Frontiers in Robotics, Automation and Control Part 3 pptx

Automatic Speaker Recognition by Speech Signal

53
took one of two forms: a speech spectrogram or a contour voiceprint (Baken & Orlikoff,
2000). The more commonly used form consists of a representation of a spoken utterance in
which time is displayed on the horizontal axis, frequency on the vertical axis and spectral
energy as the darkness at a given point.
At present, the increase in commercial application opportunities has resulted in increased
interest in speaker recognition research. The main commercial application for speaker
recognition seems to be speaker verification used to the physical entry of a person into a
secured area, or the electronic access to a secured computer file or licensed databases. Such
voice-based authorization is often a part of a security system that also includes the use of
PIN number, password, and other more conventional means. The most immediate challenge
in voice-based authorization is a caller authentication over the telephone network that will
be accurate enough so that financial transactions could take place under it aegis. Car access
is yet another popular area where voice-based security systems are gained ground. Some
automobile manufactures are testing a speaker identification system to control door locks
and ignition switches. An interesting twist to this application is that the ignition switch can
be programmed not to work if the driver is under the influence of drugs or alcohol, since
intoxication is detectable in the speech signal.

7. Acknowledgement

This work was supported by the Czech Ministry of Education in the frame of the Research
Plan No. MSM 0021630513 “Advanced Electronic Communication Systems and
Technologies”.

8. References

Baken, R. J. & Orlikoff, R. F. (2000). Clinical Measurement of Speech and Voice, Singular
Publishing Group, ISBN 1-56593-869-0, San Diego

Kepesi, M. & Sigmund, M. (1998). Automatic recognition of gender by voice, Proceedings of
Radioelektronika‘98, pp. 200-203, ISBN 80-214-0983-5, Brno, April 1998, CERM, Brno
Lancker, D.; Kreiman, J. & Emmorey, K. (1985). Familiar Voice Recognition: Patterns and
Parameters - Recognition of Backward Voices. Journal of Phonetics, Vol. 13, No. 1,
(January 1985), pp. 19-38, ISSN 0095-4470
Matsui, T.; Nishitani, T. & Furui, S. (1996). Robust methods of updating model and a-priori
threshold in speaker verification, Proceedings of IEEE Internat. Conf. on Acoustics,
Speech and Signal Processing, pp. 97-100, ISBN 0-7803-3192-3, Atlanta, May 1996,
IEEE Computer Society, Washington, DC
Rabiner, L. R. & Juang, B. H. (1993). Fundamentals of Speech Recognition, Englewood Cliffs,
ISBN 0-13-015157-2, New Jersey
Reich, A. & Duke, J. (1979). Effects of selected vocal disguises upon speaker identification by
listening. Journal of the Acoustical Society of America, Vol. 66, No. 4, (April 1979), pp.
1023-1028, ISSN 0162-1459
Sigmund, M. & Jelinek, P. (2005). Searching for phoneme boundaries in speech signal,
Proceedings of Radioelektronika 2005, pp. 471-473, ISBN 80-214-2904-6, Brno, April
2005, MJ Servis, Brno
Frontiers in Robotics, Automation and Control

54
Sigmund, M. & Mensik, R. (1998). Estimation of vocal tract long-time spectrum, Proceedings
of Elektronische Sprachsignalverarbeitung, pp. 69-71, ISSN 0940-6832, Dresden,
September 1998, w.e.b. Universitätsverlag, Dresden
Slomka, S. & Sridharan, S. (1997). Automatic gender identification under adverse conditions,
Proceedings of Eurospeech‘97, pp. 2307-2310, ISSN 1018-4074, Rhodes, September
1997, Typoffset, Patras
Titze, I. R. (1989). Physiologic and acoustic differences between male and female voices.
Journal of the Acoustical Society of America, Vol. 85, No. 4, (April 1989), pp. 1699-1707,
ISSN 0162-1459
Wu, K. & Childers, D. G. (1991). Gender recognition from speech. Journal of the Acoustical

Society of America, Vol. 90, No. 4, (April 1991), pp. 1828-1840, ISSN 0162-1459
4

Verification Based Model Localizes Faults from
Procedural Programs

"#$%%&''#( "))*+)
!"#$%&'"(&)*+),*'#-&"%)./0"(/"1)
)2$&0*($3)4(05"%60&7)*+),*'#-&"%)./0"(/"6)$(8)9'"%:0(:)./0"(/"6)
66**'%*;06&<&-:%$=<$&
1)6$+""-33$><6**'%*;(-<"8-<#?)
)D$%$/>0)
E$?06&$()

1.!Overview

,%-% /01 #02 ').#-/01 $#&'-3 /3 )0% )$ -(% *)3- /*4)+-#0- /33&%3 )$ .&++%0- +%3%#+.(
.)**&0/-56 7#05 %$$)+-3 (#8% 9%%0 -#:%0 -) /*4+)8% 3)$-;#+% 2%8%')4*%0- #02 -) 4+%8%0-
$#&'-36 <&- 3-/'' 3)$-;#+% $#&'-3 4)3% -(% *)3- .(#''%01/01 4+)9'%*3 -) 3)$-;#+% %01/0%%+36
=#&'- ').#'/>#-/)0 /3 # *)3- .(#''%01%#9'% -#3: 2&+/01 -(% 2%9&11/01 4+).%336 =#&'-
').#'/>#-/)0 /3 -(% 0%?- 3-%4 #$-%+ 2%-% /01 $#&'-3 /0 4+)1+#*36 @3% )$ $#&'- ').#'/>#-/)0 /0
.)0-+)' %01/0%%+/01 ;(%+% %01/0%%+3 )$-%0 %*4')5 -(% 4+).%2&+#' 4+)1+#**/01 4#+#2/1*6
A$-%0 .)0-+)'3 3)$-;#+% /3 3#$%-5B.+/-/.#' #02 -(&3 2%-% /)0 9&- #'3) ').#'/>#-/)0 )$ 9&13 /3
& %+*)3- /*4)+-#0-6

C(/3 .(4#-%+ *#:%3 &3% )$ #93-+# 2%4%02%0.%3 9%-;%%0 4+)1+#* 8#+/#9'%3 $)+ ').#-/01 #02
').#'/>/01 $#&'-3 /0 4+).%2&+#' 4+)1+#*36 C(% 3) .#''%2 8%+/$/.#-/)0B9#3%2 7)2%' DE<7F $)+
2%9&11/01 /3 #0 %?-%03/)0 )$ 2%4%02%0.% *)2%' $+)* G#.:3)0H3 I34% "53-%*J ;(/.( (#3
9%%0 &3%2 $)+ 8%+/$/.#-/)0 )$ K 4+)1+#*36 C(% I34% 353-%* #0#'5>%3 -(% 2%4%02%0.%3
9%-;%%0 8#+/#9'%3 )$ # 1/8%0 4+)1+#* #02 .)*4#+%3 ;/-( 34%./$/%2 2%4%02%0.%36

A-(%+;/3%J -(% 4+)1+#* $&'$/''3 -(% 34%./$/.#-/)06 L0 .#3% )$ */3*# ( -(% 4+)1+#* /3 3#/2 -)
8/)'#-% -(% 34%./$/.#-/)06 @0$)+-&0#-%'5J -(% I34% 353-%* 2)%3 0)- #''); ').#-/01 -(%
3)&+.% )$ */3*# (6 C(% E<7 %?-%023 G#.:3)0H3 /2%# -);#+23 0)- )0'5 2%-% /01 9%(#8/)+
9&- #'3) ').#'/>/01 *#'$&0 /)0/01 +%#' .#&3%6 C(% E<7 4%+$)+*3 $/?B4)/0- .)*4&-#-/)0 $)+
+%.&+3/8% /08).#-/)0 D/0 #'' .#3%3 ;(%+% ;% )9-#/0 # .5.'/. .#'' 1+#4(F6 M% 4+%3%0-%2
#'1)+/-(* #02 4+))$ $)+ $/? 4)/0- .)*4&-#-/)0 ;(/.( %03&+%3 -(#- 0) 2%4%02%0.%3 ')33
2&+/01 /-%+#-/)0 #02 ;% #';#53 +%#.(%2 $/?B4)/0- #$-%+ $/0/-% 0&*9%+ )$ /-%+#-/)036
=&+-(%+*)+% ;% 4+%3%0- 0)8%' +%3&'-3 )9-#/0%2 $+)* )&+ *)3- +%.%0- .#3% 3-&2/%36 N)-#9'5J
;(%0%8%+ )&+ 0)8%' *)2%' 2%-% 3 # 3-+& &+#' $#&'-J /- #'3) #44%#+3 -) 9% .#4#9'% )$
').#'/>/01 -(% 2%-% %2 */39%(#8/)+H3 +%#' .#&3%6

Key Words
7)2%' <#3%2 O%#3)0/01J ")$-;#+% E%+/$/.#-/)0J ")$-;#+% ,%9&11/01J =#&'- ,%-% /)0 #02
P).#'/>#-/)06

"#$%&'(#)!'%!*$+$&',)-!./&$01&'$%!1%2!3$%&#$4!

56!
2. Introduction

")$-;#+% 8%+/$/.#-/)0 /3 #0 /*4)+-#0- 4(#3% )$ 3)$-;#+% 2%8%')4*%0-6 L0 -(% '#3- 2%.#2% -(%
3)$-;#+% 8%+/$/.#-/)0 #02 -(% 2%9&11/01 .)**&0/-/%3 (#8% *#2% .)03/2%+#9'% 4+)1+%336 L0
-(/3 .(#4-%+ ;% $).&3 )0 $#&'- ').#'/>#-/)0 ;(/.( /3 9#3%2 )0 -(% #93-+# 2%4%02%0./%3 -(#-
#+% &3%2 95 -(% I34% 353-%* DG#.:3)0 QRRSF $)+ 2%-% /01 $#&'-36 C(% 8%+/$/.#-/)0B9#3%2
*)2%' $)+ 2%9&11/01 /3 #0 %?-%03/)0 )$ -(% 2%4%02%0.% *)2%' $+)* G#.:3)0T3 I34% 353-%*
DG#.:3)0 QRRSF ;(/.( (#3 9%%0 &3%2 $)+ 2%4%02%0.5 9#3%2 8%+/$/.#-/)0 )$ K 4+)1+#*36 C(%
I34% 353-%* #0#'53/3 -(% 2%4%02%0.%3 9%-;%%0 8#+/#9'%3 )$ # 1/8%0 4+)1+#* #02
.)*4#+%3 -(%* ;/-( -(% 34%./$/%2 2%4%02%0.%36 L0 .#3% )$ # */3*# ( -(% 4+)1+#* /3 3#/2 -)
8/)'#-% -(% 34%./$/.#-/)06 A-(%+;/3%J -(% 4+)1+#* $&'$/'3 -(% 34%./$/.#-/)06 @0$)+-&0#-%'5J -(%
I34% 353-%* 2)%3 0)- #''); ').#-/01 -(% 3)&+.% )$ # */3*# (6 L0 -(% $)'');/01 ;% %?-%02

G#.:3)0T3 /2%# -);#+23 0)- )0'5 2%-% /01 */39%(#8/)&+ 9&- #'3) ').#'/>/01 -(%
*#'$&0 /)0/01 +%#' .#&3%6

I'-()&1( 4+)1+#* 3'/./01J #3 # '/1(-;%/1(- -%.(0/U&%J (#3 3%%0 3& %33$&' #44'/.#-/)0 /0
$#&'- ').#'/>#-/)0 DI1+#;#' %- #'6J QRRVJ =+/->3)0 %- #'6J QRRRJ P5'% #02 76 M%/3%+ QRWXJ /-3
2/3.+/*/0#-/)0 '/:% 7<", DM)-#;# YZZYF6 L0 D[&4%+ QRWRJ M/%'#02 YZZQF -(% #&-()+3
%*4')5 -(% 0)-/)0 )$ 2%4%02%0.%3 $)+ $#&'- ').#'/>#-/)06 L0 .)0-+#3- -) '# %+ #44+)#.( ;% 2)
0)- %*4')5 2%-% %2 2/$$%+%0.% /0 8#+/#9'% 8#'&%3 #- # .%+-#/0 '/0% /0 .)2% 9&- &3% )$
2/$$%+%0.%3 9%-;%%0 34%./$/%2 #02 .)*4&-%2 2%4%02%0./%3 #02 -(&3 #'3) /0.)+4)+#-% -(%
3-+& &+#' 4+)4%+-/%3 )$ 4+)1+#* #02 34%./$/.#-/)06 C(&3J -(% *)2%'3 /0-+)2&.%2 /0 D[&4%+
QRWRJ M/%'#02 YZZQF .#0 0)- 2%#' ;/-( #33%+-/)03 )+ 4+%B #02 4)3- .)02/-/)03 /0 #
3-+#/1(-$)+;#+2 ;#56

