Proceedngs of the 2010 Internatonal Conference on Industral Engneerng and Operatons Management Dhaka, Bangladesh, January 9 10, 2010 Selecton of Industral Robots usng Compromse Rankng Method Vay Mankrao Athawale Department of Mechancal Engneerng, Government Polytechnc Amravat - 444603, Inda Prasenjt Chatterjee and Shankar Chakraborty Department of Producton Engneerng, Jadavpur Unversty Kolkata - 700032, Inda Abstract Selecton of an ndustral robot for a specfc engneerng applcaton s one of the most challengng problems n real tme manufacturng envronment. Ths has become more and more complcated due to ncreasng complexty, advanced features and facltes that are contnuously beng ncorporated nto the robots by dfferent manufacturers. The decson maker needs to select the most sutable ndustral robot n order to acheve the desred output wth mum cost and specfc applcaton ablty. Ths paper manly focuses on solvng the robot selecton problem usng VIKOR (Vlse Krterumska Optmzaca Kompromsno Resenje method, whch has already become a qute popular mult-crtera decson-makng (MCDM tool. One real tme example s cted n order to demonstrate and valdate the effectveness and applcablty of VIKOR method. Keywords Robot selecton, Mult-crtera decson-makng, VIKOR 1. Introducton An ndustral robot s a general purpose, reprogrammable machne wth certan anthropometrcal features. Its mechancal arm s the most mportant and vtal anthropometrcal component. Other less but stll mportant features, lke ts decson-makng capablty, capablty of respondng to varous sensory nputs and communcatng wth other machnes make t an mportant tool for dverse ndustral applcatons, ncludng materal handlng, assembly, fnshng, machne loadng, spray pantng and weldng. Control resoluton, accuracy, repeatablty, load carryng capacty, degrees of freedom, man-machne nterfacng ablty, programg flexblty, maxmum tp speed, memory capacty and suppler s servce qualty are the most mportant attrbutes to be taken nto account whle selectng a robot for a partcular ndustral applcaton. These attrbutes affectng the robot selecton decson can be classfed as objectve and subjectve attrbutes or benefcal and non-benefcal attrbutes. Objectve attrbutes can be numercally defned, such as the cost and load capacty of a robot. On the other hand, subjectve attrbutes are qualtatve n nature, e.g. vendor s servce qualty, programg flexblty etc. The benefcal attrbutes are those whose hgher values are always desrable, e.g. load carryng capacty, programg flexblty and non-benefcal attrbutes are those whose lower values are preferable, e.g. cost, repeatablty error. Whle selectng a robot for an ndustral applcaton, the decson maker needs to consder all these attrbutes, where a tradeoff between them and the robot performance measures s necessary. Selecton of the best suted robot for a gven ndustral applcaton from a large number of avalable alternatves s a typcal mult-crtera decson-makng (MCDM problem. Several approaches for robot selecton have already been proposed by the past researchers [1-11], whch nclude the applcatons of MCDM methods, producton system performance optmzaton models, computer-asssted models and statstcal models. In ths paper, a rankng of all the consdered alternatve robots s obtaned takng nto account dfferent robot selecton attrbutes and t s observed that the rankng obtaned usng the VIKOR method matches qute well wth that as derved by the past researchers, whch proves the applcablty of ths MCDM method to solve such type of complex ndustral problems. 2. Lterature Revew Bhangale et al. [1] developed a robot selecton methodology usng the technque for order performance by smlarty to deal soluton (TOPSIS and graphcal methods, and compared the relatve rankngs of the alternatve robots as obtaned usng these two methods. A codng system s employed for expressng varous robot selecton attrbutes 54
