fault detection and diagnosis in film processing plants · 2003-11-02 · hwahak konghak vol. 41,...
TRANSCRIPT
HWAHAK KONGHAK Vol. 41, No. 5, October, 2003, pp. 585-591
�� �� ����� �� ��
�������*��* ,†���**� ��**
������� �����, *�����, �� (�)790-784 �� ��� �� ��� � 31
**����(�)730-350 �� ��� ��� 93-1
(2003� 10� 10 !�, 2003� 7� 10 "#)
Fault Detection and Diagnosis in Film Processing Plants
Dong-Myung Yoon, Young-Hak Lee*, Chonghun Han*,†, Hun Sung An** and Sa Yun Chang**
Department of Chemical Engineering, *Department of Chemical Engineering and ISYSTECH Co. Ltd., Pohang University of Science and Technology
San 31, Hyoja-dong, Nam-gu, Pohang, Kyungbuk 790-784, Korea**ToraySaehan Co. Ltd., 93-1, Imsu-dong, Gumi, Kyungbuk 730-350, Korea
(Received 10 October 2002; accepted 10 July 2003)
� �
�� ��� �� � �� � �����, �� � ��� �� �� ��� �!�� !" #$� %�!� &�.
' ()*�" �+, -./0" 1�� 234 567� 89 :1, ;<!" ��=>? �@9 A� BC, DE
F�. � BC� G+� �HC, BIJ� !" �� KL MCJ�, => @NO LPP�� �Q!R S�T� UVF
�. DEF MC, �� �= => W S�T* X?!R, YZF S�T� ?[\� ]^. _` !a�. _0� bc
cd�O* ef!" S�T� gh* ij!R, �� 89� ]k]B l =>*� �L� m n, opq N &rJ
s, � =>*� ��� ^N WL� &rt, u1!a�.
Abstract − The fierce competition in the polymer film and sheet market requires the industry to satisfy much higher productquality specifications. This paper proposes a monitoring and diagnosis method based on multivariate statistical techniques
which help us reduce the amount of off-spec product. The method has been applied to an industrial web forming plant and has
proven that process faults such as the leak of polymer fluid can be early detected before it is developed into the production of
bad quality product.
Keywords: Film Process, Principal Component Analysis (PCA), Multivariate Statistical Process Control (MSPC), Fault Detec-
tion, Diagnosis
1. � �
� ����� ��, � �� ��� �� �� � ��� �
� ��� ����. !, �" ��# $ %&' ()� * +��,
- ./� 01 234 5� 6� 789+: 7;<= >? �� ��
@� AB� C�D E:F, ��GD ��� ��� BH@� 6�. I
�J ��K �" BHL MNKO� PQ, %&D RS, TPUVW X
Y ��� Z>[\]D �G^_` V' ab� ��4 ��E� P
� cd `e �f gh#� �K@f i�.
j��� %&�: Kk l7 �H�J �gm(machine direction,
MD)�: ngm(cross direction, CD) �f� o$pq�. ! %&D
r gmsF tfe� uv` w x�` @q� yx�, ngm z{�
|� quadratic programming4 Ec 0V_ E}� ~�� K�@q�[1].
�d: �" �1 gh' ngm �f� �gm �f` ��V���
`7� ��(steady state)�J ngm �f x�L ��� ��q�. I
�J � gh' ngm l7�J ��� ��� ��EY #f� uv
� V�� Q��� � Z �q�. ngm� �gmD ��L �fJ ;
� b�VsF ��E� PQ Bergh� MacGregor[2]� linear-quadratic-
gaussian �f ��4 V�E�� E��. e��e ngm� �gm �
f� ��VsF �H@� 6s� b�VsF ngm �f` ;� f
��� }¡ 6�[3,4].
ngm� �gm �f-sF� 7¢� £¤4 -¥� �� ¦D §`
¨E©F, �ª K2�J �G4 «¬E}� ~� � �� @fi�.
%&D RSL �fE� PEc gage �f` �H@q�[5], ®¯E�
°±D ²�V ³�4 �QE� PQ K�´4 ¢µ(identification), ¶
7(estimation) ��� �fE� �H` 6qs�[6,7], ·¸� �� �†To whom correspondence should be addressed.E-mail: [email protected]
585
586 ������������� ��
¹ >?�J� 7V º»�� ¼½D º¾� |� �H` ¿À@��
E��[8]. �ª �z, Á' ZD Â�� �4 ��E� PQ SVD �
f�` �Â@q�[9], w ãD l74 ¼ÄE� PQJ Â�^Z#
4 ¼�E�[10] £¤4 ÅÆE� �H` 6q�[11].