L0 -(/3 .(#4-%+ ;% $).&3 )0 ').#'/>/01 $#&'-3 /0 4+).%2&+#' 4+)1+#*3 #02 2%#'/01 ;/-( 1')9#'
8#+/#9'%36 \+).%2&+#' 4+)1+#*3 #+% 1%0%+#''5 *)+% .)*4&-#-/)0#''5 %$$/./%0- -(#- )9]%
)+/%0-%2 4+)1+#*3J 9%.#&3% -(%+% /3 '%33 )8%+(%#2 -) (#02'% #93-+# /)03 #02 -(% 2#-#
3-+& &+%3 *)+% .')3%'5 +%3%*9'% -(% (#+2;#+% -(#- *&3- *#0/4&'#-% -(%*6

C(% .(#4-%+ /3 )+1#0/>%2 #3 $)'');36 L0 "% /)0 V ;% /0-+)2&.% )&+ 8%+/$/.#-/)0 9#3%2 *)2%'
95 &3/01 *)-/8#-/01 %?#*4'%6 C(% +%3&'-3 #02 2/3.&33/)0 1/8%0 /0 "% /)0 ^ +%8%#' -(%
8%+/$/.#-/)0 9#3%2 *)2%' 4+)8/2%3 # &3%$&' *%#03 $)+ 2%-% /01 #02 ').#'/>/01 .)**)0
%++)+3 $)+ 4+).%2&+#' 4+)1+#*3 #02 4#+-/.&'#+ /0 .)0-%3- )$ 1')9#' 8#+/#9'%36 L0 "% /)0 S ;%
4+%3%0- +%'#-%2 +%3%#+.(6 =&+-(%+ *)+% ;% +%3%0- '/*/-#-/)0 )$ )&+ *)2%' /0 3% /)0 _6 =/0#''5
;% 3&**#+/>% -(% .(#4-%+6


3. Motivating Example

L0 -(/3 3% /)0 ;% %?4'#/0 -(% 9#3/. /2%# )$ ').#'/>/01 -(% $#&'- 95 .(%.:/01 ;(%-(%+ -(% 4)3-
.)02/-/)0 /3 3#-/3$5/01 )+ 0)- &3/01 -(% 8%+/$/.#-/)0 9#3%2 *)2%'6

L0 -(/3 .(#4-%+ ;% $).&3 )0 $#&'- ').#'/>#-/)0 ;(/.( /3 9#3%2 )0 #93-+# 2%4%02%0./%3 -(#-
#+% &3%2 95 -(% I34% 353-%* DG#.:3)0 QRRSF $)+ 2%-% /01 $#&'-36 I93-+# 2%4%02%0./%3 #+%
+%'#-/)03 9%-;%%0 8#+/#9'%3 )$ # 4+)1+#*6 M% 3#5 -(#- # 8#+/#9'% F 2%4%023 )0 # 8#+/#9'% 7/$$
7(#'8',1&'$%!91)(2!:$2(4!;$,14'<()!"1/4&)!8#$0!=#$,(2/#14!=#$>#10)!

5?!
# 0%; 8#'&% $)+ 7 *#5 .#&3%3 # 0%; 8#'&% $)+ F6 =)+ %?#*4'%J -(% #33/10*%0- 3-#-%*%0- F)))
G)7)H Q /*4'/%3 3&.( # 2%4%02%0.5 +%'#-/)06 `8%+5 -/*% ;% .(#01% -(% 8#'&% )$ 7 -(% 8#'&%
)$ F /3 .(#01%2 #$-%+ %?%.&-/01 -(% 3-#-%*%0-6 I0)-(%+ %?#*4'% ;(/.( '%#23 -) -(% 3#*%
2%4%02%0.5 /3 -(% $)'');/01 4+)1+#* $+#1*%0- D=/16 QFa

)))))))))))))))))))))))))))I+)J7)K)LMN)&>"()
)))))))))))))))))))))))))))))))))O)G)LP)
)))))))))))))))))))))))))))936")
))))))))))))))))))))))))))))))))O)G)MP

=/16 Q6 "/*4'% .)02/-/)0#' 9').:

L0 -(/3 $+#1*%0- 0)- #'' .(#01%3 #44'/%2 -) 7 .#&3% # .(#01% )0 -(% 8#'&% )$ FJ #'-()&1( F
2%$/0/-%'5 2%4%023 )0 76 C(% I34% 353-%* 0); -#:%3 # 4+)1+#*J .)*4&-%3 -(%
2%4%02%0./%3 #02 .)*4#+%3 -(%* ;/-( -(% 34%./$/%2 2%4%02%0./%36 L$ -(%+% /3 # */3*# (
-(% 353-%* 2%-% 3 # 9&1 #02 0)-/$/%3 -(% &3%+6 b);%8%+J -(% I34% 353-%* 2)%3 0)-
4/04)/0- -(% +))-B.#&3% )$ -(% 2%-% %2 */39%(#8/)+ -) -(% &3%+6

L0 -(% $)'');/01 4+)1+#* $+#1*%0- D=/16 YF ;% %?4'#/0 -(% 9#3/. /2%#3 &3/01 -(% $)'');/01
3*#'' 4+)1+#* ;(/.( /*4'%*%0-3 -(% .)*4&-#-/)0 )$ -(% ./+.&*$%+%0.% #02 #+%# )$ # ./+.'%6
C(% 4+)1+#* .)0-#/03 )0% $#&'- /0 '/0% Y ;(%+% # *&'-/4'/.#-/)0 95 ! /3 */33/016

!! #$% &$'%

)))))))))))))))))))))))))))))))L<))8)G)%)Q)RP)
)))))))))))))))))))))))))))))))))))))))R<)))/)G)8P))SS)T4UV)$)G)8)Q)#0P)
)))))))))))))))))))))))))))))))W<)))/)G)%)Q)%)Q)#0P))))))))))
!! #()& * + $
,
- ! . / + ,- $- !

=/16 Y6 K#'.&'#-/)0 )$ ./+.&*$%+%0.% #02 #+%#

C(%3% 2%4%02%0.%3 3)'%'5 #+% 1/8%0 95 # 3-#-%*%0- ;(%0%8%+ ;% #33&*% -(#- -(% 3-#-%*%0-
/3 .)++% D;6+6-6 -(% 2%4%02%0.%3F6 L$ # 3-#-%*%0- /3 #33&*%2 -) 9% /0.)++% J -(%
2%4%02%0.%3 #+% 0)- :0);06 M% %?4+%33 -(% '# %+ $# 95 /0-+)2&./01 # 0%; -54% )$
8#+/#9'%J -(% 3) .#''%2 *)2%' 8#+/#9'%36 7)2%' 8#+/#9'%3 #+% 8#+/#9'%3 -(#- ;)+: #3 4'#.%B
()'2%+ $)+ 4+)1+#* 8#+/#9'%36 =)+ %?#*4'%J /$ ;% #33&*% 3-#-%*%0- Y -) 9% /0.)++% J ;%
/0-+)2&.% # *)2%' -(#- 3#53 -(#- 4+)1+#* 8#+/#9'% $ 2%4%023 )0 *)2%' 8#+/#9'% "
Y
D;(%+% "
Y

/3 &0/U&%F6

C(% /2%# 9%(/02 )&+ #44+)#.( /3 -) $/02 #33&*4-/)03 #9)&- -(% .)++% 0%33 #02
/0.)++% 0%33 )$ 3-#-%*%0-3 ;(/.( 2) 0)- .)0-+#2/ # 1/8%0 34%./$/.#-/)06 L0 )&+ +&00/01
%?#*4'%J -(% 34%./$/.#-/)0 /3 1/8%0 /0 -%+*3 )$ # 4)3-B.)02/-/)06 =+)* -(/3 4)3-B.)02/-/)0 ;%
2%+/8% -(#- / (#3 -) 2%4%02 )0 % #02 #0 6 b);%8%+J ;(%0 #33&*/01 3-#-%*%0- Q #02 Y -) 9%
.)++% J ;% 2%+/8% -(#- $ 2%4%023 )0 8 #02 8 /0 -&+0 2%4%023 )0 % ;(/.( '%#23 -) / 2%4%023
)0 % 9&- 0)- )0 #06 b%0.%J -(% .)*4&-%2 2%4%02%0.% .)0-+#2/ 3 -(% 34%./$/%2 )0%6


C) 1%- +/2 )$ -(/3 /0.)03/3-%0.5J ;% */1(- #33&*% '/0% Y -) 9% $#&'-56 b%0.%J ;% .#0 .)*4&-%

-(#- / 2%4%023 )0 *)2%' 8#+/#9'% "
Y
6 M(%0 0); .)*4#+/01 -(% 34%./$/.#-/)0 ;/-( -(%
"#$%&'(#)!'%!*$+$&',)-!./&$01&'$%!1%2!3$%&#$4!

5@!
.)*4&-%2 2%4%02%0.% ;% 3&93-/-&-% "
Y
95 % #02 #0 #02 ;% .#0 0)- 2%+/8% #0 /0.)03/3-%0.5
#05*)+%6

C(% #&-()+3 )$ DM)-#;# #02 "))*+) YZZSJ \%/3.(' #02 "))*+) YZZ_J "))*+) YZZXF 4+%3%0- #
2%-#/'%2 $)+*#'/>#-/)0 )$ -(/3 /2%# #02 #'3) 4+%3%0- +&'%3 $)+ *)3- /*4)+-#0- '#01&#1%
#+-/$# 3 '/:% #0 #33/10*%0- 3-#-%*%0-J -(% /$B-(%0B%'3% 3-#-%*%0-J ;(/'% '))4 #02 4+).%2&+%36
L0 -(/3 .(#4-%+ ;% $).&3 )0 *%-()2 /08).#-/)0J 4#+#*%-%+ 3&93-/-&-/)0J +%-&+0 3-#-%*%0- #02
1')9#' 8#+/#9'%36

=/16 V6 "&* #02 4);%+ )$ /0-%1%+ 8#'&%3

L0 =/16 V ;% 3(); #0 %?#*4'% -(#- &3% # *%-()2 -) .)*4&-% -(% 3&* #02 4);%+ )$ /0-%1%+
0&*9%+36C(% 4+)1+#* .)0-#/03 )0% *%-()2 ;(/.( .)*4&-%3 -(% 0
-(
4);%+ )$ #0 /0-%1%+
0&*9%+6

L0 .)*4&-/01 -(% 2%4%02%0.%3 $)+ 4+).%2&+%3 #02 -(%/+ /08).#-/)03 ;% $/+3- .)*4&-% -(%
2%4%02%0.%3 )$ -(% 4+).%2&+% 9%/01 /08):%26 I$-%+;#+23 ;% 3&93-/-&-% -(% 4+).%2&+%T3
$)+*#' 4#+#*%-%+3 95 -(% # &#' )0%36 M% .#4-&+% +%.&+3/8% /08).#-/)03 95 .)*4&-/01 -(%
-+#03/-/8% #02 +%$'%?/8% .')3&+% )$ -(% 4+).%2&+%T3 9)25 #02 3&93%U&%0-'5 1%- +/2 )$
!"###$$#%&'#()*+,-,)*.#)/#(01 2#-&3'#
4"####5360,(#(01 #.375)8'&.#9#
:"# #,*-#,;# 1&-;#.37<#
="# #,*-# )5;#/<#

>"#### #?#
@"# #?# #
A"### #,#B# 1&-<#
C"## #8D,0'#E,#F# )5G#
H"# #9#
I"# ###.37#B#.37#J#5)8'&EK;/G<##
I"# ###$$#5) #E.37;#KG;#E.37;/G;#E.37;#5)8'&G################
4!"# ###,#B#,#J#4<## # #################
44"# #L#############################
4:"##
4="#####$$%&'M()*+,-,)*#)/#7'-D)+2#-&3'#
4@"# #,*-#5)8'&EK/;'/G#9#
4"# ###,*-#5)8'&#B#4<#
:"# ###8D,0'#E'/#N#!G#
:"# ###9#
="######## 5)8'&#B#5)8'&#O#4!<#
="###########$$,* '1+#)/#5)8'&#B#5)8'&#O#K/#
>"###########'/#B#'/#M#4<#
>"########L#
@"########&'-3&*#5)8'&<#
4A########$$#5) #9E5)8'&;5)8'&G;E5)8'&;K/G;#E5)8'&;#'/GL#
4A"####L#
4C"######
$$PQ31-,)*#!"#$%&'#()
$)
#######
4C
"
###
L

