Sie sind auf Seite 1von 79

BITS Pilani

Pilani Campus

Reliability Testing
Chapter - 13
Product Testing
• An integrated product test program consists of several types of
test each having different objectives.
• Reliability qualification tests obtain various measures of
product reliability.
• Reliability growth testing has as its objective increased
reliability for the end product.
• All product testing may provide useful reliability information,
and an aggressive failure mode, effect, and critically analysis
program will capture any relevant failure data.
• Reliability testing and safety testing are distinguished from
other tests in that they attempt to generate failures in order to
identify failure modes and eliminate them.

BITS Pilani, Pilani Campus


Reliability Life Testing
• The primary objectives:
- to obtain information concerning failures in order to quantify
reliability
- to determine whether reliability and safety goals are being
met
- to improve product reliability.

• The different types of reliability tests addressed are


- Burn-in and screen testing
- Acceptance and qualification testing
- Sequential tests
- Accelerated life testing
- Experimental design
BITS Pilani, Pilani Campus
Reliability Life Testing
• The important factors to be addressed before any reliability test:

- objective of the test

- type of test to be performed

- operating and environmental conditions

- number of units to be tested

- duration of tests

- definition of failure
BITS Pilani, Pilani Campus
Test Time Calculations
• For a constant failure rate , the cumulative test time T
may be obtained from the following table:

5
BITS Pilani, Pilani Campus
Test Time Calculations

• Cumulative test time for = Total operating time that all


CFR model units experienced on test

• For a CFR model an estimate for the MTTF is given by

MTTF = T/r (13.1)


where,
r – total number of failures.

BITS Pilani, Pilani Campus


Example

BITS Pilani, Pilani Campus


Length of Test
• For Type II testing, the length of time to complete the test
depends on :
1. The number of units being tested.
2. The number of failures to be observed.
3. Time-to-failure distribution.
• If n units are placed on test until r failures are observed, then
expected test time under CFR model:

E(test time) (13.2)


where,
TTFr,n is the test time for r failures with n units at risk

BITS Pilani, Pilani Campus


Length of Test
• If failed units are replaced immediately, so that n units are
always on test,

where,
TTRr,n - test time factor with replacement of failed units.

Observations
• The number of units needed to complete the test = n+r-1
• More the units on test, lesser the expected test time.

BITS Pilani, Pilani Campus


Length of Test
• For Type I testing, the length of time is t* and the number of
failures r is random.
• For CFR model with n units on test,

• Probability of a single unit failing by time t*, p = 1- e-t*/MTTF


 The number of failures among n units on test can be a
binomial process with mean np.
• With replacement of failed units,

BITS Pilani, Pilani Campus


Example

BITS Pilani, Pilani Campus


Burn-in Testing
• Merits
 Increases the mean residual life of components.
 Survived items will have a greater MTTF than the original items.
Probability of a failure occurring over a fixed length of time is
reduced.
 Burn-in testing may allow contract specifications to be met where
they otherwise couldn’t .
• Considerations
 Costs such as cost of testing, warrant costs, item lost due to burn-
in failures and cost of failure during operation decide whether to
use burn-in testing or not and if so, to which extent.
 Item must have a decreasing failure rate (DFR).
 Increases production lead time as well as costs.
BITS Pilani, Pilani Campus
Burn-in Testing
• Items that have failed during burn-in may be
 replaced with a new item from the same parent population which
may or may not have had some burn-in time accumulated.

 repaired to its original condition or may be minimally repaired.


- for the minimally repaired condition, if the intensity function is
decreasing, then improved reliability will result.

• Mathematical modelling of burn-in period depends on the


manner in which failures are disposed of.

• Primary determination for burn-in period is the length of test.

BITS Pilani, Pilani Campus


Burn-in Testing
• The model based on following figure assumes that only the
surviving units are utilized following burn-in.

