Aalborg Universitet
A Lifetime Prediction Method for LEDs Considering Real Mission Profiles
Qu, Xiaohui; Wang, Huai; Zhan, Xiaoqing; Blaabjerg, Frede; Chung, Henry Shu-hung
Published in:
IEEE Transactions on Power Electronics
DOI (link to publication from Publisher):
10.1109/TPEL.2016.2641010
Creative Commons License
Other
Publication date:
2017
Document Version
Accepted author manuscript, peer reviewed version
Link to publication from Aalborg University
Citation for published version (APA):
Qu, X., Wang, H., Zhan, X., Blaabjerg, F., & Chung, H. S. (2017). A Lifetime Prediction Method for LEDs
Considering Real Mission Profiles. IEEE Transactions on Power Electronics, 32(11), 8718 - 8727.
https://doi.org/10.1109/TPEL.2016.2641010
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
- Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
- You may not further distribute the material or use it for any profit-making activity or commercial gain
- You may freely distribute the URL identifying the publication in the public portal -
Take down policy
If you believe that this document breaches copyright please contact us at vbn@aub.aau.dk providing details, and we will remove access to
the work immediately and investigate your claim.
Downloaded from vbn.aau.dk on: August 09, 2022
0885-8993 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2016.2641010, IEEE
Transactions on Power Electronics
IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 1, NO. 1, JANUARY 2017 1
A Lifetime Prediction Method for LEDs
Considering Real Mission Profiles
Xiaohui Qu, Member, IEEE, Huai Wang, Member, IEEE, Xiaoqing Zhan, Student Member, IEEE,
Frede Blaabjerg, Fellow, IEEE, and Henry Shu-Hung Chung, Fellow, IEEE
Abstract—The Light-Emitting Diode (LED) has become a very
promising alternative lighting source with the advantages of
longer lifetime and higher efficiency than traditional ones. The
lifetime prediction of LEDs is important to guide the LED system
designers to fulfill the design specifications and to benchmark the
cost-competitiveness of different lighting technologies. However,
the existing lifetime data released by LED manufacturers or stan-
dard organizations are usually applicable only for some specific
temperature and current levels. Significant lifetime discrepancies
may be seen in the field operations due to the varying operational
and environmental conditions during the entire service time
(i.e., mission profiles). To overcome the challenge, this paper
proposes an advanced lifetime prediction method, which takes
into account the field operation mission profiles and also the
statistical properties of the life data available from accelerated
degradation testing. The electrical and thermal characteristics of
LEDs are measured by a T3Ster system, used for the electro-
thermal modeling. It also identifies key variables (e.g., heat sink
parameters) that can be designed to achieve a specified lifetime
and reliability level. Two case studies of an indoor residential
lighting and an outdoor street lighting application are presented
to demonstrate the prediction procedures and the impact of
different mission profiles on the lifetime of LEDs.
Index Terms—LED lighting, lifetime prediction, mission pro-
file, reliability.
I. INTRODUCTION
P
OWER Light-Emitting Diodes (LEDs) are increasingly
applied for indoor and outdoor lighting applications due
to their higher efficiency and longer lifetime compared to
the traditional lighting sources. The lifetime of LED lamps
involving LED drivers and source packages is routinely quoted
as 25,000 to 50,000 hours in the market [1]–[3]. These claimed
lifetimes are usually released by the LED manufacturers or
standard organizations. However, the customer experiences
may be different and some of the LED lamps can fail in
a considerable time ahead of the claimed life [4] [5]. The
This work is supported by the National Natural Science Foundation of
China (51677027), by the Natural Science Foundation of Jiangsu Province
(BK20141339), by Fundamental Research Funds for Central Universities of
China, by the Center of Reliable Power Electronics, Aalborg University, Den-
mark, and by the Hong Kong Research Grants Council’s project (11205115).
This paper was presented in part at the IEEE Applied Power Electronics
Conference 2016, Long Beach, CA, USA.
X. Qu is with the School of Electrical Engineering, Southeast University,
Nanjing, 210096, China. (Email: xhqu@seu.edu.cn)
H. Wang and F. Blaabjerg are with Center of Reliable Power Electronics
(CORPE), Department of Energy Technology, Aalborg University, Denmark.
