Classification of 3D Prismatic Parts Based on Global and Local Shape Information NSC88-2212-E-009-036 87/08/01---88/07/31 mcwu@cc.nctu.edu.tw ( !:"#$%&'() *&+,-./) 0123456789:;<=>? @+ABCD6ABEFGH;<=:" #$%CIJ"#$%KL=:MNO PQR(global shape information)S@+:T 5689EUV"#$%:WXYZ[ \CJYZ]^(tree matching):AB_` [\:;<=@a`56='S@+: Tb689Ec`56='dJefg +,-./hijk:@+Cl+,-. /:mnE"#$%opa:qrKL BCIUV;<=:stOPQRuMq rKLBCv@+wiWxySlb8 9@+ABz{|\}56;<=>?~ wCJ;<=: a0S
(Keywords: Simplified Skeleton, Group Technology, Neural network)
This research presents an automatic classification scheme for punched parts. The classification is performed based on part’s global shape descriptor modeled by their simplified skeletons. In the first stage, the simplified skeleton of a punched part is represented by a sequentially coded tree structure. Based on this tree representation, punched parts with similar tree structure can be clustered using a tree matching method. In the second stage, the parts within a cluster can be further classified by a back-propagation neural network classifier. The two-stage classification scheme can be used in a punched part
retrieval system aiming to facilitate the design and manufacturing tasks, which would greatly reduce the time-to-market and material inventory.
;<=EbPZ:CE + C¡¢PZ£¤¥:¦§C z¨©E56ª«:¬c®S¯ °±¨²r{³W:;<=C´ zJµ¶a0C·zJ¸¹r: Sº~+ :;<=C» ºW¼½:=@+w¾¿ÀCÁ¨Â ¹:ÃÄo»:;<=Jh½¬:Å ÆS ÇȾWÉ Ê12°¢^= >?@+[1, 2, 3, 4]CD+:12ËE _=:>?#ÌÍÎ56ÏÐÑÍ (feature recognition):ÒÓCÔÕÖ×Ø{ ÙÚÛ=:Ü¢QÝÑÍ4Þ&ß&à á£âÏãPZästÏÐCIåVl [±^=Ì@+S =ÏÐÑÍ:ABºEæç= stOPQR(local shape information):+ èyCzEstÏÐéèI´êK=: MNOP·+èSÎëìístOPQR :´îCWï12J"#$%[5]KL= : M N O P Q R (global shape information)CðJ"#$%hÎ=@+ :UVCl BcÏã:=ñòóôõ H´ö:@+[±[6, 7]SEDï12-÷r:=øùúCûüýW76 þWT5E@+:þx yCBuM@+:[±TbEDï@ +wW8 :C" OP
