Chapter 5

Algorithm for fault detection and analysis

5.1 Introduction
Recently Artificial Intelligence (AI) is being used in many applications such as
electrical motor CQntrols, robotics, fault diagnosis etc. Fuzzy model is one of the AI model,
which define its inputs and outputs as attributive values and defines the strength of the
relationship between the input and output reference sets. This capacity provides a useful
method for analyzing the motor current and detecting fault. So the fuzzy logic has emerged
as a profitable tool for the diagnosis system among the other expert systems. Recently
researchers have moved to AI techniques including fuzzy logic and neural network for
condition monitoring and fault detection of electric motors, because no detailed analysis of
fault mechanism or mathematical modeling is required in this system.
This chapter is organized into three sections. First part presents an overview of the
fuzzy logic,_ which is intended to provide enough theoretical background about the fuzzy
logic, to enable the reader to follow the discussion on the fuzzy logic approach proposed in
this thesis. Theoretical background about the fuzzy logic is included in the second part of
this chapter. End of this chapter provides a description of the fuzzy logic fault detection
approach, proposed to identify and analyse the severity of faults in three phase induction
motor.

5.2 Fuzzy logic system - an overview

In recent years, fuzzy logic is increasingly utilized in various applications, such as
fuzzy reasoning models, statistical decision making using fuzzy probability and fuzzy
entropy, fuzzy quantification theory, fuzzy mathematical programming, evaluation using
fuzzy measures, diagnosis using fuzzy relations, fuzzy control and inference, multistage
decision making using fuzzy dynamic programming, fuzzy database and information
retrieval using fuzzification functions, fuzzy expert systems etc.
Basically, the fuzzy logic is a multi-valued logic, based on fuzzy sets, which allows
intermediate values between the conventional evaluations like Yes/No, True/False, one/zero
etc. The basis of fuzzy logic is to use linguistic variables, whose values are words like pretty
cold, very hot, etc., rather than numbers. The values of linguistic variables can be
formulated mathematically and processed by computers. In this way, an attempt is made to
apply a more human-like way of thinking in the programming of computers. This unsharp
boundary characteristic, gives fuzzy logic some remarkable practical advantages in flexibility
and tolerance to imprecise data [37, 69, 107, 120].

5.3 Theoretical background

The block diagram of fuzzy diagnosis system is shown in fig, 5.1. It consists of four
basic blocks named as fuzzification, rule base, inference engine and defuzzification. The
process of converting input/output variables to linguistic labels is termed as fuzzification and
it is executed using fuzzy sets.

Output
I

Fig. 5.1 Structure of Fuzzy diagnosis system

5.3.1 Fuzzy sets

The basic principle of conventional set theory is that, an element is either a member
or not a member in a set. A set, that it is defined in this way is called crisp set because its
boundary is well defined. Fuzzy sets are sets that contain elements have varying degree of
membership. They are appropriate to present linguistic variables described by words and
whose values are qualitative or linguistic and not quantitative or numeric. For example, fig.
5.2 shows four fuzzy sets covering all values of negative sequence current magnitude

[79]
values of fault detection system and are designated to represent negligible, small , medium
and high respectively. Obviously the concepts of negligible, small , medium and high value of
negative sequence current are subjective and depend on the appl ication at hand.
Nevertheless , they offer a way of representing va lue judgments, which can be expressed
with degree of membership. In fig. 5.2, consider 0.1 val ue of negative sequence current , it
is 0% certain that this value of negative sequence cu rrent is medium, 60% certain that this
level of negative sequence current is negligible and 25% certa in that this level of negative
sequence current is small. The degrees of certainty are expressed usually on a scale of 0 to
1 and represent the degree of membership of the sets.

Small Medium High

1.0

0 0.1 0.3 0.4 o.s

Negative sequence current magnitude (A)

Fig. 5.2 Fuzzy sets of fault detection system for negative sequence current levels.

Degrees of certainty are not probabilities. Fuzziness measures the degree to which
something will belong or some condition exists. In the above example, if a value of negative
sequence current is not negligible, it is small, medium or high. On the other hand, values
between 0.4 and 0.5 A, evaluated using the relation expressed graphically in the figu re
yields, e.g., 0.45 A for a 0.3 degree of membership for both sets. If the degrees of
membership of the sets involved in the fuzzy evaluation sum to unity, these particular sets
are sa id to be complementary.

