Accelerated Life Testing (ALT)

and
Highly Accelerated Life Testing (HALT)

Product Reliability Course – ME6226

Prof Shankar Krishnapillai, IIT Madras

1
Accelerated Life Testing (ALT)

Life testing is done to estimate the time to failure of products (and thereby MTTF,
MTBF etc) by applying various types of stresses.
They are: Mechanical load, Temp, Electric current, Humidity , Fluid pressure etc.

If life testing is done with the normal expected stresses (f0) on the product, it may
take a long time to fail. The waiting time may be too much.

In ALT, the components are subjected to a higher stress (fA) on the product, causing
it to fail in a shorter time. The failure rate, MTTF could vary in a non-linear manner.

Acceleration Factor (AF) is defined as the ratio of time to failure under normal
load (t0) , divided by time to failure under accelerated load (tA) .

For example if we double the applied stress, and the component fails in half the time,
AF=2. The relationship between load and time to failure need not be linear.

The AF can be found by Weibull testing (which is one of the topics in this lecture).

2
Limitations of Accelerated Life Testing (ALT)

Extrapolation of results from high stress to normal stress is not always easy.
A lot of trials and experiments are needed.

The failure mechanisms at high and normal stress levels may be different. For
example, higher mechanical stresses may cause plastic deformation failure, but
lower loads may cause fatigue failure (assuming cyclic loads).

The MTTF, failure rates etc may behave in a non-linear manner with stress. The
underlying failure mechanism has to be understood before we apply higher stress.

3
Basic Relationships for Accelerated Life Testing ( ALT)
t0
AF = ; or t0 = AF t A
tA
Also: the reliability at t0 must be equal to reliability at t A ;
R0 (t0 ) = R A (t A ).
Let us look at the probability density distributions of time to failure,
under normal and accelerated stress.
 t 
FA  
 AF 
f0(t) fA(t)

F0 (t )

t t/AF
(time to failure, normal stress) (time to failure, accelerated stress)

 t 
From the above figures, F0 ( t ) = FA  
 AF  4
General relationships (applicable to any distribution):
d d  t  1  t 
f 0 (t ) = F0 ( t ) = FA   = f A .
dt dt  AF  AF  AF 

1  t 
fA  
0 ( t ) =
f 0 ( t ) AF
=  AF = 1   t 
A 
R0 ( t )  t  AF  AF 
1 − FA  
 AF 

Let us apply these relationships to Weibull distributions in the next

slide.

5
 For Weibull distribution:
Let  0 ,0 be the shape and scale factors of the normal stress;  A , A for
accelerated stress.
 t 
F0 ( t ) = FA  
 AF 
0 A
 t   t 
−  − 
 0   AF  A 
1− e = 1− e

Comparing the similar terms,

 0 = A and  0 = AF  A .
i.e., the shape factors will be the same, but the normal scale factor  0 will
be AF times the accelerated scale factor  A .

This implies that, eg., if the tim e to failure under normal stress is Gaussian,
time to failure for accelerated stress will also be Gaussian. But the spread
of the distributions will be different.
6
Weibull Distribution:
1  t 
0 ( t ) =  A  
AF  AF 
 −1
1   t 
=    (here  =  0 =  A )
AF  A  AF  A 

1
0 ( t ) = 
 A (t )
AF

For constant failure rate,  = 1.

1
0 = A
AF

7
 Example-1

12 samples of a product were tested to failure under accelerated conditions.

The times to failure in hours are shown below (sorted in increasing order):
Data-1: 39.4, 40.8, 47.1, 66.8, 69.3, 71.0, 77.7, 81.2, 83.3, 84.3, 142.8, 146.2.

Another 12 samples are subjected to normal load and the time to failure in hours
are shown below:
Data-2: 118.3, 122.4, 141.2, 200.3, 208.0, 213.1, 233.0, 243.7, 249.9, 253.0,
428.5, 438.6.

(1) Check wether the data shows the trend for accelerated testing.
(2) If so, find AF (acceleration factor).
(3) Find failure rate under normal load using AF.

