152 Andrew Kusiok TI. Matrix Formulation In the matrix formulation, a machine-part incidence matrix (a, ] is constructed. ‘The machine-part incidence matrix (a,,] consists of 0,1 entries, where an entry 1 (0) indicates that machine i is used (not used) to process part j. Typically, ‘when an initial machine-part incidence matrix (a,] is constructed, clusters of ‘machines and parts are not visible. Clustering algorithms allows one to transform ‘an initial incidence matrix into a more structured (Possibly block diagonal) form. ‘To illustrate the clustering concept, consider the machine-part incidence Matrix w, par number L234s | roar Y ale smscine s)he | amber ait arranging rows and columns in Matrix (1) results in Matrix (2) PRL PRD La24s ® Mei 4 1 crt wer Lia | Two machine cells (clusters) MC-1 = {2, 4}, MC-2 = {1, 3}, and two comre- sponding part families PF-1 {1, 3}, PF-2 = {2, 4, 5) are visible in Matrix (2) Clustering of a binary incidence matrix may result in the following two categories of clusters: (1) mutually separable and (2) partially separable. The two categories of clusters are presented in Matrices (2) and (3) fe2 tae) enema) 8 ' Mea Mea iat ii} Matrix (3) cannot be separated into two disjoint clusters because of part 5, which is to be machined in two cells MC-1 and MC-2. Removing part 5 from Matrix (3) result in the decomposition of Matrix (3) into two separable machine cells MC-1 = {I, 2}, MC-2 = (3, 4} and two part families PF-1 = {1,2} and6. Group Technology in Flexible Manufacturing Systems 153 PE-2 = {3, 4}. The two clusters are called partially separable clusters. To deal with the bottleneck part 5, one of the following three actions can be taken: 1. It can be machined in one machine cell and transferred to the other ‘machine cell by a material handing carrier. 2. It can be machined in a functional facility. 3. It can be subcontracted. To solve the matrix formulation of the group technology problem the following approaches have been developed: + Similarity coefficient methods * Sorting based algorithms * Bond energy algorithm * Cost-based method * Cluster identification algorithm. + Extended cluster identification algorithm A, SIMILARITY COEFFICIENT METHODS In this section, three similarity coefficient methods are discussed. The single linkage cluster analysis (SLCA) presented by McAuley (1972) is based ‘on the similarity coefficient s;, measure between two machines é and j and is computed as follows: 2, 8: (aay a) 2, Bs (aus ay) where Oif ay = a, = 1 To solve the GT problem using the SLCA approach, similarity coefficients, for all possible pairs of machines are computed. Machine cells are generated based on a threshold value of the similarity coefficient ‘To illustrate the SLCA, consider Matrix (3). The similarity coefficients 5, are computed below and depicted in Figure 5. aia susp 237078 5707 2e1154 Andrew Kusiok a t= 0.25 ais 0 sy = ou 5-0 ‘Assuming the threshold valve of the similarity coefficient s,, = 60%, from Figure 5 the following machine cells are obtained: MC-1 = {1, 2} and MC-2 = (3, 4 One of the disadvantages of the SLCA is that it fails to recognize the chaining problem resulted from the duplication of bottleneck machines. A bot- tleneck machine is one that does not allow one to decompose a machine-part incidence matrix into submatrices. For example, machine 3 in Matrix (4) does rot permit the decomposition of that matrix into two machine cells and two part families, par 123456 it ha “ ‘A way to decompose Matrix (4) into two mutually separable submatrices is to purchase an additional copy of machine 3. The latter allows one to transform Matrix (4) into Matrix (5), PEL PRD T2sase ot weap] tt © SL 11 3 a wea tid 5 en ‘Twomachine cells MC-1 = {1,2,3},MC-2 = {3,4, 5} and two corresponding part families PE-1 = { 1, 2}, PF-2 = {3, 4, 5, 6} are visible in Matrix (5) Analogously to the definition of a bottleneck machine, a bottleneck part can be defined. Use of the average linkage clustering (ALC) algorithm to overcome the chaining problem was studied by Seifoddini and Wolfe (1986). They