The Emerald Research Register for this journal is available at The current issue and full text archive

hive of this journal is available at

www.emeraldinsight.com/researchregister www.emeraldinsight.com/0265-671X.htm

FMECA of
Reliability prediction and failure electronic
mode effects and criticality devices
analysis (FMECA) of electronic
183
devices using fuzzy logic
Received 29 January 2004
E.P. Zafiropoulos and E.N. Dialynas Revised 17 May 2004
School of Electrical and Computer Engineering, Accepted 21 May 2004
National Technical University of Athens, Athens, Greece

Abstract
Purpose – The paper presents an efficient methodology that was developed for the reliability
prediction and the failure mode effects and criticality analysis (FMECA) of electronic devices using
fuzzy logic.
Design/methodology/approach – The reliability prediction is based on the general features and
characteristics of the MIL-HDBK-217FN2 technical document and a derating plan for the system
design is developed in order to maintain low components’ failure rates. These failure rates are used in
the FMECA, which uses fuzzy sets to represent the respective parameters. A fuzzy failure mode risk
index is introduced that gives priority to the criticality of the components for the system operation,
while a knowledge base is developed to identify the rules governing the fuzzy inputs and output. The
fuzzy inference module is Mamdani type and uses the min-max implication-aggregation.
Findings – A typical power electronic device such as a switched mode power supply was analyzed
and the appropriate reliability indices were estimated using the stress factors of the derating plan. The
fuzzy failure mode risk indices were calculated and compared with the respective indices calculated by
the conventional FMECA.
Research limitations/implications – Further research efforts are needed for the application of
fuzzy modeling techniques in the area of reliability assessment of electronic devices. These research
efforts can be concentrated in certain applications that have practical value.
Practical implications – Practical applications can use a fuzzy FMECA modeling instead of the
classical FMECA one, in order to obtain a more accurate analysis.
Originality/value – Fuzzy modeling of FMECA is described which can calculate fuzzy failure mode
risk indices.
Keywords Failure modes and effects analysis, Fuzzy logic, Electronic equipment and components
Paper type Research paper

Introduction
In recent years, due to the growing complexity and the high cost incurred by loss of
operation and maintenance actions, the reliability assurance activities are becoming
established as an integral part of the design process of engineering systems. Reliability
assessment provides insight into the key areas of a system and highlights potential
problem areas so that they can be dealt at the design stage of the system life cycle International Journal of Quality &
(LaCombe, 1999). Furthermore, it allows comparisons to be made among the competing Reliability Management
Vol. 22 No. 2, 2005
system designs, it provides a deeper understanding of their construction and operation pp. 183-200
and, as a result, it helps to improve system safety. Major tasks and decisions in system q Emerald Group Publishing Limited
0265-671X
design are affected by the prediction of the system reliability indices, such as the DOI 10.1108/02656710510577233
IJQRM failure rate or the mean time between failures (MTBF). These include the feasibility
22,2 evaluation where the compatibility of a design is weighed against the design reliability
requirements for acceptance, the design comparison where different parts of a system
can be compared and other necessary studies concerning cost, reliability, weight, etc.
(Birolini, 2000). Furthermore, the failure mode effects and criticality analysis (FMECA)
is a design technique to systematically identify and investigate potential system
184 (product or process) weaknesses (Teng and Ho, 1996; Hawkins and Woollons, 1998;
Department of Defense, 1980). It consists of a methodology for examining all the ways
in which a system failure can occur, potential effect(s) of failures on system
performance and safety, and the seriousness of these effects. The output from a
FMECA, conducted in a timely manner, provides invaluable input to the development
of a cost-effective preventive maintenance program and the tailoring of a focused
control plan.
However, conventional FMECA techniques still impose some limitations on
problem solving (LaCombe, 1999; Lee, 2001; Price and Taylor, 2002). The probability
of failure events is difficult to be precisely calculated and the majority of components
exhibit several failure modes. This information can be estimated using the related
data in the reliability databases, applying reliability prediction techniques or they can
even be expressed as linguistic variables (e.g. “high”, “critical”). The diversity and
inaccuracy of these data sources prohibit the comparison of technical expertise
among different engineering projects. Furthermore, the results are sensitive to the
methods being applied to estimate the failure event data and to the qualitative
judgment of the linguistic variables. Consequently, the validity of the results may be
questionable.
The purpose of this paper is to present a reliability analysis methodology, which
consists of a reliability prediction method with a proposed stress derating plan and a
FMECA method using fuzzy sets. The reliability prediction method is based on the
MIL-HDBK-217 reliability prediction procedure and provides estimations of the
components and device failure rates using a certain stress derating plan being
developed for the components of an electronic device. The reliability prediction
calculations can be executed using specialized reliability prediction software, such as
Relex 7 (Relex Software Corporation, 1998). The calculated reliability indices are used
as an input to the subsequent FMECA method, which incorporates several features of
the MIL-HDBK-1629A guideline for FMECA and the reliability database FMD91 for
failure modes (Lee, 2001; IIT Research Institute/Reliability Analysis Center, 1991).
Furthermore, a fuzzy assessment system architecture of FMECA was developed
while it was implemented into an efficient computer program using the fuzzy logic
toolbox platform of Matlab 6.1. The developed reliability analysis methodology was
applied to a typical power electronic device, such as a switched mode power supply
(SMPS), and the obtained results for the components’ failure modes criticality for the
device operation were compared with the respective results of the conventional
FMECA.

