Research Article

(wileyonlinelibrary.com) DOI: 10.1002/qre.1621 Published online 27 December 2013 in Wiley Online Library

A Tolerance Design Method for Electronic

Circuits Based on Performance Degradation
Guofu Zhai, Yuege Zhou*† and Xuerong Ye
Due to the effects of manufacturing tolerances and environmental conditions, component parameters vary and degrade
with time. This may cause performance measures of electronic circuits to deviate from design specifications. Therefore,
a tolerance design method based on performance degradation is proposed for electronic circuits, so as to improve the
robustness of output characteristics. First, sensitive components causing output fluctuation are determined via orthogonal
experiment and PSpice simulation. Then, degradation path models are established to describe the degradation process of
sensitive components. The predicted values worked out by the degradation path models are substituted into the
simulation model for Monte Carlo analysis. Besides, output characteristics and performance reliability are evaluated
according to Monte Carlo simulation. Finally, optimum allocation is carried out for component tolerances as per minimum
life cycle cost. The proposed method is illustrated by a case study of light-emitting diode (LED) driver. Copyright © 2013
John Wiley & Sons, Ltd.

Keywords: tolerance design; sensitivity analysis; performance degradation; performance reliability; life cycle cost

1. Introduction
lectronic circuit is the essential part of electronic system, and it is widely used in national defense and military, aviation and

E industry control fields. Its robustness and reliability have a direct and significant impact on the whole system. Components with
high reliability and stable performance are the foundation for a highly reliable system. In fact, due to the combined effect of
manufacturing tolerance, performance degradation and environmental conditions, performance of components always varies from
unit to unit. This directly causes performance variation of the system and affects the reliability.1–3 During the initial moments, non-
conformance of the system performance measures is defined as quality loss; the change of quality loss with time is often known
as soft failure, which indicates that the system is functional, but performance measures do not meet the design specifications any
more. However, tolerance design is an effective approach to predict and prevent the occurrence of soft failure.
Tolerance design for electronic circuits is a hot issue in both academic and industrial fields. Margherita and Giovanni2,4
discussed a tolerance design approach for the feedback compensation network of DC–DC switching converter. They
determined the tolerance ranges of crossover frequency and phase margin according to the performance and stability
constraints and obtained the feasible design solutions via Monte Carlo simulation and interval algorithm. Branas and
Azcondo5,6 studied the output power variation of resonant converters under the influence of component tolerance. The
optimum parameters of resonant network are selected by analyzing the lamp aging conditions and component tolerances.
Besides, the results are verified through Monte Carlo simulation. With the development of Electronic Design Automatic
(EDA) technology, tolerance design of electronic circuits can be realized with the assistance of EDA software. Moreover,
numerous tolerance design methods and application cases based on EDA technology have appeared, such as design for DC
hybrid contactors, driver controllers and low-pass filters.7,8
Currently, the traditional tolerance design is mainly used to improve the product quality and reduce the cost. It only focuses
on manufacturing tolerances of components, regardless of the degradation effect. In fact, during the products use period,
component degradation will also result in declination of product quality and performance, and then lead to the increase of
long-run cost (such as inspection and maintenance costs). Therefore, a tolerance design method based on component
degradation is proposed in this paper. This method tries to improve the robustness and life cycle costs of electronic circuits
by simultaneously considering component tolerance and degradation effect, and it is illustrated in detail by a case study of
LED driver. First, sensitive components are determined via sensitivity analysis based on orthogonal experiment and PSpice
simulation. The degradation path models are established to describe the performance degradation processes of sensitive
components. Then, the output characteristic and performance reliability are evaluated through Monte Carlo simulation. Finally,
according to the established tolerance allocation model, component tolerances are allocated under the condition of minimum
life cycle cost.

Harbin Institute of Technology, Harbin, China

635

*Correspondence to: Yuege Zhou, Harbin Institute of Technology, Harbin, China.

†
E-mail: [email protected]

Copyright © 2013 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2015, 31 635–643
 G. ZHAI, Y. ZHOU AND X. YE

2. Tolerance design method based on performance degradation

The method proposed in this paper covers four parts: sensitivity analysis based on orthogonal experiment, degradation path
modeling, tolerance analysis and tolerance allocation.

