An effective iterated greedy algorithm for solving a multi-compartment AGV scheduling problem in a matrix manufacturing workshop

Highlights

• A multi-compartment automatic guided vehicle scheduling problem is studied.

• An effective iterated greedy algorithm is proposed.

• Accelerations for evaluating solutions are presented.

• The proposed algorithm is the best performing against all the existing methods.

Abstract

In this paper, we address a multi-compartment automatic guided vehicle scheduling (MC-AGVS) problem from a matrix manufacturing workshop that has attracted more and more attention of manufacturing firms in recent years. The problem aims to determine a solution to minimize the total cost including the travel cost, the service cost, and the cost of vehicles involved. For this purpose, a mixed-integer linear programming model is first constructed. Then, a novel iterated greedy (IG) algorithm including accelerations for evaluating objective functions of neighboring solutions; an improved nearest-neighbor-based constructive heuristic; an improved sweep-based constructive heuristic; an improved destruction procedure; and a simulated annealing type of acceptance criterion is proposed. At last, a series of comparative experiments are implemented based on some real-world instances from an electronic equipment manufacturing enterprise. The computational results demonstrate that the proposed IG algorithm has generated substantially better solutions than the existing algorithms in solving the problem under consideration.

Introduction

In recent years, with the rapid development of manufacturing technologies and the sprouting of diverse demand of customers, a matrix manufacturing workshop with multi-variety and small-batch production characteristics has been more and more favored by manufacturing industries. As a transport tool, the automatic guided vehicle (AGV) is used to handle cutters and various production materials in the matrix manufacturing workshop. Considering the cost and efficiency, the AGV scheduling is essential since an efficient AGV scheduling can increase productivity and reduce the delivery cost [1], [2], [3], [4]. The AGV scheduling indicates that the control system plans the traveling route with the objective such as the shortest distance, least cost, or least time in order to enable AGVs to pass through a series of loading and unloading sites in an orderly manner under certain constraints. In a matrix manufacturing workshop, an AGV is equipped with multiple compartments, each of which only contains a load (i.e., cutter or various production materials) that is different from those in the other compartments. How to dispatch the AGVs to maximize the benefits is an attractive topic for researchers or practitioners.

The matrix manufacturing workshop is one of the most widely arranged intelligent workshops in modern manufacturing enterprises. It is usually divided into m sub-areas, each of which is responsible for producing a product. A fleet of homogeneous AGVs is waiting for the command from the control system and is ready to deliver cutters and production materials to n workstations (customers). Each AGV is equipped with multiple compartments, each of which is limited to a loading cutter or one production material. Each AGV starts from the depot, and passes several customers, and returns to the depot after delivery. The above characteristics illustrate that the MC-AGVS problem is a variant of the vehicle routing problem (VRP), but it is more complex due to many problem-specific characteristics such as multiple compartments, variable requirements, and multiple constraints. The VRP is an NP-hard problem [5], which means that the MC-AGVS problem is also an NP-hard problem. It is almost impossible for an NP-hard problem to get a solution by exact solution methods in a limited amount of computing time [6]. Heuristics and meta-heuristics, however, are highly appealed for solving such a problem within a reasonable time limit [7], [8], [9].

The first come first served (FCFS) method is commonly used in the modern industry. In other words, customers who send the request first obtain the delivery of AGVs first. The FCFS method is a simple heuristic to reflect the production behavior of manufacturing workshops, but this method is not dependable because it might make AGVs repeatedly rush between customers, resulting in much time wasted on the road and most of the production materials left in the compartments. Although an acceptable solution can be obtained by using the FCFS method, this solution has still a lot of room for improvement. Hence, it is important to find out a highly effective approach to improve the delivery efficiency of AGVs and reduce their total cost. So far, Li et al. [10] propose an improved harmony search algorithm to solve a problem closely related to the MC-AGVS problem. The authors introduced three indicators such as the total travel distance of AGVs, the standard deviation of AGV workload, and the standard deviation of the difference between the latest delivery time and the predicted time of customers. Inspired by them, we also adopt three important indicators after determining the goal with the total cost minimization, namely travel cost, service cost, and cost of the vehicle involved.

Despite its wide practical applicability, the MC-AGVS problem has not received much attention in the literature so far. To fill the gap, in this study, we construct a mixed-integer linear programming model and propose an effective iterated greedy (IG) algorithm to solve the problem. The proposed algorithm includes some advanced techniques, i.e., accelerations for evaluating neighboring solutions; an improved nearest-neighbor-based constructive heuristic; an improved sweep-based constructive heuristic; an improved destruction procedure; as well as a simulated annealing-like acceptance criterion. Furthermore, we choose the FCFS method, Gurobi solver, and four well-performing algorithms from the literature on the problems closely related to our problem as the competitive algorithms. And then, a computational campaign is implemented based on 110 practical instances from a real-world factory. The results show the effectiveness of the proposed IG algorithm.

The remainder of this paper is organized as follows. In Section 2, we review the literature on the problems closely related to the MC-AGVS problem. Section 3 formulates the MC-AGVS problem and establishes a mixed-integer linear programming model. In Section 4, an effective iterated greedy algorithm is presented. Section 5 reports the computational results and comparisons. Finally, the concluding remarks are provided in Section 6.