Reliability prediction of electronic devices

In order to estimate the reliability indices of an electronic device, a reliability engineer
in cooperation with a design engineer should interpret the functional block diagram
and the drawings of the device into a reliability block diagram (Teng and Ho, 1996;
Pukite and Pukite, 1998). Each component or assembly is represented in the reliability FMECA of
block diagram as a box with two states (operation or failure) and their interconnection electronic
depends on their contribution to the required system operation and on the assumptions
made by the reliability and design engineers. It is assumed that the components have a devices
constant failure rate, which means that the component reliability has an exponential
distribution over time. Furthermore, it is assumed that all the components and
assemblies are required to be in their operating state so that the system is operating. If 185
any component fails, the system fails. This is represented as a series connection of the
components in the reliability block diagram. These assumptions are expressed
mathematically with the following equations:


Pn
Y
n Yn li t
Rsys ðtÞ ¼ Ri ðtÞ ¼ e li t ¼ e i¼1 ¼ e lsys t ; ð1Þ
i¼1 i¼1

X
n
lsys ¼ li ; ð2Þ
i¼1

1
MTBF sys ¼ ; ð3Þ
lsys
where: t ¼ time of interest; Ri ðtÞ ¼ reliability of component i in time t; li ¼ failure rate
of component i; lsys ¼ system failure rate; Rsys ðtÞ ¼ system reliability in time t.
The reliability prediction stage of the reliability analysis for an electronic device is
briefly shown in Figure 1. It can be noticed that there are two main steps concerning

Figure 1.
Flowchart showing the
basic steps of the
reliability prediction
methodology
IJQRM the reliability prediction activities which are included in the phases of parts count and
22,2 part stress methods respectively (shown in bold in Figure 1). During the early design
phase of electronic devices, insufficient information is available for the exact circuit
configuration of the devices and the electrical or thermal stresses on the components. In
this case, the most applicable reliability prediction procedure is the parts count method
of the MIL-HDBK-217FN2 document (Department of Defense, 1991). The necessary
186 information in order to apply this method is the generic part types (including
complexity for microcircuits and quantities), the part quality levels and the equipment
environment. The following mathematical expression can be used for the estimation of
the system failure rate as an extension of equation (3):

X
n
lsys ¼ N i lgi pQi ; ð4Þ
i¼1

where: lsys¼ system failure rate; N ¼ number of different generic part categories for
the system; N i ¼ quantity of the ith generic part; lgi ¼ generic failure rate for the ith
generic part; pQi ¼ quality factor for the ith generic part.
Equation (5) is applied when the entire device is being used in one environment. If
the device comprises of several units operating in different environments (such as in
avionic systems with units in “airborne inhabited” AI and “airborne uninhabited” AU
environments), equation (5) should be applied to the portions of the device in each
environment (Department of Defense, 1991). These “environment-equipment” failure
rates should be added to estimate the total device failure rate. Additionally, in the cases
where the component manufacturer provides failure rate estimation in the datasheets,
this estimation can be used instead of the parts count model of the component. In some
cases, where there are no models in MIL-HDBK-217FN2 for certain components, other
reliability databases can be used, such as the NPRD-95 document (Reliability Analysis
Center, 1995).
When the procedure of reliability prediction is completed, the device reliability
specifications are reviewed in order to identify any significant deviations. If such
deviations exist, appropriate corrective actions must be decided and implemented
appropriately. There are quite few possible actions that can be made in order to
improve the device’s overall reliability performance. These actions include mainly the
selection of higher quality components or different component types, which, in most
cases, may increase the total cost. If there are no significant deviations to the device’s
expected reliability indices, the next stage of the detailed system design is initiated.
Generally, it is quite advisable to decide any appropriate corrective actions in a later
stage, when more information about system operation and reliability are available.
When most of the detailed design phase is completed and a detailed parts list
including parts’ stresses is determined, the part stress method of MIL-HDBK-217FN2
can be applied (Department of Defense, 1991). This method contains analytical models
for a broad variety of parts being used in electronic devices. These models incorporate
the effects of electrical and thermal stresses, which largely influence the failure rate of
electronic components. Therefore, derating these stresses during the detailed design
phase is mandatory in order to improve the inherent device reliability. The design
engineer of an electronic device should have a set of derating standards available for
use when selecting components for a specific application (LaCombe, 1999; Lalli et al.,
2000). The derating standards may vary from company to company, depending on FMECA of
their experience with specific components in their applications, but each company electronic
should have a well-defined set of standards to be used by all design engineers (derating
plan). For this purpose, a certain derating plan is incorporated in the developed devices
reliability prediction methodology and it is shown in Table I. The numerical values
correspond to the upper limits of the respective type of stresses being applied to the
corresponding components. More specifically, the numerical values refer to the ratio of 187
the actual value for the operating characteristic and the nominal value specified in the
component’ s documentation (Birolini, 2000).
The mathematical models of the part stress method of MIL-HDBK-217FN2
document differ significantly depending on the specific component category since they
incorporate the effects of the environment, the component quality, the electrical and
thermal stresses as well as the specific technology of the respective components.
Therefore, the implementation of an explicit derating plan during the design phase
facilitates the reliability prediction procedure while it improves considerably the field
reliability index of the device. Similarly to the parts count method, the respective
manufacturers’ estimated failure rates are used when such data are available. Finally,