2.1. Sensitivity analysis based on orthogonal experiment

The purpose of sensitivity analysis is to find sensitive parameters that significantly affect target responses. In sensitivity analysis of
electric circuits, relative sensitivity is often adopted as an evaluation index. It denotes the incidence degree of component parameter
fluctuation on output characteristics. Let y = f (x1,x2, …,xn) be the function of output characteristic y and the component parameters xi
(i = 1, 2, ⋯, n). Suppose that xi0 represents the nominal value of component, and then the relative sensitivity Syxi can be expressed as

dy
y0 ∂y x i0
Sfx i ¼ ¼ ; ði ¼ 1; 2; ; nÞ (1)
dx i
x i0
∂x i y 0

Although EDA software is extensively used in circuit design, sensitivity analysis will still consume a lot of simulation time.
Especially for a complex circuit, there are usually a large number of components and multiple output responses. Thus,
orthogonal experiment method is adopted to solve the problem. The component parameters are set as influence factors in
the orthogonal table, with 1% parameter fluctuation. According to the test sequence in the orthogonal table, circuit simulation
is conducted to calculate the corresponding output characteristic. Then, the relative sensitivity of each component is calculated
via ANOVA (analysis of variance).

2.2. Degradation path modeling

There are two types of component degradation models in literatures: degradation distribution model and degradation path
model. Degradation distribution model is based on statistical law and infers distribution characteristics of degradation amount
according to degeneration data. Degradation path model presents the changing process of performance parameters with time.
In general, component degradation path model is expressed as a regression model carrying constant or random coefficients.9
For the sake of presentation, only the degradation path model is adopted in this work. Thus, a component with the
performance characteristic V under a given environmental condition C is considered, and the generalized degradation path
model can be expressed as

V ðt Þ ¼ f ðt; V 0 ; C Þ (2)

where V0 represents the actual value of all the test units at time zero, V(t = 0) = V0. Due to the effect of manufacturing tolerances, the
actual performance characteristic V0 varies from unit to unit. The initial tolerance is denoted as ΔV0.
Conventionally, random effect and stochastic process models are two commonly used degradation path models. Random
effect model takes the difference of individual performance among various components into account. Please refer to literatures
of Meeker & Escobar,10 Lu & Meeker11 and Lu et al.12 for more details. When the individual difference is not too significant,
stochastic process model considers the mean degradation path of a fixed effect, and fluctuation of degradation data is mainly
described by the stochastic process related to time. Wiener process is one of the most common stochastic process models.
Please refer to literatures of Doksum & Hóyland,13 Yu & Tseng,14 and Tseng et al.15 for specific examples.
However, the main content of this work is to introduce the proposed tolerance design method based on performance
degradation. In another word, it is to illustrate how to improve robustness of electronic circuits via component degradation
information rather than the degradation path model. For convenience of description, only the linear random effect model is
utilized for tolerance design. For example, under a certain environmental condition, the actual degradation path of V can be
described as

V ðtÞ ¼ V 0 ð1 þ CtÞ (3)

The parameter V0 is a random variable, representing the performance characteristic at the initial time. The parameter C means the
degradation rate of a particular mechanism under a specific environmental condition. And C is also a random variable to describe the
differences among various individuals in the degradation process. Figure 1 shows some degradation paths under fixed and variable
environmental conditions.
Herein, we suppose that V0 obeys the normal distribution with E(V0) = 0, and var(V0) = σ2. According to the ‘3σ’ criterion of normal
distribution, 99.73% of values will fall into the scope of ±3σ. Thus, it is reasonable to let ΔV0 be equal to 3σ from the engineering
viewpoint. Thereby, V0 can be transformed into a standard normal random variable u according to Rosenblatt transformation V0 = μ + σu.
Then, the degradation path model can be transformed into the standard normal space (known as U-space). Refer to Literature16,17
for detailed Rosenblatt transformation principle. Then, the linear degradation path model (3) after transformation is
636

V ðtÞ ¼ ðμ þ σ uÞð1 þ CtÞ (4)

Copyright © 2013 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2015, 31 635–643
G. ZHAI, Y. ZHOU AND X. YE

Figure 1. Component degradation paths. (a) Three fixed environmental conditions. (b) Various environmental conditions

