XXVIII. MODELING AND OPTIMIZATION OF RADIO FREQUENCY IDENTIFICATION NETWORKS FOR INVENTORY MANAGEMENT

Transcription

1 XXVIII. MODELING AND OPTIMIZATION OF RADIO FREQUENCY IDENTIFICATION NETWORKS FOR INVENTORY MANAGEMENT Atpong Surya Department of Electrcal and Electroncs Engneerng Ubonratchathan Unversty, Thaland, J. Davd Porter School of Mechancal, Industral and Manufacturng Engneerng Oregon State Unversty, USA, Abstract Stock loss and out-of-stocks are outcomes of poorly desgned nventory management systems and can lead to sgnfcant revenue losses. Inventory management systems (IMSs) based on rado frequency dentfcaton (RFID) have the potental to mnmze these losses f they are properly desgned and deployed. However, the placement of RFID readers to support IMSs s often done on a tral and error bass whch s tme consumng and results n less than optmal coverage. A methodology to fnd the optmal locaton and number of RFID readers n a warehouse faclty to ensure a desred level of rado frequency sgnal coverage was developed n ths research. The results show that the proposed methodology works very well when appled to small rectangular facltes and small nverted-t facltes. However, some lmtatons exst when the method s appled to large facltes. 1 Introducton Rado frequency dentfcaton (RFID) s an automatc dentfcaton and data capture (AIDC) technology that has ganed sgnfcant attenton n recent years as a means of enhancng the traceablty of tems throughout the supply chan. Market trends ndcate a rapd growth n the demand for RFID-based technologes, wth an expected 18% compound annual growth rate for the three-year perod , culmnatng n a proected $19.3 bllon ndustry by the end of 2014 [1]. 1

2 A basc RFID system ncludes four man components: a host computer system, a RFID reader, an antenna, and a transponder (or RFID tag). RFID technology, when properly mplemented, has several benefts over more conventonal AIDC technology (.e., bar codes), ncludng trackng physcal obects n real tme and reductons n process tmes, labor, and the amount of paperwork needed n day-to-day operatons. Applcatons of RFID technology are becomng more common n the supply chan (and partcularly n warehouse facltes) to enable product trackng and to support nventory management systems (IMSs) [2-4]. A RFID-based IMS can be desgned to have localzed coverage (e.g., RFID readers nstalled ust on recevng/shppng doors) or to cover the entre warehouse faclty. A RFID-based IMS that covers the entre warehouse faclty wll result n several advantages when compared to one wth localzed coverage, ncludng: Real tme nventory trackng. Wth entre warehouse faclty coverage, an organzaton can montor and track nventory levels n real tme. A RFID system wth localzed coverage cannot track nventory levels n real tme because t s unable to track nventory tems when they are movng. Shrnkage and theft preventon. Ernst and Young [5] found that retalers lose $46 bllon annually to nventory shrnkage ncludng employee theft, shoplftng, admnstratve error and vendor fraud. A RFID-based IMS wth entre warehouse faclty coverage and contnuous trackng has the potental to reduce shrnkage, theft, and loss of nventory. Localzaton. A RFID-based IMS wth entre warehouse faclty coverage can provde the capablty of nventory/tem localzaton and/or postonng. Travel tme estmaton and warehouse layout mprovement. Wth comprehensve trackng, locaton, and dentfcaton, the movement hstory of tagged tems can be obtaned. Also, the average travel tme t takes employees to get to the desred tems durng pckng and put away operatons can be calculated. Item movement hstory may also be useful for adustng a route to access tems wth shortest path and mprovng the effcency of the warehouse faclty layout. One of the most mportant decsons when desgnng RFID networks to support warehouse operatons s where to locate the RFID readers. Locatng RFID readers n a faclty s referred n the lterature as the RFID network plannng (RNP) problem [6-9]. In practce, the placement of RFID readers s often done on a tral and error bass. Ths placement method s very tme consumng and results n a less than optmal coverage of the faclty area. Provdng an optmal soluton to the RNP problem s a complex task that may be affected by several factors such as rado frequency (RF) sgnal coverage, RF sgnal propagaton, and nterference (.e., both passve and actve). Thus, a formal methodology to optmze the placement and number of RFID readers requred to cover a faclty s needed. Crtcal steps n such a methodology are the development of a mathematcal 2