Stress type
Component Power Voltage Current Internal temperature

Resistors
Fixed 0.6 0.8
Variable 0.6 0.7
Thermistors 0.4 0.7
Capacitors
Film, ceramic 0.5 0.5
Ta (solid) 0.5 0.5
Al (wet) 0.8 0.5
Diodes
General purpose 0.5a 0.6 0.7
Zener 0.6 0.7
Transistors 0.5a 0.7 0.7
a
Thyristors, triacs 0.6 0.6 0.7
Optoelectronic devices 0.5b 0.5 0.8
ICS
Linear 0.8c 0.7d
Voltage regulator 0.7c 0.7d
Digital bipolar 0.8c 0.7d
Digital MOS 0.7 0.8c 0.7d Table I.
Derating plan applied for
Coils, transformers 0.5 the main electronic
Switches, relays 0.4-0.7e 0.7 components
(recommended values at
Connectors 0.7 0.6 0.8 ambient temperature
Notes: a breakdown voltage; b isolation voltage (0.7 for Vin); c sink current; d uJ # 100o C; e low values uA # 40o C) in electronic
for inductive loads devices
IJQRM other reliability databases can be used as a complementary tool, such as the NPRD-95
22,2 document (Pukite and Pukite, 1998).
If the obtained reliability indices do not conform to the reliability requirements of
the respective device specifications, the design and reliability engineers should
implement certain corrective actions to the design parameters of the device so that the
model exhibits better reliability indices. The design engineers may improve the system
188 reliability by reducing the electrical and thermal stresses being applied on the
components with modifications in the circuit design. Furthermore, the failure rate
estimation by using the part stress model is specialized to many component categories
and different technologies and the design engineers may select the appropriate
components with favorable failure rates. When the device reliability specifications are
satisfied, the calculated reliability indices of the components may be used for
conducting a FMECA of the device while other additional reliability and
maintainability activities can also be conducted.

Basic features of the conventional FMECA

FMECA is a reliability task performed during the design phase of an engineering
system in order to identify individual component failure modes, their causes and their
effects on system operation. The term “item” is used in the conventional FMECA and
means a functional or structural unit which is considered as an entity for
investigations. It may consist of hardware, software or human resources, since FMECA
may be implemented in any type of engineering system. The developed methodology
uses the term “item” for identifying the system components at the lowest indenture
level. The term “indenture levels” corresponds to the different levels of complexity
which describe relative complexity of system assembly or function. The FMECA is a
bottom-up analysis in a spreadsheet form and can be performed at a number of
different levels of detail. It can be used to determine the most critical failure modes and
determine how these critical failures can be minimized or eliminated. The best-known
FMECA practices are outlined in the guidelines SAEARP5580, MIL-HDBK-1629A and
IEC 60812 (1985-07) (Department of Defense, 1980; Society of Automotive Engineers,
2001; International Electrotechnical Commission, 2003). These procedures have similar
basic steps but propose different indices and frameworks for prioritization of failure
modes.
The developed methodology of a fuzzy-logic-based FMECA derives several features
of the conventional FMECA described in MIL-HDBK-1629A standard. These features
are the framework for the spreadsheet representation of the failure events data and the
indices that allow the ranking of failure modes. The basic steps of a conventional
FMECA are the following:
(1) System definition. It includes identification of internal and interface functions,
expected performance at all indenture levels, system restraints and failure
definition.
(2) Functional analysis. It illustrates the operation, interrelationships and
interdependencies of functional entities.
(3) Identification of possible failure modes and effects. All potential item and
interface failure modes are identified and their effect on the immediate function
or item and on the system are clearly defined. In the present application, the
 component failure modes data of the FMD91 have been used (IIT Research FMECA of
Institute/Reliability Analysis Center, 1991). electronic
(4) Severity rating of failure modes. It refers to the seriousness of the effect or devices
impact of a particular failure mode while the following classification of failure
effects was used:
.
Catastrophic: a failure that can result in the loss of personal life and/or
complete system loss; 189
.
Critical: a failure that can potentially cause serious personal injury and/or
significant personal damage and loss of system functionality;
.
Marginal: a failure that can result in minor injury and/or minor system
damage and degradation of system functionality; and
.
Minor: a failure not severe enough to cause any personal injury or system
damage, but resulting in the requirement for some corrective maintenance.
(5) Frequency of occurrence rating of failure modes and analysis of failure mode
criticality. Assuming that a system component is likely to fail in multiple ways,
this information is used to delineate the most critical aspects of the system
design. The reliability database FMD91 suggests for each component category
the percentage values of the occurrences of each failure mode to the total
number of failures of the components in that category. This index is referred in
the FMECA as failure mode ratio a. The failure rate li for each component can
be estimated by a reliability prediction methodology, which was described
previously. The probability that the failure mode will affect the system, given
that the failure mode occurs, should be quantified using engineering expertise.
This index is referred in the FMECA as failure effect probability b and is
classified to the following categories: “actual loss of operation” (b ¼ 1),
“probable loss” (0:1 , b , 1:0), “possible loss” (0 , b , 0:1), “no effect”
(b ¼ 0). Using the previous data, the failure mode criticality index Cm can be
calculated as:

C mj ¼ bj ·aj ·li ·t; ð5Þ

where bj, aj are the failure effect probability and the failure mode ratio
respectively for a failure mode j of a component i that exhibit a failure rate li,
while t is the total operating time. If a clear mission time does not exist, the
parameter t can be neglected from this equation. The criticality indices Cis of
component i are calculated by summating all the relevant failure mode
criticality indices, which have particular severity classification as:

X
n
C is ¼ ðbj ·aj ·li ·tÞ; ð6Þ
j¼1

where component i has n failure modes under a specific severity classification s

(s ¼ “catastrophic”, “critical”, “marginal”, “minor”).
IJQRM Modeling of FMECA using fuzzy logic
22,2 FMECA is a bottom-up analysis technique starting with known failure modes at one
level and investigating the effect on the next subsystem level. A complete analysis
must span all the levels from components to the entire system (Department of Defense,
1980; Bell et al., 1992). The criticality ranking of the components’ failure modes can be
performed using quantitative methods described in relevant standards or engineering
190 judgment (Department of Defense, 1980; Department of Defense, 1991; Society of
Automotive Engineers, 2001; International Electrotechnical Commission, 2003).
Therefore, the results may be highly dependent on the engineering judgment or the
methods being used to estimate indices such as the failure mode ratios or the
component failure rates. It is assumed that there is not a clear mission time for the
calculation of the failure modes’ criticality indices.
A fuzzy assessment model for performing a FMECA was developed and it was
integrated with the conventional FMECA methodology based on MIL-HDBK-1629 in
order to overcome the difficulty in sharing information among experts from various
disciplines and the dependence of the FMECA results on the various methods of
estimating all the indices being used. For this purpose, the information being used to
calculate the failure modes criticality indices is represented as fuzzy variables (Vas,
1999; Park, 1987). The information needed to calculate the failure modes criticality
indices Cm (equation (5)) is the failure effect probability b, the failure mode ratio a and
the component failure rate l while the mission time t is neglected.
The failure mode effect severity is fuzzified to a fuzzy variable, which is shown in
Figure 2. The triangular membership function is used to represent the failure mode
effect severity as fuzzy variable and the overlaps between adjacent membership
functions allow for a smooth interpolation of the inputs across membership functions.
Furthermore, the qualitative assessment of the severity can be expressed more
effectively since, in the cases where a strict severity classification cannot be performed,
real values in the interval of 0 to 3 can be used as inputs in the fuzzy inference module.
The severity rating of the failure modes (minor, marginal, critical and catastrophic) can
be represented with the crisp values 0, 1, 2, and 3 respectively, in order to compare the
results of the conventional and proposed fuzzy FMECA. Accordingly, the failure effect
probability b and the failure mode ratio a are fuzzified to fuzzy variables as it is shown
in Figures 3 and 4 respectively.
The fuzzification procedure of the component failure rates is slightly more
complicated. It is assumed that the failure rates of components should not be higher

Figure 2.
Fuzzy sets definition for
the failure effect severity
than 50 percent of the total system failure rate. Otherwise, corrective actions should be FMECA of
applied in order to improve the respective component failure rate. Given this electronic
assumption, a normalized component failure rate lni is calculated as:
devices
li
lni ¼ 2· ; ð7Þ
lsystem
191
where li, lsystem are the component and the system failure rates respectively. The
normalized component failure rate is the crisp value used in the fuzzy FMECA
replacing the failure rate used in the conventional FMECA and it is fuzzified to a fuzzy
variable as it is shown in Figure 5. The fuzzy variable of the normalized failure rate can
express qualitative estimations of component failure rate in accordance with estimated
system failure rate. Therefore, this fuzzy assessment system of FMECA can be applied
regardless of the method being used to estimate component failure rate. However,
further consideration should be made for the definition of equation (7) when there are
certain redundancies in the system reliability modeling.
The variables shown in Figures 2-5 correspond to the inputs of the fuzzy inference
module. The output of the fuzzy inference module is an index designated as failure
mode risk index and it incorporates the crisp criticality number calculated by the
conventional FMECA and the linguistic variable of severity of the failure effect, which
is not used in the conventional FMECA to calculate the failure mode criticality. The
fuzzy variable of the failure mode risk index is shown in Figure 6. The failure mode