8
Ans :
Weibull plots can be fitted using Data-1 and Data-2.
Weibull fit for normal loads gives:  0 =2.5, 0 = 268hr
Weibull fit for accelerated loads gives:  A =2.5,  A = 89.4 hr

(1) Since the  ' s are the same, and  ' s different, it shows the correct
trend for accelerated testing.
0 268
(2) AF = = = 3.0
 A 89.4
( Note :The AF need to be found once only. Thereafter it can be
used for all f uture accelerated tests on that particular component).

9
  −1 2.5−1
A  t  2.5  t 
(3)  A ( t ) =   =   = 3.31  10 −5 t 1.5
A  A  89.4  89.4 

1 1
0 ( t ) = 
 A ( t ) = 2.5
 3.31  10 −5 1.5
t = 2.12  10 −6 1.5
t
AF 3.0

(Note : 0 ( t ) can be calculated directly from  0 and 0 also)

10
Some empirical failure models used in ALT

(1) Temperature dependent stress

This is modelled based on the Arrhenius equation, according to which
Rate of Reaction (r )= A e − Ea / KT
A = a constant
Ea = Activation energy in eV
K = Boltzmann constant
T = Temperature in Kelvin

Constant failure rate under temperature load is modelled similarly as,

 =A e − B / T (where A and B are two constants).

11
To find constants A and B.
Under normal loading T0 , let constant failure rate be 0
0 =A e − B /T 0

Under accelerated loading TA , let constant failure rate be  A

 A =A e − B / T A

 1 1 
A − B − 
AF = =e  T0 TA 
(constant failure rate is proportional to AF ,
0
as per previously derived Weibull formulaes)

12
Procedure to find constants with Arrhenius model.
We can do two failure tests at temperatures T0 and TA . Ratio of Weibull
shape factors 0 / A gives AF . Knowing AF , we can find B as,
 1 1 
− B − 
log e AF
AF = e  T0 T A 
; or B = .
1 1
−
T0 TA

Thereafter A can be easily found by substituting B in one of the

failure rate equations,
0 =A e − B /T or  A =A e − B /T
0 A

A and B need to be found only once. They can be used to predict

 at any temperature for th at component.

13
Example-2

A PCB has a normal operating temperature of T0=21OC, i.e, 294OK.

It is subjected to two accelerated tests a high temps, data as shown below.

Case-1: T1=430OK, ηA1=254, β=1.72

Case-2: T2=450OK, ηA2=183, β=1.72

(1) Find constant B.

(2) Find the AF between the normal temperature T0 , and T2.

14
Ans :
(1) Using temperatures from Case-1 and Case-2, we can find B.

Acceleration factor between T1 and T2 is,

AF1 =  A1 / A 2 = 254 / 183 = 1.39
log e AF1 log e 1.39
B= = = 3172.
1 1 1 1
− −
T1 T2 430 450

(2) AF0 between T0 =294 O K (normal temperature) and T2 =450 O K is:

1 1  1 1 
log e AF0 = B  −  = 3172   −  = 3.74
 T0 T2   294 450 
AF0 = 42.1

15
(2). Temperature and Mechanical stresses (Eyering Model)
Eyering model can take into account one temperature stress and several
mechanical stresses.
 = AT a e − B /T e C S e C S .....
1 1 2 2

(where T is temperature, S1 , S 2 .etc are mechanical stresses.

All others (A, B, C , a ) are constants.
The term T a is usually taken as 1, as a is generally very small.)

For no thermal stress and only mechanical stress, we get

 = A ' eC S eC S
1 1 2 2

For one thermal and one mechanical stress, we get

 = A e − B /T e CS (we will deal only with this model)

16
Procedure for finding constants of B and C of  = A e − B /T e CS
Let (T0 , S 0 ) and (TA , S A ) be the different test levels.
Then,
0 = A e − B /T eCS
0 0
and  A = A e − B /TA e CS A
 1 1 
A B − 
C ( S A − S0 )
AF = =e  T0 TA 
e
0
1 1 
log e AF = B  −  + C ( S A − S 0 )
 T0 TA 

The constants B and C can be found from 4 tests, as shown in the

next example.
2 Tests with temps T0 and TA (keeping stress same)
2 Tests with stress S 0 and S A (keeping temperature same)

17
Example-3

Find the AF between (T0=294K , S0=200MPa) and (TA=400K , SA=400MPa) with the
Eyering model.