3 formulaton and the selecton of an approprate optmzaton technque to ensure that qualty solutons are obtaned wthn a reasonable computaton tme. Several researchers have attempted to solve the RNP problem. For example, Guan et al. [10] developed a computer-based model amed at optmzng the placement of RFID readers under the assumptons of omndrectonal antenna coverage (.e., crcular sgnal coverage) and free space propagaton. However, no physcal obstacles were consdered n the model. Chen et al. [11] developed a model consderng crcular antenna coverage and free space propagaton. They dd not model obstacles or the uplnk communcatons channel ether. It s clear that there are several crtcal techncal gaps and challengng ssues to be addressed when developng models to solve the RNP problem. In ths research, a methodology was developed to model and optmze the desgn of RFID networks for IMS whch mproves on the research done n [10] and [11]. The proposed methodology fnds a soluton that ensures complete RF coverage of a warehouse faclty whch would allow a RFID network to support real-tme nventory trackng and localzaton. The remander of the manuscrpt s organzed as follows. Secton 2 descrbes the research methodology. Secton 3 presents a proposed mathematcal formulaton to represent the RNP problem. Secton 4 ntroduces the heurstc optmzaton algorthms used to fnd the solutons for RNP problem. Secton 5 descrbes the expermental desgn used to dentfy factors that have the most sgnfcant effect on the qualty of the solutons. The results and dscussons are presented n Secton 6. Fnally, the man conclusons of ths research are dscussed n Secton 7. 2 Methodology The frst step of the methodology nvolved a thorough revew of the EPCglobal class 1 generaton 2 (C1G2) standard to dentfy the most crtcal parameters for RFID network plannng for warehouse nventory management. These parameters were later used n the development and as nputs of the crcular antenna coverage (CAC) mathematcal formulaton. The CAC mathematcal formulaton s based on the assumpton that the antenna attached to the RFID reader generated a crcular coverage area. When compared to pror work, the underlyng assumptons that guded the development of the CAC mathematcal formulaton make the modelng of the RFID-based IMSs more realstc and applcable to warehouse envronments. These mportant assumptons ncluded the consderaton of the uplnk communcaton channel (.e., tag-to-reader), the utlzaton of approprate RF propagaton models for the downlnk and for the uplnk communcaton channels, and the consderaton of obstacles. The heurstc optmzaton technques partcle swarm optmzaton (PSO) and genetc algorthm (GA) were used to fnd good solutons for the CAC mathematcal formulaton to ensure approprate coverage for the warehouse faclty, wth less nterference and wthn reasonable computaton tme. Several factors of the RNP problem 3

4 as well as specfc parameters of each heurstc optmzaton technque were adusted to mprove the qualty of the solutons. A desgned experment was conducted followed by a mult-factor analyss of varance (ANOVA) to dentfy those factors that had the most sgnfcant effect on the qualty of the solutons. Fnally, the most approprate RFID network desgn settngs were determned. When the specfc faclty shape, faclty sze, and locaton of obstructons are known, the developed research methodology can be appled to optmze both the number and the placement of RFID readers n the system. 3 The Crcular Antenna Coverage Mathematcal Formulaton 3.1 Assumptons and Notaton Several characterstcs of the RFID system were ncorporated nto the CAC mathematcal formulaton, ncludng crtcal operatonal parameters specfc to the RFID reader, the RFID tag, and the RF sgnal propagaton model. Also, the characterstcs of the test faclty, such as shape and sze, were consdered. The underlyng assumptons made n the development of the CAC mathematcal formulaton were: 1. The shape and sze of the warehouse faclty and the locaton of obstacles wthn the warehouse faclty are known. The area of the warehouse faclty s dscretzed nto grds of equal heght and wdth. 2. The term RFID reader refers to the combnaton of a RFID nterrogator and a sngle antenna. 3. The RFID reader and the RFID tags are complant wth the EPCglobal C1G2 standard. 4. Only one type of RFID reader and RFID tag were utlzed,.e., all RFID readers and RFID tags have the same operatng frequency, maxmum transmttng power, mnmum recevng power, antenna type, and antenna gan. Mxed brands of RFID systems and/or multple detecton frequences were not consdered. 5. The polarzaton and mpedance of the antennas match perfectly. 6. Dpole antennas are used for both the RFID reader and the RFID tag. These antennas produce a crcular coverage area n a 2D representaton. 7. The antenna for the RFID reader s placed at the center of a grd and n a manner that maxmzes sgnal coverage (e.g., on the celng of the warehouse faclty). 8. The radus of the sgnal coverage area produced by the antenna s known and can be calculated usng a RF propagaton model. a) The multpath propagaton model s used for the downlnk channel (.e., readerto-tag). b) The radar cross secton (RCS) propagaton model s used for the uplnk channel (.e., tag-to-reader). The notaton used to represent the parameters of the RFID system n the CAC mathematcal formulaton s descrbed below. 4