#
7(#'8',1&'$%!91)(2!:$2(4!;$,14'<()!"1/4&)!8#$0!=#$,(2/#14!=#$>#10)!

5A!
2%4%02%0.%3 /02&.%2 95 ').#' 8#+/#9'%36 =/0#''5J ;% #22 -()3% 2%4%02%0.%3 .#&3%2 95 -(%
4+).%2&+%T3 +%-&+0 8#'&%36
L0 =/16 V6J '/0% 0&*9%+ R ;% .#'' -(% *%-()2 #*X"% ;/-( 3)*% 4#+#*%-%+36 C(% 34%./$/.#-/)0
)$ *%-()2 /3 !"#$%&'()"$%&'(*)"#$%&'()"'+*)""#$%&'()",+*"-6 M(%0 .)*4&-/01 -(% 2%4%02%0.%3
$+)* -(% *%-()2 ;% 2%+/8% -(%3% 2%4%02%0.%3 !"#$%&'()"$%&'(*)"#$%&'()"'+*)"#'+)"'+*"

L0 -(%3% 2%4%02%0.%3 -(% 8#+/#9'% 4#/+ J"+1)"+N #+% 0)- /*4# /01 )8%+#'' 2%4%02%0.%36
") ;% (#8% $/0#' 2%4%02%0.%3 )$ *%-()2 #+% !"#$%&'()"$%&'(*)"#$%&'()"'+*"-6
M% (#8% -) *#4 $)+*#' -) # &#' 4#+#*%-%+3 $+)* -(%3% 2%+/8%2 2%4%02%0.%36

0%1232&2(3 4 56/$/7%&%$ 8'9)&2&'&2(3:
P%- 8 9% -(% 2%4%02%0.%3 )$ -(% # *%-()2 ' #02 '%- + 9% # $)+*#' 4#+#*%-%+ #02 $ -(%
.)++%34)02/01 # &#' 4#+#*%-%+6 C(% 2%4%02%0.%3 #$-%+ *%-()2 /08).#-/)0 #+% 1/8%0 95
/01"; 5/< =: > 51< =: ! ? @ " ; 5A< /: > 5A< 1: ! ? @-

I$-%+ *#44/01 4#+#*%-%+3 ;% 2%+/8%2 2%4%02%0.%3 #234)"$%&'(*)"#234)"+*") (%+% ;% $/02 #
.)0-+#2/ /)0 ;/-( -(% 4)3- .)02/-/)03 !"#234)",*)"#234)"+*)"#234)"$%&'(* "-6 L$ 4)3- .)02/-/)03
#+% .)03/3-%0- ;/-( .)*4&-%2 )0%3 -(%0 ;% /0-+)2&.% *)2%' 8#+/#9'%36 M% &3%2 &0/U&%
*)2%' 8#+/#9'% $)+ %8%+5 #33&*4-/)06 C(% ,%$/0/-/)0 Y 3-#-%3 -(#- (); -) %3-#9'/3( -(%
+%'#-/)03(/4 9%-;%%0 -(% +%-&+0 8#+/#9'%3 #02 -(% -#+1%- 8#+/#9'% )$ .#''/01 .)0-%?-6

0%1232&2(3 , 5B%&'$3 C/D'%) (1 / E%&F(?:

#I9D? c 4+).D#
Q
J #

Y
J dJ #
0
FF $
,D- c 4+).D#
Q
J #
Y
J dJ #
0
FF c
e-f % e#&D?J #F ! ,D4+).D#
Q
J #
Y
J dJ #
0
FFJ ? ! +%-&+0D4+).Ff
DQF

I9D- c 4+).D#
Q
J #
Y
J dJ #
0
FF $
,D- c 4+).D#
Q
J #

Y
J dJ #
0
FF c eD-J $
'
FJ D1J $H
'
F &1 ! gf
DYF

;(%+% - 2%0)-%3 -(% -#+1%- 8#+/#9'%J 1 1')9#' 8#+/#9'% /0 4+).D9)25F #02 +%-&+0D4+).F /3 #
$&0 /)0 +%-&+0/01 -(% +%-&+0 8#'&%3 )$ -(% 4+).%2&+% 4+).6 M(%+% I9 3();3 #90)+*#' #02
hI9 3();3 0)- #90)+*#' .)02/-/)036

C(% 2%$/0/-/)0 V 3-#-%3 (); -) %3-#9'/3( 2%4%02%0.%3 )$ -(% -;) .)03%.&-/8% 3-#-%*%0-3 )$
-(% 4+)1+#*6
'
)
0%1232&2(3 G 5H(7#()2&2(3:
g/8%0 -;) 2%4%02%0.5 +%'#-/)03 O
Q
J O
Y
! , )0 E #02 76 C(% .)*4)3/-/)0 )$ O
Q
#02 O
Y
/3
2%$/0%2 #3 $)'');3a

O
Q
%

O
Y
c

e D?J5F i (D?J>F ! O
Y
) (D>J5F ! O
Q
f "
e D?J5F i (D?J5F ! O
Q
) j(D?J>F ! O
Y
f "
e D?J5F i (D?J5F ! O
Y
) j(D5J>F ! O
Q
f

DVF

"#$%&'(#)!'%!*$+$&',)-!./&$01&'$%!1%2!3$%&#$4!

6B!
C(/3 2%$/0/-/)0 %03&+%3 -(#- 0) /0$)+*#-/)0 /3 ')3- 2&+/01 .)*4&-/01 -(% )8%+#'' 2%4%02%0.5

+%'#-/)0 $)+ # 4+).%2&+% )+ *%-()26 b%0.%J -(% $/+3- '/0% )$ -(% 2%$/0/-/)0 )$ .)*4)3/-/)0
(#02'%3 -(% .#3% ;(%+% -(%+% /3 # -+#03/-/8% 2%4%02%0.56 C(% 3%.)02 '/0% 3-#-%3 -(#- #''
2%4%02%0./%3 -(#- #+% 0)- +%B2%$/0%2 /0 5
6
#+% 3-/'' 8#'/26 L0 -(% -(/+2 '/0% #'' 2%4%02%0./%3
-(#- #+% 2%$/0%2 /0 5
6
#+% /0 -(% 0%; 2%4%02%0.5 3%- 4+)8/2%2 -(#- -(%+% /3 0) -+#03/-/8/-5
+%'#-/)06

=)+ .)*9/0/01 -(% 2%4%02%0./%3 )$ -;) .)03%.&-/8% 3-#-%*%0-3 ;% 2%$/0% -(% $)'');/01
.)*4)3/-/)0 )4%+#-)+ #3 1/8%0 /0 2%$/0/-/)0 V $)+ 2%4%02%0.5 +%'#-/)03 -) )9-#/0 -(%
$)'');/01 2%4%02%0.%3 7" #"234" 1"$%&'(#" ,")" +"*" *"1" !" #$%&'()" $%&'(*)"#$%&'()" '+*"-6 I$-%+
3&93-/-&-/01 $)+*#' -) # &#' 4#+#*%-%+ 2%+/8%2 2%4%02%0.%3 )$ '/0% 0&*9%+ R #+% !"#234)"
$%&'(*)"#234)"+*"- 9&- -(% 4)3- .)02/-/)03 #+% !"#234) ",*)"#234)"+*)"#234)"$%&'(*"-6 b%+% ;%
$/02 .)0-+#2/ /)0 9%-;%%0 9)-( 2%4%02%0.%3J 2%+/8%2 )0%3 #02 34%./$/%2 )0%36

L0 )+2%+ -) .)*4#+% # .)*4&-%2 2%4%02%0.% 3%- ;/-( -(% 34%./$/.#-/)0 ;% (#8% -) $/02 #
3&93-/-&-/)0 -(#- *#:%3 -(% .)*4&-%2 2%4%02%0.% 3%- %U&/8#'%0- -) -(% 34%./$/%2 )0%6 L$
-(%+% /3 0) 3&.( 3&93-/-&-/)0 -(% 3%-3 #+% 3#/2 -) 9% /0.)03/3-%0-6

I 3&93-/-&-/)0 ! /3 # $&0 /)0 ;(/.( *#43 *)2%' 8#+/#9'%3 -) # 3%- )$ 4+)1+#* 8#+/#9'%3J /6%6J
!a 7 ! Y
E
6 C(% +%3&'- )$ -(% #44'/.#-/)0 )$ -(% 3&93-/-&-/)0 ! )0 # 2%4%02%0.% +%'#-/)0 5 /3 #
2%4%02%0.% +%'#-/)0 ;(%+% #'' *)2%' 8#+/#9'%3 , /0 5 (#8% 9%%0 +%4'#.%2 95 !"#,*6

M% #33&*% -(#- 3-#-%*%0- R /3 #90)+*#'J ;% -#:% -(% -#+1%- 8#+/#9'% $+)* -(% #33/10*%0-
3-#-%*%0- #02 /0-+)2&.% # *)2%' 8#+/#9'% D3&*J $
R

F 6 L0 )+2%+ -) .)*4&-% 2%4%02%0.%3
;% 2%+/8% !"#234)"$
8
*)"#9)"9*)"#9)"2:%$*""- #02 -(% 3&93-/-&-/)0 8#+/#9'%3 #+% !"#",)"+)"$%&'(*"- 6 M%
0); .)*4#+% -(% 34%./$/.#-/)0 ;/-( -(% .)*4&-%2 2%4%02%0.%3 )9-#/0%2 95 3&93-/-&-/01 $
R"

;/-( !"#",)"+)"$%&' (* "-6 "/0.%J ;% .#0 0)- 2%+/8% #0 /0.)03/3-%0.5 #05*)+%J 3) '/0% 0&*9%+ I /3
# 9&1 .#02/2#-%6
")*% #33&*4-/)03 $)+ )-(%+ '/0%3 )$ *%-()2 .#''a
L$ ;% #33&*% 3-#-%*%0- G -) 9% /0.)++% -(%0 ;% .#0 -#:% -(% '%$- 8#+/#9'% $+)* -(%
#33/10*%0- 3-#-%*%0- #02 /0-+)2&.% # *)2%' 8#+/#9'% -) #++/8% #- #$%&'()" $
;
*6 I$-%+
.)*4&-/01 2%4%02%0.%3 ;/-( -(/3 *)2%' 8#+/#9'% ;% 2%+/8% !#$%&'()" $
;
*)" #'+)" '+*- #02
3&93-/-&-/)0 8#+/#9'%3 #+% !"#"$%&'()",+ )" '+*" -6 L$ ;% 0); .)*4#+% -(% 34%./$/.#-/)0 ;/-( -(%
.)*4&-%2 2%4%02%0.%3 )9-#/0%2 95 3&93-/-&-/01 $
;
95 !"#"$%&'()",+)"'+*"-) ;% .#0 0) ')01%+
2%+/8% /0.)03/3-%0.56 ") '/0% 0&*9%+ G $+)* *%-()2 /3 # 9&1 .#02/2#-%6

I33&*/01 '/0% , -) 9% /0.)++% J -(% 2%4%02%0.%3 2%+/8%2 ;/-( *)2%' 8#+/#9'% #+% !#"#$%&'()"
$%&'(*")"#$%&'()"$
"6
*)"#'+" )"'+*"*"-6 N); ;% 3&93-/-&-% -(% *)2%' 8#+/#9'% $+)* 3%- )$ 4+)1+#*
8#+/#9'%3J /6%J $
"6
"1"!"'+)",+)"$%&'("-6 I$-%+ 3&93-/-&-/01 ;% 2%+/8%2 !"#$%&'()"$%&'(*")#$%&'()"

,+*)"#$%&'()"'+*)"#' + ")"'+*"-6 L0 )+2%+ -) #''); -(% 2/+% .)*4#+/3)0 )$ 34%./$/%2 2%4%02%0.%3
;/-( -(% .)*4&-%2 )0%3J ;% /0-+)2&.% # 4+)]% /)0 )4%+#-/)0 ;(/.( 2%'%-%3 #'' 2%4%02%0.%3
$)+ 8#+/#9'%3 ;(/.( (#8% 0) /*4# )0 -(% )8%+#'' 2%4%02%0.%3J '/:% #0 /0-%+0#' 8#+/#9'% 4#/+
#'+)"'+* $+)* =/16 V6

I 4+)]% /)0 /3 2%$/0%2 )0 2%4%02%0.% +%'#-/)03 5"""7 #02 # 3%- )$ 8#+/#9'%3 <"
!
""=">"?6
C(% 4+)]% /)0 )$ 5 )0 < ;+/ %0 #3 &
I
DOF /3 2%$/0%2 #3 $)'');3 /0 #0 %U&#-/)0 Va

&
I
DOF c e D?J5F i D?J5F ' O * ? ! I
DSF
7(#'8',1&'$%!91)(2!:$2(4!;$,14'<()!"1/4&)!8#$0!=#$,(2/#14!=#$>#10)!

