Determination of the Fleet Size of AGVs with AGV Pools Using a Genetic Algorithm and Artificial Intelligence

Abstract

The utilization of low-cost AGVs in the industry is increasing every day, but the efficiency of these systems is low due to the lack of a central management system. Low-cost AGVs’ main characteristic is navigation via magnetic sensors, which they follow via magnetic tape on the ground with a low-level automation system. The disadvantages of these systems are mainly due to only one circuit assignment and the lack of system intelligence. Therefore, in this study, AGV pools were employed to determine the required AGV number. This study begins by calculating the required AGV number for each AGV circuit combination assigned to every parking station by the time window approach. Mathematical-solution-based mixed integer programming was developed to find the optimum solution. Computational difficulties were handled with the development of a genetic-algorithm-based approach to find the solutions for complex cases. If production requirements change, system parameters can be changed to adapt to the production requirements and there is a need to determine the number of AGVs. It was demonstrated that AGVs and pool combinations did not lead to any loss in production due to the lack of available AGVs. It was shown that the proposed approach provides a fleet size which requires five fewer AGVs, with a 29% reduction in the number of AGVs. The effects of system parameter changes were also investigated with artificial neural networks (ANNs) to estimate the required AGVs in the case of production requirement changes. It is necessary to determine the effect of the change in system parameters on the number of AGVs without compromising on computational cost and time, especially for complex systems. Thus, in this study, an artificial neural network (ANN), the response surface method (RSM), and multiple linear regression (MLR) techniques were used to examine the effects of the system parameter changes on the AGV number. In the present case, the ANN obtained the solution at a good rate with reduced computational costs, time, and correction errors compared to the GA, at 0.4% (ANN), 7% (RSM), and 24% (MLR). The results show that the ANN provides solutions which can be used in workshops to determine the number of AGVs and also to predict the effect of changes in system parameters.