2.3. Tolerance analysis

There are two tolerance analysis methods as for electronic circuits: worst-case analysis (WCA) and Monte Carlo analysis (MCA).18 In the
WCA method, the circuit performance is evaluated under the extreme case, so as to ensure robustness of circuit response at the
worst case. This approach can significantly improve the product reliability, but it will also lead to high manufacturing costs. MCA is
a statistical analysis technique which can evaluate output characteristics by way of random sampling method according to
distribution of component parameters. The analysis results are more approximate to actual situation. Hence, the Monte Carlo method
is adopted in this paper.
At the initial time, product quality fluctuation is mainly under the influence of component manufacturing tolerance. However, with
aging of components, the output characteristic y(Δ,t) is gradually affected by combined effects of component tolerance and
degradation. Therefore, influences of these two should be considered in tolerance design. The component parameter under any initial
tolerance and at any time can be worked out according to the degradation path models; by combining with PSpice simulation model,
the time-variant distribution of output characteristics can be obtained through Monte Carlo simulation. Suppose that the output
characteristic follows normal distribution, i.e. y ~ N(μ,σ2) as shown in Figure 2. Then, the performance reliability can be calculated
according to Equation (5).
y2 " #
1 ðy  μt Þ2
RðΔ; t Þ ¼ Prðy 1 < y ðΔ; tÞ < y 2 Þ ¼ ∫ pffiffiffiffiffiffiffiffiffiffi exp  dy (5)
y 1 2πσ t 2σ 2t

2.4. Tolerance allocation

Tolerance allocation is the process of allocating optimum tolerance for each component and it is an optimization problem.
Traditionally, only quality loss at the time zero is taken as the allocation criterion. However, as for the product subject to degrading
components, the degradation characteristics should also be considered. Performance degradation will result in a decrease of
performance reliability; thus, extra repair or maintenance cost will be caused. The extra repair or maintenance cost is defined as
unexpected failure cost. Thereby, life cycle cost can be adopted as the allocation criterion and then the tolerance allocation problem
will be converted into a cost optimization problem. Now, the key issue of tolerance allocation is to build the life cycle cost model
including manufacturing cost MC(Δ), quality loss cost QC(Δ) and unexpected failure cost FC(Δ). Next, they are established, respectively.

637

Figure 2. The probability density distribution and specification limits of output current

Copyright © 2013 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2015, 31 635–643
 G. ZHAI, Y. ZHOU AND X. YE

Manufacturing cost means the expense brought about by materials and processing techniques. It is directly associated with the
level of component tolerances. Decrease of tolerance will lead to the increase of manufacturing cost. There have been many models
at present, and the power exponential model C(Δ) = α Δβ is one of the most commonly used models.19 Then, the manufacturing cost
model is formulated as
n n
βi
MC ðΔÞ ¼ ∑ C i ðΔi Þ ¼ ∑ αi Δi (6)
i¼1 i¼1

The non-conformance of product quality at time t0 is defined as quality loss.11 The quality loss cost can be described via quality loss
function. For the target-the-best characteristic, the quality loss cost can be expressed as
n o  2  
2
QC ðΔ; t 0 Þ ¼ kE ½y ðΔ; t 0 Þ  y 0 2 ¼ k y ðΔ; t0 Þ  y 0 þ σ y ðΔ; t 0 Þ (7)

Where k represents the quality loss coefficient, y denotes the mean value of quality characteristic, y0 means the target value and σ 2y
expresses the variance of quality characteristic.
With aging of the system, both quality characteristic and performance reliability decrease with time. When the degradation
amount exceeds the failure threshold, the system is defined as failure. If the failure occurs before a scheduled time t, additional
maintenance or replacement will be required. And then, the unexpected failure cost will be caused. The failure cost is assumed to
be a constant fc, which is related to the repair, replacement or downtime. Then, the unexpected failure cost at the scheduled time
t can be calculated according to performance reliability.

FC ðΔ; t Þ ¼ f c ½1  RðΔ; tÞ (8)

The tolerance allocation model is established based on life cycle cost. In this model, the tolerance of each component is defined as
a design variable. The objective function of the model is the minimization of life cycle cost (TC).

Minimize TC ðΔ; t0 ; t Þ ¼ MC ðΔÞ þ QC ðΔ; t0 Þ þ FC ðΔ; tÞ

Subject to ΔL ≤ Δ ≤ ΔU
(9)
RðΔ; t0 Þ > Y 0
RðΔ; t Þ > Yt

Where R(Δ, t0) and R(Δ, t) are performance reliabilities at time t0 and t, respectively.

