You are on page 1of 8

New approaches for the solution of Tandem Configuration Problems in highly automated manufacturing systems

New approaches for the solution of Tandem Configuration Problems in highly automated manufacturing systems
Pierpaolo Caricato*, Doriana Gianfreda, Antonio Grieco
Dipartimento di Ingegneria dellInnovazione Universit degli Studi di Lecce Lecce, 73100, Italy

ABSTRACT The design of loop-based pathways is a problem that is frequently addressed in highly automated manufacturing systems, such as those using a material handling system based on Automated Guided Vehicles (AGVs). We extend the Single Loop Design Problem (SLDP) to the case in which two or more inter-connected unidirectional loops are to be designed. Each loop is connected with at least another loop through a transhipment point. This problem is usually known as Tandem Configuration Problem (TCP). We propose both an exact Mixed Integer Linear Programming (MILP) formulation of the problem and a heuristic algorithm based on Simulated Annealing (SA) including a new exact algorithm for the solution of the sub-problem that is often referred to as SSP (Station Sequencing Problem), where the loops have been fixed and the pickup/delivery stations are to be determined. The proposed approaches have been tested on test data available in the literature. Furthermore, new, larger test cases have been considered in order to thoroughly test the effectiveness of the proposed approaches. The exact MILP model was solved using ILOG CPLEX solver and was only able to find the optimal solution for small sized problems. The SA algorithm was validated on these problems, in order to certify its ability to find good solutions for these problems in a much smaller amount of time. The SA algorithm was also tested on larger problems and, in this case, compared with the sub-optimal solutions achieved using a less effective SLDP exact approach available in the literature. The TCP solved using SA outperformed the SLDP approach in terms of objective function value, thus proving the opportunity to consider tandem configuration as a better design strategy for large sized loop design problems.

1. INTRODUCTION The Tandem Configuration Problem (TCP) is an application of the divide and conquer principle to Automated Guided Vehicles Systems (AGVSs). Given a fixed block layout, a from-to chart representing the flows between departments and the pickup/delivery (P/D) stations, the classical TCP is based on the partitioning of all stations into a set of non-overlapping, inter-connected and unidirectional loops, each served by a single vehicle. Each loop is connected with at least another loop through a transhipment point, which allows loads to be moved between the two adjacent loops. Each department can be served by only one loop; in other words, if a department is served by a loop, it cannot be served by any other loop. The advantages of tandem AGV systems are the ease of control and the reduction in system complexity; indeed no traffic congestion or conflict occurs in a tandem AGV system, since only one vehicle operates in a loop. References [1] and [2] first introduced the tandem concept for AGV systems. In [3] they also presented an analytical model to evaluate the throughput performance of a basic component of the system and proposed a heuristic partitioning scheme to configure tandem AGV systems: this procedure is based on a variation of the wellknown set partitioning problem. Starting from the partitioning algorithm presented in [3], in [4] a comparison was conducted between AGV tandem and conventional systems. Reference [5] provided an algorithm for routing AGVs among non-overlapping closed loops within a tandem AGV system, whose control issues where addressed in [6] . Reference [7] developed a heuristic partitioning algorithm for a tandem AGV system based on the concept of variable path routing within a zone. Reference [8] proposed a partitioning procedure based on similarity coefficient between departments. The recent paper [9] developed a design methodology for tandem AGV systems with
*

Corresponding author: Tel: (+39) 0832 297251; Fax: (+39) 0832 297 279; E-mail: pierpaolo.caricato@unile.it

multiple-load vehicles in order to achieve the workload-balance between vehicles of different loops and to minimize the inter-loop flow and the flow distance. The two important issues to consider when designing a tandem layout are the partition algorithm and the determination of the transhipment point. Since the previous works consider a static partitioning of departments, In this paper we focus our attention on a dynamic partitioning of departments on the basis of the flows and a dynamic positioning of the transhipment point. The remainder of this paper is organized as follows. In section 2, the problem statement is given and a MILP formulation is presented. In section 3, an approach based on Simulated Annealing (SA) is described. In section 4, test cases are introduced and computational results are presented. In section 5, a brief summary is given, along with future research directions.