5 n Number of grds n the system, n = C R The ndex of a grd, = 1, 2,..., n The ndex of a grd where a RFID reader s placed, = 1, 2,..., n k The ndex of a grd where an nterferng RFID reader s located, k IS N Intal number of RFID readers r Radus of the sgnal coverage area of the RFID reader s antenna W Wdth of the faclty area H Heght of the faclty area C Total number of grd columns n the faclty area R Total number of grd rows n the faclty area w Wdth of grd h Heght of grd FG Set of forbdden grds, FG = { sthe forbddengrd } t RS FA TA D, D (max) IS (max) ID U, (max) U (max) IU x y Number of nstalled RFID readers RS Set of nstalled RFID readers, { x =1} = Feasble area Total area Sgnal power levels receved at grd from the RFID reader located at grd Set of maxmum sgnal power levels receved at grd from the RFID reader located at grd, N max = max, ( ) { ( )} D D, Set of RFID readers that generate nterference at grd, = 1, 2,..., C R Set of maxmum sgnal power levels receved at grd from the RFID reader located at grd, and nterfered by the neghborng RFID reader(s) located at grd k, k IS, ID ( max ) = {( D ( max) ) k } Sgnal power levels (backscattered) receved at the RFID reader located at grd from grd Set of maxmum sgnal power levels (backscattered) receved at the RFID (max) = max reader located at grd from grd, { ( )} U U, Set of maxmum sgnal power levels (backscattered) receved at the RFID reader located at grd from grd and nterfered by neghborng RFID readers located at grd k, IU ( max ) = {( U ( max) ) k } Bnary decson varable to dentfy whether or not a RFID reader s located at grd : x = 1 f an RFID readers located at grd 0 otherwse Bnary decson varable to dentfy whether or not grd s covered by at least 5

6 τ D τ U β one RFID reader: y = τd τu β Grd s covered by the RFID reader located at grd (y = 1) f and only f the followng three requrements are satsfed: 1) The downlnk communcaton from the RFID reader located at grd to grd s successfully establshed. 2) The uplnk communcaton from grd to the RFID reader located at grd s successfully establshed. 3) Grd s not a forbdden grd. Successful transmsson from the RFID reader located at grd to grd D (max) TD τ D = max, 0 D (max) TD TD s the specfc downlnk power threshold (29 mcrowatts or dbm) Successful transmsson from grd to the RFID reader located at grd,.e., the backscattered sgnal from the RFID tag to the RFID reader U (max) TU τ U = max, 0 U (max) TU TU s the specfc uplnk power threshold (0.316 nanowatts or -95 dbm) Feasble coverage of grd β = 1 f FG 0 otherwse N The set of RFID readers ( RS ) that can cover grd, N = d r FG DN UN d W1 W 2 W3 C { } The set of sgnal power levels receved at grd from the RFID reader located DN D N =, at grd for downlnk channel, { } The set of sgnal power levels receved at grd from the RFID reader located UN U N =, at grd for uplnk channel, { } The Eucldean dstance from the center of grd to the center of grd Weght assgned to the frst obectve,.e., total coverage Weght assgned to the second obectve,.e., total sgnal to nterference rato (SIR) Weght assgned to the thrd obectve,.e., total cost Cost of the RFID reader at grd 6

7 3.2 CAC Mathematcal Formulaton The obectve functon and constrants of the CAC mathematcal formulaton are as follows: Max mze : Z + W + W 2 2 n y = 1 = W1 n = 1 β n ( max ( D, ) n n ( max ( D, ) + ( max ( D, )) ) = 1 N = 1 N = 1 N k IS n ( max ( U, ) n n ( max ( U, ) + ( ( max ( U, ) ) = 1 N = 1 N = 1 N k IS k k (Coverage) (1a) (Downlnk SIR) (1b) (Uplnk SIR) (1c) Budget ( C t) + W3 (Cost) (1d) Budget Subect to: y TA 2 πr n =1 C x = N t t x N t Budget y t ; where N = { d r FG } where TA N 2 2r (2) (3) (4) (5) 3 W = 1 = 1 (6) { max( D )} 0 = (7) D (max) y 4 watts ; D (max), 0 ID (max) y DN D (max) ; ID (max) = N {( D (max)) } k (8) 7