3. Simulation of the LED driver

3.1. Introduction to the LED driver
LED driver is a constant current source used in LED lighting; its principle is shown in Figure 3. The single flyback topology is selected to
design the LED driver. Furthermore, in order to reduce interference to the power grid, a power factor correction (PFC) circuit is added
to improve power factor.
LED converts electric energy to light energy when a forward voltage is applied across the PN junction, and the light intensity is in
direct proportion to the current. Hence, the LED driver should be designed as a constant current source. Usually, maximum luminous
efficiency point exists in high-power LED, so the current under the highest luminous efficiency is the optimum driving current. As a
result, the optimal output current of this LED driver is 700 mA. In addition, LED has a very long lifetime, about one hundred thousand
638

Figure 3. Schematic of the LED driver

Copyright © 2013 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2015, 31 635–643
G. ZHAI, Y. ZHOU AND X. YE

hours in theory. However, the LED driver cannot get such a long lifetime owing to component degradation. But we expect that the
LED driver has a lifetime of five years at least. Therefore, design specifications of the LED driver are given as follows.
(1) Output current: 700 ± 15 mA;
(2) Operating temperature: 30°C–60°C;
(3) Performance reliability: greater than 0.95 after five years.

3.2. Simulation of the LED driver

Tolerance design of LED driver is based on PSpice simulation, so the design result will be affected by accuracy of the simulation model
to a large extent. The flyback transformer, which has directly impact on the results, is the most critical component during the
simulation. Therefore, it is quite necessary to establish a simulation model of the flyback transformer which can accurately reflect
the power loss and leakage inductance characteristic. Here, the dynamic Equations (10)-(13) are used to describe behavior of the
flyback transformer and build its equivalent model. The equivalent model is presented in Figure 4.
di1
u1 ¼ R ac1 i 1 þ Ll1  e1 (10)
dt
di 2
e2 ¼ R ac2 i2 þ Ll2 þ u2 (11)
dt
dΦ dim
e1 ¼ n1 ¼ Lm þ i m Rc (12)
dt dt
dΦ n2
e2 ¼ n2 ¼ e1 (13)
dt n1
Where
Φ is the main flux
im is the magnetizing current
Lm is the magnetizing inductance
Rc is the equivalent resistance of core loss
n1, n2 is the number of primary and secondary turns
u1, u2 and i1, i2 is the voltage and current of primary and secondary side, respectively
RAC1, RAC2 and Ll1, Ll2 is the equivalent resistance and leakage inductance, respectively.

PSpice model of the LED driver is shown in Figure 5. Output characteristics of the LED driver can be obtained by simulation under
various conditions. Then, the tolerance design is conducted for the LED driver based on the PSpice simulation.

4. Tolerance design of the LED driver

In the process of tolerance design for LED driver with performance degradation, the sensitive components that lead to fluctuation of
output current are determined first. Owing to performance degradation, parameter values of sensitive components, which are
obtained by the degradation path models at time t, are substituted into the circuit simulation model. Then, Monte Carlo method is
utilized to evaluate the output current volatility and predict the performance reliability. Finally, the optimal tolerances are allocated
to the components on the basis of minimum life cycle cost.
639

Figure 4. Equivalent model of the flyback transformer

Copyright © 2013 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2015, 31 635–643
 G. ZHAI, Y. ZHOU AND X. YE

Figure 5. PSpice model of the LED driver

4.1. Sensitive components identification

There are 19 component parameters for sensitivity analysis according to the circuit diagram of LED driver. These 19 components are
defined as influence factors. Each factor is endowed with two levels: nominal value and +2% deviant. According to the orthogonal
table L20(217), 20 simulation tests should be arranged for the 19 factors with two levels. Figure 6 shows the relative sensitivity of each
factor. The components with high sensitivity are defined as sensitive components. Thereby, the voltage regulator U5 (TL431) and
resistors R15, R16, R11 and R14 are determined as sensitive components, and other components with low sensitivity are defined as
non-sensitive components.