14
BITS Pilani, Pilani Campus
Burn-in Testing
• Given a reliability goal at time t0 of R0 where R(t0) , R0 and R(t)
has a DFR, a burn-in period, T, is desired such that R(t0|T) =
R0.

• For Weibull distribution

15
BITS Pilani, Pilani Campus
Example

16
BITS Pilani, Pilani Campus
Burn-in Testing
• The length of the burn-in period can also depend on costs.
• Let,
Cb – cost per unit for burn-in testing
Cf - cost per failure during burn-in
Co – cost per failure when operational
T - length of burn-in testing
t - operational life of the units.
• Assume that n units having a reliability function R(t) and
each undergoing burn-in testing are to be produced.

• The expected number of failures during burn-in = n[1-R(T)].

BITS Pilani, Pilani Campus


Burn-in Testing
• The expected number of operational failures
nR(T) [1 - R(t|T)] = nR(T) [1- R(t+T)/R(T)] = n [R(T) - R(t+T)]
• Expected total cost
E[TC] = nCbT + Cfn[1-R(T)] + Con[R(T) – R(t+T)]
• Expected cost per unit
E[C] = E[TC]/n = CbT + Cf[1-R(T)] + Co[R(T) – R(t+T)] (13.7)
• For the Weibull distribution,

BITS Pilani, Pilani Campus


Example

19
BITS Pilani, Pilani Campus
Example

20
BITS Pilani, Pilani Campus
Acceptance Testing
• To demonstrate that the system design meets performance
and reliability requirements under specified operating and
environmental conditions.

• It may be based on a predetermined sample size or on an


unspecified sample size resulting from a sequential test as
described subsequently.

• Units from the production line for testing must be selected


randomly.

21
BITS Pilani, Pilani Campus
Binomial Acceptance Testing
• Objective – to demonstrate that the system reliability at time T is
R1.

• The sample size n and maximum number of failures, r, for


acceptance are specified.

• X failures are observed by time T and if X ≤ r, then the desired


reliability is demonstrated; otherwise R(T) < R1 .

• The random variable X has a binomial distribution with


parameters n and p = 1-r.

BITS Pilani, Pilani Campus


Binomial Acceptance Testing
• If α – probability of incorrectly rejecting the reliability
specification
β – probability of incorrectly accepting the reliability
specification

• Then, to find the values for n and r that will result in high
probability of acceptance if R(T) = R1 and a low probability of
acceptance if R(T) = R2<R1,
Pr{X≤ r | R= R1} = 1 – α and Pr{X≤ r | R= R2} = β

• The operating characteristic curve for the distribution depends


on the values specified for n and r.

• The region R1<R<R2 is referred to as the indifference zone.


BITS Pilani, Pilani Campus
Binomial Acceptance Testing
• The operating characteristic curve:

BITS Pilani, Pilani Campus


Binomial Acceptance Testing
• Since X is binomial, the probability statements discussed earlier
can be written as

BITS Pilani, Pilani Campus


Example

BITS Pilani, Pilani Campus


Sequential Tests
• Efficient method for accepting or rejecting a statistical
hypothesis.

• Used in a reliability or maintainability demonstration or in


acceptance and qualification testing, not for estimating reliability
parameter.

• Since the sample size required depends on the observed times,


fewer failures may need to be generated.

• Assume a reliability parameter represented in general by ϕ has a


specification ϕ0 and an unacceptable value ϕ1.

BITS Pilani, Pilani Campus


Sequential Tests
• A null (H0) and alternate (H1) hypothesis are defined as:
H0 : ϕ = ϕ0 (the product meets the specification)
H1 : ϕ = ϕ1 > ϕ0 (the specification is not met)

• With each sequentially generated failure or repair time


t1,t2,..tr, a test static yr = h(t1,t2,..tr) is computed.

• Depending on the test static, the null hypothesis may be


accepted or rejected or judgement may be reversed.

• If judgement is reversed, the test is repeated using a new


sample time and test static until H0 is accepted or rejected.
BITS Pilani, Pilani Campus
Sequential Tests
• The criterion to accept, reject or continue sampling is based on
the probability of making an incorrect decision.