� °± Ç$VsF �H@fÈ ngm�: �gm �f� %&� t
fe� ³�� |� |É�b , �²D Êm �sF HË� :� ³�
�e �fEe� Ìp�. 9 l7D Á' ^Z#� 22 x�D ͵�
� Z 6�� Î4 ϱQ ;U, Á' ^Z# $ x�� AB ͵� @
� ^ZL ÐÑ Ò�VsF .Óf £/ÔÕE� x�L Öc;�+
E� ~�� %BE�. �× Ø� l7� |QJ� A�> >Ù(PCA)
4 <Q Z>Ú� D� z{L ÏÆKO�+ E� �H` 6q�Û[12],
Ü �HD V��Ý�J� ^Z ��D w �ÞÞ�L �}� PCAL
��Ec l7D ½7� ��L Ïep�. Üx�J� |�l7� |Q
.;� ��o(outlier)L £/ÔÕEc x�� AB͵� @� ^Z
L ß.;à�. �" �7�J áGVsF ¯¹4 EUJ â� ��Y
l74 £/ÔÕE� P� cd `e gh4 �KE� á� l74 <
Q ãäE��.
2. ����: �� ��
Ü �H�J� å �æD ��4 ��E� �Å�= PET %& l7
4 |�sF ��. �� �7�J %&� HË� :¦: tfe� ³�
4 ‘çT’��� E�Û, çTè' ��� éTD $B� eê�� Ü
�H�J AF �ëY � �� ³���. Fig. 1' b�Vµ %& ��
l74 :ìí ���, �L îæKïD {�� I� T-dieL $ðsF
:ñ Z 6�Û, òC>4 ‘ÍóC l7’�� E� ôC>4 ‘�� l7’
�� ��. � ÍóC l7' ‘feed l7’� ‘matrix sheet �l7’
sF :õfe�, �� l7' ‘forming l7’� ‘coiling l7’sF :
õf¿�.
2-1. Feed ��
ö �÷ l7' �>+ ø4 ù)p� ̀��F lúE� l7��.
�] ûü� 6� PET ø� Z>� ý. 6sU `Z>Q� Z 6�.
�L �� PQ R «D ®¯��J þ|F ®¯��. ®¯�D 2 ÿ
� eκ� È� �J` 6�Û, � È�L �7[sF� ��4 ç�
E� f� 7� ²G� �°p�e ¢µ� Z 6�. � PET ø�
feed ûüF �°E�Û, �y ¿l4 \eKO� PQ �� �f�L
��E� 6�.
2-2. Matrix sheet �� ��
�>+ ø4 %&sF -#� PQJ� �È, ��4 `Q? E�Û,
��` � �4 ��. ��� å� `QeUJ °K� £Ô` �
�ë(screw)L �} �>+` �°E� �y �E� º»å� DQ
|C>D �>+` ����. T-die�J Kk(sheet)D ��F �.e�
�9� �>+D ��` ±7V�f? E� yx�, �� �µ�J È
�L �7Ec ��L ¢µE� die �F ò� %Ô` 6f �²4 �
¦��.
2-3. Foaming ��
T-dieFCÔ :�� ��ZeL �2K� KkL ��E�, Kk ��
D �>+L ��=� n�=K� %&4 �Y ��. Z�gmsF �
=KO� PQ ��` {� :� �(roll) ��L �°KO�, Z�gm
sF �=KO� PQ %&D ��4 �7K� ngm Ê4 e:
Y ��. �y �2� 7�� casting drum� lú@� ²D È�L �
7Ec Z 6�, ��= 7�� £ÔD ��, � ÓC� lú@� ²
D È�� ��sF ×ëf � Z 6s�, n�= 7�� KkD � �
� ��� 6f �7@� �v �m ¦�L <Q Z 6�.
2-4. Coiling ��
ºe�sF ��(T/U) l7�J �Å �=� ����Ô %&� º�
Ô� ��F Ï e�, �T(W/D) l74 ¦o� ��D Ã! "��
#�F $} �%&� %���.
3. � �
3-1. ��� ��(PCA; principal component analysis)
_& l7 ^Z#' ;< �ÞÞ�` �� yx�, �" Þ'�4 �
�Ec Z ( «D ^ZL T ) «D A�>(PC; principal component)
sF :ìÓ� �� `¨E�. �" ghsF >ÙE� �h4 PCA��
E�Û, � PCAL ��Ec Û�ÔL ÅÆ(reduction), £¤Õ(modeling),
��o Ïe(outlier detection), ̂Z ¬*(variable selection), >æ(classification)
��� +�(prediction) � Z 6�[13].