4.2. Degradation path models of sensitive components

In this paper, parameters of sensitive components are known as random variables due to quality fluctuation. There are five variables which
are U5, R15, R16, R11 and R14 denoted as V1, V2, V3, V4 and V5, respectively. The tolerances of sensitive components ΔVi (i = 1, 2, ⋯, 5) are
defined as design variables. All the design variables are assumed to be i.i.d. normally distributed, and the following formula can be
obtained according to ‘3σ’ criterion
 
ΔV 1 ΔV 2 ΔV 3 ΔV 4 ΔV 5
p¼ ; ; ; ; ¼ ½σ 1 ; σ 2 ; σ 3 ; σ 4 ; σ 5  (14)
3 3 3 3 3

Suppose that C in the degradation model (4) obeys normal distribution, and the degradation rate ci of every component is
independent of each other. Then,
640

Figure 6. Relative sensitivity of component parameters

Copyright © 2013 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2015, 31 635–643
G. ZHAI, Y. ZHOU AND X. YE

V i ðtÞ ¼ ðμi þ σ i ui Þð1 þ ci t Þ for i ¼ 1; … 5 (15)

The nominal values of the five sensitive components are

u ¼ ½ μ1 ; μ2 ; μ3 ; μ4 ; μ5  ¼ ½2:5V; 3:9Ω; 4:7Ω; 15kΩ; 10kΩ (16)

In the researches of Johannes20 and Young KS,21 degradation rate is a constant value. In this research, the degradation rate of U5 is
assumed to be normal distribution c1 ~ N (1.0 104, 3.3 106), and the degradation rates of the resistors R15, R16, R11 and R14 are
supposed to follow the distribution of c2 = c3 = c4 = c5 ~ N (1.5 104, 1.0 105). Then, the degradation models of sensitive
components are expressed as
8
> V 1 ðt Þ ¼ ð2:5 þ σ 1 u1 Þð1 þ c1 t ÞV
>
>
>
>
< V 2 ðt Þ ¼ ð3:9 þ σ 2 u2 Þð1 þ c2 t ÞΩ
>
V 3 ðt Þ ¼ ð4:7 þ σ 3 u3 Þð1 þ c3 t ÞΩ (17)
>
>
>
> V ðt Þ ¼ ð15 þ σ u Þ ð1 þ c t ÞkΩ
>
> 4 4 4 4
:
V 5 ðt Þ ¼ ð10 þ σ 5 u5 Þð1 þ c5 t ÞkΩ

where ui(i = 1, 2, ⋯, 5) is i.i.d. standard normally distributed.

4.3. Tolerance analysis considering performance degradation

Five sensitive components are identified through sensitivity analysis, including four resistors and a voltage regulator. In practical
application, the tolerance levels can only be selected from the existing specifications on the market. Generally speaking, resistors have
six tolerance levels: 5%, 2%, 1%, 0.5%, 0.2% and 0.1%, capacitors have three tolerance levels: 5%, 10% and 20%, and the voltage
reference TL431 has three tolerance levels: 0.5%, 1% and 2%. When the tolerance levels of non-sensitive resistors are set as 5%
and non-sensitive capacitors as 20%, these tolerance levels will compose a discrete search space of tolerance design. Finally, the
optimal tolerances of all components will be selected from this discrete space.
The initial quality of output current is largely affected by the component tolerances. However, with the component aging, the
output current Iout(Δ,t) will be under combined effects of tolerance and degradation. The predicted values of sensitive
components, which are calculated according to the degradation path models, are substituted into the PSpice simulation to work
out the output current. The time-variant distribution of output current can then be obtained via Monte Carlo simulation. Here, the
component tolerances in Table I are taken as an example to illustrate the tolerance analysis. The time-variant distribution
characteristic of output current is explained in Figure 7, and both the mean value and deviation change with time in the
degradation process.
As for design specifications of the LED driver, the fluctuation of output current is less than 15 mA and the performance reliability is
higher than 0.95. Performance reliability of the LED driver is the probability for the output current to be within the specification limits
during a specific period under the stated conditions. The target value of output current is 700 mA, so the maximum and minimum of
the specification limits are 715 mA and 685 mA, respectively. Assume that the output current follows the normal distribution that is
 
Iout eN μt ; σ 2t . The performance reliability can be calculated according to Equation (18) when the specification limits are set as
y1 = 685 mA and y2 = 715 mA.
" #
y2 1 ðIout  μt Þ2
RðΔ; tÞ ¼ Prðy 1 < Iout ðΔ; tÞ < y 2 Þ ¼ ∫y1 pffiffiffiffiffiffiffiffiffiffi exp  dIout (18)
2πσ t 2σ 2t

The performance reliability is the function of component tolerances and operating time. At any moment t, the performance
reliability varies with component tolerances. The tolerance design considering performance degradation is to find the optimal
component tolerances based on minimum life cycle cost, while the performance reliability is higher than 0.95 after five years, as
shown in Figure 8.