• We may reject a correct null hypothesis (called type I error) or


we may accept a false null hypothesis (called type II error).

Pr { reject H0|ϕ0} = α and Pr{accept H0|ϕ1} = β


α – the producer’s risk
β – the consumer’s risk

• A test static yr is formed from the ratio of the likelihood function


formed under H1 to that formed under H0 .
BITS Pilani, Pilani Campus
Sequential Tests
• If yr ≤ A, H0 is accepted.

• If yr ≥ B, H0 is rejected.

• In conducting a sequential test, α,β,ϕ0 and ϕ1 must be specified.


• If A < yr < B, then the test continues by generating another
sample.
BITS Pilani, Pilani Campus
Exponential Case
• For the exponential distribution f(t) = λe –λt ,
H0: λ≤ λ0
H1: λ = λ1 > λ0
• The continuation region is represented by

where, ti – time to failure of the ith unit tested.


• Taking logs and rearranging,

BITS Pilani, Pilani Campus


Example

32
BITS Pilani, Pilani Campus
Example

33
BITS Pilani, Pilani Campus
Binomial Testing
• No assumption concerning the failure distribution is necessary.

• Test is based on a binomial process.

• The hypotheses to test are


H0: R(t0) = R0
H1: R(t0) = R1 < R0
• If n units are tested until time t0 and that y survivors are
observed,

34
BITS Pilani, Pilani Campus
Binomial Testing
• Therefore, the continuation region is found from

• After taking logarithms and performing some algebra,

• Number of survivors yn ,among n units at risk is observed.


• If yn is less than or equal to lower bound, H0 is rejected.
• If yn is equal to greater than upper bound H0 accepted.
• If yn is in continuation region, another unit is tested until t0.
35
BITS Pilani, Pilani Campus
Example

36
BITS Pilani, Pilani Campus
Example

37
BITS Pilani, Pilani Campus
Maintainability Demonstration
• The hypotheses for binomial sequential test to perform
maintainability demonstration :
H0 : H(t0) = P0
H1 : H(t0) = P1 < P0
where,
H(t) – CDF of the repair distribution
P0 – fraction of repairs to be completed within t0 time units.
P1 – unacceptable fraction of repairs to be completed within
time t0.
• If the number of repairs yn among n attempts completed within
time t0 equals or exceeds the upper bound, then H0 is accepted.
• If yn is less than or equal to lower bound, then H0 is rejected.
38
BITS Pilani, Pilani Campus
Maintainability Demonstration
• Parameter under the alternative hypothesis may take on a
range of values.
• The farther these values from the hypothesized value ϕ0, the
smaller will be the value of type II error β .
• A plot of the probability of Type II error Vs the value of ϕ
under the alternate hypothesis generates the following OC.

BITS Pilani, Pilani Campus


Accelerated Life Testing
• The amount of time available for testing is considerably less
than the expected lifetime of the component.

• For highly reliable components, testing under normal


conditions would generate only few failures within a
reasonable time period.

• In order to identify design weakness during testing,

1. Increase the number of units on test


2. Accelerate the number of cycles per unit of time
3. Increase the stresses that generate failures.
BITS Pilani, Pilani Campus
Number of Units on Test
• The fraction savings in test time by having n units, rather than r
units at risk, when r failures are desired can be found using the
expected test time table.
• Let

• Then the percent savings is 100(1- fr,n)


• If failed units are replaced, then fr,n = r/n .
• For the Weibull failure distribution,

• If failures are replaced, then fr,n = (r/n)1/β


BITS Pilani, Pilani Campus
Example

42
BITS Pilani, Pilani Campus
Accelerated Cycling
• Assume
- no new failure modes are introduced as a result of
increasing the number of cycles per unit of time.
- failures occur due to cycling only.