8 { max( U )} 0 = (9) U (max) y 4 watts ; U (max), 0 IU (max) y UN U (max) ; IU (max) = N ( 0,1) ; = { 1, 2, n } ( 0,1) ; = { 1,2, n } x..., y..., {( U (max)) } As equaton 1 shows, the CAC mathematcal formulaton s a maxmzaton problem that conssts of three obectve functons combned usng a weghted sum or lnear combnaton wth specfc weght values. All obectves are normalzed; therefore, ther values range from zero to one. Equaton 1a represents the frst obectve, whch s to maxmze the area of the faclty covered by the nstalled RFID reader network. The bnary decson varable y s calculated as the product of three terms,.e., y = τd τu β. The term τ D ndcates whether or not the downlnk sgnal transmsson from RFID reader to grd s successful. If the maxmum sgnal power level receved at grd s greater than or equal to the downlnk power threshold (.e., 29 mcrowatts or dbm), the communcaton lnk s establshed successfully and the value of τ D s set to one. Otherwse, t s equal to zero. The term τ U ndcates whether or not the uplnk sgnal transmsson from grd to a RFID reader located at grd s successful. If the maxmum sgnal power level receved at a desred RFID reader located at grd s greater than or equal to the uplnk power threshold (e.g., nanowatts or -95 dbm), the communcaton lnk s establshed successfully and the value of τ U s set to one. Otherwse, t s equal to zero. Fnally, the term k (10) β ndcates whether or not the RFID reader can be placed at grd. If grd s not a forbdden grd, the value of β s equal to one; otherwse, t s set to zero. Therefore, a partcular grd s consdered as covered by a RFID reader f and only f the decson varables τ D, τ U, and β are all equal to one. The second obectve, represented by equaton 1b and equaton 1c, s to mnmze nterference. Snce the term sgnal-to-nterference rato (SIR) s used as the measure of nterference, mnmzng nterference means maxmzng SIR. Equaton 1b reflects the SIR n the downlnk channel, whereas equaton 1c quantfes the SIR n the uplnk channel. The total SIR s then calculated as the summaton of the downlnk SIR and the uplnk SIR. The thrd obectve s to mnmze the mplementaton cost of the RFID network. However, the overall obectve functon (or ftness functon) seeks a maxmzaton obectve. To transform the obectve of mplementaton cost nto a maxmzaton obectve, the rato of the budget mnus the cost of the nstalled RFID readers dvded by the budget can be used to calculate the cost of mplementaton, as shown n equaton 1d. 8

9 Constrant (2) ensures that f x 1, at least one RFID reader n the set N covers grd ( = 1) ( y = 0) N y. On the other hand, f N = φ (.e., x = 0 ), grd wll not be covered. In addton, the number of RFID readers that cover grd must not be greater than the number of nstalled RFID readers n the system (.e., x y t ). The lower bound and the upper bound for the total number of RFID readers nstalled s defned by constrant (3). Ths constrant allows the PSO and GA algorthms to search for the optmum number of RFID readers requred n the system that would result n the best value of ftness. Constrant (4) mples that the number of grds where a RFID reader s nstalled must be equal to the total number of RFID readers nstalled. Constrant (5) specfes that the mplementaton cost of the RFID reader network must be less than or equal to the avalable budget. Constrant (6) specfes that the lnear combnaton of weghts must not exceed one. Constrant (7) ensures that f grd s covered (.e., y = 1), the maxmum sgnal power levels receved at grd from the RFID reader located at grd must not be greater than the 4-watt EIRP lmtaton set by the Federal Communcaton Commsson (FCC). In constrant (8), the set DN N conssts of the sgnal power levels receved at a partcular grd from the RFID readers located at the grd(s) N for the downlnk communcatons channel. Only one sgnal power element n ths set s consdered as the maxmum sgnal power level (.e., D (max) ), whereas the other sgnal power elements (f any) are consdered as nterference. Therefore, the maxmum value of sgnal nterference power affectng a partcular grd can be calculated by summng all elements n the set DN and then subtractng the maxmum sgnal power level D (max). Thus, constrant (8) specfes that f grd s covered (.e., y = 1), the level of nterference affectng ths grd can vary between zero and DN D (max). Constrants (9) and (10) mplement smlar concepts as that of constrants (7) and (8) but appled to the uplnk communcatons channel. In ths case, the maxmum sgnal power level and the sgnal power levels receved at a partcular grd from the RFID readers located at the grd(s) N are represented as U (max) and UN, respectvely. N N N 4 Heurstc Optmzaton Algorthms Fndng solutons to the CAC mathematcal formulaton s an NP-hard problem. Therefore, an automated tool was developed n MATLAB to mplement two heurstc optmzaton algorthms based on PSO and GA to fnd good solutons wthn a reasonable 9