6C!
b);%8%+ ;(%0 #33&*/01 3-#-%*%0- 0&*9%+3 $+)* *%-()2 L1R1W1Y ;% )9-#/0 -(+%%
2/#10)3%36 P/0% 0&*9%+3 L1 R1W #+% 3#/2 -) 9% $#&'-5J 9&- '/0% 0&*9%+ ^ 2/2 1%- 3&93-/-&-/)0 3)
'/0% 0&*9%+ @ /3 0)- $#&'-56

0%1232&2(3 J 5K$%/&7%3& (1 LD(9/D C/$2/9D%):
C) )9-#/0 2%4%02%0.%3 $)+* 1')9#' 8#+/#9'%3 ;% #+% 2%#'/01 ;/-( -(% $)'');/01 $%#-&+%3

A.!BC%DEC"FE(9EDC'2"94$EG:"HC%DEC"FE(9EDC'2"

L$ # 1')9#' 8#+/#9'% 2%4%023 &4)0 1')9#' 8#+/#9'% /0 # 4+)1+#* -(%0 ;% &3% 3/*/'#+ +&'%3 -)
2%+/8% 2%4%02%0.%3 $+)* 3/*4'% 3-#-%*%0-36

=)+ #0 #33/10*%0- 3-#-%*%0- H"1"E"I"H -(% 2%4%02%0.%3 #+% #H)"E*)"#H)"H*6

6.!J%GEC"FE(9EDC'"94$EG:"HC%DEC"FE(9EDC'2"
"
L$ # 1')9#' 8#+/#9'% 2%4%023 )0 # ').#' 8#+/#9'% )$ # *%-()2 #02 -(% +%-&+0 8#+/#9'% #'3)
2%4%023 &4)0 -(% 3#*% ').#' 8#+/#9'% -(%0 ;% .#0 .)*4&-% 2%4%02%0.%3 #3a P%- / 9% -(%
2%4%02%0.%3 )$ -(% *%-()2 4 #02 '%- C 9% # ').#' 8#+/#9'%J H -(% .)++%34)02/01 1')9#'
8#+/#9'% #02 , 9% -(% +%-&+0/01 8#+/#9'% 6 C(% 2%4%02%0.%3 #$-%+ *%-()2 /08).#-/)0 /3 1/8%0
95 !"#H)",*"K"#H)C*""-J ;(%+% , /3 #,""
!
""(':3(L* 6

;.!M%(4EC"FE(9EDC'2"94$EG:"HC%DEC"FE(9EDC'2"/'$'L/'LG'2"
"
P%- / 9% -(% 2%4%02%0.%3 )$ # *%-()2 4 #02 '%- + 9% # $)+*#' 4#+#*%-%+J E 9% #0 # &#'
4#+#*%-%+ #02 H" -(% .)++%34)02/01 1')9#' 8#+/#9'%6 C(% 2%4%02%0.%3 #$-%+ *%-()2
/08).#-/)0 /3 1/8%0 95 !"#H)"E*"K"#H)"+*"
!
""/"-6

L$ ;% #33&*% #0 /08).#-/)0 -) 9% #90)+*#' ;% /0-+)2&.% # 3/01'% 8#+/#9'% $)+ %8%+5
) &++%0.% )$ # .%+-#/0 4+).%2&+%6 =)+ +%.&+3/8% /08).#-/)03 D/0 #'' .#3%3 ;(%+% ;% )9-#/0 #0
.5.'/. .#'' 1+#4(F ;% (#8% -) 4%+$)+* # $/?B4)/0- .)*4&-#-/)06 L0 )+2%+ -) 1&#+#0-%% -(#- -(%
.)*4&-%2 2%4%02%0.%3 /0.+%#3% *)0)-)0/.#''5 ;6+6-6 -(% 3&93%- +%'#-/)0 '/:% 2
0
c 2
0
k Q6
K)*4&-/01 $/?%2 4)/0- ;% #22 -(%3% 2%4%02%0.%3 -) )8%+#'' 2%4%02%0.%36

M2N- J 3();3 -(% $/?%2 4)/0- .)*4&-#-/)06

"#$%&'(#)!'%!*$+$&',)-!./&$01&'$%!1%2!3$%&#$4!

6D!



=/16 ^6 I'1)+/-(* )$ =/?%2 \)/0- K)*4&-#-/)0
!

L0


M2N- J 8
M
1) 81) 8
E%"5
#+% 8#+/#9'%3 ;(/.( #+% 3-)+/01 4#/+3 )$ 2%4%02%0.%3J ;(%+% 8 .)*4&-%3
0%; 2%4%02%0.%3J 8
E%"5
3-)+%3 4+%8/)&3 2%4%02%0.%36 8 .)*4+/3%3 )$ 9').: 2%4%02%0./%3
$+)* ;(/'% '))4 OF2D% H ?( ; P @ %3?6 C(% +%-&+0 3-#-%*%0- 3-)+%3 )8%+#'' 2%4%02%0.%3 /0-)
8 #$-%+ $/02/01 $/?B4)/0-6 =&0 /)0 -(0*( #223 9)-( 2%4%02%0.%3 +
%&'R#
"#E+
%&'R
#!#+

!
G6 C(%
.)*4)3/-/)0 )4%+#-)+ ! %03&+%3 -(#- 0) /0$)+*#-/)0 /3 ')3- 2&+/01 .)*4&-/01 -(% )8%+#''
2%4%02%0.5 +%'#-/)06 @0/)0 )4%+#-)+ /0 -(% #'1)+/-(* 3();3 -(#- 2%4%02%0./%3 #+% #22/01
#$-%+ %8%+5 /-%+#-/)0 )$ '))46 C(% .)02/-/)0 )$ -(% 8*) Z[) X>03" '))4 %03&+%3 -(#- ;(%0%8%+
4+%8/)&3 #02 0%; 2%4%02%0.%3 #+% 3#*% ;% +%#.(%2 $/?B4)/0- ;/-( $/0/-% 0&*9%+ )$
/-%+#-/)036 I$-%+ +%#.(/01 $/?B4)/0- ;% (#8% -) -%+*/0#-% '))4 #02 #22 .)*4&-%2
2%4%02%0.% 4#/+3 /0-) )8%+#'' 2%4%02%0.%3 86

KF%($%7 4 5M2A%? 6(23& H(7#'&/&2(3 KF%($%7:
C(% $/?%2B4)/0- .)*4&-#-/)0 #'1)+/-(* .)*4&-%3 # $/?B4)/0- $+)* +%4%-/-/8% /08).#-/)0
;/-(/0 # $/0/-% 0&*9%+ )$ /-%+#-/)036

N(%%+." M% 4+)8% -(/3 -(%)+%* /0 -;) 3-%43a =/+3- ;% 4+)8% -(#- -(% 2%4%02%0./%3 #+%
/0.+%#3/01 *)0)-)0/.#''56 "%.)02 -(%3% 2%4%02%0./%3 3()&'2 9%.)*% %U&#' ;/-(/0 $/0/-%
0&*9%+ )$ /-%+#-/)03 #- )0% 4)/0- ;(/.( /3 # $/?B4)/0-6

+! ,%4%02%0./%3 #+% /0.+%#3/01 *)0)-)0/.#''5 /6% ,0 2
/kQ
- 2
/
6
=+)* -(% #9)8% #'1)+/-(*J ;% :0); -(#- 2
/kQ
c 2
/
"# 9+
,
# !# "L6 C(% .)*4&-%2
2%4%02%0.%3 )$ 9').: 3-#-%*%0-3 #+% 3-)+%2 /0 26 <%.#&3% )$ -(% &0/)0 )4%+#-)+ /- /3

)98/)&3 -(#- 2
/kQ
- 2
/
6 C(/3 E+
,
#!#"G #+% -(% 0%; 2%4%02%0.%3 ;(/.( #+% #22%2 -)
-(% )'2 2%4%02%0.5 3%-6 C(/3 '%#23 -) # *)0)-)0/.#''5 /0.+%#3/01 #*)&0- )$
2%4%02%0.%36
+! =/? \)/0- K)*4&-#-/)0 /6%6 (0 2
/kQ
c 2
/
6
M% :0); -(#- 3%- )$ 8#+/#9'% 8 /3 $/0/-%6 b%0.%J 2
/
c E % E /3 $/0/-% #02 -(% &44%+
9)&02 )$ -(% 2%4%02%0.5 .)*4&-#-/)06 =+)* -(/3 $)'');3 -(#- /-%+#-/)0 #0 / %?/3-3
;(%0 2
/kQ
c 2
/
6
!
"#$%&#2#8D,0'#S#+)#9TL#'*+#
'%&$%&2#U#EU'5'*+'*('.G#
,*,-,10,V'#+#B#+!#
,*,-,10,V'#+
%&'R
#B#*300#

+)#
9#
####+
%&'R
#B#+#
####+#B#+
%&'R
#"#E+
%&'R
#!#+
!
G#
L#
8D,0'#EW+"'Q310.E+
%&'R
GG#
100+'5#B#E100+'5#!#+G#
#
7(#'8',1&'$%!91)(2!:$2(4!;$,14'<()!"1/4&)!8#$0!=#$,(2/#14!=#$>#10)!

6E!

DM/%'#02H3 YZZQ D=/?B4)/0- K)*4&-#-/)0FF K)*4'%?/-5 )$ .)*4&-/01 2%4%02%0.%3 $+)*
;(/'% '))4 -) +%#.(%2 $/?B4)/0- /0 # $/0/-% 0&*9%+ )$ /-%+#-/)036

C(%)+%* Q 3();3 -(% .)*4'%?/-5 )$ ;(/'% '))4 $)+ .)*4&-/01 -(% 2%4%02%0.%3 &3/01 -(%
#9)8% #'1)+/-(*6 L0 lM/%'#02 YZZQmJ -(% #&-()+ 4+)8%3 -(% -(%)+%* 2%4%02/01 &4)0 -(%
0&*9%+ )$ 8#+/#9'%3 &3%2 /0 2%4%02%0.%3 3%-6 C(/3 %?#*4'% +%4+%3%0-3 -(% ;)+3- .#3%6

L0 =/16 S ;% .#'' *%-()2 +%% +%.&+3/8%'56 b%+% ;% 3(); -(#- -(% 2%4%02%0.%3 )$ +%.&+3/8%'5

*%-()23 #+% /0 -(/3 $#3(/)0 ;/-( &3/01 $/?%2 4)/0- .)*4&-#-/)06 C(% 2%4%02%0.%3 )$ .#''/01
*%-()2 +%% (#3 $)'');/01 2%4%02%0.%3 !" #O),*)"#P)"O*)" #P)",*)" #('2)"P*)"#('2)" O*)" #('2)",*"-6 C(%
2%$/0/-/)0 ^ %03&+%3 -(#- ;% 3&93-/-&-/)0 )$ ').#'J 1')9#' 8#+/#9'%3 #+% 2%+/8%2 .)++% '56M%
&3% $/?%2 4)/0- #'1)+/-(* -) $/02 8
(
)G)8
(
)H)L6 L0 -(% +**"F$'#3" *%-()2 '/0% 0&*9%+ S (#3 #0
#33/10*%0- 3-#-%*%0- &)G)+**J$1)\N -(#- .#''3 # *%-()26 N); ;% (#8% -) 3&93/-&-% $)+*#' /0-)
# &#' 4#+#*%-%+3 $+)* .)*4&-%2 2%4%02%0.%3 )$ .#''/01 *%-()26 I$-%+ 3&93-/-&-/)0 ;%
2%+/8%2 $)'');/01 2%4%02%0.%3 J&1)$N1J&1)\N1J&1)%"6N6


=/16 S6 O%.&+3/8%'5 .#'' $)) $&0 /)0


4. Example applying on fixed point computation

M% 3(); -(% 3*#'' %?#*4'% -) $/02 $/?%2 4)/0- )8%+ -+#03/-/8% +%'#-/)036 L0 #0 %?#*4'%
4+)1+#* /0 =/16_ ;% .)*4&-% 2%4%02%0.%3 3-%4 95 3-%4 $#3(/)0 -) 3(); -(#- (); ;% +%#.(
$/?%2 4)/0- 95 &3/01 -(% #'1)+/-(* /0


M2N- J6 C(% 9)25 )$ ;(/'% '))4 /-%+#-%3 / -/*%36

4"####5360,(#,*-#/))'K1750'#9#
:"# ## ,*-#-;#1;#6<#
="# ## ?#
>"# ## ?#
@"#### ## -#B#/))E1;#6G#

@"# ## $$9E-;1G;E-;6G;E-;&'.GL# #
# 5&,R1-'#,*-#/))E,*-#K;#,*-#XG#9# #
A"### ## ,*-#&'.B!<#
C"## ## ,*-#VB4<#
H"# ## ,/#E#K#F#!#G#
I"# #### ###X#B#K<#
# ## '0.'#
4!"# #### ###
V#B#/))EKM4;#XG<# # #################
44"# ## &'.#B#V#J#X<#############################
4:"## ## &'-3&*#&'.<#
4:"####9((!)*+!,-+!).+!*-+!).+!,-+!)/01+! +!)/01+!*-+!)/01+!,-2!
4:"##
L
#
"#$%&'(#)!'%!*$+$&',)-!./&$01&'$%!1%2!3$%&#$4!