2. PROBLEM STATEMENT A block layout is a schematic representation of a physical layout, in which departments are represented by cells, i.e. rectilinear but not necessarily convex polygons, each characterized by its edges and vertices. The material handling system is represented by a subset of edges, selected as a guide path for the AGVs, along with a subset of vertices, selected as pickup and/or delivery stations. Due to both economical and technical reasons, the pickup and delivery functions are commonly grouped in combined pickup and delivery stations (P/D stations). As stated in the introduction section, the problem that is named TCP in the literature considers a given block layout, well known material flows among cells and given P/D stations for each cell. We consider a more general problem in which the position of P/D stations is not given in advance. The presented formulation can also be used to represent the classical TCP problem, fixing the P/D stations positions a priori. The considered problem can be formulated as a Mixed Integer Linear Programming (MILP) problem. The objective is to minimize the total material flow in the system on the basis of the flows between different departments. We assume that the guide path has to contain at least one edge of each department and is made up of N loops. Consequently, N-1 transhipment points are to be determined. Adjacent loops can be connected only through a single transhipment point and cannot have any vertex in common apart from this one. Each department must have exactly one P/D station, that has to be positioned in one of the vertices of the same department; if a department is served by a loop, it cannot be served by the other loops. No hypothesis is made regarding the shape and the size of departments. Hence, the problem is characterized by the following information: C set of departments; E set of edges in the layout; V set of vertices in the layout; set of the couples of departments between which there is a non-zero flow; T set of the possible transhipment points; Vc set of the vertices in department c; EIi set of the edges going in vertex i; EOi set of the edges going out vertex i; lij length of the edge (i,j); fcd material flow from department c to department d, where

( c, d ) .

The formulation uses five sets of variables:

New approaches for the solution of Tandem Configuration Problems in highly automated manufacturing systems

real variables xijcd which represent the fraction of flow from department c to department d that passes through edge (i,j); binary variables yij which are equal to one if and only if edge (i,j) belongs to the loop; binary variables zi which are equal to one if and only if vertex i belongs to the loop; binary variables zci which are equal to one if and only if vertex i is the pickup/delivery point for department c; binary variables wi which are equal to one if and only if vertex i is a transhipment point.

The formulation is then:

min
s.v.

ij cd (i, j)E (c, d)

l f

x ijcd

(2.1)

y ij + y ji 1

(i, j) E
c C

(2.2) (2.3)

( i , j )Ec

y
ci

ij

z
iVc

=1

c C
c C i Vc

(2.4)

zci zi

(2.5) (2.6)

( i , j )EiO jVc

y
i

ij

( j ,i )EiI jVc

ji

zci

c C i Vc

w
iT

=N i V i V \ T 1 + wi 1 + wi
+ zi

(2.7)

wi zi wi = 0

(2.8) (2.9)

( j ,i )EiI

y y

ji

i V
(2.10)

( i , j )EiO

ij

i V
(2.11)

( j , i )E iI

ji

( i , j )E iO

ij

4 wi

i V
(2.12)

( j ,i )EiI

ji

i V
(2.13)

( i , j )EiO

ij

zi

i V
(2.14)

zi + z j 2 yij

(i, j ) E
jicd

(2.15)

( i , j )E iO

ijcd

( j , i )E iI

= zci zdi

(c, d ) i V
(2.16)

( c ,d )

ijcd

| | yij

(i, j ) E
(2.17)

xijcd [0,1] yij [0,1] zi [0,1] zci [0,1]