defined the similarity coefficient between any two clusters as an average of the similarity coefficient between all elements of the two clusters. To solve the group tech nology problem, Seifoddini and Wolfe represented the machine-part incidence6. Group Technology in Flexible Manufacuring Systems 155 o 25 50 18 Simi ta 1 Machine Number Figure 5. A tre of similanty coefficient ‘matrix using a binary machine code. The total number of intercellular movements (ICM) between two machine cells MC-i and MC-j is computed as follows: Tem, = 3, 8 (as vs) lifla, #0 Btw = {ddan Note that 8: (vi4, vx) = 1 indicates that part k requires processing in both machine cells MC-i and MC, and thus part k isa bottleneck part. This problem can be solved either by removing part & from the two machine cells or adding an identical machine, say g, to each machine cell. The average linking clustering algorithm is illustrated below. Consider the example illustrated in Figure 3 where two clusters: were obtained. The intercellular movement is computed as follows: [wud = 11,1, 0, 0, 11 bal = 10,011, 11 ICM: = 3 8s (uy vo) OF 04040411 ‘The only part which requires processing in both machine cells MC-1 and MC- is part 5 with corresponding 8 (vis, v5) = 1 ‘Another interesting similarity coefficient method was studied by De Witte156 Andrew Kusiok (1980). He designed a clustering algorithm advocating the concept of some ‘machines to be included in more than one machine cell. He divided all the available machines into (1) primary machines, (2) secondary machines, and (3) tertiary machines. To analyze the relationship between these machines three 24 23 22 21-29 decimal weight equivalent ete 1 202 machine io 3 number Ot Step 2. Sorting the decimal weights in decreasing order results in the following matrix: pan umber Pe es ro 2 rod 4 machine 11a) dumber roi da Step 3: Repeating the above steps for each column produces the following matrix: i a In the above matrix two separable clusters are visible. The matrix obtained is identical to Matrix (2). ‘The ROC algorithm was further extended in King and Nakornchai (1982) and Chandrasekharan and Rajagopalan (1986). In the latter paper, Chandrasek- hharan and Rajagopalan improved the ROC algorithm incorporating “block and slice” method and the hierarchical clustering method. Another interesting sorting-based algorithm, the direct cluster algorithm (DCA), was developed by Chan and Milner (1982). The DCA consists of the following steps: Step 1. Determine the total number of 1s in each row and column in the rmachine-part incidence matrix. Step 2. Sort each row in increasing order corresponding to the total number of Is, ‘Step 3. Sort each column in decreasing order corresponding to the total number of Is.158 Andrew Kusiak Step 4, Repeat the above steps until the position of each element in each row and column does not change. Based on the final outcome of the above steps, some rows and columns in the ‘machine-part matrix are to be rearranged until satisfactory clusters are obtained, C, BOND-ENERGY ALGORITHM McCormick ef al. (1972) developed an interchange clustering algorithm called the bond-energy algorithm (BEA). The BEA secks to form a block diagonal form by maximizing the measure of effectiveness which is defined as follows: ME = 12 3 3 aylayes + ayer + ary + dienyl ‘The ME is illustrated below. omer are) 1 yit racine 2 mactine 2/) | mimber 3 mime 3} 11a i att ME BEA Algorithm Step 1. Set j = 1. Select one of the columns arbitrarily Step 2. Place each of the remaining n ~ j columns, one at a time, in each j + 1 position, and compute each column's contribution to the ME. Place the column that gives the largest incremental contribution to the ME in its best location. Increase j by 1 and repeat the above steps until j = 7. Step 3. When all the columns have been placed, repeat the procedure for the rows ‘The bond-energy algorithm applied to Matrix (1) is illustrated in Example 2 Example 2 Step I, Setj = 1. Select column 2. ‘Step 2. Placing each of the remaining columns in each of the j + 1 Positions, its contribution