Flexible Job-Shop Dynamic Scheduling Considering Multi-AGV Path Planning

Flexible job shop scheduling is the key to high efficiency,flexibility and low cost in manufacturing industry.There are dynamic uncertainties and complexity of resource scheduling in the actual manufacturing process of workshop.Dynamic uncertainty means that in the actual production process of the workshop,the environment of the workshop is complex and changeable,and dynamic events such as urgent workpiece insertion,machine failure and dynamic arrival of the workpiece occur from time to time,making the original scheduling scheme operate inefficiently and even interrupted.The complexity of resource scheduling refers to that in flexible job shop scheduling,not only the workpiece processing scheduling needs to be carried out,but also the AGV(automatic guide car)used for moving parts needs to be scheduled.The handling path of AGV and the workpiece processing scheduling interact,and the overall allocation of flexible job shop resources makes the determination of scheduling scheme more complicated.By timely processing the dynamic events in the process of processing,the limited resources such as machine and AGV in the flexible job shop are coordinated and allocated,which has certain theoretical significance and engineering application value for improving the stable implementation of the production scheduling scheme in the flexible job shop,ensuring the delivery time of the workpiece and improving the overall performance of the manufacturing system.This paper mainly focuses on flexible job-shop dynamic scheduling considering multi-AGV path planning.The main research contents are as follows:(1)Based on the production process and characteristics of multi-AGV path planning flexible job shop,the process flexibility,constraint complexity,multi-objective and dynamic characteristics of dynamic shop scheduling were analyzed.The flexible job shop dynamic scheduling problem considering multi-AGV path planning was decomposed into three sub-problems: dynamic scheduling time determination,workpiece,machine and AGV scheduling and multi-AGV path planning.Considering the constraints of workshop layout,machine failure,dynamic arrival of workpiece,urgent workpiece insertion,AGV railless,no collision between AGVs,workpiece process flexibility,The mathematical model of dynamic scheduling problem was established by taking minimizing the total delay of workpiece,the balance of machine utilization,the balance of AGV utilization,minimizing the maximum deviation of completion time and minimizing the sequence of working machines as scheduling objectives,and minimizing the handling time as path planning objectives.(2)The flexible job-shop dynamic scheduling method based on D3QN(dual double depth Q network)algorithm is studied.Based on the dynamic scheduling method process of the shop,the dynamic scheduling strategy driven by both events and variable cycle is adopted.When the urgent dynamic event of machine failure or urgent job insert occurs,the event-driven strategy is used to trigger scheduling.If no urgent dynamic event occurs in the shop,the variable cycle driven strategy is used to trigger scheduling.The variable cycle scheduling is realized according to the frequency of the workpiece’s dynamic arrival to the workshop and the processing load of the workshop.When the shop system clock meets the variable cycle driving time,the cycle driving scheduling is triggered.Scheduling agents are designed to realize resource scheduling of flexible job shop.In order to fully express the state of the shop environment and improve the generalization of scheduling algorithm,9 scheduling state features are extracted according to the job information,machine information and AGV information of the shop as the observation state space of scheduling agents.Average process completion rate,average workpiece completion rate,standard deviation of workpiece completion rate,simulated delay rate and actual delay rate relative to workpiece.Average machine utilization and standard deviation of machine utilization relative to machine.The average utilization rate of AGVs and the standard deviation of utilization rate of AGVs are related to AGVs.According to the scheduling objective,eight scheduling decisions are made by the combination of three kinds of rules: job sorting selection,machine sorting selection and AGV sorting selection.Four layers of reward functions are designed to judge the decision actions of scheduling agents based on actual delay rate,simulated delay rate,average machine utilization rate and average AGV utilization rate,in order to realize the mapping relationship between reward function and scheduling objective function.In the training process of the scheduling agent,the experience priority playback mechanism and the target network weight soft update mechanism are used to improve the training speed and stability of the scheduling agent,and the interaction mechanism between the scheduling algorithm and the path planning algorithm is designed.Then,the multi-AGV path planning problem is studied,the types of multi-AGV conflicts and the path planning process are analyzed,and the path planning method based on MADDPG(multi-agent depth deterministic strategy gradient)algorithm is adopted.Combined with the position of AGV itself,target position and other AGVs,the observation state space of the agent in the path planning environment is designed.The Gumbel-Softmax sampling method is used to output the discrete action space of the strategy gradient algorithm.A non-sparse reward function is designed to guide the training of the agent according to the minimization of the AGV transport time target and the non-collision constraint between AGVs.The positive reward is obtained when the AGV reaches the target position or is close to the target position,and the negative reward is obtained when the AGV collids,moves away from the target position or is stationary.The path planning case and dynamic scheduling case are used to test the algorithm respectively to verify the feasibility of the algorithm.(3)The flexible job shop scheduling prototype system is developed,analyze the production requirements and processes of the flexible job shop,design the system framework,each functional module and database.The prototype system is developed based on Py Qt5 platform,including the function realization of landing module,shop information management module,scheduling algorithm module,dynamic scheduling module and scheduling path display module,so as to provide meaningful guidance for the production scheduling plan of flexible job shop.