! :=Ëc5Cc=+ CWB\}4>?@+:S Îë7þC0125 6;<=:>?@+BSD6AB:w %\z@a6¬(ST56¬(FJ Wu Wang o:B[8]_;<= :OP p!a" #PCúJ Wu Chen o:B[5]GH$%P: "#$%STb6¬(EJ012o34 :"#$%KLBC_"#$%KLaY Z[\ST&6¬(E'@+£(Î@ 'C_Wé`YZ[\:"#$%@a `5'ST6¬(Ek@+C^`5' ó:;<=Jefg+,-./Î@+: C^"#$%h5):@+S*Î +,-./:mn+ÎqrCoJc@+ ,C»F_"#$%:YZ[\paq r:KLA-SåV.ú@+:[±ûü 0AB/0ëJ1@+AB:7þ S 3.1 012o3:;<=@+ABE_; <=:MNOPQRJ"#$%KLCð UV"#$%^;<=hb89:@+S @+:T5689EJYZ[\]^(tree matching):ABC_"#$%:YZ[\ é`:;<=@a`5'STb689ð {efg(Back-Propagation)+,-./ ^`5';<=:"#$%@+S ;<=@':ABEFGH"#$% :WXYZ[\(coded tree structure)Cv2 5"#$%¨W5:YZ[\KLBC ðJlWXYZ[\h77]^C_W é`WXYZ[\:;<=3a`5'S ·©E4@':EåV[\:+è ySWXYZ[\:GHABCT5EY å:56C7ã8:9:CTbð;< oW9:56=XPaWXYZ[ \ST&ð_25å9::>?U=X @a56ABPaWXYZ[\CJ] ^AB:ABCa;<=:@'S 3.2 789=@+:Tb689E^` 56='k@+S@+:UVE" # $ % : q r K L B C v { e f g (Back-Propagation)+,-./[9]hÎ@+ :S*Îefg+,-./E56D E-(supervised)+,-./CD+,-./F+F-÷56GH:÷CÁzJ Î@+:SI+,-./:mn +Îqr:P-CoJc@+,»º_" #$%:YZ[\p!aJqr:A-S "#$%paqrKLBW&=Jº hCT5=JE7ã$%9::Fú= XCTb=JEJ5(AKêK9::« yKCT&=JE_êK9::«yKU L9:=X@B¾C©HÄ"#$%: qrKLBSc012dC"#$%:9 :=XEMWXYZ[\7ãCJN4 O9:«yK:ãPS 012J6«yKL25å9:o Q:«yKCDï«yRS9::Ü¢ TUQRS{D6«yVW:AK ("(«yK)CoX4:qrKLBzJY ZÎZ¾"#$%:YZ[\SD6« y:ãP°NS
1.
T56«yE9::é^ó8[S2.
Tb6«yE9::é^\[S3.
T&6«yE9:]9::^_S4.
T6«yÎ$9:o`:aS D6«y"#$%:bc¬dÏ yzJ 56^]ST56«yêK$9 :¬dbc:eCTb6«y¬dbc :\WCT&6«yE¬dbc:f qCT6«yE¬dbcfg:aS HÄoW9::6«yKúC_2 569::«yKUL9::Fú=X@ B©E"#$%:qrKLBS cv{efg+,-./,CºFG HD6+,-./CÁ¨^mnqr @+Chk:GH÷ijæ Lippmann :" kl[9]S012v{:m0=´EM; <=d5GCEM"#$%:WXYZ [\®Cn:Eº®CMIWûo p§:m0CGm^@+[±: qrSTb89@+:ñòE`5':; <=C`5';<=sWé`:WXYZ [\C*l`5':m0="#$% :WXYZ[\·ºé`S012JWX YZ[\ÎtuCuM9:9`:a \CPa´`:"#$%hÎm0CJ NEm0=:®v=S
1.
w9:]9::`¨c]9 ::7x£dS2.
m0=:9::\[yãC9: Î 1CTb 9:Î 0.5CT& 9: Î 0.25CUXz{S3.
m0=|9::ó8[é`C} Î 1S4.