Fig. 2� ,� l7 ^ZL 2 ÅsF Ec Û�ÔD >(L :ìí
���J, ̀ ) w z{L -� gm� |Ec Å4 7DE� ��4 ö
Fig. 1. Diagram of PET film process. Fig. 2. Geometric interpretation of principal component analysis.
���� �41� �5� 2003� 10�
�� �� ����� �� �� � �� 587
�÷ A�>(PC 1)�� ��. ��� � Å� .ÊK� 7�Ê /#4
�0f(score) 1Ô�� ��. �2 T�F PC 1� �þE�3 � �2
sF w z{L -� gm� |Ec R �÷ A�>(PC 2)4 44 Z
6�. �" ØsF z{` w 5sF PCL 7D :6 Z 6�Û, -#
f¿ 5J|F £¤4 ¼ÄE� 7�` �� yx� ò7D ) « PC
-sF 9îD £8 ^ZL |9VsF :ì: Z 6�. Û�Ô À;,
XL ZØsF ê³EU �2� ,�.
(1)
c�J i� 2 <D µ=�, j� 2 ^ZD µ=��� xij� i�÷ <
D j�÷ ^Z /��, X� � xij /#4 £.>' À;��. a� PC
D «Z, T� PC#4 £. >' À;, P� PC*F ^ZD Êm�4
:ìÓ� `$o(loading) À; ��� E� ?{(residual) À;4 :ì
í�. tk, pk ��� ek� 22 k�÷ �0f 1Ô, `$o 1Ô ���
?{ 1Ô��. @, XL ù {Í ��J À;D A(TPT)sF 1�� Z
6�Û, �� Te a«D Å� .ÊK� /� � `$oD ¬�Vµ B
-sF K�´4 C�� Z 6�� �4 D×��.
3-2. �� �� �� ��(MSPC; multivariate statistical process
control)
PCA� DQ HQ¿ £¤' �Y PC lï� ?{ lïsF :õf ê
³@� 2 HïD <�V �DoL �l[sF� £¤ |½ l7 �E
4 >Ù� Z 6Y ��. £¤4 �DsF, � R lï4 :õ� P�
µ=�F� T2(Hotelling’s T2 statistic)� SPE(squared prediction error)
L Á� ����.
T2� <� 7� ���J Fº: Gf¡ 6�`L :ìÓA� Zo
F, �2� ,� � Z 6�[14].
(2)
S−1' TD l>� À;D sF, S−1=((TTT) / (n−1))−1��, t=xTP(PTP)−1
��. T2D �P �f¬(UCL; upper control limit)' �2� ,�.
(3)
c�J n� < Z��, Fα(a, n−a)� αD <�V \DZD4 `e�,
a� n−aD +\�L `e� F >(D �� eÎ��. T2� |� <
D j�÷ BÆ, xjD 2 ^Z �c�(contribution)� �2 ØsFCÔ
��� Z 6�[15].
(4)
�� ½Q, SPE� £¤�D �{L :ìÓ� µ=�F �2� ,� ê
³� Z 6�.
(5)
��� �2 ØsFCÔ �� QWE� ��¬4 �4 Z 6�.
(6)
c�J h0=1−((2θ1θ3)/(3θ22) ), θ1=tr(V), θ2=tr(V2), θ3=tr(V3), V� ED
l>� À;��, zα� h0� ,' CûL -� (1−a)D =HHï4 `
e� êD7Ã^Z��.
SPElim, α;� w /4 -� ��(abnormality)� Ïe@q�U, SPE
� |� 2 ^ZD �c�(contribution)L ½þQIsF� f� ^Z`
��� `) Êm4 Á� ×J�e Z 6�. l7^Z j� |� �c
�, cjSPE� �2� ,�[15].
(7)
c�J, � £¤� DQ +�� BÆ C>�� ej� ?{��.
l7 ��' T2: SPED <�V �� /;� w /4 -� �®µ`
L ¢µEc ÏeE�, Ïe� ��4 ¿TE� PQJ� Ïe KÎ�
JD �c�L ;� �� %ZV��. �K KQ l7D XZ! Á' ^
Z �Þ�4 �}Ec T ) «D PCF �ÅE�, �Å� lï�J l
7 �E4 £/ÔÕE� PQ SPE� T2 µ=�L ����. £/ÔÕ
4 <Q 7 KÎ�J ��� Ïe@U Ïe� ��D 2 ^Z* �c
�L <Q ��4 ¿T� Z 6�.