[80]
5.3.2 Membership functions
A membership function is a curve, often given the designation of IJ, which defines
how the input space is mapped to a degree of certainty between 0 and 1. The input space is
referred as a universe of discourse. The universe of discourse for the previous fuzzy sets
corresponds to the domain of negative sequence current. The membership function itself
can be an arbitrary curve whose shape is defined as a function of suitable behavior to
assure simplicity, convenience and efficiency.
A classical set might be expressed as: A = {xI x > 6}. A fuzzy set is an extension
of a classical set. If X is universe of discourse and its elements are denoted by x, then a

fuzzy set A in X is defined as a set of ordered pairs as A = {x, Jl A (x) I x E X}, where

JlA(x) is called membership function (or MF) of x in A. The membership function maps each
element of X to a membership value between 0 and 1. The fuzzy logic tool box includes
several types of built in membership functions. These membership functions are built from
some basic functions, such as piecewise linear functions, Gaussian distribution function,
sigmoid curve and quadratic and cubic polynomial curves.

1,0

··\<"o'""'•z~4<'""--76=;i;.a"''=7,1o·
91Wml'

.. . . .
0 10 o:. .z 4 6 8 .10 l' '4' 6 8· 10
.[Link]{ zm( s!lll

Fig. 5.3 Structure of membership functions

The simplest membership functions are formed using straight lines. The triangle
membership function is a collection of three points forming a triangle and it has a function

[81]
name 'trimf. The trapezoidal membership function 'trapmf has a flat top and a truncated
triangle curve. These. straight line membership functions have the advantage of simplicity,
and are used in this approach. Two· membership functions are built on the Gaussian
distribution curve, they are single Gaussian curve and a two - sided composite of two
different Gaussian curves. Their function names are 'gaussmf' and 'gauss2mf respectively.
Sigmoid function is either open left or right. Asymmetric and closed lillembership function
can be synthesized using two sigmoid functions. The membership functions related to the
..
polynomial based curves are Z curve, S curve and Pi curve, all named because of their
shape. Their function names are Zmf, Smf and Pimf. The standard shapes of some
membership functions are shown in fig. 5.3.

5.3.3 Fuzzy reasoning

The theory of fuzzy reasoning provides a powerful frame work for reasoning in the
face of imprecise and uncertain. Heart of this theory is fuzzy ope,ration, which is the
representation of propositions as statements assigning fuzzy sets as value to variable.

Fuzzy operators I
The fuzzy logical reasoning is based on a super set of standard :Boolean logic, which
involves operators, such as fuzzy intersection or conjunction (AND), fuzzy union or
disjunction (OR) and fuzzy compliment (NOT). The classical operators for these functions
are: AND= min, OR= max and NOT= additive complement.
In general, the intersection of two fuzzy sets 'A' and 'B' is specified by a binary
mapping 'T' and their union operation is specified by a binary mapping 'S'. They aggregate
two membership functions as follows:

(5.1)

(5.2)

IF • THEN rules
Fuzzy logic is based on the rules to form "IF ..........THEN" t~at convert an input
fuzzy set to an output fuzzy set. A single fuzzy If- Then rule assume the form of, if x is A
then Y is 8, where, A and 8 are linguistic values defined by fuzzy sets on the ranges such
as universe of discourse, X and· Y respectively. 'If part of the rule, "x is A" is called

[82]
antecedent or premise, while 'then' part of the rule "Y is 8" is called consequent or
conclusion. The antecedent of the rule can have multiple parts, in which case all parts of
antecedent are calculated simultaneously and resolved to a single number using logical
operators described in the preceding section, e.g.
If the negative sequence current is high AND positive sequence current is high
THEN ......... The consequent of the rule can have multiple parts, in which case all
consequents are affected equally by the result of antecedent. In general, one rule alone is
not effective, two or more rules may be needed in the inference process. The output of each
rule is a fuzzy set. The output fuzzy sets for each rule are aggregated into a single output
fuzzy set. Finally the resulting set is defuzzified and resolved into single number.