4.4. Tolerance allocation

The optimum tolerance levels are allocated to sensitive components in order to improve the robustness of output characteristic and
minimize the life cycle cost. U5, R15, R16, R11 and R14 are defined as the sensitive components. Their tolerances are design variables in
the tolerance allocation model. Thus, the manufacturing cost associated with the tolerances can be formulated as

Table I. Tolerance of the sensitive components

Test ΔU5 ΔR15 ΔR16 ΔR11 ΔR14
641

1 2% 1% 1% 1% 1%

Copyright © 2013 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2015, 31 635–643
 G. ZHAI, Y. ZHOU AND X. YE

Figure 7. Degradation distribution characteristic of output current

1
0.9
0.8
Performance reliability

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 20 40 60 80 1.00
Time(months)

Figure 8. Performance reliability

MC ðΔÞ ¼ C ðΔU5 Þ þ C ðΔR15 Þ þ C ðΔR16 Þ þ C ðΔR11 Þ þ C ðΔR14 Þ

  (19)
¼ 1:6785 þ 0:00027  Δ1:538U5 þ 0:0058  Δ0:7107
R15 þ Δ0:7107
R16 þ Δ0:7107
R11 þ Δ0:7107
R14

The non-conformance of the output current at initial time t0 is defined as quality loss. The target value of output current is 700 mA.
The quality loss cost is described as Equation (20), and the coefficient k is assumed to be 0.5.
h i  2 
QC ðΔ; t0 Þ ¼ 0:5  E ðIout ðΔ; t0 Þ  700Þ2 ¼ 0:5  Iout ðΔ; t 0 Þ  y 0 þ ðσ Iout ðΔ; t0 ÞÞ2 (20)

The failure cost fc is assumed to be 200, and then the unexpected failure cost after five years can be derived as

FC ðΔ; t ¼ 5Þ ¼ 200  ð1  RðΔ; t ¼ 5ÞÞ (21)

The life cycle cost is the sum of manufacturing cost MC(Δ), quality loss cost QC(Δ, t0) and unexpected failure cost FC(Δ, t). The
tolerance allocation model is as follows

Minimize TC ðΔ; t0 ; tÞ ¼ MC ðΔÞ þ QC ðΔ; t 0 Þ þ FC ðΔ; t Þ

Subject to 0:1% ≤ ΔR15 ; ΔR16 ; ΔR11 ; ΔR14 ≤ 5%
0:5% ≤ ΔU5 ≤ 2% (22)
RðΔ; t0 Þ > Y 0 ¼ 1  3:4  106
RðΔ; t ¼ 5Þ > Yt ¼ 0:95

According to the six-sigma quality policy (i.e. 3.4 defects per million) at the initial time t0, Y0 is equal to 1–3.4 106; and for five
years performance reliability policy,22 Yt is 0.95. The proposed tolerance allocation model is solved via the discrete search algorithm
by adopting MATLAB software. The optimum tolerances of the sensitive components are presented in Table II. The tolerances of the
642

non-sensitive resistors and capacitors are set as 5% and 20%, respectively.

Copyright © 2013 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2015, 31 635–643
G. ZHAI, Y. ZHOU AND X. YE

Table II. Optimal tolerance allocation

Sensitive parameters ΔU5 ΔR15 ΔR16 ΔR11 ΔR14
Optimal tolerance 0.1% 0.2% 0.5% 0.2% 0.5%

5. Conclusions
A tolerance design method based on performance degradation is proposed for electronic circuits in this paper. The method tries to
improve output robustness and reduce life cycle cost by constructing connection of performance reliability with component
tolerance and degradation effect. Moreover, unexpected failure cost is introduced into the life cycle cost model, and tolerance design
model is established on the basis of performance degradation and the corresponding unexpected cost. The optimum tolerances can
then be allocated to components under the minimum life cycle cost. Finally, this method is validated by a case study of LED driver.
The proposed method provides a tool to optimize the performance reliability at the initial design. In the future, more effective and
practical results can be further acquired by utilizing more accurate degradation models associated with failure mechanism to describe
the performance degradation process of components.