10 computaton tme. Both PSO and GA have been appled prevously to the RNP problem and have produced good solutons wthn reasonable computaton tme. The PSO algorthm was developed as an nerta-weghted PSO algorthm [12]. Three maor parameters must be set approprately n an nerta-weghted PSO: nerta ( w), the local cogntve component ( c 1 ), and the global socal component ( c 2 ). If the values of w, c 1, and c 2 are set correctly, the partcles n the nerta-weghted PSO can converge very well. Fgure 1 shows the steps of the PSO algorthm mplemented n ths research. START Set nput parameters followng EPC global standard and propagaton models. Calculate values for r, a, and b Set level for factors faclty shape, faclty sze, grd sze, and power levels Intalzaton: # partcles, # RFID readers, length of a partcle, stoppng crtera, ntal values of gbest andpbest Generate ntal solutons wth random postons and veloctes = 0 Determne the modfed antenna coverage area and evaluate the ftness functon pbest = ftness YES Ftness > pbest? NO pxbest = current poston gbest = ftness YES Ftness > gbest? next teraton = +1 gxbest= partcle s array ndex and value NO Adust the postons and veloctes of partcles v(,)=w*v(,)+c1*rand*(gxbest()-x(,) +c2*rand*(pxbest-x(,) x(,)=x(,)+v(,) : round up to nearest nteger x > LV? NO YES x = LV x < 1? NO YES x = 1 Stoppng crtera met? YES NO Dsplay results and plot graphs STOP Fgure 1: Steps of the nerta-weghted PSO algorthm. 10

11 The requred steps to apply the GA to the CAC mathematcal formulaton are summarzed n the flowchart depcted n Fgure 2. Fgure 2: Steps of the GA algorthm. 11

12 5 Expermental Desgn A 2 k full factoral expermental desgn was utlzed to evaluate the effect that dfferent factors have on the effectveness of the developed PSO and GA algorthms to fnd the optmal number and locaton of RFID readers n a warehouse faclty. The PSO and GA algorthms were mplemented usng MATLAB verson The effectveness of the PSO and GA algorthms was measured by means of a value of ftness (.e., the expermental response). Fve factors (.e., k = 5) were nvestgated n ths experment. Each factor was tested at two levels (.e., hgh and low). The expermental factors and ther levels are summarzed n Table 1. Table 1: Expermental controlled factors n the expermental desgn. Factors Level of the Factor Hgh (+) Low (-) A. Faclty Shape Non-rectangular shape Rectangular shape Inverted-T (J's [13]) (Tang's [14]) B. Faclty Sze Large Small C. Grd Sze 1 meter 0.25 meter D. Power Levels RFID readers have RFID readers have the dfferent power levels same power levels E. Optmzaton Algorthm PSO GA The hgh level for the man factor faclty sze used the orgnal szes for the facltes obtaned from [13] (.e., 80 meters by 60 meters) and [14] (.e., 75 meters by 30 meters). The small faclty szes (.e., low level) were obtaned by scalng down the orgnal faclty sze whle mantanng the faclty shape. The sze of the small facltes was approxmately one-nnth of that of the large facltes. A total of 32 treatment combnatons were created usng the dfferent levels of the expermental factors. Three replcatons were performed for each treatment combnaton. Therefore, a total of 96 expermental runs were conducted n MATLAB. 6 Results and Dscusson 6.1 Results of the Statstcal Analyses based on Fve RFID Network Desgn Factors A mult-factor ANOVA was used to determne whch man factors (.e., the fve RFID network desgn factors) and nteractons had an effect on the qualty of the ftness. The man factors and nteractons were consdered to be statstcally sgnfcant f ther p- 12