to the ME value is computed below:6, Group Technology in Flexible Manufacturing Stems 159 Column 4 is placed in the j + 1 position. Repeating the same procedure for the remaining columns leads to the following col- umn order: {2, 4, 5, 1, 3). Step 3. Repeating the above steps for rows results in the following matrix: part umber Pe Ts machine number La ‘The value of ME for the above matrix is 9 Slagle ef al. (1975) developed a clustering algorithm based on the BEA and the shortest spanning path (SPS) algorithm. Their concept was then extended jn Bhat and Haupt (1976). They developed an algorithm where the matching between any two rows (columns) of the incidence matrix is calculated as follows: mo = 3, low ~ al The Bhat and Haupt (1976) algorithm is similar to the McCormick etal (1972) algorithm. The difference between the two algorithms is that the Bhat and Haupt algorithm permutes rows and columns of matrix A ~ AT as opposed to permuting rows and calculating matchings in matrix A in the McCormick er al. algorithm, One can note that an entry m, of matrix (m} represents a distance between rows i and jin matrix A, D. COST-BASED METHOD Askin and Subramanian (1987) developed a clustering algorithm which considers the following manufacturing costs: 1. Fixed and variable machining cost 2. Setup cost 3. Ptoduction cycle inventory cost160 Andrew Kusiak 4. Work-in-process inventory cost 5. Material handling cost The algorithm consists of three stages. In the first stage, parts are classified using a coding system. In stage two, an attempt is made to develop a feasible ‘grouping of parts based on the manufacturing costs. In stage three, the actual layout of machine cells is analyzed. Kusiak and Chow (1987a) applied the concept presented in Iri (1968) to develop the cluster identification (Cl) algorithm. The cluster identification al- {gorithm allows one to check the existence of mutually separable clusters in @ binary machine-part incidence matrix provided that they exist, CI Algorithm Step 0. Set iteration number k= 1 Step 1. Select any row # of incidence mateix A“ and draw horizontal line fy through it (A'® reads: matrix A at iteration &). Step 2. For each entry “I” crossed by the horizontal line h, draw a vertical line v, Step 3. For each entry horizontal line hy. crossed once by a vertical line v, draw a ‘Step 4. Repeat steps 2 and 3 until there are no more crossed once entries “1” in AW. All crossed twice entries “I” in A“ form machine cell MC-k and part family PF-k. Step 5. Transform the incidence matrix A” into A'*” by removing rows and columns corresponding to all the horizontal and vertical lines drawn in steps 1 to 4. Step 6. If matrix A” = 0 ( where 0 denotes a matrix with all elements equal to zero ), stop; otherwise set k = k + I and go to step 1 Note that the cluster identification algorithm scans each element of matrix A two times. Since there are mn elements in matrix A, its time computational complexity is O(n), that is, the number of elementary computer operations is of the order of mn. ‘Application of the cluster identification algorithm for solving the group technology problem is illustrated in Example 36. Group Technology in Flexible Manufacturing Systems 161 Example 3 Consider the following machine-part incidence Matrix (6). par umber 12345678 oan 4 ay 1 © 5 1 ' as a1 1 machine 5| tou 1 umber @| 1 doviea 1 Step 0. Ser iteration number k = 1 Step 2. Row 1 of Matrix (6) is selected and horizontal line hy is drawn. ‘The result of step | and step 2 is presented in Matrix (7), Step 2. Three vertical lines v2, v5, and vs are drawn. Ate 1 o Asaresult of drawing these three vertical lines, five new crossed- conce entries “1” are created in Matrix (7), that is, entries (5, 3), (5, 5), and entries (7, 2), (7, 3), and (7, 5) Step 3. Two horizontal lines hs and fry are drawn through the crossed- once entries of Matrix (7), as shown in Matrix (8). 