and a mert value s used to rank the robots n the order of ther sutablty for a gven ndustral applcaton. Goh et al. [2] proposed a revsed weghted sum decson model that can take nto account both the objectve and subjectve attrbutes whle selectng the ndustral robots. Khouja and Booth [3] used a statstcal procedure known as robust fuzzy cluster analyss that can dentfy the robots wth the best combnaton of specfcatons based on varous performance parameters. Khouja [4] developed a two-phase decson model for solvng the robot selecton problems. In the frst phase, data envelopment analyss (DEA s employed for dentfyng the robots wth the best combnaton of vendor specfcatons based on the robot performance parameters. In second phase, a mult-attrbute decsonmakng (MADM method s appled to select the best robot from those as dentfed n the frst phase. Zhao et al.[5] combned a mult-chromosome genetc algorthm wth frst-ft bn packng algorthm for the optmal robot selecton and workstaton assgnment problem for a computer ntegrated manufacturng system. Baker and Tallur [6] proposed a robot selecton methodology based on cross effcences n data envelopment analyss (DEA wthout consderng the crtera weghts or the decson maker s preferences. Goh [7] appled the analytc herarchy process (AHP for robot selecton that can smultaneously consder both the objectve and subjectve attrbutes. Parkan and Wu [8] demonstrated the applcatons and nterrelatonshp of the operatonal compettveness ratng (OCRA and TOPSIS methods n a robot selecton problem and compared ther performances wth other approaches. It s observed that both these methods are strongly nterrelated, and ther performance measurements and decsonmakng processes nvolve the same mathematcal treatment though they have ther apparent structural dfferences. Rao and Padmanabhan [9] employed the dagraph and matrx methods for evaluatng and rankng of the alternatve robots for a gven ndustral applcaton, usng the smlarty and dssmlarty coeffcent values. Kahraman et al. [10] developed a herarchcal fuzzy TOPSIS method to solve the mult-attrbute robot selecton problems. Karsak [11] ntroduced a decson model for robot selecton based on qualty functon deployment (QFD and fuzzy lnear regresson methods whle ntegratng the user demands wth the techncal characterstcs of the robots. Although a number of research works have already been presented by the past researchers on robot selecton problems, but stll there s a need for a smple as well as systematc approach/mathematcal tool to gude the decson makers to select and dentfy the best suted robot from a gven set of alternatves, because a wrong selecton may often negatvely contrbute to the productvty and flexblty of the entre process. In ths paper, an attempt s made to dscover the potentalty and applcablty of VIKOR (a compromse rankng method whle selectng the most sutable robot for a gven ndustral applcaton. 3. Compromse Rankng Method The VIKOR (the Serban name s Vlse Krterumska Optmzaca Kompromsno Resenje whch means multcrtera optmzaton (MCO and compromse soluton method was manly establshed by Zeleny [12] and later advocated by Oprcovc and Tzeng [13-14]. Ths method s developed to solve the MCDM problems wth conflctng and non-commensurable (crtera wth dfferent unts attrbutes, assug that compromse can be acceptable for conflct resoluton, when the decson maker wants a soluton that s the closest to the deal soluton and the alternatves can be evaluated wth respect to all the establshed attrbutes. It focuses on rankng and selectng the best alternatve from a fnte set of alternatves wth conflctng crtera, and on proposng the compromse soluton (one or more. The compromse soluton s a feasble soluton, whch s the closest to the deal soluton, and a compromse means an agreement establshed by mutual concessons made between the