Dual track and segmented single track bidirectional loop guidepath layout for AGV systems

Abstract

In this study, we develop integer programming models for simultaneous design of the guidepath track and location of the pickup and dropoff stations for automated guided vehicle systems. Two alternative bidirectional topologies are compared: a dual track loop, and a single track loop partitioned into non-overlapping segments, each served by a single vehicle. We show that the empty vehicle trips have a significant impact on the quality of design in both topologies. More importantly, we identify the difference between empty flow node balancing for a dual track loop with the empty flow edge balancing on a single track loop subject to segmentation. Our work presents an economic decision making model for the trade-off analysis between the costs associated with the additional shop floor space of a dual track loop and the increased vehicle fleet size of a segmented single track loop.

Introduction

Automated guided vehicle (AGV) systems have been implemented in a large variety of industries such as aerospace, automotive, chemical, electronics, food and beverage, plastic, and textile as well as in inter-modal container ports. AGV systems result in better production planning and control, safety, cost reduction, and flexibility according to the Material Handling Industry of America (MHIA, 1999). While the number of AGV systems implemented in the United States still lags significantly behind those in Europe and Asia, the last few years have seen a consistent growth. The number of new installed systems has grown from 43 in 1999 to 73 in 2002 (a 70% increase), and the number of new vehicles has increased from 222 to 386 (a 72% increase) during the same time period. This trend is based mainly on the increase in the efficiency as well as on the cost reductions in the installation and operations (MHIA, 2002). Unit load AGVs carrying a single load at-a-time are highly maneuverable and need a small portion of the shop floor for operation. They form the largest segment of the AGV industry (MHIA, 1999). Unit load AGVs have bidirectional movement capabilities and do not need to execute U turns.

One of the main tasks in designing an efficient and economically viable AGV system includes the concurrent determination of the vehicle guidepath tracks layout and the location of the pickup (P) and dropoff (D) stations, assuming a given block layout of the work-centers. The block layout of our example with work-centers $C_1\text{,}\dots \text{,}{C}_8$ along with the node identification numbers and edge lengths appears in Fig. 1a. The required material flow between pairs of work-centers during the planning horizon is shown in Fig. 1b. Because our material handling devices are unit load AGVs, the elements of the material flow matrix correspond to the required number of loaded trips from the P station of the work-center in the row to the D station of the work-center in the column.

The edges on the boundaries of the work-centers are potential elements for the vehicle’s guidepath in a general AGV system track layout. A general guidepath layout (a) occupies a substantial portion of the shop floor, (b) has many intersections which may reduce the average velocity of the vehicles due to frequent stops, and (c) requires complex vehicle dispatching, vehicle routing, and traffic control rules and algorithms. The operational complexities of the general guidepath layouts are reduced in loop-based guidepath layout covering at least one edge of each work-center.

A dual track bidirectional loop (DTBL), as illustrated in Fig. 2a, is composed of two single track unidirectional loops (STULs) with interconnections at the P and D stations. A single track bidirectional loop (STBL), as illustrated in Fig. 2b, has the advantage of requiring about half the foot print on the shop floor of the equivalent DTBL. A STBL with more than one vehicle is not operationally feasible since the vehicles may be blocked in face-to-face deadlocks. The requirements to avoid or resolve such vehicle deadlocks would add significant complexity to the vehicle dispatching and control system. A simpler method to avoid deadlocks is to partition the loop into non-overlapping segments, where each segment is served by a single vehicle (Sinriech et al., 1996). However, on a segmented STBL (SSTBL), as shown in Fig. 1c, a load in transportation from origin to destination may require one or more transshipments between the vehicles servicing adjacent segments.

We consider the problems of the concurrent design of (i) a dual track bidirectional loop, or (ii) a segmented single track bidirectional loop, both concurrent with the design of the P and D stations. It is assumed that the required number of loaded vehicle trips during the planning horizon are known. This is essentially a simplification of the real design problem, where the required numbers of trips are based on forecasting and are highly stochastic. Our methodology can be extended to stochastic design through the use of scenarios, albeit with a significant increase in problem size and computational cost. The same is true for extending the basic model to include time windows. The objective is to minimize the fleet size, i.e. the required number of vehicles to execute the material handling requests during the planning horizon. The cost of the fleet of the vehicles and the space occupied cost of the guidepath are the main determinants in the total cost of the system.

The remainder of the paper is organized as follows: Section 2 reviews previous related research results. In Section 3, we first discuss the importance of incorporating empty vehicle trips in the objective function of the DTBL model, and then develop the optimization model for simultaneous design of the DTBL and P and D station locations. Computational results for the DTBL model are reported in Section 4. In Section 5, we first identify the difference between empty flow node balancing for a dual track loop with the empty flow edge balancing on a single track loop subject to segmentation. We then present a two-phase algorithm to design a SSTBL and report on the computational comparison of DTBL and SSTBL. Conclusions and the possible directions for future research follow in Section 6.