• Define
xn- number of cycles per unit of time under normal operating
conditions.
xs – number of cycles per unit of time under accelerated
conditions.
tn – time to failure under xn cycles per unit of time.
ts – time to failure under xs cycles per unit of time.

BITS Pilani, Pilani Campus


Accelerated Cycling
• xntn = xsts , since the number of cycles to failure is the same for
both normal and accelerated conditions.

• For the Weibull and exponential distributions

• Therefore βs = βn = β, and

BITS Pilani, Pilani Campus


Example

BITS Pilani, Pilani Campus


Constant-Stress Models
• Assumptions
- at higher stress levels the same failure mechanisms will be
present and act in the same manner as at normal stress levels.
- failures will happen more quickly.
- only a transformation of time scale is observed
- no new failure modes are introduced.
• The simplest case of accelerated stress testing assumes a
linear acceleration effect over time.
tn = time to failure under normal stress
ts = time to failure at high stress level
then
tn = AF x ts
BITS Pilani, Pilani Campus
Constant-Stress Models
• AF – acceleration factor to be determined.
• Therefore CDF of the distribution,

• PDF

• Hazard rate function

BITS Pilani, Pilani Campus


Constant-Stress Models
• If the failure rate at the accelerated stress level is constant,
then the failure rate under normal stress will also be constant.
• For Weibull failure distribution

• Therefore only characteristic life is affected by the linear


accelerated stress testing.
• For the Weibull failure law,

48
BITS Pilani, Pilani Campus
Example

49
BITS Pilani, Pilani Campus
Example

50
BITS Pilani, Pilani Campus
Other Acceleration Models
Arrhenius model
• When failures are accelerated primarily as a result of an
increase in temperature,

where,
r – reaction process rate
A and B – constants
T – temperature in Kelvin.
• Acceleration factor may be determined from,

BITS Pilani, Pilani Campus


Arrhenius model
• The constant B can be estimated by testing at two different
stress temperatures and computing the AF on the basis of
the fitted distributions.

where AF = θ1/θ2
θi – scale parameter at the stress level
corresponding to Ti

BITS Pilani, Pilani Campus


Example

BITS Pilani, Pilani Campus


Eyring Model
• Allows for additional stresses and can be derived from quantum
mechanics.
r = ATae–B/TeCS
where,
r – process rate
A,a,B and C – constants
T – temperature in Kelvin
S – second stress.
• The first exponential factor and its coefficient account for the
temperature and , except for Ta, behave as Arrhenius model.
• The second exponential factor accounts for a non thermal
stress and additional factors for additional stress can be added.

BITS Pilani, Pilani Campus


Eyring Model
• If a is close to zero, Ta factor will be close to 1 at all temperature
and in the absence of a second stress, it’s similar to Arrhenius
model.
• To apply Eyring model, constants must be estimated.
• For this four data points at two different temperature levels and
two different stress levels are required.
• The AF for this model,

BITS Pilani, Pilani Campus


Degradation Models
• For products with performance variation with time, by
extrapolating degradation in performance over time, time to
failure may be predicted.

• Examples of degradation process can be corrosion, crack


propagation and the shelf life of pharmaceutical products.

• The simplest relationship is linear with y = a- bt where,


y – performance measure
a,b – constants
t- amount of time the product is exposed at a constant stress
level.

BITS Pilani, Pilani Campus


Degradation Models
• If yf is the level at which a failure occurs, then the time to failure
tf,

• It may be interpreted as the median or mean of the failure


distribution.

• Testing n units and fitting separate regression lines to each unit


will generate a sample of n predicted failure times.

BITS Pilani, Pilani Campus


Example

58
BITS Pilani, Pilani Campus
Example

BITS Pilani, Pilani Campus


Cumulative Damage Models
• A component damage is leading to continuous failure
accumulation and the damage rate depends only on the
amount of damage and not on any past history.
• Then

where, ti – amount of time at stress level i


Li – expected lifetime at stress level i
• To apply this model, consider a normal stress level(t1) and an
other high stress level (t2),

BITS Pilani, Pilani Campus


Cumulative Damage Models
• The Eq. (13.29) represents ‘the failure line’ and any
combination of stress times (t1,t2) that lie on the line will result
in a failure.