13 value was less than The mult-factor ANOVA table for the value of ftness based on fve RFID network desgn factors s shown n Table 2. Table 2: Mult-factor ANOVA results based on fve RFID network desgn factors. Source Sum of Squares Df Mean Square F-Rato P-Value MAIN EFFECTS A: Faclty shape B: Faclty sze C: Grd sze D: Power levels E: Algorthm INTERACTIONS AB AC AD AE BC BD BE CD CE DE RESIDUAL TOTAL (CORRECTED) Table 2 shows that the RFID network desgn factors faclty shape, faclty sze, power levels, and optmzaton algorthm have a statstcally sgnfcant effect on the value of ftness (p-value < 0.05). The only RFID network desgn factor that does not have a statstcally sgnfcant effect on the value of ftness s grd sze. Fgure 3 depcts Fsher s least sgnfcant dfference (LSD) nterval plots (at a 95% confdence level) of all RFID network desgn factors to llustrate how they nfluence the performance of the system n terms of the value of ftness. 13

14 (a) (b) (c) (d) (e) Fgure 3: LSD nterval plot of the RFID network desgn factor: (a) faclty shape, (b) faclty sze, (c) grd sze, (d) power levels, and (e) optmzaton algorthm. 15

15 6.2 Results of the Statstcal Analyses for Fxed Faclty Shape and Fxed Faclty Sze Further analyses were performed to gan a better understandng of the effect of a subset of the RFID network desgn factors f the faclty shape, faclty sze, and locaton of obstructons were known. Thus, only the RFID network desgn factors grd sze, power levels, and optmzaton algorthms were entered as the nput for the mult-factor ANOVA and the Fsher s LSD nterval plots. For a specfc faclty shape and faclty sze, the best confguratons for each layout and the resultng value of ftness are summarzed n Table 3. Table 3: The best confguraton for a specfc faclty shape and faclty sze and ts ftness. Best Confguraton Small Rectangular Faclty Small Inverted-T Faclty Large Rectangular Faclty Large Inverted-T Faclty Grd sze or 1 1 Power levels Optmzaton PSO or GA PSO or GA GA GA algorthm Value of Ftness Coverage SIR Cost As Table 3 shows, the confguraton that resulted n the hghest value of ftness nvolved a faclty wth a rectangular shape (Tang's faclty [14]), a small faclty sze (10 meters by 25 meters), a small grd sze (0.25 meter), a sngle power level of RFID readers, and usng ether the PSO or the GA. For llustraton purposes, Fgure 4 depcts the graphcal results that were obtaned for the small rectangular faclty wth the analyss tool developed n MATLAB. The black dots n Fgure 4 represent RFID tags that were assumed to be placed n the small rectangular faclty at the center of all canddate grds, whereas the larger black dots at the center of the coverage areas ndcate the optmum locatons for the RFID readers, as determned by the PSO and the GA algorthms. In ths case, 116 out of 131 RFID tags were covered whch resulted n a value for the coverage obectve of The obectve of SIR (.e., the summaton of downlnk SIR and uplnk SIR) and the obectve of cost were and 0.40, respectvely. The computaton tme of the PSO algorthm was seconds, whereas the computaton tme of the GA was seconds. 16

16 Fgure 4 clearly shows that three RFID readers cannot provde entre coverage for the small rectangular faclty. Therefore, the optmum locaton and optmum number of RFID readers needed to cover the entre small rectangular faclty (.e., coverage of the center of all canddate grds) were further nvestgated. The number of requred RFID readers must be n the range between three to fve, whch s derved from constrant (3) of the CAC mathematcal formulaton. The mprovement n coverage can ncrease the value of ftness. However, the cost obectve decreases as the number of RFID readers ncrease and the nterference tends to ncrease as well wth more RFID readers because of a denser envronment. Therefore, a confguraton wth complete coverage does not always guarantee the hghest value of ftness. Fgure 4: The placement of RFID readers n a small rectangular faclty (10 meters by 25 meters) usng the best confguraton of RFID network desgn factors. Fgure 5 depcts an alternatve placement of RFID readers n the small rectangular faclty. In ths partcular case, fve RFID readers was the optmum number to ensure entre coverage, whch resulted n a value of ftness of The value for coverage was one; for SIR; and zero for cost. The zero cost means that no savngs were realzed wth ths RFID system snce the maxmum number of RFID readers allowed (.e., fve) were used n ths scenaro. 17