Figure 3.
Fuzzy sets definition for
the failure effect
probability b

Figure 4.
Fuzzy sets definition for
the failure mode ratio a
IJQRM
22,2

192
Figure 5.
Fuzzy sets definition for
normalized component
failure rate

Figure 6.
Fuzzy sets definition for
the failure mode risk index

risk index RFMi of component i can be interpreted as a priority index for corrective
actions to system design. Another index that can prioritize corrective actions is the
item risk index Ri, which is introduced as the mean value of the failure mode risk
indices RFMi.
The expert’s knowledge about the influence of the fuzzy input variables of
Figures 2-5 to the item criticality for the system operation, expressed by the fuzzy risk
index of Figure 6, can be represented in the form of if-then rules. These rules comprise
the knowledge base of the fuzzy inference module. A single fuzzy if-then rule assumes
the form:
If x is A then y is B ðzÞ;
where A and B are linguistic values defined by fuzzy sets on the ranges (universes of
discourse) X and Y respectively. The if-part of the rule “x is A” is called the antecedent
or premise, while the then-part of the rule “y is B’ is called the consequent or conclusion.
It must be noted that the antecedent is an interpretation that returns a single number
between 0 and 1, whereas the consequent is an assignment that assigns the entire fuzzy
set B to the output variable y. The number (z) in the parentheses above represents a
weight factor between zero and one that can be applied to the rule if desired. The
weights are used to describe the uncertainty of expert’s assessment on the rules. All the
rules that have any truth in their antecedent will “fire” and contribute to the fuzzy
conclusion set. If the antecedent is true to some degree of membership, the consequent
is also true to that same degree. This point leads a natural way to combine multiple FMECA of
qualitative assessments. In the fuzzy assessment system of the component failure electronic
mode risk index, a fuzzy rule base was developed that includes rules such as:
devices
.
If (failure mode ratio is remote) and (normalized component failure rate is
moderate) and (failure effect probability is low) and (failure mode severity is
catastrophic) then (failure mode risk index is moderate). (1)
193
.
If (failure mode ratio is frequent) and (normalized component failure rate is very
high) and (failure effect probability is actual loss) and (failure mode severity is
critical) then (failure mode risk index is high). (1)
The fuzzy inference uses the method of min-max implication-aggregation inference.
The defuzzification of the output is made using the center of area method. The process
of fuzzification, inference and defuzzification is demonstrated in Figure 7 by using a
generic example with two inputs and a single output. Three rules (1-3) are fired
applying nine membership functions (A1-A3, B1-B3, C1-C3) while (z1), (z2), (z3) are the
weight factors for these three rules. The process has the following five stages where the
inputs are x, y (x, y are crisp values):
(1) Fuzzification of inputs:
.
fuzzification of the variables in rule 1 results in: A1(x), B1(y);
.
fuzzification of the variables in rule 2 results in: A2(x), B2(y);
.
fuzzification of the variables in rule 3 results in: A3(x), B3(y).
(2) Application of fuzzy operation (and ¼ min, or ¼ max):
.
fuzzy operation for rule 1 is: max (A1(x), B1(y));
.
fuzzy operation for rule 2 is: min (A2(x), B2(y));
.
fuzzy operation for rule 3 is: max (A3(x), B3(y)).

Figure 7.
Generic example of fuzzy
inference process
IJQRM (3) Application of implication method (min):
22,2 . implication for rule 1 is: C1 ¼ min (C1(max (A1(x),B1(y))));
.
implication for rule 2 is: C2 ¼ min (C2(min (A2(x),B2(y))));
.
implication for rule 3 is: C3 ¼ min (C3(max (A3(x),B3(y))));
(4) Application of aggregation method (max). The result of the aggregation method
194 is the area E ¼ maxðz1 *C1, z2 *C2, z3 *C3).
(5) Application of defuzzification (the center of area COA). The results of the
defuzzification is the center of the area E which is calculated as:
R
EðxÞxdx
C¼ R : ð8Þ
EðxÞdx

Reliability analysis of an SMPS

General aspects of the SMPS
The SMPS is a power electronic device which has the topology presented in Figure 8.
This power supply is practically a dc-to-dc converter where the ac input voltage is
rectified by a bridge of diodes and filtered by capacitors (Mohan, 1989). A voltage
“chopper” module based on a single power MOSFET converts the rectified voltage to
ac pulse voltage and feeds the primary coil of the transformer. The ac pulse voltage has
a high frequency (20-300 kHz) and, consequently, the transformer is small, light and
inexpensive. The ac voltage at the secondary coil of the transformer is rectified and
filtered in order to supply the load. A pulse width modulation (PWM) circuit controls