12345678 Ae 1 ®)162 Step 4. Step 5. Andrew Kush Since the entries (5, 8) and (7, 8) of Matrix (8) are crossed once, the vertical Tine vy is drawn as shown in Matrix (9) 12345678 : ® if-ns Ade i ~~ -1} my Since there are no more crossed-once entries“, all the crossed: twice entries “I” of Matrix (9) form machine cell MC-1 5, 7} and part family PF-1 = {2, 3, 5, 8} Matrix (9) is transformed into Matrix (10). 1467 Fiata (10) Are (a\i ols ot In the second iteration (k = 2), steps I to 4 are performed on Matrix (10). This iteration results in incidence Matrix (11) and ‘machine cell MC-2 = {2, 4} and part family PF-2 = {1,6} Lae? 1 i i ay : In the third iteration (K = 3) Matrix (12) is generated AG 3 piety wo. ofits «2 From this matrix, MC-3 = {3, 6} and PF-3 = {4, 7} are obtained. “The final clustering results illustrated in Matrix (13). The layout corresponding to Matrix (13) is illustrated in Figure 6.6. Group Technology n Flexible Mansfactring Systems 163 MANUFACTURING SYSTEM ‘MACHINE CELL MC-1 Machines :1,5.7 Pars © 2.3.5.8 MACHINE CELL, ‘MACHINE CELL MC2 MCS Machines : 2,4 Machines + 3,6 Paws 147 Figure 6. A layout of the manuactaring sytem from Example 2, y wer { ¥ aa 3 wea {3 mes { 3 ‘The computational experience showed that the cluster identification algorithm is very efficient. To show the efficiency of this algorithm, consider the clustering of a matrix man = 60 X 80. It takes only 0.06 s to identify clusters by the developed algorithm. To cluster the same matrix by Bhat and Haupt (1976) algorithm, more than 300 s are required. F, EXTENDED CLUSTER IDENTIFICATION ALGORITHM Consider the following machine-part incidence Matrix (14) part number «aay machine umberAndrew Kasih Using any of the existing clustering algorithms, Matrix (14) can be trans- formed into a more structured Matrix (15), part number Saisie as) art MCL ay), ‘machine my omer wer ST LIE Matrix (15) cannot be separated into two disjoint submatrices because of part 3, which is to be machined in two machine cells MC-1 and MC-2 Removing part 3 from Matrix (15) results in the decomposition of Matix (15) into two mutually separable machine cells MC-1 {1, 3} and two part families PF-1 = {5, 2} and PF-2 2, 4}, MC-2 1, 4) ‘The extended cluster identification algorithm is developed to deal with overlap- ping parts. It takes advantage of both the expertise of a production manager regarding possible grouping of machines and parts and also the efficiency of the cluster identification algorithm, Extended CI Algorithm Step 0. Step 1 Step 2. Step 3 Set iteration number k = 1 Select those machines (rows of matrix A'!) which, based on the user's expertise, are potential candidates for inclusion in machine cell MC-k, Draw horizontal lines h, through each row of matrix ‘AM corresponding to these machines. In the absence of the user's expertise any machine can be selected. For each column in A”? corresponding to the entry “1” crossed by any of the horizontal lines A, draw a vertical line ¥, For each row in A corresponding to the entry “I” crossed by the Vertical ine v,, drawn in step 2, draw a horizontal line A, Based on the machines corresponding to all the horizontal lines drawn in step 1 and step 3, a temporary machine cell MC-k is formed. Ifthe production manager's expertise indicates that some of the machines cannot be included in the temporary machine cell MC-k’, erase the coresponding horizontal lines in matrix ‘A, Removal of these horizontal lines results in machine cell MC-k. Delete from matrix A parts (columns) which are to be manufactured on at least one of the machines already included in MC-K. Place these parts on the list of parts to be manufactured6, Grow Technology n Fleuble Manufactaring Systems 165 Step 4. Step Step 6. ina functional machining facility. Draw a vertical line v, through cach crossed once entry “I” in A® which does not involve any ‘ther machines than those included in MC-k. For all the crossed twice entries “I” in A®, form a machine cell MC-K and a part family PF, ‘Transform the incidence matrix A® into A“*"” by removing all the rows and columns corresponding to the rows and columns included in MC-K and PF-k, respectively If matrix A“*"” = 0 (where O denotes a matrix with all elements ‘equal 10 zer0), stop; otherwise