4. � � ��
4-1. ��� �� � �� ���� �� �� ��
V� �Ý�� %& l7�J 38bï 1> ï!sF �7� Û�ÔL
��E��, � l7 Û�Ô� ÍóC l7D ^Z(tags)̀ 87«, ��
l7D ^Z` 31«��. � Û�ÔD 9î um�4 ç�E� PQ, ³
)D >��f K�´sFCÔ �' ÍL(raw) Û�Ô� PCAL V�
E��. � �� Fig. 3� ,� �Y R ÊsF :õf¡ :ìM�Û, Ê
�� C>�J ¯¹ ^_` 6q24 ¢µ� Z 6q�.
4-2. � ! � ���"� �# $% ��
9î Û�ÔL |�sF � £¤4 `e� £8 Û�Ô� V�E�
�' VBEe N�. ��� �ª uv�e E:D �GF �fO��
yx��. �LJ ¯¹ ��` ^_[� I� �P4 :Q 22 £
¤ÕEc? E�Û, Ü �H�J� PC 1� PC 2� |� �0f {k�
�J R±sF �P_E� gh4 ��E��. � gh' Sv ïzE
� �× V�(reasonable)E�� }¡ 6s©F[16], Fig. 3� ,� 9
î l74 R �PsF :ñ Z 6�.
�P� :T¿ �\L >ÙE� PQ, 2 �PD PC 1�>4 ½þE
c Fº: w {�L ;��e UV;à�. Fig. 3�J 2 Û�ÔD K
24 UV;U, W7 �P' +��, �ª7 �P' �¹� PET %&
���, R �PD {�L T2 statistic4 ��� �c� {kF ;
U Fig. 4� ,�. êDìÂD uv, +��' 7µm�J 14µm RSF
�fÓ� �¹�' 12 um�J 188 um RSF ��E� yx�, +��
4 ��� uv Ñ X' RSL �fÓ� P� ¯¹4 ��. � Î4 �
Y4 <Q ¢µ� Z 6�Û, Fig. 4�J ÍóC l7D ö �÷ ��
X x iji 1=
n
∑j 1=
m
∑ TaPaT E tk
k 1=
a
∑ pkT ek
k 1=
a
∑+=+= =
T2 tTS 1– t=
TUCL2 n 1–( ) n 1+( )a
n n a–( )----------------------------------Fα a n a–,( )=
cjT2
tTS 1– xjpj PTP( )1–
[ ]T
=
SPE ek2
k 1=
a
∑=
SPElim α, θ1 1 θ2h01 h0–
θ12
------------ zα 2θ2h0
2( )θ1
----------------------------+–
1h0
----
=
cjSPE ej( )2 xj x̂j–( )2= =
x̂j
Fig. 3. Plot of the first two PCs corresponding to the total data (thirty-eight days).
HWAHAK KONGHAK Vol. 41, No. 5, October, 2003
588 ������������� ��
È�` ��� w Êm4 ×o�, �� l7�J� n�= l7D È�
^Z` `) w ��4 ��. @, ½þV X' +�� %&4 �� PQ
J� ÿ' È�F �>+L ��KO�(Fig. 5(a)), ��� KkL ÿ'
È��J �=[4 ¢µ� Z 6�(Fig. 5(b)).
�� 2 ��� I� £/ÔÕ4 E� PQ, 2 �P� |Ec PCA
L V�Q;U Fig. 6� ,' ��L �4 Z 6�. Fig. 64 ;U, �
Ô` PC 2 gmsF Z ìÍ£�4 E� 6�. ��' PC 2 gm� [
' z{L -� PC 1 gm;� Ñ Á' ��D ͵4 -� 6�� �
4 D×��. á�F PC 1 gm� |Q Û�Ô >(` ½��Vµ u
m� :ì\ �' CE �E ¯� yx�q�. ��� PC 2 gm� |
Q w `$oL -� ^Z#� l7 ^_� w Êm4 ×]4 Z 6
�Û, ��4 Fig. 7� Fig. 8�J ¢µ� Z 6�. Fig. 7�J ���µ
� ^ �÷ %ÔD ̂ Z#� w /4 ̀ e�Û, Fig. 8�J 2 TP(unit)
D |êVµ È� ¶�L �KQ;U, 7«D TP l7 $ �� �µ�
^ �÷ %ÔD È�` �|VsF �Y ^_��.