Figure 8.
A forward type SMPS
controlled by a pulse
width modulation circuit
the output voltage of the power supply and produces the firing pulses for the transistor FMECA of
switch accordingly while protects the converter from output overcurrent.
electronic
Reliability prediction of the SMPS devices
The parts count method was applied firstly to the SMPS in order to estimate the
components failure rates. The device’s operating environment is characterized as
“ground benign” according to the MIL-HDBK-217FN2 document (Department of 195
Defense, 1991). Ground benign is defined as “nonmobile, temperature and humidity
controlled environment readily accessible to maintenance while laboratory
instruments and test equipment, medical electronic equipment, business and
scientific computer complexes, missiles and support equipment in ground silos are
included”. All the reliability prediction calculations were made using the Relex 7
reliability prediction software, which provides great flexibility for graph generation
and “what-if” studies (Relex Software Corporation, 1998). The next stage of the
reliability prediction considers the necessary data for the implementation of the part
stress method, such as the electrical and thermal stresses on the components during
the device operation. Several parameters concerning the electrical and thermal
stresses where estimated using engineering judgment and the assumption of “worst
case” conditions of the derating plan presented in Table I. The components and
device failure rate using the parts count and part stress method are presented in
Table II.
The parts count and the part stress method provide different results for the
reliability indices of the device. In general, the failure rates calculated by the parts
count method are higher than the ones obtained by the part stress method. This

Failure rate (failures/million hours)

Component designation Parts count Part stress

D1, D2, D3, D4 0.019606 0.015123

C6 0.001710 0.016618
C7 0.001710 0.010315
R9, R10 0.030736 0.026686
R8, R11 0.003658 0.004783
Q1 0.217627 0.308058
T1 0.070834 0.115776
D5 0.128989 0.099494
D6, D7 0.015479 0.027044
C8 0.001710 0.004890
C9 0.001710 0.005800
C10 0.001710 0.012690
L 0.000097 0.000158
R1, R2, R3, R4, R5, R6, R7 0.003658 0.003690
C1 0.001710 0.001613
C2 0.001710 0.002441
C3 0.001710 0.002294 Table II.
C4 0.001710 0.002112 Failure rates for the
C5 0.001710 0.002441 SMPS and its
SG1525 0.095144 0.078881 components using the
Optocoupler 0.147576 0.244204 parts count and the part
Device 0.881148 1.111133 stress
IJQRM happens because the aim of the parts count method is to estimate roughly the failure
22,2 rates since it is applied during the early design stages and few design parameters are
only required. However, there are cases that the failure rates resulting from the part
stress method are higher due to the impact of the temperature rise from the ambient
temperature during operation. In some cases, the mathematical models for certain
component categories are rather pessimistic. The optocoupler component in the SMPS
196 is such an example, where most of the failure rate estimations given by semiconductor
manufacturers are much lower. Generally, the results obtained by the part stress
method are nearer to the field reliability indices of the device since it incorporates many
design parameters of the device and, especially, the electrical and thermal stresses
applied on the components.
In Figure 9 the system and component failure rates are presented graphically in
order to show the relative contribution of the components to the system failure rate,
according to both reliability prediction methods. It should be noted that the failure rate
calculations using the part stress method showed that transformer T1 has a greater
impact on the system failure rate than that was calculated by the parts count method.
However, the components with the highest failure rates are the same regardless of the
method being used to calculate the failure rate and these components’ failure modes
will be studied thoroughly by the FMECA.

Conventional and fuzzy failure mode effects and criticality analysis of the SMPS
The conventional and fuzzy FMECA were applied to the SMPS. The full FMECA of
the SMPS includes the failure events data for the 33 components of the device. The
fields in the respective spreadsheet of FMECA are item designation, failure mode,
local effect, next level effect, end level effect, severity, failure effect probability b,
failure mode ratio a, failure rate l and failure mode criticality Cm. The next level
effects are the effects of the component to its subsystem operation and the end level