set k = k + 1 and go to step 1. {As can be seen, computational complexity ofthe extended cluster identification algorithm is slightly higher than the cluster identification algorithm, mainly due to step 3. In the case when the clustered matrix has the block diagonal structure, ‘computational time complexity of the extended cluster identification algorithm reduces to O(mn), that is, equal to the complexity of the cluster identification algorithm, ‘The extended cluster identification algorithm is illustrated in Example 4 Example 4 Given the machine-part incidence, Matrix (16) determines mutually sep- arable machine cells and part families part number 123436789 01 ' 1 1 16) 1 machine umber ‘The user's expertise indicates that machines 1 and 4 should be included in the machine cell MC-1 and machine 3 and 5 should be included in the machine ‘example. Step 0. Step 1 cell MC-2. Some other expertise required is indicated later in the Set iteration number k = 1 ‘Since the user's expertise indicates that machines 1 and 4 should bee included in machine cell MC-1, two horizontal lines A, and ‘hg are drawn, as shown in Matrix (17),Andrew Kuaiah 123456789 00 Ife cesseee F 5 a an AO» aha peered ly 5 1 1 aa 1 ria 1 Ate in ew Step 2. For columns 1, 2, 3, 6, and 7 crossed by the horizontal fines fy and hg in Matrix (17), five vertical lines v1, v2, vs, ¥oy and vp are drawn as shown in Matrix (18), 123456789 0 sesee Jeo 1 1 hy (18) 1 Wi ee Step 3. Three horizontal Lines, hs, he, and ltr, are drawn through rows: 2, 5, and 7, coresponding to the crossed-once elements “I” of Matrix (18). The result of this operation is shown in Matrix (19), ety Gs 6) 78 011 ag) Based on all the crossed-twice elements of matrix (19) a tem- porary machine cell MC’-1 which ineludes machines {1, 2. 4, 6, 7} is formed. This temporary machine cells evaluated by the user. Assume thatthe user's expertise indicates that machines 2 and 6 cannot be included in MC'=1. Since machines 2 and 6 are excluded, the horizontal lines fi and fh, ate ereased from Matrix (19). This leads to the deletion of columns (parts) 1. 4,{6 Group Technology in Flexible Manufacturing Systems 167 Step 4. Step 5 Step 6. and 9 from Matrix (19). Note that each of the three parts 1, 4, and 9 is to be manufactured on at least one of the machines 2 and 6 lines fy and fh, in Matrix (19) and at least one of the machines 1, 4, and 7 which form machine cell MC-1 (see Matrix (20) and step 4), ‘The three parts 1, 4, and 9 are placed on the list of parts to bbe manufactured in the functional machining facility. Since there are no more crossed-once entries “1,” no vertical line is drawn, As a result of the above process, Matrix (20) is obtained. 235678 101 1 2 3 aia foe 20) 5 é 5 ‘The crossed-twice entries "I" of Matrix (20) indicate machine cell MC-1 = { 1, 4, 7 } and part family PF-1 = {2, 3, 6,7}. Matrix (20) is transformed into Matrix (21) tod en Set k = k + 1 = 2 and go to step 1. The second iteration (k 2) resulted in machine cell MC-2 = {2, 3, 5, 6 } and part family PF-2 = {5, 8, 10, 11 }, as shown in Matrix (22). wn ina | 5 @) ioe168 Andrew Kush ‘The final matrix result generated by the extended cluster iden- tification algorithm is presented in Matrix (23), PF Pra 236758 wnl49 iw mera] 1 1 1 jyui4 rap 2| T 7] 1 mc 3] 1 3 ra 6 estees pera Machine cell MC-1 and MC-2 and corresponding part families PF-1 and PF-2 are visible in Matrix (23), The corresponding layout is illustrated in Figure 7. ‘The functional machining facility in Figure 7 should involve a set of ‘machines which are able to perform operations on parts 1, 4, and 9, However, these machines do not have to be identical to machines 4, 7, 2, and 6 [see Matrix (23)}, initially designated for machining these three parts. Apart from the ap- proach for manufacturing parts 1, 4, and 9 which is illustrated in Figure 7, three other alternative approaches are possible: 1. Process plans for parts 1, 4, and 9 might be modified, so that each of these parts could be machined in one of the two existing machine cells MC-1 or MC-2. MANUFACTURING SYSTEM MACHINE CELL MACHINE CELL MC Mc? Machines + 1.4.7 Machines © 2,3,5.6 Pane B67 Pans 9 58,10, 17 FUNCTIONAL MACHINING FACILITY 149 Figure 7. Layout of the manufacturing system fom Examele 3.6, Grow Technology in Flexible Manufacturing Systems 169 2. Design of parts 1, 4, and 9 might be modified, so that the resulting process plans could fit the existing machine cell MC-1 or MC-2. 