References
1. Jeang A. Combined parameter and tolerance design for quality via computer experiment-A design for thermoelectric micro actuator. IEEE
Transactions on Electronics Packaging Manufacturing 2008; 31:192–201.
2. del Casale MDL, Femia N, Lamberti P, Mainardi V. Selection of optimal closed-loop controllers for DC-DC voltage regulators based on nominal and
tolerance design. IEEE Transactions on Industrial Electronics 2004; 51:840–849.
3. Savage GJ, Tong D, Carr SM. Optimal mean and tolerance allocation using conformance-based design. Quality and Reliability Engineering
International 2006; 22:445–472.
4. Petrone G, Spagnuolo G. Tolerance design of controllers for switching regulators. IEEE Transactions on Aerospace and Electronic Systems 2004;
40:661–674.
5. Branas C, Azcondo FJ, Bracho S. Desigh of LCpCs resonant inverters as a power source for HID lamp ballast applications. IEEE Transactions on Industry
Applications 2005; 41:1584–1593.
6. Branas C, Azcondo FJ, Bracho S. Study of output power variation due to component tolerances in LCsCp resonant inverters applied to HPS lamp
control. IEEE Transactions on Industrial Electronics 2004; 51:122–131.
7. Shujuan W, Youtao S, Hui Z, Guofu Z. Method of reliability tolerance design based on EDA technology and its application on DC hybrid contactor.
Journal of Zhejiang University Science A 2007; 8:426–433.
8. Junyou S, Rui K. Research on method of circuit tolerance analysis based on EDA technology. Journal of Beijing University of Aeronautics and
Astronautics 2001; 27:121–124.
9. Huang W, Dietrich DL. An alternative degradation reliability modeling approach using maximum likelihood estimation. IEEE Transactions on
Reliability 2005; 52:310–317.
10. Meeker WQ, Escobar LA. Statistical Methods for Reliability Data. New York: Wiley, 1998.
11. Lu CJ, Meeker WQ. Using degradation measures to estimate a time-to-failure distribution. Technometrics 1993; 35:161–174.
12. Lu JC, Park J, Yang Q. Statistical inference of a time-to-failure distribution derived from linear degradation data. Technometrics 1997; 39:391–400.
13. Doksum KA, Hóyland A. Model for variable-stress accelerated life testing experiments based on Wiener processes and the inverse Gaussian
distribution. Technometrics 1992; 34:74–82.
14. Tseng ST, Tang J, Ku IH. Determination of burn-in parameters and residuals life of highly reliable products. Naval Research Logistics 2003; 50:1–14.
15. Yu HF, Tseng ST. Designing a screening experiment for highly reliable products. Naval Research Logistics 2002; 49:514–526.
16. Rachwitz R, Fiessler B. Structural reliability under combined random load sequences. Computers and Structures 1978; 9:489–494.
17. Hohenbichler M, Rachwitz R. Nonnormal dependent vectors in structural reliability. Journal of the Engineering Mechanics Division 1981;
107:1227–1238.
18. Muthu P, Dhanalakshmi V, Sankaranarayanasamy K. Optimal tolerance design of assembly for minimum quality loss and manufacturing cost using
metaheuristic algorithms. International Journal of Advanced Manufacturing Technology 2009; 44:1154–1164.
19. Dong Z, Hu W, Xue D. New production cost-tolerance models for tolerance synthesis. Journal of Engineering for Industry 1994; 116:199–206.
20. Johannes A, Jaya S, Aarnout CB. A method for reliability optimization through degradation analysis and robust design. Proceeding of Annual
Reliability & Maintainability Symposium, USA, 2003.
21. Young KS, Gordon JS. A new sample based approach to predict system performance reliability. IEEE Transactions on Reliability 2008; 57:322–330.
22. Ryu, D, Chang, S. Novel concepts for reliability technology. Microelectronics Reliability 2005; 45:611–622.

Authors' biographies
Guofu Zhai received his PhD from the Harbin Institute of Technology in 1998. He is currently a Professor in the Department of
Electrical Engineering at Harbin Institute of Technology (HIT). His main research interests include electrical contacts and reliability
and testing techniques of electrical apparatus.
Yuege Zhou received his BS and MS degrees from HIT in 2009 and 2011, respectively. He is currently working toward the PhD degree
in the Department of Electrical Engineering at HIT, Harbin, China. His main research interests are reliability design and prognostic of
electronic systems.
Xuerong Ye received his BS and MS degrees from HIT in 2005 and 2009, respectively. He is currently an Associate Professor in the
Department of Electrical Engineering at HIT. His main research interests include reliability testing of relay, design for reliability,
643

prognostics, and health management for electronics.

Copyright © 2013 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2015, 31 635–643