6F!
L0 4#+-/.&'#+ ;% #+% /0-%+%3-%2 -) .)*4&-% 2%4%02%0.%3 $+)* -(% 9)25 )$ '))4 #02
3&**#+/>%2 -(% '))4 2%4%02%0.%36 A98/)&3'5 #'' 8#+/#9'%3 #+% &3%2 /0 -(% -(% 9)25 )$ '))4
J$1) \1) /1)8) $(8) 0NJ ;(/.( 2%4%023 )0 8#+/#9'% 0J 9%.#&3% 0 #44%#+3 /0 -(% .)02/-/)0 )$ ;(/'%
'))46 M% 2/2 0)- 3(); -(% .)*4&-%2 2%4%02%0.%3 )$ / 8#+/#9'% /0 =/16_ #02 -(% 1+#4(H3 =/16
X6 C(% 2%4%02%0.%3 )$ / #+% %?4'/./-'5 .)*4&-#9'% #02 2) 0)- *#:% #05 .(#01%3 2&+/01
/-%+#-/)0 )$ '))46 C(/3 /3 -(% +%#3)0 ;% #+% 0)- 4+%3%0-/01 -(%3% 2%4%02%0.%36 C(%
2%4%02%0.%3 )$ )-(%+ 8#+/#9'%3 #+% 4+%3%0-%2 /0 -(% 9%');6 C(% #'1)+/-(* 4+%3%0-%2
4+%3%0-%2 %#+'/%+ )&-'/0%3 -(% *%-()2 )$ .)*4&-/01 2%4%02%0.%3 $+)* +%.&+3/8% /08).#-/)06

C(% 2%4%02%0.%3 )$ -(% 8#+/#9'%3 #n 9n . #02 2 #+% 2%4/ %2 /0 -(% $)'');/01
.#3%3a
+! D/ c ZFa #02 2

Z
c eD#J 9FJ D9J .FJ D.J 2FJ D2J %Ff
+! D/ c QFa #02 2
Q
c eD#J 9FJ D9J .FJ D.J 2FJ D2J %FJ D.J %FJ D#J .FJ D9J 2Ff
+! D/ c YFa #02 2
Y
c eD#J 9FJ D9J .FJ D.J 2FJ D2J %FJ D.J %FJ D#J .FJ D9J 2FJ D#J 2FJ D9J %Ff
+! D/ c VFa #02 2
V
c eD#J 9FJ D9J .FJ D.J 2FJ D2J %FJ D.J %FJ D#J .FJ D9J 2FJ D#J 2FJ D9J %FJ D#J %Ff
+! D/ c ^Fa #02 2
^
c eD#J 9FJ D9J .FJ D.J 2FJ D2J %FJ D.J %FJ D#J .FJ D9J 2FJ D#J 2FJ D9J %FJ D#J %Ff

+! C(% 1/8%0 34%./$/.#-/)0 $+)* =/16 _ #02 .)*4&-%2 2%4%02%0.%3 #+% %U&#'6 ") ;%
+%#.(%2 $/? 4)/0- ;/-( $/0/-% 0&*9%+ )$ /-%+#-/)06 M% $/02 2
V
c 2
^
/3 # $/?B4)/0-
;(%+% 9)-( 2%4%02%0.%3 #+% %U&#'6 M(/.( %03&+%3 2
0
c 2
0kQ
6
I'' 8#+/#9'%3 )$ #33/10*%0- 3-#-%*%0-3 #+% 2%4%02 &4)0 8#+/#9'% / . D#J 9J .J 2F6

=/16 _6 `?#*4'% 4+)1+#* 3();/01 -+#03/-/8% 2%4%02%0./%3 /0 # '))4


C(% 1+#4( 3();3 -(% 2%4%02%0.%3 )$ -(% ;(/'% '))46 C(%+% #+% $/8% 0)2%3 /0 g+#4(
/0.'&2/01 /0.)*/01 #02 )&-1)/01 %21%36 <#3%2 )0 C(%)+%* Q ;% .#0 4+)8% -(#- #$-%+ # $/0/-%
0&*9%+ )$ /-%+#-/)03 ;% +%#.(%2 $/?%2 4)/0-6 ;% %?4'#/0 #3 $)'');3a

,%4%02%0.%3 #+% /0.+%#3/01 *)0)-)0/.#''5 /6%6 ,0 2
/kQ
- 2
/
6 M% 4+%3%0- -(% #'1)+/-(* /0

86QH ;5/< 9:<59< *:<5*< ?:< 5?< %:< 5*< %:< 5/< *:<59< ?:< 5/< ?:< 59< %:<5/< %:@

#

4"#5360,(#,*-#YD,0'PK1750'#9#
:"## ,*-#1;#6;#(;#+;#'<#
="## ,*-#,#B#!<#
>"## """"#
@"## """"#
A"## 8D,0'#E,#/#@G#
C"## 9#
H"## # 1#B#6<#
I"## # 6#B#(<#
4!"## # (#B#+<#
44"## # +#B#'<#
4:"## # ,#B#,#J#4<#
# L#
####L#
7(#'8',1&'$%!91)(2!:$2(4!;$,14'<()!"1/4&)!8#$0!=#$,(2/#14!=#$>#10)!

65!

M2N- J $)+ .)*4&-/01 2%4%02%0.%3 )$ 9').: 3-#-%*%0-36 <%.#&3% )$ -(% &0/)0 )4%+#-)+ /- /3
)98/)&3 -(#- 2
/kQ
- 2
/
6 C(/3 '%#23 -) # *)0)-)0/.#''5 /0.+%#3/01 #*)&0- )$ 2%4%02%0.%36
;(/.( /3 4+)8%0 /0 C(%)+%* Q6 I$-%+ .#'.&'#-/01 0 ;% $/02 $/?B4)/0- ;/-(/0 $)&+ /-%+#-/)030 c
^6 C(%+%$)+%J ;% +%#.(%2 #- $/?%2 4)/0- #$-%+ # $/0/-% 0&*9%+ )$ /-%+#-/)03 ;(/.( /3 2%3.+/9%2
/0 -(% #'1)+/-(*6 C(% 1+#4( 3();3 -(% 1+#4(/.#''5 +%4+%3%0-#-/)0 )$ $/02/01 $/?%2 4)/0-6

=/16 Xa C(% ,/#1+#4( )$ -(% M(/'% P))4 $+)* =/16_
5. Experimental Results and Discussions

C(% 4+)4)3%2 *)2%' (#3 9%%0 /*4'%*%0-%2 /0 G#8# &3/01 -(% `.'/43% 4'#-$)+*6 L0 -(/3
3% /)0J ;% 4+%3%0- -(% %?4%+/*%0-3 -(#- %8#'&#-% -(% +%3&'- &3/01 2%4%02%0./%3 /0 G#8#
4+)1+#*3 ;/-()&- &3/01 )9]% B)+/%0-%2 $%#-&+%36 `?4%+/*%0-3 ;%+% 4%+$)+*%2 )0 # L0-%'
\%0-/&* ^ M)+:3-#-/)0 DV gboJSQY 7< 7%*)+5F +&00/01 g%0-)) P/0&? Dg%0-)) <#3%
"53-%* E%+3/)0 Q6^6RJ [%+0%' 8%+3/)0 Y6_6SF6 C(% +%3&'-3 #+% +%4)+-%2 /0 C#9'% Q6

=)+ 8#+/)&3 %?#*4'%3 4+)1+#*3J ;% /0-+)2&.%2 # 3/01'% $#&'-J #02 #$-%+;#+23 .)*4&-%2 #''
3/01'%B$#&'- 2/#10)3%36 C#9'% Q 4+%3%0-3 %*4/+/.#' +%3&'-3 )$ 4+)1+#*3 ;/-( *%-()236 M%
.)03/2%+%2 *%2/&* 3/>%2 4+)1+#*36 C(% 3%.)02 .)'&*0 3();3 -(% '/0%3 )$ .)2% $+)* Y_ -)
SZR6 C(% -(/+2 .)'&*0 .)&0-3 -(% 0&*9%+ *%-()23 /0 -(% 4+)1+#*36 C(% $)&+-( .)'&*0
+%4)+-3 -(% 0&*9%+ )$ 2/#10)3/3 .#02/2#-%36 C(% S
-(
.)'&*0 1/8%3 -(% 0&*9%+ )$ /04&-
8#+/#9'%3 #02 -(% '#3- .)'&*0 3();3 -(% 0&*9%+ )$ )&-4&- 8#+/#9'%36

L0 C#9'% Q -(% -%3-%2 4+)1+#*3 .)03/3-/01 )$ 3/*4'% #02 *&'-/4'% 3-#-%*%0-3J '))43J *%-()23
#02 1')9#' 8#+/#9'%36 \+)1+#* 34%./$/.#-/)0 .)03/3-3 )$ #'' 8#+/#9'%3 /6%6J /04&- 8#+/#9'%3 #02
)&-4&- 8#+/#9'%36 L0 =/16 ;% 3(); #'' -(%3% 4+)1+#*3 ;/-( */0/*&*J *#?/*&* #02 #8%+#1%
)$ 2/#10)3%36

L0 =/16 W ;% 4+%3%0-%2 # 1+#4( )$ 4+)1+#*3 ;/-( -(% 0&*9%+ )$ $#&'-3 2%4%02/01 )0 )&-4&-
8#+/#9'%3 D4+%3%0-%2 /0 C#9'% QF6 M% &3%2 (&02+%2 /-%+#-/)0 $)+ %8%+5 4)33/9'% .)*9/0#-/)0
)$ )&-4&- 8#+/#9'%36 L- 3();3 -(% */0/*&*J *#?/*&* #02 *%#0 )$ 2/#10)3%3 /0 +%34% -)
"#$%&'(#)!'%!*$+$&',)-!./&$01&'$%!1%2!3$%&#$4!

66!
-(% )&-4&- 8#+/#9'%3 #02 -(% $#&'-36 L0 #'' 1+#4(3 $&'' '/0% +%4+%3%0- -(% */0/*&*J 2#3( '/0%
-(% *#?/*&* #02 2) %2 '/0% -(% 0&*9%+ )$ 2/#10)3%3 .#02/2#-%36


6$(N$/7) RSH E%&F(?) 02/N-T( U3#'&VCWB S'&#'&VCWB
I22%+ SQ V _ QW QS
I22&'3%C/*% VXW YQ QZ RW S^
`U&#-/)0 Y_ ^ ^ QV S
7#-(=&0 /)03 SZR YY V WZ SX
7%-()2C%3-Q ^Y V QY Q^ QQ
7%-()2C%3-Y XS S Y YY Q_
7%-()2C%3-V ^_ S V YZ Q^
6$(N$/7) RSH E%&F(?) 02/N-T( U3#'&VCWB S'&#'&VCWB
7%-()2C%3-^ YQW QS V SV ^^
C#9'% Q6 ,/#10)3/01 .#02/2#-%3 )9-#/0%2 95 #0 /0-+)2&./01 # 3/01'% $#&'-

N)-%J /0 -(% 1+#4( ;% .)03/2%+ )0'5 -()3% 2/#10)3%3 ;(/.( (#3 .)0-+#2/ /)06 C(/3 *%#03
-(#- ;% 0%8%+ 4/.: 8#'&%3 ;(/.( '%#2 -) 0) .)0-+#2/ /)06 =&'' '/0% /02/.#-%3 -(#- ;(%0 ;%
/0.+%#3% -(% 0&*9%+ )$ )&-4&- 8#+/#9'%3 &3%2 /0 -(% 34%./$/.#-/)0J #02 -(%0 -(% 0&*9%+ )$

2/#10)3/3 /0.+%#3%36 C(% +%3&'-3 /02/.#-% -(#- )&+ #44+)#.( /3 $%#3/9'% $)+ 2%-% /01 #02
').#'/>/01 +%#' .#&3% )$ */39%(#8/)&+6 C(% +%3&'-3 4+%3%0-%2 -(%+% 3)'%'5 3-%* $+)*
4+).%2&+#' 4+)1+#*36

6. Limitation

@0'/:% 4+%8/)&3 #44+)#.(%3 DM/%'#02 YZZQJ 7#5%+6 M p "-&*4-0%+ 7 YZZVF J -(% 2%9&11/01
#44+)#.( /0-+)2&.%2 /0 -(/3 .(#4-%+ 34%./#''5 /0-%023 ').#'/>/01 3-+& &+#' $#&'-36 L0 #0
# )&0- )$ -(/3J ;% $).&3 -(/3 2/3.&33/)0 )&+ *)2%'H3 ;%#:0%33%3 /0 2%-% /01 #02 ').#'/>/01
-(%3% $#&'-36

7(#'8',1&'$%!91)(2!:$2(4!;$,14'<()!"1/4&)!8#$0!=#$,(2/#14!=#$>#10)!