3. Pants 1, 4, and 9 might be manufactured in the two existing machine cells MC-1 and MC-2 without any changes in the design or process planning. This approach is applicable as long as the flow of parts among, different machine cells (in the above case MC-1 and MC-2) is relatively low, Finally, if user is not satisfied with the machine cells and part families generated, hhe can repeat the entire computational process. Initiating the extended cluster identification algorithm with another machine(s) in step | may result in a different configuration of groups of machines and parts ‘Another extension of the cluster identification algorithm named the cost analysis algorithm is presented in Kusiak and Chow (1987b). The cost analysis algorithm allows one to solve the group technology problem with known subcontracting costs. The overlapping parts are removed from the incidence matrix in such a way that the total sum of subcontracting costs is minimized II. Mathematical Programming Formulation Most mathematical programming models developed in GT consider a distance ‘measure dj, between parts { and j. The distance measure di is a real valued symmetric function obeying the following axioms: + Reflexivity d, = 0 + Symmetry dy = dy, + Triangle inequality dy = dp + dyy “The most commonly used distance measures are the following 1. Minkowski distance measure (Arthanari and Dodge, 1981) 4, = (3, bow ~ anh” where r is a positive integer and n is the number of parts. ‘Two special cases of the above measure are widely used: + Absolute metric measure (for r = 1) + Euclidean metric measure (for r = 2)170 Andrew Kusiak 2, Weighted Minkowski distance measure (Arthanari and Dodge, 1981) ‘There are two special cases: + Weighted absolute metric measure (for + Weighted Euclidean metric measure (for r 3. Hamming distance measure (Lee, 1981) is 3 8 (anon) where af lifes # a 8 (an ay) = { omens Distance measures are also referred to as dissimilarity measures. Some similarity measures were discussed inthe section on similarity coefficient meth- ods. In this section, the following mathematical programming models are dis- cussed: + p-Median mode! + Generalized p-median mode! * Quadratic programming model A. THE p-MEDIAN MODEL The p-median model is used to group parts into p part families. In order to consider the p-median model, the following parameters and a variable are defined (Kusiak, 1985): m= number of machines number of parts p= number of part families diy = distance measure between parts i and j 1 if part # belongs to part family j O otherwise6. Group Technology in Fleuble Manufacturing Systems m ‘The objective function of the p-median model is to minimize the total sum of distances between any two parts / and j min 3 E dy xy (24) subject to Zaxy= 1 forall an 05) Zaye (26) xy Sx forall n j n 7 ay = OL focalie ty. 28) j .n Constraint (25) ensures that cach part belongs to exactly one part family. Con- straint (26) specifies the required number of part families. Constraint (27) ensures that parti belongs to part family j only when this part family is formed. The last constraint (28) ensures integrality. Note that, in the p-median model, the required number of part families, p, is specified intially. The p-median model (24) through (28) is illustrated in Example 5. Example § Given the machine-part incidence matrix (1), form p = 2 part families ‘and their corresponding machine cells. For the machine-part matrix (1), the following matrix of Hamming distances d,, is obtained: part number 12345 o404g 40401 (ay)=3]0 4043] px 4040.1] number 31310 Solving the p-median model (24) to (28) for the above matrix and p provides the solution below.m Andre Kusiok Based on