

2.5 Distribution Assumptions

2.5.1 The test plans discussed in this Chapter are based on the following assumptions
regarding up times and down times:
a) The up times are negative exponentially distributed
b) The down times are gamma distributed

The probability density functions of the negative exponential and gamma distributions are
discussed in Pt D Ch 7. The test plans are only valid if these assumptions are true for the
equipment under test. Any assumptions regarding the distribution of up times and down
times should be based on previous experience, and in some cases supported by information
about the physics of failure. If an equipment has a constant failure rate, the up times will be
negative exponentially distributed. This is normally true for equipment which is not subject
to early life failures or wear out failures, or repairable equipment made up of many
components. Within a complex repairable equipment each component has its own failure
characteristic, and may wear out and be replaced many times before others reach the wear out
stage. Thus a complex repairable equipment never wears out, if components are replaced on
failure. However, if no information is available about the equipment failure time
distributions, the exponential may be used as an approximation, and checked using the Chi-
Square or Kolmorgorov-Smirnov Test (see PtDCh7).

Issue 1
Chapter 9
Availability Demonstration Plans
2.5.2 The Availability Demonstration test plans illustrated in this Chapter are based on the
assumption of a gamma distribution for down times. BS 5760 1 states that this distribution is
used for mathematical convenience and is not critical to the test plans, and that repair rate
functions may be approximated to a gamma distribution, even if the true distribution is nearer
to a log-normal. The shape of the gamma distribution is dependent on the shape parameter
‘p’. When p = 1, the gamma distribution results in a negative exponential distribution and as p
increases the probability density function tends to a lognormal distribution. Details of the
gamma distribution and the effect of the shape parameter on the probability density function
are illustrated in PtDCh7.

2.5.3 At the end of the demonstration, the up times must be scrutinised for trends. If the up
times have an increasing or decreasing trend, the steady-state availability assumption will not
be valid. This does not necessarily mean that the equipment is unacceptable. For example, if
the equipment passes the test and the up times are tending to increase, obviously the
equipment is acceptable. The consumer and producer may agree to undertake a trend and
distribution test, or agree the assumptions before the demonstration test and thereby avoiding
any disagreements of interpretation.

3 STATISTICAL TEST PLANNING

3.1.1 The statistical test plans result in either an accept or reject decision, as to whether the
equipment satisfies the Availability requirement, given the associated risk levels, and  A
test plan should be designed using the following steps:
a) The plan should include the Availability requirement, Availability measure (i.e.
Operational, Intrinsic, Effective), the classification of states and times, failure
definitions, the consequences of rejection and the acceptable value of steady state
unavailability.
b) The consumer and producer need to agree the shape parameter ‘p’, for the assumed
gamma distributed down times. Estimations of the mean up time and mean down
time must be determined, such that a predicted equipment unavailability can be
calculated.
c) Acceptable risk levels, and , are selected and the test parameters determined in
accordance with the appropriate test plan.

4 DEMONSTRATION TEST PLANS

4.1 Fixed Number of Failures

4.1.1 The plan designs a demonstration for a minimum number failures (n) which satisfies
the following inequalities:

F1− α (2pn,2n) x F1− β (2n,2pn) ≤ D(1 − U 0 ) / (1 − DU 0 ) (1)

Where

F1− α the 1- fractile of the F-distribution (see Table 1)

Page 4
 Applied R&M Manual for Defence Systems
Part D – Supporting Theory
F1− β the 1- fractile of the F-distribution (see Table 1)

p = shape parameter for gamma distribution down times

n = number of failures

D = discrimination ratio

U0 = minimum acceptable unavailability

4.1.2 The accept/reject decision is as follows:

Reject if DT/UT > F1− α (2pn,2n) x U 0 / (1 − U 0 )

Where :

DT = accumulated down time

UT = accumulated up time

Accept otherwise

4.1.3 Example: The minimum acceptable Availability for an equipment is 90.0% and the
producer’s target Availability is 95.0%. Therefore, U 0 = 0.05 (producer’s target
unavailability) and U 1 = 0.1 (minimum acceptable unavailability). The down time (gamma
distribution) shape parameter, p = 2. The consumer’s risk (  = 0.10 and the producer’s risk
( ) = 0.10.