Fuzzy inference system

Fuzzy inference is the process of mapping from a given input space (A, B) to an
'
output space (R) based on "If A, Then 8" rules. The technique used for obtaining the output
space is known as the fuzzy implication operation and they are valid for all values of
x EX andy E Y. The mapping provides a basis from which decisions can be made.

Several forms of implication operators named as Mamdani's model, Sugano model and
Tsukanto model etc. exist but Mamdani's model is most commonly used. This operator is
equivalent to fuzzy cross product of fuzzy sets of A and 8, and it is expressed as:

JlR(x,y) = min[.uA(x),p8(x)] (5.3)

Aggregation of fuzzy rules

Most rule based systems involve more than one rule. The process of obtaining the
overall consequent from the individual consequents contributed by each rule in the rule base
is known as aggregation. This aggregation is also called min - max aggregation. According
to this operation, aggregation strategies are classified into two categories named as
conjunctive system of rules and disjunctive system of rules.

Conjunctive system of rules

In this case, the rules are connected by "AND" connectives. Thus the aggregated
output y is found by the fuzzy intersection of all individual rule consequents, y', where i=1, 2,

[83]
 '1 d 2 d ' I 2 r
3, .... ,r y=y anY an ......y ory=y ny n .......... ny ,whichisdefinedbythe
membership function.

f.1y (y) = min1\ f.1y , (y ), f.1y , (y ) ........ .f.1y , (y )) for yEY (5.4)

Disjunctive system of rules

For the cases of a disjunctive system of rules where the satisfaction of at least one
rule is required, the rules are connected by "OR" connectives. In this case, aggregate output
is found by the fuzzy union of all individual rule contribution through

y= y I or y 2 ................or y r
or

y = / u y 2 u ......... u y' which is defined by the membership function

'
!J,(y) =max l,u,. (y),!J,,(Y), ........ .IJ,.. (y)j for y e Y (5.5)

The overall fuzzy inference process described above can be explained graphically.
For graphical explanation, a simple case is considered as an example and its overall fuzzy
inference process is divided into three sections. In the first section, general definition of the
problem is described, the complete mathematical model description of the considered case
is formulated in the second section and the graphical representation of the inference
process is described in the last section. The system is described by a simple two. rule
system, where each rule comprises of two antecedents and one consequent. This
formulation is similar to dual input single output fuzzy system. The graphical procedures for
dual inp\lt single output fuzzy system can be easily extended and will hold for fuzzy rule
bases with any number of antecedents or inputs and consequents or outputs.
The mathematical formulation of the dual input single output system described
above can be expressed as follows: A fuzzy system with two non interactive inputs x 1 and x2
(antecedents) and a single output y (consequent) is described by a collection of two
linguistic IF ...... THEN propositions in Mamdani's form;
If x, is A,' and x, is A; Then y' is B' for k = 1,2 .... (5.6)

where A,' and ~ are the fuzzy sets representing the k1h antecedent' pairs,

and Bk is the fuzzy set representing the k1h consequent. Based on equation (5.5) and
for a set of disjunctive rules, the aggregated output for the two rules will be given by.

[84]
 p; (y) =;" [min [u;, (input (i), p ~'(input (j)] k =I, 2 (5.7)
The graphical interpretation of equation (5. 7) is shown in fig. 5.4. In this ftgure A11
I
and A 12 refers to the first and second fuzzy antecedent of the first rule respectively, and
symbol [Link] to the fuzzy consequent of the first rule. A21 and A22 refer to the first and
second antecedents and B2 refers to the fuzzy consequent of the second rule.

,. RUie·i.
•

ln~ut (j) .'

v
~pply.&~ggregration
(Dl~j_uncti~q ~ ·~a·x)'

"
.'

Fig. 5.4 Graphical representation of Mamdani inference method

The minimum function in equation (5.7) is illustrated in fig. 5.4. The ante'cedent pairs
given in general rule structure for this system are connected by a logical "And" connectives
as in equation (5.4 ). The minimum [Link] value of the antecedent propagates through
the consequence and truncates the membership function for the conse,quent of each rule.
This graphical inference is done for each rule. Then, truncated membership functions for
each rule are aggregated using the max operation, which results in an aggregate
membership function comprised of the outer envelope of the individual truncated
memberships from each rule. Finally a crisp value for the aggregated output, i.e., an

[85]
appropriate defuzzification technique, needs to be applied to determine the value of 'y' [116,
134].