(i, j ) E (c, d ) (i, j ) E i V c C i Vc

(2.18) (2.19) (2.20) (2.21) (2.22)

wi [0,1]

i V

In this formulation, the objective function (2.1) minimizes the total material flow within the block layout. Since loops must be unidirectional, each edge can be in a loop in only one of the two possible directions (2.2). Moreover, each department must be touched by the loop at least in one edge (2.3) and must have exactly one P/D station (2.4). A vertex of a department can be a P/D point if and only if it belongs to the loop (2.5), otherwise the department would result isolated. If a vertex is a P/D point for a department, then at least one edge (of the same department) containing the considered vertex must be in the loop (2.6). The number of transhipment points is equal to a fixed number N (2.7) and a vertex can be a transhipment point if and only if it belongs to the loop (2.8). Binary variable wi is equal to zero for all vertices that cannot be transhipment points, being the intersection of less than four edges (2.9). Each vertex that is not a transhipment point can have at most one edge going in the vertex (2.10) and at most one edge going out the vertex (2.11); on the contrary, each vertex that is a transhipment point must have exactly two edges going in and two edges going out the vertex (2.12). If a vertex is in the loop, then at least one edge going in the vertex (2.13) and one edge going out the vertex (2.14) must be in the loop. Finally, if an edge is in the loop, the two vertices which identify it must be in the loop (2.15). The set of constraints (2.16) represents flow preservation. This constraint does not represent expressly the case in which vertex i does not belong to the vertices of one or both departments; in this case, corresponding variable is not defined and is replaced by zero, maintaining the validity of the constraints. The amount of flow which can transit in a generic edge of the layout is equal at most to the total amount of material flows in the whole layout (2.17). As it will be further analyzed when discussing computational results, this MILP model can be solved using a commercial MILP solver such as ILOG CPLEX. However, the complexity of the problem is highly dependent on the considered problems dimensions, making it possible to optimally solve only very small problems. Hence, the need to introduce a heuristic approach for larger problems.

New approaches for the solution of Tandem Configuration Problems in highly automated manufacturing systems

3. A SIMULATED ANNEALING APPROACH Simulated Annealing is a general purpose meta-heuristic that can be tailored to solve many combinatorial problems. In order to use SA to solve a particular problem, a thorough comprehension of the problem is needed to better adapt the general SA algorithm to the considered problem. Tailoring SA on a particular problem means: defining what a solution for the problem is determining a first feasible solution defining a neighbourhood of a solution, including a specialized distance function among solutions introducing an objective function to compare solutions

3.1.1 DEFINITION OF SOLUTION Let N be the number of loops that are to be determined, in the SA approach we can consider as fixed the position of the transhipment points, since the several possible positions of the transhipment point can be iteratively evaluated. Under this assumption, a solution can be described as a vector composed of a number of zones: the variables in the first N zones indicate whether a given oriented edge belongs to the first, second, and up to the N-th loop or not the variables in the following zone represent the positions of the P/D stations for each cell

3.1.2 INITIAL SOLUTION The creation of a first, feasible solution to start up the SA main algorithm is an important aspect for the effectiveness of the whole algorithm, since it can deeply influence the following steps of the execution. We developed a local search algorithm to determine a first feasible solution. We show the proposed procedure for the case of two loops. Given the transhipment point, two departments having this point and without common edges are selected. The algorithm starts from a generally infeasible solution given by the perimeters of the two departments and then moves through pivoting operations (described in the following section) on a random department sequence. The objective of the local search is to maximize the number of departments with at least one edge on one of the loops.