(Email: hwa@et.aau.dk; fbl@et.aau.dk)
X. Zhan and H. S. H. Chung are with Center for Smart Energy Conversion
and Utilization Research, City University of Hong Kong, Hong Kong. (Email:
xiaoqingzhan1011@gmail.com; eeshc@cityu.edu.hk)
failure could be induced either by the LED drivers or by the
LED source packages. The discrepancies between the claimed
lifetime and the field operation experiences are mainly due to
the following reasons [6], [7]:
1) The definition of the specified lifetime of LED lamps
is vague. A necessary lifetime definition should include
at least four aspects: a) operation conditions; b) end-
of-life criteria; c) required minimum reliability at the
end of the specified lifetime; d) confidence level of the
specified lifetime.
2) The claimed lifetime is usually tested or predicted under
a specific temperature and current level. The environ-
mental and operational conditions in field operation may
vary within the operation specifications of the LED
lamps, or even exceed the specifications for severe users.
3) The lifetime mismatch between the LED drivers and the
LED packages may occur. Sometimes, the lifetime of
LED packages is misused as the claimed lifetime of the
whole LED lamps.
The LED lamps could fail due to the following reasons: 1)
failure of LED drivers; 2) catastrophic failure of LED package;
and 3) wear out failure due to long-term lumen depreciation
and color shift [8]. The level of lumen depreciation is usu-
ally used as an end-of-life criteria. For color quality critical
applications, the color shift level is also used as an additional
criteria. Fig. 1 (a) illustrates the definition of the time to
failure L
p
of an LED individual. For example, L
70
is the
time when the lumen is maintained at 70% of its initial value.
With a more stringent requirement on lumen maintenance, the
lifetime is shortened (e.g., L
90
is less than L
70
for a specific
LED). Nowadays, the L
70
or L
85
criteria are usually used for
commercial and residential outdoor applications and L
90
is
for residential indoor applications [9]. In some applications
without the stringent lumen requirement, L
50
is also used as
a design criteria.
It is known that the L
p
lifetime varies among LED samples
even with the same part number from the same manufacturer
due to the variances in materials, process control, etc [10].
Therefore, the percentile lifetime B
X
for a population of
LEDs is of more interest with X% of failures as a result of
gradual loss of luminous flux. Fig. 1 (b) shows the definition
of B
X
lifetime based on the required minimum reliability
level R (= 1 − X%) at the end of the specified lifetime.
For example, B
10
lifetime means the time when 10% of the
LEDs fail (i.e., with a reliability R = 0.9), and B
1
lifetime
0000–0000/00$00.00
c
2017 IEEE
0885-8993 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2016.2641010, IEEE
Transactions on Power Electronics
2 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 1, NO. 1, JANUARY 2017
means the time when 1% of the LEDs fail. Accordingly, L
p
B
X
lifetime refers to the time when X% of the LEDs have the
lumen output below p% with respect to their initial values.
The choices of p and X are application-dependent. L
p
B
X
lifetime is more legitimate to declare the lifespan of the LED
package [11]. It is also applicable for LED drivers to evaluate
the reliability level. The reliability curve can be plotted using
these L
p
data arranged by a specific rank method to define
the cumulative percentage of the population. Among different
data rank methods as discussed in [12], the median rank is
corresponding to a confidence level of 50%. It is also possible
to obtain the reliability range with certain Confidence Bounds
(CBs) as shown in Fig. 1 (b) with other data rank methods. For
example, the 2-sided 90% CBs have the top CB and the bottom
CB curve to provide 5% and 95% confidence respectively.
These statistical properties are necessary to define the lifetime
of LED lamps.
Degradation testing is usually performed to obtain the time
to failure L
p
of each individual LED sample. The industry
standard LES LM-80 [13] requires a minimum of 6,000 hours
of degradation testing. LED manufacturers usually conduct
the test for 6,000 hours up to 10,000 hours. Based on the
available lumen degradation data, the time-to-failure of each
testing sample is projected by an exponential curve-fitting
extrapolation as described in the standard IES TM-21 [14].