Figure 9.
Components and SMPS
failure rates according to
the parts count and the
part stress method
effects are the effects of the component on the system operation. The SMPS can be FMECA of
divided into two subsystems, which are the dc-dc converter subsystem and the PWM electronic
subsystem. The dc-dc converter subsystem is the main module of the device that
produces the output voltage and its component failure modes have the same next devices
level and end level effects. The PWM subsystem controls the MOSFET switch and
protects the device operation from overcurrents. The failure modes of the components
in the PWM subsystem have next level effects influencing these two operations and 197
end level effects influencing the dc voltage output of the device. The component
failure modes were calculated using the FMD91 reliability database (IIT Research
Institute/Reliability Analysis Center, 1991).
Furthermore, the normalized component failure rates were calculated and the crisp
values of the normalized component failure rates, the failure mode ratios, the failure
effect probabilities and the severity of failure mode effects were used as input to the
fuzzy inference system that was developed in Matlab 6.1. The results of the
conventional and fuzzy logic based FMECA for the most critical components of the
SMPS are presented in Table III while the corresponding item criticality indices and
the item fuzzy risk indices are summarized in Table IV.
According to the results of the reliability prediction and the FMECA, the most
critical components are the optocoupler, the MOSFET transistor switch Q1, the
transformer T1, the diode UES1306 connected at the primary coil of the transformer
and the PWM IC circuit SG1525. According to the reliability database FMD91, there is
only one single failure mode for the optocoupler described as “device failure”.
Furthermore, the transistor switch Q1 exhibits high crisp criticality indices in three
severity categories and this is reflected to a high fuzzy item risk index. The diode
UES1306 and the IC SG1525 exhibit a higher criticality index for critical severity
classification than the transformer T1. However, these components do not have any
failure modes being rated with catastrophic or marginal failure effects while
transformer T1 has such failure modes. These facts are reflected to the results of these
two components that have a lower item fuzzy risk index compared with the
corresponding indices of transformer T1.
The item fuzzy risk index incorporates the effects of all the criticality indices being
used in the conventional FMECA. During the design phase of electronic products, the
four item criticality indices for the respective severity categories can be replaced by
the item fuzzy risk index for prioritizing design modifications in order to prevent the
occurrence of specific component failure modes. The criticality indices of the most
critical components can be reduced by using other component types which could
have lower failure rates (and, probably, higher cost), by modifying the component
derating plan in order to decrease their failure rates or by applying redundancy to
decrease the total failure rate and the failure effect severity of several failure modes.
Furthermore, the representation of the variables as fuzzy sets allows imprecise data
to be used in a consistent manner by exploiting the engineering knowledge being
used to construct the fuzzy system architecture. For example, there are certain
deviations of the failure rate estimations and failure mode ratios using different
reliability prediction procedures and reliability databases. In some cases, the
probability of failure events is expressed in a linguistic way such as “likely” or “very
high”. These deviations and ambiguities for the values of variables can be mitigated
 22,2

device
198
IJQRM

Table III.

the most critical

Conventional and fuzzy
logic-based FMECA for

components of the SMPS

Failure
Item mode Local effect Next level effect End level effect Severity A B C D E F G

Optocoupler Device Loss of operation Loss of PWM Output loss Critical 1 1 0.244204 0.244204 2 0.439559 0.55
failure control
Q1 Short Output loss Output loss Output loss Catastrophic 1 0.51 0.308058 0.157110 3 0.554493 0.75
(transistor) circuit (short-circuit) (short-circuit) (short-circuit)
Output low No switching (output Output loss Output loss Critical 1 0.22 0.308058 0.067773 2 0.554493 0.35
loss)
Parameter Degradation of Degraded output Degraded Marginal 1 0.17 0.308058 0.052370 1 0.554493 0.35
change switching operation output
Open No switching (output Output loss Output loss Critical 1 0.05 0.308058 0.015403 2 0.554493 0.35
circuit loss)
Output No switching (output Output loss Output loss Critical 1 0.05 0.308058 0.015403 2 0.554493 0.35
high loss)
T1 Open No transforming Output loss Output loss Critical 1 0.42 0.115776 0.048626 2 0.208393 0.35
(transformer) circuit
Short Short circuit Output loss Output loss Catastrophic 1 0.42 0.115776 0.048626 3 0.208393 0.55
circuit
Parameter Degraded output Degraded output Degraded Marginal 1 0.16 0.115776 0.018524 1 0.208393 0.15
change output
UES1306 Short Output loss Output loss Output loss Critical 1 0.49 0.099494 0.048752 2 0.179086 0.33
(diode) circuit
Open Output loss Output loss Output loss Critical 1 0.36 0.099494 0.035818 2 0.179086 0.30
circuit
Parameter Minor Minor Minor Minor 1 0.15 0.099494 0.014924 0 0.179086 0.02
change
SG1525 (IC) Improper Degraded PWM Degraded PWM Output loss Critical 1 0.77 0.078881 0.060738 2 0.141983 0.35
output control control
No output Output loss-no PWM Output loss-no Output loss Critical 1 0.23 0.078881 0.018143 2 0.141983 0.24
control PWM control
Notes: A ¼ Crisp failure effect prob. b; B ¼ Failure mode ratio a; C ¼ Failure rate l (frs/million hrs); D ¼ Failure mode criticality Cm; E ¼ Fuzzy
severity number; F ¼ Fuzzy normalized failure rate ln; G ¼ Fuzzy risk index RFM
using fuzzy sets while the variables are represented in a more realistic and flexible FMECA of
manner. electronic
devices
Conclusions
The reliability assurance activities are becoming an important part of the design of 199
electronic devices. This paper presents a reliability analysis methodology, which
consists of a reliability prediction methodology and a subsequent fuzzy FMECA
methodology. The objective of the reliability prediction methodology is to estimate the
component and system failure rates using the available operating data in every stage
of the electronic devices design. It is based on the MIL-HDBK-217FN2 reliability
prediction procedure and it developed a certain derating plan for the component
electrical and thermal stresses. The calculated component and system failure rates are
used as inputs in the subsequent fuzzy FMECA methodology. FMECA is a design
technique to systematically identify and investigate potential system (product or
process) weaknesses. All the components failure modes and their effects on the system
are identified and the severities of the effects are classified qualitatively. The
component failure mode criticality is estimated using failure data from various sources
or engineering expertise. However, the diversity and inaccuracy of these data sources
prohibits the comparison of technical expertise among different engineering projects.
Furthermore, the results are sensitive to the methods being applied to estimate the
failure event data and to the qualitative judgment of the linguistic variables. Thus, a
fuzzy assessment system was developed for the FMECA in order to limit the
dependence of the methodology on the data sources being used and to support the
comparison of technical expertise. The parameters affecting the failure mode criticality
are modified to fuzzy variables and a novel index of item failure mode risk is calculated
as a fuzzy variable. The conventional and the fuzzy FMECA methodology were applied
to a power electronic device such as a SMPS. The components and device failure rates
were calculated and these parameters were used as an input to the FMECA. The
conventional FMECA based on MIL-HDBK-1629 was implemented and the failure
mode criticality numbers were calculated. The parameters used in the FMECA were
then fuzzified and the fuzzy failure mode and item risk indices were calculated. The
results of the conventional and the fuzzy FMECA proved to be quite similar and
identify the same most critical components of the device. Possible corrective actions
were proposed when the device reliability requirements are not satisfied by the
FMECA results.