alternatves. The followng multple attrbute mert for compromse rankng s developed from the L p -metrc used n the compromse programg method [15]. L p, [ m ] [ (m ] 1/p M p = (w j (m max (m max (1 j = 1 1 p ; = 1,2,..., N where M s the number of crtera and N s the number of alternatves. The m values (for = 1, 2,...,N; j = 1,2,...,M denote the values of crtera for dfferent alternatves. In the VIKOR method, L1, and L, are used to formulate the rankng measure. The procedural steps for VIKOR method are hghlghted as below: Step 1: Identfy the major robot selecton crtera for a gven ndustral applcaton and short-lst the robots on the bass of the dentfed crtera satsfyng the requrements. A quanttatve or qualtatve value s assgned to each dentfed crteron to construct the related decson matrx. 55
Step 2: a After short-lstng the robots and development of the decson matrx, detere the best, (m max and the worst, (m values for all the crtera. b The weghts or relatve mportance of the consdered crtera are estmated usng analytc herarchy process (AHP or any other method. c Calculate the values of E and F. M E 1, = w j = [(m max m ] [(m ] (2 j 1 m F, = Max of { w j [(m max m ] [(m ]} j= 1,2,..., M (3 Eqn. (2 s only applcable to benefcal attrbutes (whose hgher values are desrable. For non-benefcal attrbutes (whose lower values are preferable, the term [(m max m ] n Eqn. (2, s to be replaced by [m (m ]. Hence, for non-benefcal attrbutes, Eqn. (2 can be rewrtten as: M E 1, = w j = [(m (m ] [(m ] (4 j 1 d Calculate P values. P = v((e E - /(E -max E - + (1 v((f F - /(F -max F - (5 where E -max and E - are the maxmum and mum values of E respectvely, and F -max and F - are the maxmum and mum values of F respectvely. v s ntroduced as weght of the strategy of the majorty of attrbutes (or the maxmum group utlty. The value of v les n the range of 0 to 1. Normally, the value of v s taken as 0.5. The compromse can be selected wth votng by majorty (v > 0.5, wth consensus (v = 0.5 or wth veto (v < 0.5. e Arrange the alternatves n the ascendng order, accordng to the values of P. The compromse rankng lst for a gven v can be obtaned by rankng wth the P measure. The best alternatve s the one havng the mum P value. The VIKOR method s an effectve MCDM tool, specfcally applcable to those stuatons when the decson maker s not able, or does not know to express hs/her preference at the begnnng of the decson-makng process. The resultng compromse soluton can be accepted by the decson maker because t provdes a maxmum group utlty of the majorty and a mum ndvdual regret of the opponent. The compromse solutons can be the base for negotatons, nvolvng the decson maker s preference on crtera weghts. The VIKOR results depend on the deal soluton, whch stands only for the gven set of alternatves. Incluson (or excluson of an alternatve can affect the VIKOR rankng of the new set of alternatves. 4. Illustratve Example In order to demonstrate and valdate the applcaton of the above-mentoned MCDM method for solvng the robot selecton problem, a real tme example s cted. Ths example [1] deals wth the selecton of the most sutable robot for some pck-n-place operatons where t has to avod certan obstacles. Performance of an ndustral robot s often specfed usng dfferent attrbutes. Repeatablty, accuracy, load capacty and velocty are observed to be the most mportant attrbutes affectng the robot selecton decson. Among these, repeatablty and accuracy are the most confusng attrbutes. Repeatablty s the measure of the ablty of a robot to return to the same poston and orentaton over and over agan, whle accuracy s the measure of closeness between the robot end effectors and the target pont, and can usually be defned as the dstance between the target pont and the center of all ponts to whch the robot goes on repeated trals. It s easer to correct poor