However, TM-21 uses the average degradation value of the
LED samples for the further projection, which ignores the sta-
tistical properties and therefore the reliability information can
not be obtained from the TM-21 procedure. The degradation
testing presented in IES LM-80 is usually performed under
several specific conditions, that are typical constant driving
currents and at least three cases of ambient temperatures
(55
◦
C, 85
◦
C and one selected by the manufacturers). The
driving current depends on the user profiles and driving
schemes. The ambient temperature may vary with time and
is geographically dependent. Only one constant current and
temperature based reliability prediction method can not take
into account a realistic mission profile with loading variations
[5], [15]. Therefore, there are still gaps between the degrada-
tion testing data and the practical applications like:
1) The specific reliability information (with a certain confi-
dence level or confidence bounds) and the corresponding
lifetime model are not readily available. A comprehen-
sive analysis on those testing data is needed.
2) The mapping of the reliability under the specific accel-
erated testing conditions to under field conditions (i.e.,
long-term mission profiles) is missing.
To overcome the above issues, this paper proposes an
advanced lifetime prediction method concerning the long-term
field operation mission profiles and the statistical properties
of the life data available from the accelerated degradation
testing. The mission profile dependent lifetime models has
been analyzed in the conference paper [16], based on the
degradation testing data. In this paper, the electrical and
thermal characteristics of LEDs are experimentally measured
by a Thermal Transient Tester (T3Ster) system, used for the
electro-thermal modeling. More temperature steps are used to
Φ
t
90%
1
70%
90
L
70
L
p%
p
L
Lumen Flux
Degradation
(a) Time to failure L
p
of an individual LED
R
t
0.99
0.9
1
B
10
B
1-X%
X
B
0
Reliability Curve
1
Top Confidence Bound
Bottom
Confidence Bound
(b) B
X
lifetime for a population of LEDs
Fig. 1. Two LED lifetime criteria, where (a) L
p
is defined as the time
when p% of the initial output lumen of an LED is maintained, and (b) B
X
is defined as the time when X% of the LEDs have the lumen output below
p% of their initial values.
obtain the temperature-dependent electro-thermal parameters.
A feedback implementation system of the junction temperature
to update the electro-thermal parameters is built to acquire the
operation point for the accurate lifetime prediction. With the
improved electro-thermal models and the lifetime prediction
method, some key variables for thermal design and lifetime-
matching of LED drivers in different field conditions can easily
be identified to achieve a specified lifetime and reliability. Two
case studies of an indoor residential lighting application and an
outdoor street lighting application are presented to demonstrate
the prediction procedures and the impact of different mission
profiles on the lifetime of LEDs. The proposed method can
also be extended to the prediction of the LED drivers and the
entire LED lighting systems.
Specifically, Section II introduces the comprehensive life-
time models involving the statistical properties and the re-
liability information of life data. Based on these models, an
advanced lifetime prediction method is then presented in detail
to map the lifetime from the testing condition to the field
condition in Section III. Two case studies are demonstrated and
evaluate the performance of the proposed method in Section
IV. Section V concludes the paper.
II. LIFETIME MODELS AND DEGRADATION TESTING DATA
ANALYSIS
Since LEDs are basically p-n junctions, the emitted lumen
flux and intensity are proportional to the concentration of
carriers [17]. The concentration of carriers depends on the
current density and junction temperature, which results in
LED output lumen, color chromaticity, and the forward voltage
characteristics also varying with these two stresses. Hence, a
0885-8993 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2016.2641010, IEEE
Transactions on Power Electronics
QU et al.:A LIFETIME PREDICTION METHOD FOR LEDS CONSIDERING REAL MISSION PROFILES 3
generally accepted Black model in (1) is used to describe the
Time to failure under different stresses [18], [19].
Time to failure = A
0
J
−n
e
E
a
k
B
T
, (1)
where A
0
is a constant, J is the current density, n is a scaling
factor, E
a
is the activation energy in unit of eV, k
B
is the
Boltzmann constant, and T is the absolute temperature in
Kelvin.
The model in (1) describes the impact of current and
temperature on the lifetime of LEDs. Therefore, L
p
lifetime,
defined as time to failure for an LED individual, follows
this model. Moreover, B
X
lifetime based on a population
of L
p
lifetime data also follows this equation to specify the
reliability of an LED population. The parameters of A
0
, n
and E
a
are usually obtained according to the accelerated
testing data. n and E
a
are material-dependent, which can be
assumed constant for a given type of LEDs with a given failure
mechanism. Hence, (1) can be rearranged as (2) and (3).