Fig. 4. Contribution plot identifying the difference between two groups (a) principal variables of the first extruder, (b) principal variables of the cross-direction process.
Fig. 5. Temperature change (a) of the first extruder, (b) of the cross-directional process.
Fig. 6. Plot of the first two PCs (a) corresponding to group one data, (b)corresponding to group two data.
���� �41� �5� 2003� 10�
�� �� ����� �� �� � �� 589
4-3. �� &�' ()
b�VsF çT ³�' çTD �æ: ð2�� |� 7;` _B�
4 `� 6f, �L Û�ÔF ZÇE�` âe N�. �" uv, �G ^
Z� �ÞÞ�` w ^Z#� faY b���e Þ»[sF� ½7�
��L Ïe� Z 6�. v¬, l7� 7�sF c9� yD Û�ÔL
��Ec £¤Õp�. � £¤� çT� �� Û�Ô� � çT� l
74 e:È îæKï� QWE� Û�ÔL .ÊEc çT� �E�
eL ß.;à�. � �� Fig. 9�J ;d�, .Ê� Û�Ô` çT 5
>9CÔ �f ��L ef:� K�QJ çT ³�� :ì\ Û�Ô
� ��oF Ïe@� �4 f� Z 6q�.
4-4. �� &�' *+
��o` :ì:U, �yD �c� {kL <Q ½7���D ͵4
>ÙQ � Z 6�. b�VsF ��Ó�J È�� ��D ^_` �
� yx�, � eÎ�J çTD ͵� Á4 ���� +���. �d:
� l7D uv� 6«g 9 º»å {�� Iª ��ë þîF ��
D ��4 Îã� ����, �±E� gh4 <Q +�Ee Ìph l
7�J x�Î4 f� Z 6q�. �c� �L¸�J È� ^_D r
� ih ���µ� ^ �÷ %ÔL $ÎVsF UV;à�Û, � ^Z
Fig. 7. Contribution plot corresponding to PC two (a) calculated from group one data, (b) calculated from group two data.
Fig. 8. Temperature changes of the first extruder, the first filter, the sec-ond extruder, the gear pump, the second filter, the melting line,the third filter, and the die.
Fig. 9. Real-time monitoring and fault detection in PC space: Film-bro-ken case.
HWAHAK KONGHAK Vol. 41, No. 5, October, 2003
590 ������������� ��
l of
g
ilm
ust
es,”
et-
rol
ca-
atic
#� Þ'Ec ��o` Êm4 524 ¢µ� Z 6q�. Fig. 104 ;
U, Û�Ô` £¤D 7� Ê� #f64 y� �' 2D /4 `e
h �� �µ ^Z` çT ³�� `�jGZ3 w 2D /4 `eY
k4 Z 6�. l)�D µÔlL <Q .Ü ��, � çT� �
� KÎ� �>+D õZ ³�� 6q24 ¢µ� Z 6q�.
5. �
Ü Òx�J� ^Z#� w �ÞÞ�L `¿�� Î4 �}Ec, 9
î l74 |�sF PCAL V�Q ;à�. �y Û�Ô� PC 1 | PC
2 �0f {k ��J ��*F �P4 ef :ìM�. � 22D �
P� PCAL V�Ec ½m� gradeF ��@� Û�Ôn� £. £
¤4 -#q�. ��� �oY �f¿ £¤� ��� bfM4 yD Û
�ÔL .ÊE� £¤D ��¬4 ef:�e UVIsF� ��4 Ï
e� Z 6q�. � çT� :ìM4 y� � �F �9 Û�ÔL .
Ê[sF�, çTD AB ͵� @� µ+L ß4 Z 6q�. Ü É�
�ÝD uv, �� �µD È� ^Z` çT ³�4 9Ec �Y ^
_E��, �y �� �µ�J �>+` p� ³�� 6q24 ¢µ�
Z 6q�.
� °± �� l7� |� �H� Á'Û ½Q, ÍóC l7�e .v
�� %& l7 9î� |� �H` CNE��. Ü Òx�J� 9 l
74 |�sF �� ³�4 ÏeE� x�` @� AB ^ZL âY ß
4 Z 6q�. ��� Á' ^ZL `¿ l7� ½7� ��F c9@
� KÎ4 `K_E�s�, �L <Q � °± �� ³�D AB� Í
µ� ./�� éTEc Æq! ��Eh TP�J x�Î4 ß4 Z
6q�.