accuracy than repeatablty and thus, repeatablty s generally assumed to be a more crtcal attrbute. Load capacty s the maxmum load that a manpulator can carry wthout affectng ts performance. Load capacty of a robot s related to ts acceleraton and speed, and s a functon of manpulator acceleraton and wrst torque. Maxmum tp speed s the speed at whch a robot can move n an nertal reference frame. Memory capacty of a robot s measured n terms of number of ponts or steps that t can store n ts memory whle traversng along ts predefned path. Manpulator reach s the maxmum dstance that can be covered by the robotc manpulator so as to grasp the objects for the gven pck-n-place operaton. Although t s usually assumed that the specfed robot selecton attrbutes are mutually ndependent, n general, performance 56
parameters provded by dfferent robot manufacturers are not smultaneously achevable. Furthermore, t s qute dffcult to establsh the functonal relatonshp between those robot selectons attrbutes. Hence, makng ths assumpton ntroduces a rsk of selectng a robot that may fal to provde the requred performance. In ths example [1], fve dfferent robot selecton attrbutes are consdered as load capacty (LC, maxmum tp speed (MTS, repeatablty (R, memory capacty (MC and manpulator reach (MR, among whch load capacty, maxmum tp speed, memory capacty and manpulator reach are the benefcal attrbutes, whereas, repeatablty s a nonbenefcal attrbute. Thus, the robot selecton problem conssts of fve crtera and seven alternatve robots, as gven n Table 1. Table 1: Quanttatve data for dfferent robots [1] Sl. No. Robot LC (kg MTS (mm/s R (mm MC MR (mm 1. ASEA-IRB 60/2 60 2540 0.40 500 990 2. Cncnnat Mlacrone T3-726 6.35 1016 0.15 3000 1041 3. Cybotech V15 Electrc Robot 6.8 1727.2 0.10 1500 1676 4. Htach Amerca Process Robot 10 1000 0.20 2000 965 5. Unmaton PUMA 500/600 2.5 560 0.10 500 915 6. Unted States Robots Maker 110 4.5 1016 0.08 350 508 7. Yaskawa Electrc Motoman L3C 3 177 0.10 1000 920 The problem of selectng the best suted ndustral robot for the gven pck-n-place operaton s solved usng the VIKOR method. At frst, the best and the worst values of all the crtera are dentfed. Rao [15] estmated the crtera weghts as w LC = 0.036, w MTS = 0.326, w R = 0.192, w MC = 0.326 and w MR = 0.120 usng analytc herarchy process (AHP and these weghts are used here for the VIKOR method-based analyss. Now, the values of E and F are calculated usng Eqns. (2 or (4 and (3 respectvely, as gven n Table 2. Table 2 also shows the values of P for v = 0.5 and the compromse rankng lst of the consdered alternatve robots. The canddate robots are arranged n ascendng order, accordng to the values of P. The best choce of robot for the gven pck-n-place operaton s robot 3 (Cybotech V15 Electrc Robot. Cncnnat Mlacrone T3-726 s the second choce and the last choce s robot 7 (Yaskawa Electrc Motoman L3C. Rao [15] obtaned a rankng of the alternatve robots as 3-2-7-1-4-6-5 usng the TOPSIS method, whereas, VIKOR method derves a compromse rankng of robots as 3-2-4-1-5-6-7 (Spearman s rank correlaton coeffcent = 0.8333. It s observed that n VIKOR method, the frst and second best choce of robots reman the same. Table 2: E, F and P values for alternatve robots Robot E F P Rank 1 0.5700 0.3075 0.7473 4 2 0.3511 0.2103 0.1034 2 3 0.3420 0.1845 0 1 4 0.5118 0.2125 0.3314 3 5 0.7069 0.3075 0.9348 5 6 0.6910 0.3260 0.9782 6 7 0.6974 0.3260 0.9870 7 Whle calculatng P values, the value of v s usually taken as 0.5 [15], but actually ts value les between 0 and 1. Table 3 shows the comprse rankngs of the alternatve robots for two extreme values of v = 0.1 and v = 0.9. In both the cases, the best choce of robot (Cybotech V15 Electrc Robot does not change, although the rankng of the alternatve robots changes slghtly. 57