L
p
(I
F
, T
J
) = A
p
I
−n
F
e
E
a
k
B
T
J
, and (2)
B
X
(I
F
, T
J
) = A
X
I
−n
F
e
E
a
k
B
T
J
, (3)
where I
F
is the LED driving current proportional to the current
density, and T
J
is the junction temperature of LEDs. Although
A
p
and A
X
are dependent on the different L
p
and B
X
criteria,
(2) and (3) have the same Acceleration Factors (AF).
AF(n, E
a
) =
(
I
F
I
F 0
)
−n
e
E
a
k
B
(
1
T
J
−
1
T
J0
)
. (4)
Here, (I
F 0
, T
J0
) is the initial stress level, whilst (I
F
, T
J
) is
the accelerated stress level. To solve factors of n and E
a
in
(4), the time-to-failure data from at least three different stress
levels are required.
With this information, a case study based on an LM-80
test report [20] for Lumileds Luxeon Rebel LEDs [21] will
show how to establish the models of (2) and (3), where the
data in the LM-80 report are experimentally measured by the
manufacturers. Weibull distribution is the most widely used
to process the lifetime data in reliability engineering [12],
which is adopted here to analyze the reliability information
of LEDs. The report [20] provides multiple accelerated life
testing conditions with stress levels of I
F
from 0.35 A, 0.5
A, 0.7 A, to 1 A and air temperature T
a
from 55
◦
C, 85
◦
C,
105
◦
C, to 120
◦
C. There are 25 samples in each test to ensure
the accurateness of the results, lasting for at least 9,000 hours.
To solve n and E
a
in (4), at least three different stress levels
of I
F
and T
J
are randomly chosen. Here, the degradation
data and fitted curves at three stress levels of I
F
and T
J
with (0.35 A, 129
◦
C), (0.7 A, 74
◦
C) and (1 A, 112
◦
C) are
plotted by software tool ReliaSoft [22] and shown in Fig. 2.
The data points are provided by LM-80 report, which are
measured every 1,000 hour for 25 samples in the accelerated
testing. Then the fitting curves are projected by an exponential
extrapolation according to TM-21 procedure. With two end-
of-life criteria L
70
and L
90
, two groups of end-of-life L
p
data
can be read directly in Fig. 2. It should be noted that TM-21
uses the average lumen value of the samples for the further
projection, which ignores the statistical properties and provides
70
L
90
L
3000 10000 100000
1.10
1.00
0.06
Normalized Output Lumen
Degradation vs. Time
Degradation
Data Points
200000
Time (Hr)
(a) I
F
=0.35 A, T
a
=120
◦
C, and T
J
=129
◦
C
70
L
90
L
3000 10000 100000
1.10
1.00
0.06
Degradation vs. Time
D
egradation
Data Points
300000
Time (Hr)
Normalized Output Lumen
(b) I
F
=0.7 A, T
a
=55
◦
C, and T
J
=74
◦
C
70
L
90
L
3000 10000 100000
1.10
1.00
0.06
Degradation vs. Time
D
egradation
Data Points
Time (Hr)
Normalized Output Lumen
(c) I
F
=1 A, T
a
=85
◦
C, and T
J
=112
◦
C
Fig. 2. Lumen degradation curves with L
70
and L
90
lifetime criteria under
different LED operating conditions of (a), (b) and (c). X-axis is time in hour
of logarithmic scale and Y-axis is the normalized output lumen with respect
to the initial lumen at t=0.
limited information. Therefore, the long-term lumen projection
is done for each sample in the proposed method.
Each group of L
p
data is then arranged in sequence and
ranked by the algebraic approximation of the Median rank in
(5) [12].