Item criticality indices (conventional FMECA) according to

severity
Item Catastrophic Critical Marginal Minor Item fuzzy risk index
Table IV.
Optocoupler Not available 0.244204 Not available Not available 0.55 Item criticality indices
Q1 0.157110 0.098579 0.052370 Not available 0.43 and item fuzzy risk
T1 0.048626 0.048626 0.018524 Not available 0.35 indices for the most
UES1306 Not available 0.084570 Not available 0.014924 0.22 critical components of the
SG1525 Not available 0.078881 Not available Not available 0.29 SMPS device
IJQRM References
22,2 Bell, D., Cox, L., Jackson, S. and Schaefer, P. (1992), “Using causal reasoning for automated failure
modes and effects analysis (FMEA)”, Annual Reliability and Maintainability Symposium,
Las Vegas, NV, pp. 343-53.
Birolini, A. (2000), Reliability Engineering: Theory and Practice, Springer-Verlag, Berlin.
Department of Defense (1980), Procedures for Performing a Failure Mode Effects and Criticality
200 Analysis, Vol. MIL-HDBK-1629A, Department of Defense, Washington, DC.
Department of Defense (1991), Reliability Prediction of Electronic Equipment,
Vol. MIL-HDBK-217F, Department of Defense, Washington, DC.
Hawkins, P.G. and Woollons, D.J. (1998), “Failure modes and effects analysis of complex
engineering systems using functional models”, Artificial Intelligence in Engineering,
Vol. 12, pp. 375-97.
IIT Research Institute/Reliability Analysis Center (1991), Failure Mode Data, Vol. FMD91, IIT
Research Institute/Reliability Analysis Center, New York, NY.
International Electrotechnical Commission (2003), Analysis Techniques for System Reliability –
Procedure for Failure Mode and Effects Analysis (FMEA), Vol. IEC 60812 (1985-07),
International Electrotechnical Commission, Geneva.
LaCombe, D. (1999), Reliability Control for Electronic Systems, Marcel Dekker, New York, NY.
Lalli, V.R., Malec, H.A. and Packard, M.H. (Eds) (2000), Reliability and Maintainability Training,
Vol. NASA-TP-2000-207428, National Aeronautics and Space Administration, Cleveland,
OH.
Lee, B.H. (2001), “Using Bayes belief networks in industrial FMEA modeling and analysis”,
Annual Reliability and Maintainability Symposium, Philadelphia, PA, pp. 7-15.
Mohan, N. (1989), Power Electronics: Converters, Applications and Design, John Wiley & Sons,
New York, NY.
Park, S.K. (1987), “Fuzzy apportionment of system reliability”, IEEE Transactions on Reliability,
Vol. R-26 No. 1, pp. 129-32.
Price, C.J. and Taylor, N.S. (2002), “Automated multiple failure FMEA”, Reliability Engineering
and System Safety, Vol. 76, pp. 1-10.
Pukite, J. and Pukite, P. (1998), Modeling for Reliability Analysis, IEEE Press, New York, NY.
Relex Software Corporation (1998), Relex 7 Reliability Prediction Software, Relex Software
Corporation, Greensburg, PA.
Reliability Analysis Center (1995), Non-electronic Parts Reliability Data, Vol. NPRD-95, Reliability
Analysis Center, Rome, NY.
Society of Automotive Engineers (2001), Recommended Failure Mode and Effects Analysis
(FMEA) Practices for Non-Automobile Applications, Vol. SAEARP5580, Society of
Automotive Engineers, Warrendale, PA.
Teng, S.H. and Ho, S.Y. (1996), “Failure mode and effects analysis: an integrated approach for
product design and process control”, International Journal of Quality & Reliability
Management, Vol. 13 No. 5, pp. 8-26.
Vas, P. (1999), Artificial-Intelligence-Based Electrical Machines and Drives, Oxford Science
Publications, Oxford.