6?!
=/+3-J -(% 4+)4)3%2 *)2%' /3 0)- ').#'/>/01 $#&'-3 .#&3%2 95 # %++)0%)&3 -#+1%- 8#+/#9'%6 =)+
%?#*4'%J -(% $)'');/01 .)2% 30/44%- #33/103 -(% 8#'&% )$ 8#+/#9'% =) -) -;) 2/$$%+%0-
8#+/#9'%3J 0#*%'5J F #02 7<

4- = " X !! )F('D? 9% A " X
,- = " X

,&% -) D3-+& &+#'F %++)+ /0 '/0% QJ ;% )9-#/0 J71) =N #3 # 3/01'% 2%4%02%0.%J 9&- )&+
34%./$/.#-/)0 %0&*%+#-%3)Z)JF1)=N1)J71)=N) [ #3 2%4%02%0.%36 "/0.% ZJ71)=N[ /3 0)- 3&4%+ 3%- )+ %U&#'
-) ZJF1)=N1)J71)=N)[J ;% :0); 3)*% -(/01 *&3- 9% ;+)016 C(&3J )98/)&3'5J ;% .#0 2%-% -(/3
9&16 L0 ').#'/>/01 -(% 2%-% %2 */39%(#8/)&+H3 .#&3%J ;% #33&*% 3-#-%*%0- Q -) 9% #90)+*#'
#02 )9-#/0 ]
L
)G)Z)JF1)!
L
N[<)]

R
)G)ZJF1)=N[1)$(8)]
L
)")]
R
)GZJF1)=N[6 C(/3 +%3&'-3 3();3 -(#- ;% .#0 0)-
$/02 #05 3&93-/-&-/)0 ( -(#- +%*)8%3 -(% 1/8%0 .)0-+#2/ /)06 E/+-&#''5J ;% .#0 0)- .)++%
)&+ 4+)1+#* 95 *)2/$5/01 3)'%'5 -(% +/1(-B(#02 3/2% )$ 3-#-%*%0- Q6

I'');/01 $)+ *)2%' 8#+/#9'% '%$- (#02 3/2% ;% .#0 1%- +/2 )$ -(/3 2%$/./%0.5a OQ c e D!
L
J >Ff O
Y
J
c e D5J >F fJ ;(/.( 5/%'23 -) O
Q
) O
Y
c e D!
L
J >FJ D?J >F f6)I44#+%0-'5J -(% 3&93-/-&-/)0 (D)!
L
F c ?
$&'$/'3 eD!
L
J >FJ D5J >Ff /3 3&4%+ 3%- )+ %U&#' -) ZJF1)=N1)J71)=N[)#02 -(&3 ;% .#0 ').#'/>% -(/3 $#&'-6
b);%8%+J -(/3 /3 #44+)8%2 3)'%'5 /0 -(% +#-(%+ #+% .#3% -(#- -(% 3&93-/-&-%2 8#+/#9'% 2)%3 0)-
#44%#+ /0 #05 )$ -(% 3-#-%*%0-3 D)0 -(% +/1(- (#02 3/2%F 3&93%U&%0- -) 8#+/#9'%3
3&93-/-&-/)06 C(&3J -(/3 #44+)#.( /3 0)- #44'/.#9'% /0 # 4+# /.#' 3% /016

=&+-(%+*)+% )&+ *)2%' /3 )0'5 #44'/.#9'% -) #'/#3B$+%% 4+)1+#*36 C(/3 /33&% /3 $&+-(%+
2/3.&33%2 /0 DG#.:3)0 QRRSF6 L0 3&**#+5 ;% .)0.'&2% -(#-J 4+)8/2%2 -(% $#&'- #44%#+3 )0
-(% +/1(- (#02 3/2% )$ #0 #33/10*%0-J )&+ *)2%' #'');3 $)+ .)++% /01 -(%3% :/02 )$
3-+& &+#' $#&'-3 #3 ')01 #3 ;% .#0 #'3) 2%-% -(/3 $#&'- /0 -%+*3 )$ -(% 34%./$/%2
2%4%02%0.%36



















"#$%&'(#)!'%!*$+$&',)-!./&$01&'$%!1%2!3$%&#$4!

6@!

=/16 W6 "%03/-/8/-5 I0#'53/3 )$ I'' \+)1+#*3 )$ C#9'% Q6

7. Related Research

C(% #&-()+ )$ DG#.:3)0 QRRSF 4+%3%0-3 ;)+: ;(/.( /3 .')3%3- -) -(% ;)+: 4+%3%0-%2 (%+%/06
C(/3 ;)+: %*4')53 #93-+# 2%4%02%0.%3 $)+ 2%-% /01 +#-(%+ -(#0 $)+ ').#'/>/01 # $#&'-6
=&+-(%+*)+% /0 D[&4%+ QRWRJ M/%'#02 YZZQF -(% #&-()+3 %*4')5 -(% 0)-/)0 )$ 2%4%02%0.%3
$)+ $#&'- ').#'/>#-/)06 L0 .)0-+#3- -) -(% '# %+ #44+)#.( ;% 2) 0)- %*4')5 2%-% %2
2/$$%+%0.%3 /0 8#+/#9'% 8#'&%3 #- # .%+-#/0 '/0% /0 .)2% 9&- *#:% &3% )$ 2/$$%+%0.%3 9%-;%%0
34%./$/%2 #02 .)*4&-%2 2%4%02%0.%3 #02 -(&3 #'3) /0.)+4)+#-% -(% 3-+& &+#' 4+)4%+-/%3 )$
4+)1+#* #02 34%./$/.#-/)06

L0 -(% +%.%0- 4#3- -(% #&-()+3 DM)-#;# YZZZJ M)-#;# %-J #'6 QRRRJ [)%9 #02 M)-#;# YZZ^J
7#5%+ #02 "-&*4-0%+ YZZVF 2%8%')4%2 *)2%'3 $)+ 2/$$%+%0- '#01&#1%3 #- 8#+/)&3 #93-+# /)0
'%8%'3 /0 -(% *)2%'B9#3%2 .)0-%?-6 L0 1%0%+#'J #93-+# *)2%''/01 #44+)#.(%3 3#.+/$/.% 2%-#/'
/0 $#8)&+ )$ .)*4&-#-/)0#' .)*4'%?/-5 ;(%+%#3 *)+% 2%-#/'%2 8#'&%B'%8%' *)2%'3 DM)-#;#

YZZYJ \%/3.(' #02 M)-#;# YZZVF 4+)8/2% # &+#-% $#&'- ').#'/>#-/)0 .#4#9/'/-/%3 9&- )0 -(%
)-(%+ (#02 +%U&/+% .)03/2%+#9'% .)*4&-#-/)0#' +%3)&+.%3 /0 -%+*3 )$ 34#.% #02 .)*4&-/01
4);%+6

I'-()&1( 4+)1+#* 3'/./01J #3 # '/1(-;%/1(- -%.(0/U&%J (#3 3%%0 3& %33$&' #44'/.#-/)0 /0
$#&'- ').#'/>#-/)0 DI1+#;#' %-J #'6J QRRVJ =+/->3)0 %-J #'6J QRRRJ P5'% #02 M%/3%+ QRWX F /-3
2/3.+/*/0#-/)0 '/:% 7<", DM)-#;# YZZZF6 L0 D[&4%+ QRWRJ M/%'#02 YZZQF -(% #&-()+3
%*4')5 -(% 0)-/)0 )$ 2%4%02%0.%3 $)+ $#&'- ').#'/>#-/)06 L0 .)0-+#3- -) '# %+ #44+)#.( ;% 2)
0)- %*4')5 2%-% %2 2/$$%+%0.% /0 8#+/#9'% 8#'&%3 #- # .%+-#/0 '/0% /0 .)2% 9&- &3% )$
2/$$%+%0.%3 9%-;%%0 34%./$/%2 #02 .)*4&-%2 2%4%02%0./%3 #02 -(&3 #'3) /0.)+4)+#-% -(%
3-+& &+#' 4+)4%+-/%3 )$ 4+)1+#* #02 34%./$/.#-/)06 C(&3J -(% *)2%'3 /0-+)2&.%2 /0 D[&4%+
QRWRJ M/%'#02 YZZQF .#0 0)- 2%#' ;/-( #33%+-/)03 )+ 4+%B #02 4)3- .)02/-/)03 /0 #
3-+#/1(-$)+;#+2 ;#56
7(#'8',1&'$%!91)(2!:$2(4!;$,14'<()!"1/4&)!8#$0!=#$,(2/#14!=#$>#10)!

6A!

C(% #&-()+3 )$ DM)-#;# YZZZJ =+/%2+/.( %-J #'6J QRRRJ M/%'#02 YZZQF 3)'%'5 *#:% &3% )$
.)0.+%-% 8#'&%3 /0 /0.)+4)+#-/01 .)++% 0%33 /0$)+*#-/)06 C(%3% *)2%'3 2) 0)- #''); -#:/01
#28#0-#1% )$ #+9/-+#+5 +%'#-/)03(/43 9%-;%%0 3%8%+#' 8#+/#9'%3 )+ 8#+/#9'%3 #02 .)03-#0-36
C(% #&-()+ D"-&*4-0%+ YZZQF 3();3 -(#- ').#'/>/01 3-+& &+#' $#&'-3 +%U&/+%3 %?4')/-/01
2%3/10 /0$)+*#-/)0 '/:% #33%+-/)03J #02 4+% #02 4)3- .)02/-/)036

A-(%+ #44+)#.(%3 '/:% DG)(03)0 QRW_F $).&3 )0 0)8/.% 4+)1+#**%+3 #02 *#:% &3% )$
*%-()23 -(#- (%'4 -) $/02 $#&'-3 /0 -(% .)2% 95 .)*4#+/01 -(% .)2% ;/-( 4+%B34%./$/%2
4+)9'%* $)+*&'#-/)036

7. Conclusion

L0 -(/3 .(#4-%+ ;% %?-%02 #02 4+%3%0- -(% 0)8%' *)2%' ;(/.( 2%-% #02 ').#'/>% +%#' $#&'-3

$+)* 4+)1+#*3J .)*4+/3/01 *%-()23 /08).#-/)03 #02 1')9#' 8#+/#9'%36 C(% .)*4&-#-/)0 )$
2%4%02%0.%3 $+)* +%.&+3/8% /08).#-/)0 ;% $)&02 -(#- %8%+5 /-%+#-/)0 (#3 #22%2 0%;
2%4%02%0.%3 #02 -(% 0&*9%+ )$ 2%4%02%0.%3 /0.+%#3/01 *)0)-)0/.#''56 L0 )+2%+ -)
1&#+#0-%% -(#- -(% .)*4&-%2 2%4%02%0.%3 #+% /0.+%#3/01 *)0)-)0/.#''5 ;6+6-6 /-%+#-/)03J ;%
$/02 $/?%2B4)/0- ;(%+% #'' 2%4%02%0.%3 #+% %U&#' ;/-( $/0/-% 0&*9%+ )$ /-%+#-/)036 M%
4+%3%0-%2 #0 #'1)+/-(* #02 4+))$ $)+ $/?%2 4)/0- .)*4&-#-/)0 ;(/.( %03&+%3 -(#- 0)
2%4%02%0.%3 ')33 2&+/01 /-%+#-/)0 #02 ;% #';#53 +%#.(%3 $/?B4)/0- #$-%+ # $/0/-% 0&*9%+ )$
/-%+#-/)036 7)+%)8%+J -(% #44+)#.( /3 2/$$%+%0- -) )-(%+ #8#/'#9'% 2%4%02%0.5B*)2%'3 #02
4+)8/2%3 9% %+ +%3&'-3 $)+ *%2/&* 3/>%2 4+)1+#*36

I $&-&+% +%3%#+.( .(#''%01% /3 -(% $)+*#' #02 %*4/+/.#' %8#'&#-/)0 )$ -(% *)2%''/01
#44+)#.(%3 ;(%0 #44'5 /- -) +%#' )9]% B)+/%0-%2 4+)1+#*36