Median rank r
j
=
j − 0.3
N + 0.4
, (5)
where j is the order number of the sequenced L
p
data,
j ∈ [1, N ] and N is the total number of failure (i.e., the size of
L
p
data). The rank r
j
is actually the probability to failure for
0885-8993 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2016.2641010, IEEE
Transactions on Power Electronics
4 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 1, NO. 1, JANUARY 2017
the j
th
LED. With these ranks and the corresponding L
p
group
at one stress level, the probability to failure line for this stress
level can be generated via ReliaSoft ALTA (Accelerated Life
Testing Analysis) degradation. Figs. 3 (a) and (b) illustrate the
unreliability function F (t) (i.e., probability to failure function)
at each operating stress level with 50% confidence level. The
unreliability curves with the different confidence levels could
also be plotted upon the application requirements. In Figs. 3
(a) and (b), most data points of the three stress groups are
fitted to the Weibull distribution reasonable well. Few data
points outside of the probability lines due to the measurement
error or LED sample variation can be dismissed here. With
the three probability lines, B
X
satisfying F (B
X
) = X% can
be obtained. The probability lines follow the two-parameter
Weibull distribution and the cumulative failure F (t) is de-
scribed as
F (t) = 1 − R(t) = 1 − e
−(
t
η
)
β
, (6)
where t is time, β is the shape parameter, and η is the scale
parameter of characteristic life B
63.2
(i.e., the life at which
63.2% of the tested samples fail) at each stress condition. For
the wear-out failure, β > 1. With the same failure mechanism,
β is assumed constant under different stress levels within the
physical limits [12]. In Figs. 3 (a) and (b), six well fitted
curves show good consistence on β, n and E
a
, where the
discrepancies are caused by the distribution variation. Besides,
the probability lines under the different stress levels can be
readily plotted in the figure with the same β, n, E
a
and
different η, such as two lines at the stress level of I
F
and
T
J
with (0.7 A, 25
◦
C, i.e., 298 K) in Figs. 3 (a) and (b)
seperately.
With the known n and E
a
, substituting any B
X
value at one
stress (I
F
, T
J
) into (3), A
X
can be solved. Two groups of n
and E
a
are derived from the experimental accelerated testing
data, and can be validated by the good consistence. Thus, any
percentile B
X
lifetime with different failure rate X% can be
derived using the above mentioned method. Here, B
10
and B
1
distributions based on L
70
and L
90
criteria are given in (7) and
Fig. 4 as examples, which are valuable for further mapping the
reliability information under field operational mission profiles,
and discussed in the next section.
ln B
10 L70
(I
F
, T
J
) = 3.956 − 0.57 ln I
F
+
2588
T
J
ln B
1 L70
(I
F
, T
J
) = 3.628 − 0.57 ln I
F
+
2588
T
J
ln B
10 L90
(I
F
, T
J
) = 2.558 − 0.698 ln I
F
+
2636
T
J
ln B
1 L90
(I
F
, T
J
) = 2.221 − 0.698 ln I
F
+
2636
T
J
(7)
III. EXPERIMENTAL CHARACTERIZATION OF
ELECTRO-THERMAL PROPERTIES OF LEDS AND MISSION
PROFILE BASED LIFETIME PREDICTION
From Fig. 4, I
F
and T
J
are the key factors to predict the
lifetime and reliability in the LED lighting applications. In
the field operation, the LED driving current I
F
depends on
7.16
β
=
0.223
a
E
=
0.57
n
=
10000 100000
99
50
Unreliability, F(t)=1-R(t)
1000000
Time (Hr)
10
5
1
Probability-Weibull
S
tress Level:
(K) (A)
J F
T I
298 0.7
347 0.7
402 0.35
385 1
Data Points
Probability Line
X
X
B
η
63.2
(a) With L
70
criteria
6.96
β
=
0.227
a
E
=
0.698
n
=
10000 100000
99
50
Unreliability, F(t)=1-R(t)
Data Points
1000000
Time (Hr)
10
5
1
Probability-Weibull
Probability
Line
Stress Level:
(K) (A)
J F
T I
298 0.7
347 0.7
402 0.35
385 1
(b) With L
90
criteria
Fig. 3. Unreliability curves under different stress levels based on the L
p
data
given in Fig. 2. X-axis is time in hour of ln(t) scale and Y-axis is percentage
in the scale of
ln ln(
1
1−F (t)
).
(A)
F
I
(K)
J
T
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
300
320
340
360
380
400
(hour)
X
B
6
10
5
10
4
10
3
10
Distribution
X
B
10 _L 70
B
1_L70
B
1_L90
B
10 _L 90
B
Fig. 4. B
X
lifetime vs. forward current I
F
and junction temperature T
J
.
the user profiles (e.g., indoor or outdoor occasion for dif-
ferent lumen requirements, dimming schemes, and periodical
operational hours per day, month or year, etc.). The junction
temperature T
J
, which is affected by the ambient temperature
T
A
, power loss in chip, and thermal distribution of materials,
