Abstract: The battery of AGV will affect the efficiency of the production system. For example, if the system knows that the battery power of an AGV is not sufficient to perform a task, it will assign the task to another AGV or assign the AGV to a charging station. For the production workshop using AGV for material handling, one of the bottlenecks to increase production capacity is the available production time of AGV. The method of increasing production capacity by increasing the number of AGVs is very expensive, and the layout of the production workshop is mostly compact. Too many AGVs may cause the system to be more blocked and reduce production efficiency. In current industrial production applications, the battery is managed as follows: when the AGV continues to operate until its power level drops below a certain threshold, it goes to the charging station to fill it up to 100%. In this decision, if the system is in an idle period, that is, the AGV in the system is idle and the battery power does not reach the threshold, limited by the power management method, the AGV cannot effectively use the idle time to go to the charging station. If the system is in a busy period, that is, the AGV in the system is in a full load state, when the AGV power level drops below the threshold, the current position of the AGV may be accidentally far away from the nearest charging station, which will increase the time required to go to the charging station to charge and return to perform the remaining tasks, or at this time, the task volume in the system is large, and the AGV is performing an emergency order task, which cannot consume a lot of time to wait for full power before continuing to perform the remaining tasks. AGV can make a more sensible decision, that is, not fully charged, but it still can complete the task of emergency orders. After the system busy period, AGV can decide to continue charging.
Therefore, this paper studies how to deal with a sudden increase in capacity demand in the production workshop by changing the charging mode of AGV without changing the number of AGVs and improve the production efficiency of the workshop. A general AGV charging model is proposed. This model studies the most widely used lead-acid battery as the research object, and considers AGV in both full load and non-full load state, aiming at the AGV automatic charging mode lacking in previous studies. The model aims at the shortest total time for all tasks to complete, and makes decisions on the time, placement, and duration of AGV charging tasks in its task sequence. The AGV's automatic charging model has three decision problems to be solved: (1)Whether charging is required after the i task site. (2)Which charging station to go to be charged? (3)How long to be charged? There are two possible situations in the production that need to be considered at the same time: (1)The system is relatively busy, the AGV is in full load operation, and the tasks in all task sequences are continuously performed. (2)AGV is in a non-full load operation state, and there is idle time in the process of executing the task sequence. In order to solve the model, a heuristic algorithm (PSO-NCMO) combining particle swarm optimization algorithm and nonlinear constrained multivariable optimization algorithm is proposed, and the improved particle evolution framework is used to optimize the model.
The experimental data come from a real electrode production workshop of a new energy enterprise. The data of small, medium, and large AGV task scales in the workshop are selected for simulation experiments. By comparing the optimization results of the industrial benchmark method and the proposed PSO-NCMO algorithm, the results show that when the AGV is at full load, the PSO-NCMO algorithm reduces the total execution time of the task including charging by an average of 23.55% compared with the benchmark. When the AGV is at non-full load, the algorithm can prioritize the AGV to use the idle time to go to the charging station. When the initial power and minimum power limit of AGV are changed, AGV ends the task with as little residual power as possible, which shows that the charging model and algorithm have strong power management effect and have certain versatility. The charging time can be adjusted according to the actual needs of the enterprise. The limitation of the text is that it does not consider the faults and congestion during AGV driving. In the future, we will use deep learning technology to establish a prediction model during AGV driving, making the AGV power consumption and charging model more suitable for actual production.

Key words: AGV, battery management, particle swarm optimization, auto-charging

摘要： 为提高车间的生产效率,在不改变AGV数量的前提下应对生产车间产能需求的突然增加,提出一种通用的AGV自动充电任务模型。该模型以最小化AGV任务总执行时间为目标,对AGV充电任务序列的持续时间及地点进行决策。为求解该模型,提出一种融合粒子群算法和非线性约束多变量优化算法的启发式算法(PSO-NCMO),使用改进的粒子演化进化框架对模型进行优化求解。实验数据来源于某新能源企业电极片生产车间,选取车间内部三种任务规模的数据进行仿真实验。通过将工业基准法和所提出的PSO-NCMO算法的优化结果进行对比,结果表明:相对于工业基准,PSO-NCMO算法将包括充电在内的满负荷AGV工作任务总执行时间平均降低了23.55%,对于非满负荷的工作情况,算法能够主动引导AGV在任务空窗期前往充电站进行自动充电,且能够确保AGV在任务执行阶段维持电量安全阈值,避免由于电量不足导致任务中断。当改变AGV的初始电量和最低电量限制时,AGV均以尽可能少的剩余电量结束任务,这表明充电模型及算法有较强的电量管理效果,能够依据企业实际作业需求进行调整。

关键词: 自动导引车, 电量管理, 粒子群算法, 自动充电