8. Acknowledgments

C(/3 ;)+: ;#3 3&44)+-%2 95 -(% b/1(%+ `2&.#-/)0 K)**/33/)0J L3'#*#9#2J \#:/3-#0 &02%+
/-3 +%3%#+.( #02 2%8%')4*%0- $&02/016 C(% #&-()+ ;)&'2 '/:% -) -(#0: \+)$6 =+#0> M)-#;#
#02 <%+0(#+2 \%/3.(' $)+ -(%/+ 8#'&#9'% .)**%0-3 #02 #28/.% 2&+/01 *5 3-#5 #- g+#>
@0/8%+3/-5 )$ C%.(0)')15J I&3-+/#6


8. References

\%/3.('6 < p M)-#;# =6 DYZZVF6 7)2%'B<#3%2 ,/#10)3/3 )+ O%#3)0/01 $+)* =/+3- \+/0./4'%3J
I999)I(&"330:"(&).76&"'6J QWDVFJ D7#5BG&0% YZZVF 4#1% 0&*9%+3 DVYqVXFJ QS^QBQ_XY6
\%/3.(' <6n "))*+) "6 p M)-#;# =6 DYZZ_F6 I93-+# ,%4%02%0.% 7)2%' /0 ")$-;#+%
,%9&11/016 E%*/""80(:6) *+) &>") L^&>) I(&"%($&0*($3) _*%?6>*#) *() E%0(/0#3"6) *+) !0$:(*606)
J!O@M`N J =+#0.%6
G#.:3)0 ,6 DQRRSF6 I34% a ,%-% /01 <&13 ;/-( I93-+# ,%4%02%0.%3J B,a)C%$(6$/&0*(6)*())
))))) .*+&X$%")9(:0(""%0(:)$(8)a"&>*8*3*:7J ^DYFJ DI4+/' QRRSF 4#1% 0&*9%+3 DQZRBQ^SFJ

QZ^RBVVQ?6
M/%'#02 ,6 DYZZQF6 7)2%' <#3%2 ,%9&11/01 )$ G#8# \+)1+#*3 @3/01 ,%4%02%0.%36 E>!))
"#$%&'(#)!'%!*$+$&',)-!./&$01&'$%!1%2!3$%&#$4!

?B!
) C>"606J) b0"(($) 4(05"%60&7) *+) C"/>(*3*:71) ,*'#-&"%) ./0"(/") !"#$%&'"(&1) I(6&0&-&") *+)
I(+*%'$&0*().76&"'6)JLcYN[1)!$&$\$6")$(8)B%&0+0/0$3)I(&"330:"(/")U%*-#)DQW^rYFJ E/%00#J
I&3-+/#6
[)%9 ,6 p M)-#;# =6 DYZZ^F6 L0-+)2&./01 I'/#3 L0$)+*#-/)0 /0-) 7)2%'B<#3%2 ,%9&11/016
E%*/""80(:6)*+)L `&>)9-%*#"$(),*(+"%"(/")*()B%&0+0/0$3)I(&"330:"(/")J9,BIN J LA" \+%33J 44
WVVBBWVXJ E#'%0./#J "4#/06
M)-#;# =6DYZZZF6 A0 -(% O%'#-/)03(/4 9%-;%%0 7)2%'B<#3%2 ,%9&11/01 #02 \+)1+#*
"'/./01J B%&0+0/0$3) I(&"330:"(/"J QVSDQBYF D=%9+&#+5 YZZYFJ 4#1% 0&*9%+3 DQY^qQ^VFJ
ZZZ^BVXZY6
M)-#;#6 =DYZZYF6 ,%9&11/01 b#+2;#+% ,%3/103 &3/01 # E#'&%B9#3%2 7)2%'J B##30"8)
I(&"330:"(/"J Q_DQF J DG#0&#+5B=#9+&#+5 YZZYF 4#1% 0&*9%+3 DXQqRYFJ ZRY^B__R?6
M)-#;# =6 p "))*+) " DYZZSF6 @3/01 #93-+# 2%4%02%0./%3 /0 2%9&11/016 E%*/""80(:6)*+))
) Ld&>)I(&"%($&0*($3)_*%?6>*#)*()e-$30&$&05")]"$6*(0(:)e]@MfJ 446 YVBBYWJ I&3-+/#6
M)-#;# =6 p "))*+) " DYZZSF6 =#&'- P).#'/>#-/)0 <#3%2 )0 I93-+# ,%4%02%0./%36
E%*/""80(:6)*+)&>")Lc&>),*(+"%"(/")*()I(&"%($&0*($3),*(+"%"(/")*()I(8-6&%0$31)9(:0(""%0(:)
g)h&>"%)B##30/$&0*(6)*+) B##30"8) I(&"330:"(&).76&"'6) JI9BSBI9)RMMfNJ P% &+% N)-%3 /0
I+-/$/./#' L0-%''/1%0.% DPNILFJ446 VSXBVSRJ "4+/01%+ E%+'#1J L-#'56
=+/%2+/.( g6n "-&*4-0%+ 76 p M)-#;# =6DQRRRF6 7)2%'B9#3%2 ,/#10)3/3 )$ b#+2;#+%
,%3/103J B%&0+0/0$3)I(&"330:"(/"J QQQDQBYFJ 4#1% 0&*9%+3 DVqVRFJ ZZZ^BVXZY6
I1+#;#'J b6n I6 ,%*/'')n O/.(#+2 p b6 "4#$$)+2n `&1%0% DQRRVF6 ,%9&11/01 ;/-( 250#*/.
3'/./01 #02 9#.:-+#.:/01 ")$-;6 E%$/&0/") $(8) 9F#"%0"(/"J YVD_FJ DG&0% QRRVFJ 4#1%
0&*9%+3 D SWRq_Q_F J ZZVWBZ_^^6
P5'% G6O6 p M%/3%+6 7 DQRWXF6 I&-)*#-/. \+)1+#* <&1 P).#-/)0 95 \+)1+#* "'/./016
E%*/""80(:6)*+)&>") R(8)I(&"%($&0*($3),*(+"%"(/")*(),*'#-&"%6)$(8)B##30/$&0*(6J 446 WXXBB
WWYJ <%/1/01 D\%:/01FJ K(/0#6

G)(03)0 M6 P%;/3 DQRW_F6 L0-%0-/)0B<#3%2 ,/#10)3/3 )$ N)8/.% \+)1+#**/01 `++)+3J L0a
a*%:$()D*++'$(JD`26FJ 4#1% 0&*9%+3 DVVVF P)3 I'-)36
M%/3%+ 76 DQRW^F6 \+)1+#* "'/./01J I999) C%$(6$/&0*(6) *() .*+&X$%") 9(:0(""%0(:J QZD^FJ DG&'5
QRW^F J D^VRB^^RF ZBWRXRQBQ^_B_6
"-&*4-0%+ 76 DYZZQF6 @3/01 ,%3/10 L0$)+*#-/)0 -) L2%0-/$5 "-+& &+#' ")$-;#+% =#&'-3J
E%*/""80(:6)*+)&>")LY&>)B-6&%$30$()i*0(&),*(+"%"(/")*()B%&0+0/0$3)I(&"330:"(/"J 446 ^XVB^W_J
P)02)0J "4+/01%+BE%+'#1J @[6
=+/->3)0 \6n g5/*)-(5 C6n [#*:#+ 76 p "(#(*%(+/ N6 DQRRRF6 g%0%+#'/>%2 I'1)+/-(*/.
,%9&11/01 #02 C%3-/01J E%*/""80(:6) *+) &>") B,a) .IUEjB2kdL) ,*(+"%"(/") *()
E%*:%$''0(:)j$(:-$:")!"60:()$(8)I'#3"'"(&$&0*(1)44 VQXqVY_J C)+)0-)J K#0#2#J
[&4%+ O)06 L6 DQRWRF6 ,%4%02%0.5 ,/+% %2 ').#'/>#-/)0 )$ 3)$-;#+% 9&136 C"/>(0/$3) ]"#*%&
ILBCO QZSVJ 7LC IL P#9J 7#5 @"I6
7#5%+ M6 p "-&*4-0%+ 76 DYZZVF6 `?-%02/01 ,/#10)3/3 -) ,%9&1 \+)1+#*3 ;/-(
`?.%4-/)036 E%*/""80(:6) *+) &>") Lc&>) I999) I(&"%($&0*($3) ,*(+"%"(/") *() B-&*'$&"8)
.*+&X$%") 9(:0(""%0(:) JB.9N1) a*(&%"$31) I999) ,*(+"%"(/"6) *() B-&*'$&"8) .*+&X$%")
9(:0(""%0(:J 44 Y^ZBBY^^J 7)0-+%#'J K#0#2#6
"))*+)6 "6 DYZZXF6 @3/01 I93-+# ,%4%02%0.%3 -) P).#'/>% =#&'-3 $+)* \+).%2&+#'
\+)1+#*36 E%*/""80(:6)*+)&>")Rf&>)/*(+"%"(/")*()E%*/""80(:6)*+)&>")Rf&>)IB.C9!)I(&"%($&0*($3)
a-3&0@,*(+"%"(/"l)$%&0+0/0$3)0(&"330:"(/")$(8)$##30/$&0*(6)JBIBN1)44BQWZBQWSJ L0039+&.:J I&3-+/#6

5

Neural Networks Applied to Thermal Damage
Classification in Grinding Process

Marcelo M. Spadotto, Paulo Roberto de Aguiar, Carlos C. P. Sousa,
Eduardo C. Bianchi
Sao Paulo State University – Unesp – Bauru Campus
Brazil

1. Introduction

Throughout the course of human progress from prehistoric times until now the
technological world was characterized by the development and improvement of new
methods to control the environment. One of the most obstacle to overcome in order to reach
the complete automation of machining process within the integrated and flexible
manufacturing systems is the development that can be named a non-human-assisted
machining, that is, a process in which the moment for tool change, the tool change itself and
the change of the grinding conditions no longer need the human being assistance. Thus, the
development of monitoring and control systems in real time is of great importance. High
temperatures in grinding process are the main source of thermal damages to the ground
surface, which is a visible manifestation in steels known as grinding burn. Depending on the
temperature reached in the grinding zone a burn degree on the part surface can be observed
which is due to temper color from very thin oxide layers. One of the challenges found in the
implementation of intelligent grinding process is the automatic detection of surface burn of
the parts. Several systems of monitoring have been assessed by researchers in order to
control the grinding process and guarantee the quality of the ground parts. However,
monitoring techniques still fails in certain situations where the phenomenon changes are not
completely obtained by the employed signals.
Several monitoring systems which use force or power and acoustic emission sensors have
been assessed by researchers to control surface burn in grinding (Aguiar et al.,2002; Aguiar
et al., 1998; Kwak & Song, 2001; Wang et al., 2001; Kwak & Ha, 2004; Dotto et al., 2006;
Aguiar et al., 2006a; Aguiar et al., 2006b). However, those techniques still need
improvements where the phenomenon variations are not entirely acquired by the signals
used.
High temperatures in grinding are the main source of thermal damages to the ground
surface. A visible manifestation of this damage in steels is grinding burn – a discoloration of
the ground surface often visible directly to the naked eye or brought out by etching of the
surface. Depending on the temperature reached in the grinding zone a burn degree on the
part surface can be observed which is due to temper color from very thin oxide layers on the

part surface. This layer of ferrous material is composed of Fe
2
O
3
, Fe
3
O
4
, and FeO membranes
Frontiers in Robotics, Automation and Control

72
from the free surface. At the onset of a grinding burn, the grinding force and the rate of
wheel wear increase sharply, and the surface roughness deteriorates (Kwak & Song, 2001;
Badger & Torrance, 2000). Other type of thermal damage is referred to as rehardening burn,
which is caused by a metallurgical phase change in the material when the grinding
temperature exceeds the austenizing temperature, creating a thin layer of hard, brittle,
untempered martensite. To further exacerbate the problem, rehardening burn is also
accompanied by secondary residual stress, because the newly formed material has a greater
density than the original material (Badger & Torrance, 2000)
In this study, the neural network has been applied to classifying the burn degrees obtained
on the surfaces of the parts in grinding. The parameters of acoustic emission (AE), power
signals and others derived from these signals have been used as the inputs of the neural
network. The characterization of the surface quality of the ground parts was done by visual
analyses with the naked eye and also by the software developed ( Dotto, 2004).
What makes this work distinguished from others is the use of grinding parameters as input
to the neural network, which have not been tested yet in burn classification by neural
networks. Besides, a high sampling rate data acquisition system was employed to acquire
the raw acoustic emission and cutting power.