Table 3: Rankngs of robots for dfferent values of v P (v = 0.1 Robot/Rank P (v = 0.9 Robot/Rank 0.8451 ASEA-IRB 60/2 (4 0.6494 ASEA-IRB 60/2 (4 0.1661 Cncnnat Mlacrone T3-726 (2 0.0407 Cncnnat Mlacrone T3-726 (2 0 Cybotech V15 Electrc Robot (1 0 Cybotech V15 Electrc Robot (1 0.2242 Htach Amerca Process Robot (3 0.4387 Htach Amerca Process Robot (3 0.8826 Unmaton PUMA 500/600 (5 0.9870 Unmaton PUMA 500/600 (7 0.9956 Unted States Robots Maker 110 (6 0.9608 Unted States Robots Maker 110 (6 0.9974 Yaskawa Electrc Motoman L3C (7 0.9766 Yaskawa Electrc Motoman L3C (5 5. Conclusons The cted example demonstrates the potentalty, applcablty and smplcty of the compromse rankng method n solvng robot selecton decson-makng problems. The method can ncorporate the decson maker s preferences regardng the relatve mportance of dfferent robot selecton attrbutes. As the measures of the quanttatve as well as qualtatve attrbutes and ther relatve mportance are used together to rank the alternatves, the VIKOR method provdes a better evaluaton of the alternatves. It can make a compromse rankng of the alternatve robots from a fnte set of alternatves for a gven problem. The results derved usng ths MCDM method almost match wth those as obtaned by the past researchers. Ths compromse rankng method can also be used for any type of decsonmakng problems, nvolvng any number of quanttatve and qualtatve attrbutes, and any number of alternatves. In order to facltate the applcaton of VIKOR method for solvng varous MCDM problems, the related flow logc presentng ts mplementaton module may be developed. References 1. Bhangale, P. P., Agrawal, V. P. and Saha, S. K., 2004, Attrbute based Specfcaton, Comparson and Selecton of a Robot, Mechansm and Machne Theory, 39, 1345-1366. 2. Goh, C.-H., Tung, Y.-C. and Cheng, C-H, 1996, A Revsed Weghted Sum Decson Model for Robot Selecton, Computers & Industral Engneerng, 30, 193-199. 3. Khouja, M. and Booth, D. E., 1995, Fuzzy Clusterng Procedure for Evaluaton and Selecton of Industral Robots, Journal of Manufacturng Systems, 14, 244-251. 4. Khouja, M., 1995, The Use of Data Envelopment Analyss for Technology Selecton, Computers & Industral Engneerng, 28, 123-132. 5. Zhao, L., Tsujmura, Y. and Gen, M., 1996, Genetc Algorthm for Robot Selecton and Work Staton Assgnment Problem, Computers & Industral Engneerng, 31, 599-602. 6. Baker, R. C. and Tallur, S., 1997, A Closer Look at the Use of Data Envelopment Analyss for Technology Selecton, Computers & Industral Engneerng, 32, 101-108. 7. Goh, C.-H., 1997, Analytc Herarchy Process for Robot Selecton, Journal of Manufacturng Systems, 16, 381-386. 8. Parkan, C. and Wu, M.-L., 1999, Decson-makng and Performance Measurement Models wth Applcatons to Robot Selecton, Computers & Industral Engneerng, 36, 503-523. 9. Rao, R. V. and Padmanabhan, K. K., 2006, Selecton, Identfcaton and Comparson of Industral Robots usng Dgraph and Matrx Methods, Robotcs and Computer-Integrated Manufacturng, 22, 373-383. 10. Kahraman C., Çevk, S., Ates, N. Y. and Gülbay, M., 2007 Fuzzy Mult-crtera Evaluaton of Industral Robotc Systems, Computers & Industral Engneerng, 52, 414-433. 11. Karsak, E. E., 2008, Robot Selecton usng an Integrated Approach based on Qualty Functon Deployment and Fuzzy Regresson, Internatonal Journal of Producton Research, 46, 723-738. 12. Zeleny, M., 2002, Multple Crtera Decson Makng, McGraw Hll, New York. 13. Oprcovc, S. and Tzeng, G. H., 2004, Compromse Soluton by MCDM Methods: A Comparatve Analyss of VIKOR and TOPSIS, European Journal of Operatonal Research, 156, 445-455. 14. Oprcovc, S. and Tzeng, G. H., 2007, Extended VIKOR Method n Comparson wth Outrankng Methods, European Journal of Operatonal Research, 178, 514-529. 15. Rao, R. V., 2007, Decson Makng n the Manufacturng Envronment usng Graph Theory and Fuzzy Multple Attrbute Decson Makng Methods, Sprnger-Verlag, London. 58