To find the constants B and C these 4 special tests were conducted:

Test-1: at T0=294K, η1=500

Test-2: at TA=400K, η2=350 (stress held constant for Test-1 and Test-2)

Test-3: at S0=200MPa, η3=200

Test-4: at SA=400Mpa, η4=100 (temp held constant for Test-3 and Test-4)

18
Ans:
Let AF1 be the acceleration factor between Test-1 and Test-2.
(i.e between temps 294K and 400K at same stress).
1 500
AF1 = = = 1.43
2 350
 1 1 
log e 1.43 = B  −  + C 0
 294 400 
B = 396

Let AF2 be the acceleration factor between Test-3 and Test-4.

(i.e between stresses 200MPa and 400MPa at same temp.).
3 200
AF2 = = =2
4 100
log e 2 = B  0 + C  (400 − 200)
C = 3.47  10 −3
19
( problem solution cont ' d )

Once B and C are known, the acceleration factor AF between any two
combinations of temperature and mechanical stress can be easily found.

Here we want AF between T0 , S 0 and TA , S A.

Hence,
1 1 
log e AF = B  −  + C ( S A − S 0 )
 T0 TA 
 1 1 
 + 3.46  10  ( 400 − 200 )
−3
= 396  −
 294 400 
=1
AF = 2.72

20
Highly accelerated life testing (HALT)
and
Highly accelerated stress screeing (HASS)

21
Highly accelerated life testing (HALT) and Highly accelerated
stress screening (HASS).

HALT is done to test the reliability of a prototype and expose the weakness.
HASS is done for quality control of a final manufactured product.
But their principle is the same. We will discuss HALT first.

HALT is a “brute force” method compared to ALT. There is no attempt to find

acceleration factor, or to extrapolate time to failure etc. ALT is a highly complex and
analytical process, requiring extensive Weibull testing etc. This is not needed in HALT.

In HALT, extreme values of load are applied, sometimes exceeding design limits, but
for short periods of time, iteratively in steps. Components which survive the test are
accepted. Checking survivability of the prototype is the main aim.

Combinations of extreme stresses (temp, humidity, vibrations, mechanical loads,

voltages etc) are applied in specific sequence to check the survivability.

22
The aim of HALT is to find the following information of the prototype:

(a) Find destruct limits.

(b) find operational or functional limits.

(c) get information about possible weakness in design.

(d) initiate various failure modes which were not previously thought.
(loose contacts, delamination, short circuit, plastic failure, melting etc).

At the design stage, a few prototypes are put through HALT. The weaknesses
discovered in HALT are rectified and the tests are repeated.

23
Stress limits used in HALT
Any product (mechanical or electrical) could have 4 stress limits. Consider the
case of a Printed Circuit Board (PCB) subjected to temperature stress. Let us
assume the PCB is designed to work at temps from 10OC to 40OC.

LOSL - lowest operational stress limit (this is 10OC)

UOSL - upper operational stress limit (this is 40OC)

LDSL - lowest destruct stress limit (i.,e lower side temperature at which it is
permanently damaged, say 0OC. )
UDSL - upper destruct stress limit (i.,e higher side temperature at which it is
permanently damaged, say 50OC. )

In the case of the PCB, the stress limits in increasing order are 0OC, 10OC, 40OC,
50OC.

The functional range of the PCB is 10OC to 40OC i.,e between LOSL and UOSL.

The destruct margin (i.,e the dangerous operating range) is 0OC to 10OC on the
lower side, and 40OC to 50OC on the higher side. In HALT testing, stresses in all
these ranges are applied.

24
 Destruct Destruct
Margin Margin
Lower side Higher side
pdf f (stress)

Functional range
(or operational range,
or field stress)

LDSL LOSL stress UDSL

UOSL
0OC 10OC 40OC 50OC

Refer the above figure.

The values of the 4 stress limits (LDSL to UDSL) are uncertain (because of design and
manufacturing variations) and lie on a probability density distribution, with a mean
and s.d.

For HALT we have to fix the stress limits of functional range and destruct margins
based on the the above distribution.

25
 Destruct Destruct
Margin Margin
Lower side Higher side
pdf f (stress)

Functional range
(or operational range,
or field stress)

LDSL LOSL stress UDSL

UOSL

For simplicity and for a preliminary analysis, we can ignore the distributions and
take the mean values of LDSL to UDSL which are indicated by dashed lines.