D(1 - U 0 ) / ( 1 - D U 0 ) = (2)(0.95)/(0.9) = 2.111

Determine the number of failures (n) which ensures the inequality equation [1] is satisfied.
The result of F1−α (2pn,2n)
 x F1− β (2n,2pn) can be determined from the F-distribution in Table
1. For example, for 5 failures F1−α (2(2)(5),2(5)) x F1− β (2(5),2(2)(5)) = F1−α ( 20,10) x
F1− β (10,20) = 2.2 x 1.94 = 4.268. Calculating relevant values:

Number of Failures (n) 10 18 19 20

F1−α (2pn,2n) x F1− β (2n,2pn) 3.749 2.146 2.103 2.059

Therefore, the test should run until 19 failures and restorations have occurred and reject if :

DT/UT > F1−α (2pn,2n) x U 0 / ( 1 - U 0 )

ie DT/UT > 0.0784.

Issue 1
Chapter 9
Availability Demonstration Plans

0.80 fractiles

v2
v1 2 4 6 8 10 20 30 40 60 120
2 4.00 4.24 4.32 4.36 4.38 4.43 4.45 4.46 4.46 4.47
4 2.47 2.48 2.47 2.47 2.46 2.44 2.44 2.44 2.43 2.43
6 2.13 2.09 2.06 2.04 2.03 2.00 1.98 1.98 1.97 1.96
8 1.98 1.92 1.88 1.86 1.84 1.80 1.78 1.77 1.76 1.75
10 1.90 1.83 1.78 1.75 1.73 1.68 1.66 1.65 1.64 1.63
12 1.85 1.77 1.72 1.69 1.66 1.61 1.59 1.58 1.56 1.55
14 1.81 1.73 1.67 1.64 1.62 1.56 1.53 1.52 1.51 1.49
16 1.78 1.70 1.64 1.61 1.58 1.52 1.49 1.48 1.47 1.45
18 1.76 1.67 1.62 1.58 1.55 1.49 1.46 1.45 1.43 1.42
20 1.75 1.65 1.60 1.56 1.53 1.47 1.44 1.42 1.41 1.39
30 1.70 1.60 1.54 1.50 1.47 1.39 1.36 1.35 1.33 1.31
40 1.68 1.57 1.51 1.47 1.44 1.36 1.33 1.31 1.29 1.26
60 1.65 1.55 1.48 1.44 1.41 1.32 1.29 1.27 1.24 1.22
120 1.63 1.52 1.45 1.41 1.37 1.29 1.25 1.23 1.20 1.17

0.90 fractiles

v2 v1 2 4 6 8 10 20 30 40 60 120
2 9.00 9.24 9.33 9.37 9.39 9.44 9.46 9.47 9.47 9.48
4 4.32 4.11 4.01 3.95 3.92 3.84 3.82 3.80 3.79 3.78
6 3.46 3.18 3.05 2.98 2.94 2.84 2.80 2.78 2.76 2.74
8 3.11 2.81 2.67 2.59 2.54 2.42 2.38 2.36 2.34 2.32
10 2.92 2.61 2.46 2.38 2.32 2.20 2.16 2.13 2.11 2.08
12 2.81 2.48 2.33 2.24 2.19 2.06 2.01 1.99 1.96 1.93
14 2.73 2.39 2.24 2.15 2.10 1.96 1.91 1.89 1.86 1.83
16 2.67 2.33 2.18 2.09 2.03 1.89 1.84 1.81 1.78 1.75
18 2.62 2.29 2.13 2.04 1.98 1.84 1.78 1.75 1.72 1.69
20 2.59 2.25 2.09 2.00 1.94 1.79 1.74 1.71 1.68 1.64
30 2.49 2.14 1.98 1.88 1.82 1.67 1.61 1.57 1.54 1.50
40 2.44 2.09 1.93 1.83 1.76 1.61 1.54 1.51 1.47 1.42
60 2.39 2.04 1.87 1.77 1.71 1.54 1.48 1.44 1.40 1.35
120 2.35 1.99 1.82 1.72 1.65 1.48 1.41 1.37 1.32 1.26