Figure 2: Initial solution construction algorithm illustrated

3.1.3 NEIGHBOURHOOD A neighbourhood of a given solution can be defined as the set of feasible solutions that can be reached from the given solution through a number of moves. A move can be defined as a pivoting operation over a feasible loop l on a given department c that defines a new loop l such that the edges of c that are on l do not belong to l, and the edges of c that are not on l belong to l. Pivoting can be better explained graphically as showed in figure 2 for the determination of the initial solution. The second layout in the figure is obtained through a pivot operation between cell 2 and the first (upper) loop. Since tandem configuration consists of two or more loops, when a department is selected, the pivoting operation between this department and the first loop is made; if the resulting solution is feasible, then it is accepted, otherwise the same operation will be made with the other loops. In TCP, there it has no practical sense to define a metrics in order to determine the distance between two solutions; indeed the modification of the solution in the neighbourhood can determine a change of part of the guide path or only a re-positioning of the P/D points and so it is not possible to define a coherent concept of neighbourhood in the sense of distance. A reasonable, constructive concept of distance can be expressed by the number of pivoting operations required to move between two solutions. This practical metrics has been adopted in the proposed SA approach. After obtaining a feasible guide path, the P/D points must be fixed. The sub-problem that has to be addressed that has to be solved is well known in the literature and can be described as follows. Given a block layout, the fromto chart representing material flows between departments and a tandem guide path which is made up of two or more loops and which contains at least one edge for each department, the Station Positioning Problem (SPP) consists of establishing the position of pickup/delivery points. In other words, the loops have been fixed and the pickup/delivery stations are to be determined. The objective is to minimize the total material flow in whole system. We propose two approaches for the solution of the SPP: a heuristic approach and an optimal algorithm. The heuristic approach can be summarized as follows: the flows between departments are sorted in decreasing order for each flow the P/D points of all couples of communicating departments get fixed as follows if the P/D points of the two departments are already fixed, then next flow is analysed if the P/D points are not fixed for both departments, then they are chosen on the loops in order to minimize the distance between the selected points if the P/D point of only one department is fixed, then the P/D point of the other department is fixed in order to minimize the distance between the selected points

We also adopted an exact approach for the solution of the SPP that is based on a linearization of the problem. The approach is not straightforward and will be deeply discussed in a future paper. Basically, it consists in unrolling the loops into rectilinear segments and using a MILP formulation of the problem achievable through such linearization. 3.1.4 OBJECTIVE FUNCTION The objective function to be pursued is the minimization of the total material handling flows among departments. For each couple of departments the contribution to the objective function is given by the known flow between the two cells multiplied by distance between their P/D points. Since tandem configuration consists of two or more loops connected through a transhipment point, we must consider that, if the communicating departments are served by different loops, the distance between the P/D points is obtained by adding the distance between the P/D point of source department and the transhipment point with the distance between the transhipment point and the P/D destination department.

New approaches for the solution of Tandem Configuration Problems in highly automated manufacturing systems

4. COMPUTATIONAL RESULTS The algorithm based on SA was coded in C# and executed on a Intel Pentium Processor (1.5 GHz). The MILP formulation was solved using CPLEX 9.0. A maximum of 7200 seconds time limit was forced. We tested the algorithm on instances generated in order to explore several scenarios in terms of complexity of the layout and of flows network. We used block layouts with a number of departments between 10 and 30 and from-to charts with a number of non-zero elements between 10% and 100%. In table 1, we report the computational results obtained by using MILP formulation and SA algorithm for the TCP. The instance column provides the name of the experiment, where L? indicates a particular block layout, while F? denotes a specific flows matrix. CPLEX O.F. contains the objective function achieved with the MILP formulation, while SA O.F. is the objective function obtained using the SA approach. The % Err column is the percentage error of the SA value compared with the CPLEX value. The CPLEX time is the time (in seconds) used by CPLEX, while SA time is the time consumed by the SA algorithm. When the time limit imposed has been reached by the CPLEX solver, the CPLEX O.F. column contains the best feasible solutions value achieved; in this case, the % ERR value can be negative, indicating that the SA solution is better than the sub-optimal solution found using the time limit arrested CPLEX branch and bound solver.
Table 1: Computational results of MILP vs. TS

Instance L1F1 L2F1 L2F2 L4F1 L4F2 L4F3 L5F1 L5F2 L5F3 L5F4 L5F5 L6F1 L6F2 L6F3 L6F4 L6F5 L6F6 L6F8

CPLEX O.F. 3811 65118 748886 114760 1524298 1531382* 92277 292237 315852 623739 792443* 73898* 98161* 258968* 217283* 655471* 766079* 1606316*

SA O.F. 3811 69525 769988 119712 1616703 1584253 100752 306879 327888 648478 736598 75126 98349 256081 213702 656541 731379 1430889