26
Conducting a HALT test

1. First finalize the stress limits for functional (LOSL to UOSL) and destruct ranges
(LDSL to UDSL).

2. HALT test is first carried out (Phase-1) in the functional range of stress only.

3. In case of the PCB, the temperature can be increased in steps of 5OC

from LOSL (10OC) to UOSL (40OC), holding several minutes at each step
(this is called the “dwell period”). Once temp reaches UOSL, rapid stress
recycling is done i.e, the cycle is again started from LOSL to UOSL.
UOSL

recycling
Temp

LOSL start

time
27
4. If there is another additional stress (say vibration excitation in g levels), those
stresses (LOSL, UOSL g values) can also be applied with a shaker. Combined
vibration and temperature stresses can be stepwise applied, making sure
that the worst combinations are tested.

One method is to hold the vibration stress (g) at a constant level and then vary
the temperatures from LOSL to UOSL. Then the vibration stress is increased to the
next g level and held constant, and same procedure is repeated.

5. After Phase-1 test, the damages/failures are assessed and modifications are
made. A lot of unexpected failure modes like cracking, loose connections,
unequal expansions, de-laminations etc will come to light via Phase-1.

6. The modifed component is subjected to Phase-2 (destructive zone) of HALT

(next slide).

28
 HALT test (Phase-2)
UDSL

UOSL
recycling

Temp

LOSL

LDSL start

time
In HALT testing Phase-2, stresses in the full range from LDSL (0OC) to UDSL (50OC) are
applied i.e covering the destruct range and functional range (as per figure).
The same thermal recycling, and also combination of thermal and vibration stress
can be used as in Phase-1.
29
After HALT testing Phase-2, lessons are learnt from the performance of the
product in the destruct range and functional range. Failures are noted, and the
product undergoes final modifications.

The improved design has to again pass the HALT test.

30
 HALT Test chamber

Humidifier Heating element

Electronic
Product Control unit

HALT Shaker
Test chamber Computer

HALT testing is carried out in test chambers where different stresses can be
applied iteratively. In the above figure, vibration, humidity and thermal loads
can be applied. An electronic control unit connected to a computer with software
for user interface and programming, is used to carry out the test cycles.

31
 HASS
HASS is “Highly Accelerated Stress Screening”. It is a quality control procedure
for factory produced items. High stress values are applied on the product, and
survivors are “passed”.

It is also a burn-in test to eliminate parts with high failure rate (ref. bathtub curve).

Stress cycling in HASS is carried out in the same way as HALT.

HASS is carried out in Phase-1 and Phase-2. The stress ranges are indicated below.
HASS Phase-1
Semi-destruct range

HASS Phase-2
Functional range

stress
LDSL LOSL UOSL UDSL
32
 For HASS Phase-1 the product is subjected to stresses which are mid-way
between the operational (functional) stress and destruct stress. The full destruct
stress is not applied because it may damage the component. The Phase-1
stress range is fixed such that, it is sufficient to reveal flaws in the sample, but not
too severe to damage the sample.

For HASS Phase-2, the same product is subjected to stresses in the functional
range only, to detect any faults in normal operational conditions.

Eg. Light bulbs are usually subjected to HASS before leaving the factory. For a
typical light bulb, LOSL is 150V and UOSL is 250V. The destruct voltages could be
100V and 300V. Phase-1 may be between 125V-275V. Phase-2 may be between
150-250V. Vibration stress (in g values) may be applied to reveal loose connections
or other manufacturing defects.

Decrease in Life after HASS.

Since the product is stressed beyond operational limits , its lifetime is reduced.
Hence HASS is a compromise between quality assurance and product lifetime.
An optimum HASS procedure has to be devised which will ensure quality control
and at the same time have minimum effect on product life time.

33
HASA (Highly Accelerated Stress Audit).

Once a product has been manufactured for long periods, with several improvements
in manufacturing process, materials, training etc., the quality of the product reaches a
guaranteed level. There is no need to conduct HASS on each and every component.

In HASA some 5 components out of 1000 are randomly picked and subjected to
HASS. The remaining components may be subjected to mild and quicker testing or
even non-destructive evaluation. Thus there is saving of time and labour and
improvement of the product lifetime.

34
END

35