BITS Pilani, Pilani Campus


Cumulative Damage Models
• To determine the value for L2, test the component at the high
stress level until failure (L2).

• To determine a second point on the line, test the component


first for some time t1 at normal stress level and then at high
level until failure occurs at time t2.

• Then L1, time to failure under normal stress, is found from

BITS Pilani, Pilani Campus


Step Stress Models
• Testing begins with normal stress

• Stress is increased after a period of time

• Stepwise stress increase is continued until all the test units fail.

• The increase in stress is assumed to be equivalent to a linear


change in the time scale.

• More complex models than the constant-stress models.

63
BITS Pilani, Pilani Campus
Experimental Design
• Efficient collection and analysis of data that will maximize the
obtained information.
• The factors and their values to be investigated are identified
wrt their effect on dependent variable.
• The objective of the experiment may be to identify critical
factors, or to estimate the effect the selected factors have on
the response variable or both.
• A particular experimental design consisting of a statistical
model for collection and analysis of data is selected.
• The design will identify the factors, levels, number of
replications and the use of blocking

BITS Pilani, Pilani Campus


Experimental Design
• Let us discuss the use of factorial designs in identifying factors
that significantly affect a reliability or maintainability parameter.
• A factorial experiment consists of the collection and
simultaneous evaluation of data at all combinations of the levels
of the factors being investigated .
• If k factors are being considered, each at m different levels,
then a single replication will consist of mk experiments.
• To overcome the requirement of large number of experiments,
the number levels and factors must be kept small.
• The advantage of factorial designs is the ability to measure the
effect of the interaction two or more factors have on the
response variable.

BITS Pilani, Pilani Campus


Experimental Design
• Mathematical model for a two-factor factorial experiment is

66
BITS Pilani, Pilani Campus
Experimental Design
• Since the factor effects are assumed to be deviations from the
overall mean,

• The statistical hypotheses of interest are

67
BITS Pilani, Pilani Campus
Experimental Design
• To test these hypotheses an analysis of variance (ANOVA) is
performed.

• Independent estimates of population variance (factor mean


square) from the data is computed.

• If a factor is not significant, its variance estimate should not


differ significantly from a pure population mean square( the
mean square for error).

• A significant factor would have a larger mean square than the


mean square for error.
68
BITS Pilani, Pilani Campus
Experimental Design
• Te ratio of the factor mean square over the mean square for
error forms an F distribution.
• The larger the computed F static, the more likely the factor is
significant.
• A comparison with a tabulated F distribution will establish the
critical value at a given level of significance.
• A table summarizing the result of analysis when the factor
levels are determined by the experimenter( a fixed-effects
model) rather than being randomly selected from a parent
population (a random-effect model).
• For the fixed effects model, conclusions are valid only for the
factor levels considered.
69
BITS Pilani, Pilani Campus
Experimental Design

70
BITS Pilani, Pilani Campus
Experimental Design
• In the table,

and

71
BITS Pilani, Pilani Campus
Experimental Design
• Where the notations

72
BITS Pilani, Pilani Campus
Example

73
BITS Pilani, Pilani Campus
Example

74
BITS Pilani, Pilani Campus
Example

75
BITS Pilani, Pilani Campus
Example

76
BITS Pilani, Pilani Campus
Example

77
BITS Pilani, Pilani Campus
Competing Failure Modes
• To distinguish among failure modes during reliability testing.

• If the failure modes are mutually independent, they can be


separately analyzed by treating them as multiply censored data.

• The failure times of all the failure modes except the failure
mode under investigation would be considered censored times.

• Different methods are then applied to determine an acceptable


theoretical reliability model.

78
BITS Pilani, Pilani Campus
Thank You!

79
BITS Pilani, Pilani Campus

Das könnte Ihnen auch gefallen