m0=-~pú´Hâm0 =é`C:=zC*; <=W"@S J% 1.a 7å9:é9:"#$%Î CT5 :9:faCw9:z JWN|&6`C®6m0 =C-~p 180 [~ dW&6E? :C*l.úW°% 1.b :&6m0= ®(Sa1, Sa2, Sa3)S 0125W&'m0=C25' GH56+,-./CT5'W&6m 0=(Sa1, Sa2, Sa3) C@a&6´`:+ @Î Ga1, Ga2, Ga3STb'6= (Sb1—Sb9)@;<Î Gb1 Gb9 + CT&'(Sc1—Sc14)E Gc1 Gc14 6+S 0125Gë 55 6;<=hÎ@+: 0CD 55 6;<=U"#$%:WX YZ[\zJ@a&6='CT56 = ' W 6 ø ù = (W1—W4) (° % 2.a) T b6='W 26 6øù= (W5—W30)T&6='W 25 6øù =(W31—W55)S øù=:"#$%paqrKLB ú@U='mn&6ôCaGH :+,-./C256mnqr: Ë´Ä 1 ÕHÄ0m4qrS0 12UL@+:[±_øù=Mc% 2.bC256øù=:+EUKdm4 qr:A!.Ê7ãS 3.3 c@+:÷d×zJC;< =ºWC=:"#$%© 5å9:CD9: 5å:P7ã =o«:='C°;<= W17 W19 Ë*ÎW5å@ÄTb6= 'SEDï0cëC*lÛM NO¾D76=T5=' :T&+=·+èS012^D 34ë567ABC(ÎÏÐ@ BSo:ÏÐ@BE_÷£÷k: 9:ÍÎ=:ÏÐCF_DïÏ Ðð=:@+S c012dCÏÐ: ãEJÎ :A-ã76¡K(threshold)C@E9 ::T56«y:KTb6«y:KC ¢9::\[£ó8[|£:¡K C@+wFDï9:ÍÎÏÐC FDï9:úð^$=@ +S°øù=Û W17¤8C_Tb6« y:¡KÎ 0.1CDï=zHÄcT 5='N:@+[±(K 1)S 1. b. a. a. b. 2.# K 1. ÏÐ:=@+[± c012dCv{ÊÔzJã7( J´`:¡KCHÄ56=c´ `='d:@+[±SDï@+[±z JUL{¥:´`¸Jv{C°=: R¦§zJÛ[\¨ :='@+ [±56SÏÐ@BW©:MstO PQRC7ÏÐ^0@+w:q rC´ªv@+wiWxyI@+[ ±:Ø{iΫ¬S ! "# 0123456789:;<=@+ ABCcD6ABd;<=ºFpaÇè :" #PCðJ" #P:"#$ %KL=:MNOPQR(global shape information)S@+:T5689EUV" #$%:WXYZ[\CJYZ]^(tree matching):AB_`[\:;<=@a` 56='S@+:Tb689Ec`5 6='dJefg+,-./U"#$ %pa:qrKLBhijk:@+S D6b89@+wºWJN6 ®T5Cb89@+zJõHéè [´`:@+C¯¸@+[±:v{x ySTbCb89@+zJ56+, -./:°±C@+[±i"/ST&C ²÷012:ÏÐ@BC@+w¨M MNstOPQRc´`:=' Hį=³¢:@+STCm0 =EJ´ã:µ®C´¶Î:q r¦@+[±STC012v{·" :"#$%KLBC=:+èy¦#Î 9¸yC¯¸@+uM:xyS $%&'
1. Kaparthi, S., and Suresh, N., A neural network system for shape-based classification and coding for rotational parts. International Journal of Production Research, 29, 1771-1784 ( 1991).
2. Chakraborty, K. and Roy, U.,
“Connectionist models for part-family classifications,” Computers Industrial Engineering, 9, 189-198 (1992).
3. Kao, Y. and Moon, Y. B., “A unified group technology implementation using the back-propagation learning rule of neural network,“ Computers Industrial Engineering, 20(4), 425-437 (1991).
4. Liao, T. W. and Lee, K. S., ”Integration of a feature-based CAD system and an ARTI neural network model for GT coding and partfamily forming,” Computers Industrial Engineering, 26(1), 93-104 (1994).
5. Wu, M. C., and Chen, J. R., A skeleton approach to modelling 2D workpieces, Journal of Design and Manufacturing, 4, 229-243 (1994).
6. Wu, M. C., and Jen, S. R., “Global shape information modeling and classification of 2D workpieces,” International Journal of Computer Integrated Manufacturing, 7(5), 261-275 (1994).
7. Cheok, B. T., Zhang, Y. F., and Leow, L. F., “A skeleton-retrieving approach for the recognition of punch shapes,” Computers in Industry, 32, 249-259 (1997).
8. Wu, M. C., and Wang, J. T., “An algorithm for converting the contour of 2D workpiece into a rectilinear polygon,” Computers in Industry, 29, 197-208 (1996).
9. Lippmann, R. P., “An introduction to computing with neural nets,” IEEE ASSP Magazine, April, 4-22 (1987).