%Err 0% 6.7% 2.8% 4.3% 6.1% 3.4% 9.2% 5% 3.8% 4% -7% 1.7% 0.2% -1.1% -1.6% 1.16% -4.5% -11%

CPLEX time 0.7 22 3779 34 4825 Stop 7200 1283 1673 3377 11744 Stop 7200 Stop 7200 Stop 7200 Stop 7200 Stop 7200 Stop 7200 Stop 7200 Stop 7200

SA time 87 328 1812 804 4001 6371 1072 3650 2628 3383 4036 3990 4254 5899 6013 6454 7054 7112

The exact MILP approach can only be used with relatively small instances of the problem: the larger problem solved to optimality is the L5F4 problem, that corresponds to a 15 cells block layout with a 80% dense flow matrix (20% of possible flows are null). In these cases, the results achieved using the SA approach are always less than 10% far from the optimal value, confirming the quality of the proposed approach. The following, larger cases, include very large problems, such as the L6F8 that is a 30 cells block layout with a 100% dense matrix (all cells require material flow with any other cell in the block layout). In these cases, the objective achieved with the SA approach is often better than the one achieved using the time limit arrested CPLEX solver.

Furthermore, the time needed for the SA approach to end its heuristic search grows much slowly than the time needed by the CPLEX solver. Hence, the quality of the results achieved using the time limit arrested CPLEX solver are due to become worse and worse as the problem dimension rises, as can also be seen in the proposed results.

5. SUMMARY We proposed a new mathematical formulation for the Tandem Configuration Problem as a mixed-integer linear programming model. We also proposed a heuristic algorithm based on Simulated Annealing to solve this problem within reasonable amounts of time. We tested the proposed approaches on several cases chosen in a range of small, average and large problem instances. The exact approach allows to find optimal solutions to small instances of the problem, while the heuristic approach can be used for larger problems. The achieved results proves the effectiveness and quality of the proposed approach. Further developments will investigate the possibility to use other metaheuristic frameworks, such as Tabu Search or Genetic Algorithms as alternative heuristic approaches. A wider experimental campaign will also be developed in order to better valuate the performances of the heuristic approach under an even wider range of test circumstances.

6. REFERENCES
[1] [2] [3] [4] [5] [6] [7] [8] [9] Y. A. Bozer and M. M. Srinivasan: Tandem configurations for AGV systems offer simplicity and flexibility, Industrial Engineering, Vol. 21, No. 2, pp. 23-27, 1989. Y. A. Bozer A. Yavuz and M. M. Srinivasan:Tandem configurations for automated guided vehicle systems and the analysis of single vehicle loops, IIE Transactions, Vol. 23, No. 1, pp. 72-82, 1991. Y. A. Bozer and M. M. Srinivasan: Tandem AGV systems: A partitioning algorithm and performance comparison with conventional AGV systems, European Journal of Operational Research, No. 63, pp. 173-191, 1992. N. Gaouar, L. Ghomri and Z. Sari: Comparaison entre un systme AGV Tandem et un systme AGV conventionnel: www.univ-tlemcen.dz J. T. Lin, C. C. K. Chang and W. C. Liu: A load-routing problem in a tandem configuration AGVs, International Journal of Production Research, No. 32, pp. 411-427, 1994. J. T. Lin and P. K. Dgen: An algorithm for routing control of a tandem automated guided vehicle system, International Journal of Production Research, No. 32, pp. 2735-2750, 1994. W. Yu and P. J. Egbelu: Design of a Variable Path Tandem Layout for Automated Guided Vehicle Systems, Journal of Manufacturing Systems, Vol. 20, No. 5, 2001. A. Aarab, H. Chetto and L. Radouane: Flow Path Design for AGV Systems, Studies in Informatics and Control, Vol. 8, No. 2, 1999. Y. Ho and P. Hsieh: A machine-to-loop assignment and layout design methodology for tandem AGV systems with multiple-load vehicles, International Journal of Production Research, Vol. 42, No. 4, pp. 801-832, 2004.

You might also like