5.3.4 Defuzzificaton
· As discussed previously, there may be situations where the output of a fuzzy
process needs to be a singular quantity of precise quantity. There are many defuzzification
techniques but primarily only three of them are in use. These defuzzification techniques are
named as maximum defuzzification techniques, centroid ·defuzzification techniques and
weighted average defuzzification techniques. Perhaps, the most popular defuzzification
approach is the centroid method. This method is also known as center of gravity or center of
area defuzzification method. This technique was developed by Sugeno in 1985. The
centroid defuzzitication technique can be expressed as

• [Link]).ydy
y = - {5.8)
[Link]).dy
Where J denotes algebraic integral, y' is the defuzzification output, Jlc (y) is the

aggregate membership function and 'y' is the output variable. This method is shown in fig.
5.5.

~~· . . .
/. - _··....·.
r=_j ·.

'
.
.
'
~
·,

y* y
Fig. 5.5 Graphical representation of the centroid method

[86]
5.4 Fuzzy logic based fault detection
The fuzzy logic system explained in the previous session provides a robust and
sensitive [Link] mechanism of electrical faults and mechanical faults in a three phase
induction motor. The robustness of fault detection system is exhibited consistently and
accurately detects the faults under different fault conditions. The sensitivity is manifested in
the capacity of judging and discriminating between an incipient fault and healthy condition.
The diagnostic can be developed by' integrating human knowledge and experience with
fuzzy set and fuzzy rules, which helps in the formulation of the knowledge database. The
induction motor condition can be diagnosed using compositional rule of fuzzy inference.
In this work, four fuzzy logic algorithms are proposed for the detection and analysis
of electrical faults, mechanical faults and switching fault of AC three phase induction motor
drive systems. First algorithm is used for the detection of electrical fault. Second and third
algorithms are developed to detect and diagnosis of mechanical faults like rotor broken bar
fault of three phase squirrel cage induction motor and inter turn short circuit faults of three
phase slip ring induction motor and last algorithm is used for the detection and diagnosis of
switching faults of three phase AC induction motor drive system.
All the system inputs and output variable are defined using fuzzy set theory. The
motor condition 'MC' is chosen as the output variable. The detection process is
i'\Ccomplished using a fuzzy logic system with a fuzzy membership function. The
membership functions are formulated with the help of simulation results of the machine
under healthy condition and different fault conditions. These membership functions
established are only valid for the system studied and the system considered must be
adjusted for each different machine.
The fuzzy rule system provides the antecedent and consequent statements of fuzzy
logic. These statements provide the condition of the fault being monitored given the
linguistic operating range of inputs. The fuzzy rules represent a combination of the
qualitative heuristic knowledge .of the system and the quantitative description of motor
conditions. The consequence provides the quantitative information about the motor using
description of healthy or normal and different fault conditions. The implication operator used
in fuzzy inference system is based on Mamdani's form and is constituted by a number of
conjunctive rule~?. As membership functions, these rules are also formulated on the
experience and meticulous observations of the behaviors exhibited by the machine under
different faulty and healthy conditions during simulation. The defuzzification process is
performed using the centroid method.

[87]
~.5 Conclusion

In this chapter, theoretical study of fuzzy logic system and fuzzy logic based fault
detection scheme is presented. Mainly, fuzzy logic system consists ·of four sections named
as fuzzification, fuzzy rule base, inference engine and defuzzification. Fuzzification is used
to convert the crisp input variable into the linguistic variables for fuzzy set operation. These
fuzzy set operations are used to design the fuzzy rules, which are used to convert the input
fuzzy set to an output fuzzy set. Finally the resulting fuzzy output set is defuzzified and
resolved into a single number for representing the condition of electrical machine.
In fuzzy logic fault detection scheme, the membership functions and fuzzy rules are
formulated using the observations collected from the simulation of the machine under
healthy condition and different fault conditions. Electrical fault detection and analysis of
three phase induction motor are dealt in the next chapter.

[88]