2. Grinding Burn Monitoring

High temperatures generated in the grinding zone can cause several types of thermal
damage to the part, for instance grinding burn in the case of steels, excessive tempering of
the superficial layer with possible rehardening and increase of the brittleness, undesirable
residual stress of tension, decrease of fatigue-life performance and micro-cracks. The
decrease of grinding power is needed in order to minimize the restriction of thermal
damages. This can be obtained by utilizing a softer grinding wheel or a rougher dressing
operation but both present disadvantages (Malkin, 1989)
Grinding burn occurs during the cutting process when the amount of energy generated in
the contact zone produces an increase of temperature enough to provoke a localized phase
change in the material of the part. Such occurrence can be visually observed by the
discoloration of the part surface (Malkin, 1989; Kwak & Song, 2001; Kwak & Ha, 2004). The
burning is expected to occur when a critical temperature is exceeded in the grinding zone.
He estimated a temperature rise of 720°C for burning to occur (Malkin, 1989).
Burn in steels is characterized by a visible bluish temper color on the ground surface. In
steel, due to the burning phenomenon, the temper color changes from light brown to dark
brown to violet to blue, in that order, depending on the severity of burn (Malkin, 1984;
Nathan et al., 1999; Badger & Torrance, 2000; Liu et al., 2005).
The root mean square value (RMS) of the acoustic emission signal has been the main
parameter studied in previous grinding researches over a frequency band carefully selected.
This signal has been a good parameter because it is rich in sound waves carrying a lot of
useful information (Lee et al., 2006; Liu et al., 2006).
Aguiar et al. (2002) has demonstrated in their investigation that the combination of the
acoustic emission (RMS) signal and the electric power signal of the electric motor that drives
the wheel can provide meaningful parameters to indicate when grinding burn takes place.
From the combination of these signals they obtained the parameter referred to as DPO for
burn detection in grinding. This parameter consists of the relationship between the standard
Neural Networks Applied to Thermal Damage Classification in Grinding Process

73
deviation of the RMS acoustic emission and the maximum value of the electric power in the
grinding pass. The equation (1) describes the mentioned parameter.


maxEA
DPO S P= (1)

Where S
EA
is the standard deviation of the RMS acoustic emission; and P
max
is the maximum
value of the electric power.
In the hope of more sensitivity to detect grinding burn, Dotto et al. (2006) has proposed a
new parameter referred to as DPKS, which also utilizes the RMS acoustic emission and the
power signals. This parameter is defined according to equation (2).


4
1
(()()) ()
im
i
DPKS POT i S POT S EA
=
=
=−
⎛⎞

⋅
⎜⎟
⎝⎠
∑
(2)

Where i is the power index that varies from 1 to m samples per grinding pass; m is the
number of samples of the pass; POT(i) is the instantaneous value of the power; S(POT) is the
standard deviation of the power in the pass; S(EA) is the standard deviation of the RMS
acoustic emission in the pass.
The statistics known as Constant False Alarm Rate (CFAR) and Mean Value Deviance
(MVD) were employed successfully for detection of grinding burn (Wang et al. 2001; Aguiar
et al., 2006b). The equation 3 represents the CFAR and the equation 4 the MVD.

1
0
1
0
()
M
k
k
cpl
M
k
k
X
TX
X
ν

ν
−
=
−
=
=
⎛⎞
⎜⎟
⎝⎠
∑
∑
(3)

Where X
k
is the k-th is the magnitude-squared FFT bin, υ is a changeable exponent and 2M is
the total number of FFT bins (due to conjugate symmetry, only half of the magnitude-
squared FFT bins need be interrogated). Respectively υ =1 and υ=∞ correspond to the energy
detector and max{X
k
}; 2<υ<3 provides good performance in a wide range.

1
0
1
() log
M
mvd
k
k

X
TX
M
X
−
=
=
⎡
⎤
⎢
⎥
⎣
⎦
∑
(4)

Where
X
is the mean value of
{
}
k
X
; M and X
k
have the same meanings as in the CFAR
statistic.

3. Neural Network and its Application to Grinding

Neural networks are composed of many non-linear computational elements operating in
Frontiers in Robotics, Automation and Control

74
parallel fashion. Neural networks, because of their massive nature, can perform
computations at a higher rate. Because of their adaptive nature in using the learning process,
neural networks can adapt to changes in the data and learn the characteristics of input
signals. Learning in a neural network means finding an appropriate set of weights that are
connection strengths from the elements of this layer to the elements of a next layer (Kwak &
Ha, 2001).
There are three layers in a network, namely the input layer (which receives input from the
outside world), the hidden layer (between the input and the output layers) and the output
layer (the response given to the outside world) (Nathan et al. 1999). The neurons of different
layers are interconnected through weights. Thus, a neural network is constituted by
processing elements at different layers, interconnections between them, and the learning
rules that define the way in which inputs are mapped on to the outputs. The usefulness of
an ANN comes from its ability to respond to an input pattern in a desirable fashion, after
the learning phase. As such, the processing units receive inputs and perform a weighted
sum of its input values using the connection weights given initially by the user. This
weighted sum is termed the activation value of the neuron, given by:

1
N
ii
i
uwx
θ
=
=
+

∑
(5)

where w
ij
is the weight interconnecting two nodes i and j; x
i
is the input variable; and u is the
threshold value. During the forward pass through the network, each neuron evaluate an
equation that expresses the output as a function of the inputs. Using the right kind of
transfer function is therefore essential. A sigmoidal function can be used for this purpose,
and is given by:

()
1
()
1
u
fx
e
−
=
+
(6)

Depending on the mismatch of the predicted output with the desired output, the weights
are adjusted by back-propagation of error, so that the current mean square error (MSE)
given by the following equation is reduced:

()

2
11
1
2
NK
nk nk
nk
MSE b S
NK
==
=−
∑∑
(7)

where N is the number of patterns in the training data, K is the number of nodes in the
network, b
nk
is the target output for the n-th pattern and s
nk
is the actual output for the n-th
pattern.
Still, it should be noted that the MSE itself is a function of the weights, as the computation of
the output uses them. During this learning phase of the network the weights and the
threshold values are adapted in order to develop the knowledge stored in the network. The
weights are adjusted so as to obtain the desired output. The problem of finding the best set
of weights in order to minimize the discrepancy between the desired and the actual
Neural Networks Applied to Thermal Damage Classification in Grinding Process

75
response of the network is considered as a non-linear optimization problem (Nathan, 1999).

The most popularly used learning algorithm, namely the back-propagation algorithm, uses
an interactive gradient-descent heuristic approach to solve this problem. Once the learning
process is completed, the final set of weight values is stored, this constituting the long term
memory of the network, which is used later during the prediction process.
Previous investigations have proved the efficiency of the artificial neural networks in the
prediction of grinding burn (Wang et al., 2005; Kwak & Song, 2001; Kwak & Ha, 2001;
Nathan et al., 1999; Aguiar et al., 2005, Spadotto et al., 2006). Thus, this technique is very
promising and can also be applied successfully to industrial automation in a flexible and
integrated fashion.

4. Methodology and Results

A surface grinding machine from Sulmecânica manufacturer, Brazil, model RAPH-1055 was
used in the grinding tests. The grinder was equipped with an aluminum oxide grinding
wheel, from Norton Manufacturer, Model ART-FE-38A80PVH. A fixed acoustic emission
sensor from Sensis manufacturer, model DM-42, placed near the workpiece and an electrical
power transducer for measuring the electrical power consumed from the three-phase
induction motor that drives the wheel were employed.
The workpieces for the grinding tests consisted of laminated bars of steel SAE 1020 ground
in the shape of a prism with 150mm length, 10mm width and 60mm height. The grinding
process took place along the workpiece length.
The power transducer consists of a Hall sensor to measure the electric current and a Hall
voltage sensor to measure the voltage at the electric motor terminals. Both signals are
processed internally in the power transducer module by an integrated circuit, which
delivers a voltage signal proportional to the electrical power consumed by the electric
motor. The acoustic emission and the power signal are further sent to the data acquisition
board from National Instrument, model PCI-6011, which is installed onto a personal
computer. The LabVIEW software was utilized for acquiring the signals and storing them
into binary files for further processing and analysis. The acoustic emission sensor used has a
broad-band sensitivity of 1.0 MHz. Its amplifier also filtered the signal outside the range of

50 kHz to 1.0 MHz. Figure 1 shows the schematic diagram of the grinding machine and
instrumentation used.


Fig. 1. Experimental setup.
Frontiers in Robotics, Automation and Control

76
The tests were carried out for 12 different grinding conditions, and subsequently the burn
degrees (no-burn, slight burn, medium burn, and severe burn) could be visually assessed for
each workpiece surface. Dressing parameters, lubrication and peripheral wheel speed were
adequately controlled in order to ensure the same grinding condition for each test. The
workpiece speed was set up at 0.033 m/s and the wheel speed at 30 m/s. The latter was
maintained constant by adjusting the frequency of the induction motor on the frequency
inverter, as the grinding wheel had its diameter decreased along the tests. The G-ratio,
which is the volume of material removed per unit volume of wheel wear (Malkin, 1989), was
set to 1, maintaining the dressing condition the same for all the tests. A water-based fluid
was used with 4% concentration. Each run consisted of a single grinding pass of the
grinding wheel along the workpiece length at a given grinding condition to be analyzed.
The acoustic emission and cutting power signals were measured in real time at 2.0 millions
of samples per second rate, and then stored onto binary data files for further processing. It is
important to mention that the raw acoustic emission signal was acquired instead of the root
mean square generally used.
The digital signal processing phase started after all the 12 tests were carried out and the data
files stored. The digital signal processing of acoustic emission and power generated 7 new
statistics as previously described, that is, the parameters DPO and DPKS, and the statistics
CFAR and MVD. Seven structures were used for the neural network implementation as
shown in Table 1. It can be noted in this table that besides the signals and statistics
aforementioned the depth of cut a was also used as input of every structure.

Structure Inputs Structure Inputs
I Pot, AE, a V CFAR, a
II DPO, a VI AE, a
III DPKS, a VII Pot, a
IV MVD, a
Table 1. Neural network structures.

In this work, the back-propagation algorithm of neural networks, which is one of the
learning models, was used. The following parameters were also found more suitable:
downward gradient training algorithm; all data in the neural networks were normalized;
training for 1000 epochs; square mean error value of 10-5. Cross-validation was used to
estimate the generalization error of the model.
The outputs of the neural network was configured in a binary way according to the degree
of burn obtained, that is, 0001 for no-burn, 0010 for slight burn, 0100 for medium burn, and
1000 for severe burn.
Each statistic was represented by a vector of 3000 samples for each test subsequently the
digital processing of the acoustic emission and power signals. Initially, a visual analysis was
carried out by naked eye on the part surfaces. Then, a quantification of the grinding burn on
every part surface was done by a specific software for that purpose, which assessed the
surface of a given part regarding the burn level through its digitalized picture (Dotto, 2004).
Thus, a precise characterization of the burn levels occurred on the part surfaces was
achieved. From the results of this characterization, input vectors were separated and
Neural Networks Applied to Thermal Damage Classification in Grinding Process

77
assigned to the corresponding type of burn. The input vectors were again divided into
training, validation and test vectors. Then, the process of optimization for the neural
network was carried out.
The architecture of the neural networks was determined according to the tests of number of
neurons of the hidden layer, learning rate and momentum. As the problem dealt in this

work consists of pattern classification, only a simple hidden layer was chosen (Haykin,
1994). With learning rate equal to 1 and momentum fixed to zero the neurons of the hidden
layer were varied at steps of 5 up to 50 neurons. The optimum number of neurons of the
hidden layer was estimated according to the mean square error of the validation set of each
structure calculated in the training phase. Figure 2 illustrates the comparison of errors
obtained for each number of neurons considered for the structure 2 (DPO and depth of cut).


Fig. 2. Mean square error versus number of neurons of the hidden layer for structure 2.

In order to obtain the best values for learning rate and momentum the training was carried
out for all aforementioned structures by varying these parameters. The proceeding for
choosing the best pair momentum and learning rate was performed by fixing initially a
value for momentum, varying the learning rate from 0.1 to 0.7 with step of 0.1. After the
error curves were obtained, the pair momentum and learning rate was chosen based on the
curve which presented the smallest oscillation with the smallest number of epochs. This
process was repeated for values of momentum from 0.2 to 0.7, with step of 0.1. Thus, the
best six pairs were obtained in the end of this process, choosing among these pairs the one
which presented the smallest oscillation with the smallest number of epochs.
Figure 3 illustrates the process of choice of the pair momentum and learning rate from
comparisons between two curves of the mean square error in function of the number of
epochs. These curves were obtained following the training of the structure 4.