0.95 fractiles

v2 v1 2 4 6 8 10 20 30 40 60 120
2 19.00 19.25 19.33 19.37 19.40 19.45 19.46 19.47 19.48 19.49
4 6.94 6.39 6.16 6.04 5.96 5.80 5.75 5.72 5.69 5.66
6 5.14 4.53 4.28 4.15 4.06 3.87 3.81 3.77 3.74 3.70
8 4.46 3.84 3.58 3.44 3.35 3.15 3.08 3.04 3.01 2.97
10 4.10 3.48 3.22 3.07 2.98 2.77 2.70 2.66 2.62 2.58
12 3.89 3.26 3.00 2.85 2.75 2.54 2.47 2.43 2.38 2.34
14 3.74 3.11 2.85 2.70 2.60 2.39 2.31 2.27 2.22 2.18
16 3.63 3.01 2.74 2.59 2.49 2.28 2.19 2.15 2.11 2.06
18 3.55 2.93 2.66 2.51 2.41 2.19 2.11 2.06 2.02 1.97
20 3.49 2.87 2.60 2.45 2.35 2.12 2.04 1.99 1.95 1.90
30 3.32 2.69 2.42 2.27 2.16 1.93 1.84 1.79 1.74 1.68
40 3.23 2.61 2.34 2.18 2.08 1.84 1.74 1.69 1.64 1.58
60 3.15 2.53 2.25 2.10 1.99 1.75 1.65 1.59 1.53 1.47
120 3.07 2.45 2.18 2.02 1.91 1.66 1.55 1.50 1.43 1.35

Table 1: F-Distribution

4.2 Fixed Time Test Plan

4.2.1 This test plan designs a demonstration with a duration (T) using the expression below.
The test will result in duration of at least 15 times the estimated mean up time (m u ) which is
required input for this plan. Pt C Ch 36 describes Reliability Prediction.

Page 6
 Applied R&M Manual for Defence Systems
Part D – Supporting Theory
2
⎧ ⎡ u (1 − DU 0 ) D ⎤ ⎫
⎪ u 1-α 1 − U 0 + ⎢ 1-β ⎥⎪
⎪
⎪ ⎢
⎣ 1 − U 0 ⎥⎦ ⎪⎪
T = m u (1 + p −1 ) ⎨ ⎬ (2)
⎪ (D − 1) ⎪
⎪ ⎪
⎪⎩ ⎪⎭

Where:

T = test duration

mu = the estimated mean up time

p = shape parameter for gamma distribution down times

u1−α the 1- fractile of the standardised normal distribution

u1−β the 1- fractile of the standardised normal distribution

U0 = minimum acceptable unavailability

D = discrimination ratio

Values for u1−α and u1−β are illustrated in Table 2.

1-α u1 - α

1-β u1 - β

0.80 0.842

0.90 1.282

0.95 1.645

Table 2: Standardised Normal Distribution

4.2.2 The accept/reject decision is as follows:

Reject if DT/(UT + DT) > U lim

Where :
DT = accumulated down time

UT = accumulated up time

Issue 1
Chapter 9
Availability Demonstration Plans
[u1−α D(1 − U 0 ) + u 1−β D (1 − DU 0 ) ]
U lim = U0 x
[u 1−α (1 − U 0 ) + u 1−β D (1 − DU 0 ) ] (3)

Accept otherwise.

4.2.3 Example: Using the same values as for the example for the fixed number of failures
plan in Section 4.2.3. Substituting these values into equation 2:
2
⎧ ⎡ 1.282 (1 − 2.0.05) 2 ⎤ ⎫
⎪ 1.282 .95 + ⎢ ⎥⎪
−1 ⎪ ⎣ .95 ⎦⎪
T = mu (1 + 2 ) ⎨ ⎬
⎪ (1) ⎪
⎪ ⎪
⎩ ⎭
2
⎧ 1.2495 + 1.6317 ⎫
= mu (1.5) ⎨ ⎬
⎩ 1 ⎭

= 12.45 mu

Therefore, for an estimated mean up time of 100 hours, the test duration should be 1,245
hours. Reject if the DT/(UT+DT) is greater than:

⎡ 1.282.2(.95) + 1.282 2 (1 − 2.0.05) ⎤

U lim = 0.05 ⎢ ⎥
⎣ 1.282(.95) + 1.282 2 (1 − 2.0.05) ⎦

⎡ 2.4358 + 1.6317 ⎤
= 0.05 ⎢ ⎥
⎣ 1.2179 + 1.6317 ⎦

= 0.07137

4.2.4 An alternative fixed time test plan, designs a demonstration with a test duration (T)
based on the relationship between T and the discrimination ratio, D, given in Tables 3, 4 and
5. The test should only be applied when the ratio of mean down time and mean up time is
less than 0.05. Pt C Ch 36 and Ch 37 detail Reliability and Maintainability prediction
respectively.