17 Fgure 5: The placement of RFID readers n a small rectangular faclty (10 meters by 25 meters) usng the best confguraton of RFID network desgn factors wth entre coverage. 7 Conclusons The methodology developed to fnd the optmal locaton and number of RFID readers requred n a warehouse faclty to support nventory management worked very well wth small rectangular facltes and small nverted-t facltes. The PSO and GA algorthms were able to obtan better values for the ftness functon (.e., maxmze the RF coverage, mnmze the level of nterference and mnmze the mplementaton cost) for a faclty wth a rectangular shape, a small sze and a sngle power level of RFID readers. In general, the GA algorthm produced better values of ftness than the PSO algorthm. Ths mght be due to the fact that the problem n ths research s a dscrete problem. Roundng the updated value of poston (.e., x) n the PSO algorthm to the nearest nteger was the approach used to make t a dscrete PSO. A more sophstcated dscrete bnary PSO may be needed to obtan better (or comparable) results than those obtaned wth the GA. The PSO and GA algorthms (but especally the PSO) dd not work very well wth the large faclty sze scenaros used n ths research. Some parameters of the algorthms, the weghts of each obectve, and the termnaton crtera wll be further nvestgated to mprove on these results. 18

18 References [1] RNCOS Industry Research Solutons., Global RFID Market Forecast to Retreved from (2012). [2] Angeles, R., RFID Technologes: Supply-Chan Applcatons and Implementaton Issues, Informaton Systems Management, 22, (2005). [3] Chande, A., Dhekane, S., Hemachandra, N., and Rangara, N., Pershable nventory management and dynamc prcng usng RFID technology, Sadhana, 30, 2-3, (2005). [4] Jahao, W., Zongwe, L., and Wong, E. C., RFID-enabled trackng n flexble assembly lne, Internatonal Journal Of Advanced Manufacturng Technology, 46,1-4, (2010). [5] Ernst, and Young. Ernst and Young Study Estmates Retalers Lose $46 Bllon Annually to Inventory Shrnkage; Employee Theft Is Bggest Problem, Busness Wre, (2003). [6] Gampaolo, E. D., Forn, F., and Marrocco, G., "RFID-Network Plannng by Partcle Swarm Optmzaton," Proceedngs of the Fourth European Conference on Antennas and Propagaton, Aprl 12-16, Barcelona, Span, 1-5 (2010). [7] Gupta, N., and Iyer, S., RFIDPlanner - A Coverage Plannng Tool for RFID Networks. Asan Intl Moble Computng Conference (AMOC 2007), 1-8. Calcutta, Inda (2007). [8] Bhattacharya, I., & Roy, U. K., Optmal Placement of Readers n an RFID Network Usng Partcle Swarm Optmzaton, Internatonal Journal of Computer Networks & Communcatons (IJCNC), 2, 6, (2010). [9] Reza, A. W., & Geok, T. K., Investgaton of Indoor Locaton Sensng va RFID Reader Network Utlzng Grd Coverng Algorthm, Wreless Personal Communcatons, 49, 1, (2009). [10] Guan, Q., Lu, Y., Yang, Y., and Yu, W., Genetc Approach for Network Plannng n the RFID Systems, Sxth Internatonal Conference on Intellgent Systems Desgn and Applcatons (ISDA 2006), (2006). [11] Chen, H., Zhu, Y., Hu, K., and Ku, T., RFID Network Plannng usng a Mult- Swarm Optmzer, Journal of Network and Computer Applcatons, 34, 3, (2011). [12] Clerc, M., and Kennedy, J., "The partcle swarm - exploson, stablty, and convergence n a mult-dmensonal complex space," IEEE Transactons on Evolutonary Computaton, 6, 1, (2002). [13] J, Z., Sarkar, T. K., and L, B.H., "Methods for optmzng the locaton of base statons for ndoor wreless communcatons," IEEE Transactons on Antennas and Propagaton, 10, (2002). [14] Tang, K. S., Man, K. F., and Ko, K. T., Wreless LAN Desgn usng Herarchcal Genetc Algorthm. Proceedngs of the Seventh Internatonal Conference on Genetc Algorthms, (1997). 19