the definition of x, two part families are formed: PF-1 = 3} and PF-2 = {2, 4, 5}. For the two part families, the corresponding two ‘machine cells are determined from Matrix (I): MC-1 = {2, 4} and MC- 2-13) ‘One can easily notice that the solution to the p-median mode! is identical to that solution obtained using the matrix formulation [see Matrix (2)] B. GENERALIZED p-MEDIAN MODEL Model (24) to (28) was developed under the assumption that each part i corresponds only to one set of machine operations, known as a process plan, To relax this constraint, Kusiak (1987) modified model (24) to (28) so that it permits one to consider for each part more than one process plan. In addition, a production cost is associated with each process plan. The objective is to ‘minimize the total sum of distance measures and production costs. To present the generalized p-median model the following notation is used: Fis the set of process plans for part number k, k= 1... ., 1, p the required number of part families, d., the distance measure between process plans / and J. and c, the production cost of process plan j min SD day + Seay 9) subject to DV Ht fork. (0) Dawe GD By ys for all i wn G2) i " =O forall i = 1)... 40 (3) j Constraint (30) ensures that for each part (a set of process plans) exactly one process plan is selected. Constraint (31) imposes the upper bound on a number fof part (process) families. Constraints (32) and (33) correspond to constraints (27) and (28) in the model (24) to (28), respectively. Model (29) to (33) is illustrated in Example 6.6, Group Technology n Fleble Manufacturing Systems m3 Example 6 Given the machine-part incidence Matrix (34) with the corresponding vec tor of production costs (35), form p = 2 part families and corresponding ‘machine cells using model (29) to (33). pan number re ee ee process plan mumber 123 45 67 FS on oo porioad 1 ita t 2 machine i 11 | 3 number mH road ja S tate etterts eee] G5) ‘The Hamming distances d., are shown in Matrix (36). process plan 123456789 wn fo m2 22222037)1 P w2o=22020202342 © woo 2222222143 2 iat a 20) oe e242 ee (lee 242e0202021]5 § 36) i= |222420e0221/6 § 20224 002423/7 tae ala eee a) Oneal a 2022024=e021/9 7 0242222220e/10 4 Bars 1r131teo)un All entries in Matrix (36) equal to a denote that, for each Fi, only ‘one process plan is t0 be selected. Solving the model (29) to (33) for the above data with the LINDO software produces the following solution: 29 = ete = Lome =o ‘This solution can be interpreted as follows: PE-1 = {2,7} and PF-2 59,11}. For these two part families from Matrix (34), two machine cells are obtained: MC-1 = {2,4} and MC-2 = {1,3}174 Andrew Kush ‘The final result is shown in Matrix (37). par par family PR-1 family PE-2 se 1 en C. QUADRATIC PROGRAMMING MODEL Kusiak er al. (1986) developed a formulation of the GT problem that allows fone to deal with the restricted number of clusters and cluster sizes. The for- ‘mulation can be expressed in the form of the following 0-1 quadratic program- ming mode!: = m, for all Pp (40) jah P Constraint (39) ensures that each part belongs fo exactly one part family. Con straint (40) imposes that part family j contains exactly m, parts. Constraint (41) ensures integrity. The model was solved by an eigenvector-based algorithm (Kusiak et al., 1986) IV. Graph Formulation In the graph formulation, the incidence matrix (a,)] is represented by a graph. The following three type of graphs can be used: bipartite, transition, and bound- ay.6. Group Technology in Flexible Manufacturing Systems 15 PARTS. MACHINES: [igure 8. Biparive graph coresponcing to Matix (1) A, BIPARTITE GRAPH In the bipartite graph, one set of nodes represents parts and the other ‘machines. In order to discuss the application of bipartite graphs to GT, consider the graph in Figure 8 representing the incidence matrix (1). It can be seen that the bipartite graph in Figure 8 decomposes into two disjoint bipartite graphs shown in Figure 9. PARTS — MACHINES PARTS MACHINES SUBGRAPH 1 SUBGRAPH 2 Figure 9. Two dicint bipartite graph forthe graph in Figue 8.