4.2.5 The accept/reject decision is as follows:

Reject if DT/(DT+UT) > U lim

where U lim is a function of the duration (T) from Tables 4, 5 and 6.

Accept otherwise.

4.2.6 Example: Using the same values as for the example for the fixed number of failures
plan in Section 4.2.3. Using the values of D = 2, p = 2, α = β = 0.10, Table 4 gives:

Page 8
 Applied R&M Manual for Defence Systems
Part D – Supporting Theory
T/ mu - 17.78.

U lim /U 0 = 1.387

Assuming mu = 100 hours, the test duration should be 1,778 hours. The producer's target
unavailability (U 0 ) is 0.05, and therefore a reject decision is DT/(UT+DT) is greater than
0.06933.

p=1
T*/m u α = β = 0.05 α = β = 0.10 α = β = 0.20
D Ulim /U0 D Ulim /U0 D Ulim /U0
1.0 24.73 3.92 15.01 2.91 7.37 1.86
1.2 20.01 3.64 12.39 2.75 6.30 1.83
1.4 16.88 3.42 10.64 2.63 5.56 1.80
1.6 14.66 3.24 9.38 2.52 5.03 1.77
1.8 13.01 3.10 8.44 2.44 4.63 1.74
2.0 11.73 2.98 7.70 2.36 4.31 1.71
2.5 9.52 2.75 6.42 2.22 3.74 1.65
3.0 8.12 2.58 5.59 2.11 3.37 1.61
3.5 7.15 2.45 5.01 2.03 3.10 1.57
4.0 6.44 2.34 4.58 1.96 2.90 1.54
5.0 5.46 2.19 3.98 1.86 2.61 1.49
6.0 4.82 2.07 3.58 1.78 2.42 1.45
7.0 3.46 1.99 3.29 1.72 2.28 1.42
8.0 4.02 1,92 3.07 1.67 2.17 1.40
9.0 3.75 1.86 2.90 1.63 2.08 1.37
10.0 3.53 1.81 2.76 1.60 2.01 1.36
15.0 2.88 1.65 2.33 1.49 1.78 1.29
20.0 2.53 1.56 2.10 1.42 1.65 1.26

Table 3: Discrimination Ratio (D) and Rejection Limit for Test Plan 3
with Shape Parameter (p) = 1

Issue 1
Chapter 9
Availability Demonstration Plans

p=2
T*/mu α = β = 0.05 α = β = 0.10 α = β = 0.20
D Ulim/U0 D Ulim/U0 D Ulim/U0
1.0 18.94 3.47 11.74 2.69 6.01 1.87
1.2 15.43 3.22 9.78 2.54 5.19 1.81
1.4 13.11 3.04 8.46 2.42 4.63 1.76
1.6 11.45 2.89 7.52 2.33 4.22 1.72
1.8 10.22 2.77 6.80 2.25 3.91 1.69
2.0 9.26 2.67 6.25 2.18 3.66 1.66
2.5 7.62 2.47 5.28 2.06 3.22 1.60
3.0 6.56 2.33 4.65 1.96 2.93 1.55
3.5 5.83 2.22 4.20 1.89 2.72 1.51
4.0 5.29 2.13 3.87 1.83 2.56 1.48
5.0 5.54 2.00 3.40 1.74 2.33 1.44
6.0 4.05 1.91 3.09 1.67 2.17 1.40
7.0 3.70 1.84 2.87 1.62 2.06 1.37
8.0 3.43 1.78 2.70 1.58 1.97 1.35
9.0 3.22 1.73 2.56 1.54 1.90 1.33
10.0 3.06 1.69 2.45 1.51 1.84 1.31
15.0 2.54 1.56 2.10 1.42 1.65 1.26
20.0 2.26 1.48 1.92 1.36 1.55 1.22