� �
� Òx' 2003r� IMT2000 �¹(���û: 00015993)� Rs�t
21�¹� DEc eÍ@qu/�.
����
a : total number of principal components
cjSPE : contribution to the SPE for process variable [j]
cjT2 : contribution to the T2 statistic for process variable [j]
E : residual matrix for [X]
ej : residual of the data of process variable [j]
ek : kth residual vector for [X]
Fα(a, n-a) : F-distribution with a and n-a degrees of freedom in leve
significance [α]
i : index for events
j : index for process variables
m : number of variables
n : number of samples
P : loading matrix for [X]
pk : kth loading vector for [X]
S : covariance matrix of [T]
SPE : squared prediction error
T : score matrix for [X]
T2 : Hotelling’s T2 statistic
tk : kth score vector for [X]
V : covariance matrix of [E]
X : process data set
xij : obtained data
xi : column vector [i]
xj : jth element of [x]
: part of the element predicted by the model
zα : standardized normal variable with (1-α) confidence limit
�,�- ./
α : level of significance
����
1. Boyle, T. J., “Control of Cross-Direction Variations in Web Formin
Machines,”Can. J. Chem. Eng., 55, 457-461(1977).
2. Bergh, L. G. and Macgregor, J. F., “Spatial Control of Sheet and F
Forming Processes,”Can. J. Chem. Eng., 65, 148-155(1987).
3. VanAntwerp, J. G., Featherstone, A. P. and Braatz, R. D., “Rob
Cross-directional Control of Large Scale Sheet and Film Process
J. Proc. Cont., 11, 149-177(2001).
4. Braatz, R. D., Tyler, M. L. and Morari, M., “Identification and
Cross-Directional Control of Coating Processses,”AIChE J., 38,
1329-1339(1992).
5. Rawlings, J. B. and Chien, I.-L., “Gage Control of Film and She
Forming Processes,”AIChE J., 42, 753-766(1996).
6. VanAntwerp, J. G. and Richard, D., “Fast Model Predictive Cont
of Sheet and Film Processes,”IEEE trans. control syst. technol., 8(3),
408-417(2000).
7. Braatz, R. D., Ogunnaike, B. A. and Featherstone, A. P., “Identifi
tion, Estimation, and Control of Sheet and Film Processes,”in Proc.
of the IFAC World congr., Tarrytown, NY, 319-324(1996).
8. Beecher, A. E. and Bareiss, R. A., “Theory and Practice of Autom
Control of Basis Weight Profiles,”Tappi, 53(5), 847-852(1970).
9. Hovd, M., Braatz, R. D. and Skogestad, S., “SVD Controllers for H2−,
x̂j
Fig. 10. Contribution plot (a) calculated from the data during seven-teen minutes before the film-broken, (b) calculated from thefilm-broken data.
���� �41� �5� 2003� 10�
�� �� ����� �� �� � �� 591
n-
ntri-
ss
H�−, and µ-optimal Control,”Automatica, 33, 433-439(1996).
10. Featherstone, A. P. and Braatz, R. D., “Inupt Design for Large-Scale
Sheet and Film Processes,”Ind. Eng. Chem. Res., 37, 449-454(1998).
11. Russell, E. L. and Braatz, R. D., “Model Reduction for the Robust-
ness Margin Computation of Large Scale Uncertain Systems,”Com-
put. Chem. Eng., 22, 913-936(1998).
12. Gilmour, J. and Wang, L., “Detection of Process Abnormality in
Food Extruder Using Principle Component Analysis,”Chem. Eng.
Sci., 57, 1091-1098(2002).
13. Wold, S. Kim, E. and Geladi, P., “Principal Component Analysis,”
Chem. and Int. Lab. Sys., 2, 37-52(1987).
14. Harold, H., “Analysis of a Complex of Statistical Variables into Pri
cipal Components,”J. Educat. Psychol., 24, 498-520(1933).
15. Westerhuis, J. A., Gurden, S. P. and Smilde, A. K., “Generalized Co
bution Plots in Multivariate Statistical Process Monitoring,” Chem.
and Int. Lab. Sys., 51, 95-114(2000).
16. Sebzalli, Y. M. and Wang, X. Z., “Knowledge Discovery from Proce
Operational Data Using PCA and Fuzzy Clustering,”Eng. appl. artif.
intell., 14, 607-616(2001).
HWAHAK KONGHAK Vol. 41, No. 5, October, 2003