Table 4: Discrimination Ratio (D) and Rejection Limit for Test Plan 3
with Shape Parameter (p) = 2
p=5
T*/mu α = β = 0.05 α = β = 0.10 α = β = 0.20
D Ulim /U0 D Ulim /U0 D Ulim /U0
1.0 15.37 3.14 9.64 2.52 5.05 1.83
1.2 12.72 2.93 8.19 2.38 4.48 1.77
1.4 10.87 2.77 7.15 2.27 4.04 1.72
1.6 9.54 2.64 6.38 2.19 3.71 1.68
1.8 8.55 2.54 5.80 2.11 3.45 1.61
2.0 7.78 2.45 5.36 2.05 3.25 1.61
2.5 6.47 2.28 4.57 1.94 2.89 1.55
3.0 5.62 2.16 4.06 1.85 2.65 1.51
3.5 5.03 2.07 3.70 1.79 2.47 1.47
4.0 4.59 1.99 3.43 1.73 2.34 1.44
5.0 3.98 1.88 3.04 1.65 2.15 1.40
6.0 3.58 1.80 2.79 1.60 2.02 1.36
7.0 3.29 1.73 2.60 1.55 1.92 1.34
8.0 3.07 1.68 2.46 1.51 1.84 1.32
9.0 2.90 1.64 2.34 1.48 1.78 1.30
10.0 2.76 1.61 2.25 1.46 1.73 1.28
15.0 2.32 1.49 1.96 1.37 1.57 1.23
20.0 2.09 1.42 1.80 1.32 1.48 1.20

Table 5: Discrimination Ratio (D) and Rejection Limit for Test Plan 3
with Shape Parameter (p) = 5

4.3 Sequential Test Plan

Page 10
 Applied R&M Manual for Defence Systems
Part D – Supporting Theory
4.3.1 This test plan designs a demonstration with a duration which depends on the observed
Availability of the equipment during test. After each repair, a decision is made as to whether
the test should bed terminated or continued. The decision limits depend on the number of
failures (n), that have occurred up to that instant of time.

The decision is based on the following rules:

Reject if DT/UT > Re(n) x U 0 /(1 − U 0 ) (4)

Accept if DR/UT < Ac(n) x U 0 /(1 − U 0 ) (5)

The test continues if neither of the above inequalities determine a reject or accept decision.

Where DT = Accumulated downtime

UT = Accumulated up time

U 0 = Producer's target unavailability

Re(n) =
[D − G(n)] if n > ln [(1 − β)/α]
p[G(n) − 1] p.ln(D)

else Re(n) = ∞

Ac(n) =
[D − H(n)] if n > ln [1 − β/α ]
[H(n) − 1] p.ln(D)

else Ac(n) = 0.

Where G(n) = D1/(1+ p) [α/(1 − β) ]1/(n + np)

H(n) = D1/(1+ p) [(1 − α)/β ]1/(n + np)

4.3.2 Example. Using the same data as before, i.e. U 0 = 0.05, α = β = 0.1, D = 2 and p
= 2. Table 6 below illustrates an example of how a demonstration test can be monitored and
an accept/reject decision determined. The first failure occurred after 60 hours, with 17 hours
downtime, the second failure after 344 hours of uptime, followed by 28 hours downtime, etc.
The table shows the values calculated for each parameter to make a decision, based on the
rules in equations 4 and 5. Note, ND signifies no rejection is possible in this case, as the
value for Re(n) is either negative or infinity.

Table 6: Calculated Parameters for Sequential Test Plan Examples

Issue 1
 Applied R&M Manual for Defence Systems
Part D – Supporting Theory

LEAFLET D9/0

REFERENCES

1 BS 5760: Section 10.3: 1993. Reliability of Systems, Equipment and Components.

Part 10 - Guide to Reliability Testing. Section 10.3 - Compliance Test Procedures for
Steady-State Availability.

Issue 1
Chapter 9, Leaflet 0
Availability Demonstration Plans

Intentional Blank Page

Page 2