Research On Multi-resource Cooperation Scheduling And Optimization For Intelligent Flexible Job Shop

In order to cope with uncertain market changes and fierce competition,lots of manufacturing enterprises that aim to improve quality,increase efficiency and reduce cost,vigorously promote the deep integration of the new generation information technology and industrialization system,such as industrial robot,industrial Internet of Things,artificial intelligence,etc.,and reshape the brand advantage of “made in China” with intelligent manufacturing and personalized customized production.Among them,the Intelligent Flexible Manufacturing System(IFMS)have become one of the important paths for the transformation and upgrading of manufacturing enterprises because of its high-flexibility,which strengthens the system to effectively adapt to increasing product complexity and customized production mode.Different from traditional flexible manufacturing systems,IFMS is composed of the unified information control system,the automated material handling system and the digital processing equipment that are inter-connected through the industrial Internet of Things technology,and independently cooperating to complete the task,and ultimately realize the automation,digitalization and intelligence of production and operation.The automated material handling system is mainly composed of automatic guided vehicles(AGVs)that can assist in achieving autonomous production.The classical flexible job shop scheduling(FJSP)theory and method cannot fully adapt to the operation management requirements of new job processes,production processes and intelligent equipment systems for overall scheduling,intelligent decision-making and system optimization.It is urgent to explore the theory and method of multi-resource collaborative scheduling and optimization of intelligent flexible job shop based on the engineering practice of IFMS,so as to support the intelligent flexible job shop to form autonomous decision-making,intelligent scheduling and system optimization capabilities.Based on the classical FJSP theory,this paper focuses on the flexible job shop collaborative scheduling and optimization problem with transportation resources(FJSPT),and proposes a set of multi-resource collaborative scheduling and optimization models and methods suitable for the industry practice of IFMS.It mainly includes four critical contents: the complexity analysis and mechanism modeling of FJSPT under intelligent architecture,the joint optimization of FJSPT and multi-AGV conflict-free route planning,the dynamic scheduling of FJSPT considering new job insertion,and the end-to-end scheduling decision of FJSPT based on deep reinforcement learning method.The main work of this paper is as follows:(1)The complexity analysis and mechanism modeling of FJSPT under intelligent architecture.In order to reveal the in-depth impact of multi-AGV on the work flow,production organization and scheduling of intelligent flexible job shop,the interaction between multiple subsystems in intelligent flexible job shop,the coupling relationship between production and transportation activities and the dynamic characteristics of the system are analyzed.The mechanism model of multi-resource collaborative scheduling problem for intelligent flexible job shop is constructed,which is divided into mathematical model that aims to achieve collaborative scheduling and disjunctive graph model based on sequential decision.The two types of model can guide the extension of the mechanism model and the innovative application of reinforcement learning methods,respectively.Accordingly,meta-heuristic algorithms and deep reinforcement learning methods can be used for model and problem solving.In view of the mature logic and paradigm of meta-heuristic algorithms in the industry,this paper does not list them separately.The basic principle and paradigm of deep reinforcement learning method for solving FJSPT are given separately.(2)Joint optimization of FJSPT and multi-AGV conflict-free route planning(CFRP).To solve the problem that the main decision of flexible job shop scheduling and the sub-decision of multi-AGV transportation system cannot be optimized synchronously and the decision time delay,this paper considers the interaction between the main problem FJSPT and the sub-problem CFRP and differentiates the different effects of AGV distribution sharing and following dispatching strategies on scheme,and build a bi-level programming model to realize the synchronous decision and optimization between collaborative scheduling(including jobs,machines,and AGVs)and AGV distribution route planning in flexile job shop,namely Bi-FJSPT-CFRP model.The model consists of the upper FJSPT main model with the objective of minimizing Makespan and the lower CFRP sub-model with the objective of minimizing travel duration.The differences of the two AGV dispatching strategies are reflected in the dispatching constraints for AGV in the FJSPT model.A bi-level algorithm framework,based on self-learning genetic algorithm(SLGA)that introduce Q-learning into its program and Dijkstra with slied time windows(Dijkstra TW)is designed for solving the complex problem.The effectiveness analyses of the model and algorithm demonstrate that the bi-level programming model can effectively improve the operation efficiency of the system,and the designed Q-learning mechanism can improve the search ability and convergence speed of the algorithm.Sensitivity analyses indicate that the change of the number of jobs and AGVs has a significant impact on the Makespan and resource utilization rate of the system,but there are certain thresholds.The scenarios of different AGV dispatching strategies and AGV configuration suggestions in production practice are given.(3)Dynamic scheduling of FJSPT considering new job insertion.In view of the frequent dynamic events that the new jobs are inserted unexpectedly,in order to make up for the lack of active adjustment mechanism for dynamic disturbances generated by the existing production scheduling system,the dynamic scheduling problem of FJSPT considering new job insertion,namely DFJSPT problem,is studied in this paper.The event-driven proactive-reactive dynamic scheduling method is adopted to deal with the dynamic disturbances in this problem.At the same time,considering the multi-criteria management requirements in practical job shop,a two-stage multi-objective mixed integer programming model is constructed to decide the initial scheduling and rescheduling of DFJSPT problem.The first stage aims to minimize Makespan and maximize the workload balance of equipment to achieve the deterministic initial scheduling of DFJSPT.The second stage introduces the objective of minimizing system instability to achieve the reactive rescheduling of DFJSPT with new job insertion while ensuring the stability of the system.For the difficulty of solving the two-stage multi-objective model,a multi-objective adaptive large-scale neighborhood search algorithm(MOALNS)is innovatively designed by constructing several destroy and repair operator supporting multi-objective optimization and the acceptance criterion based on the dominance.The effectiveness of the model and algorithm is verified by the experience analysis.According to the multi-objective equilibrium analysis,the operation management insights for the dynamic scheduling decision of production practice are proposed.(4)The end-to-end scheduling decision for FJSPT based on deep reinforcement learning method.In order to make up for the problems of complex modeling and inefficient heuristic solution of model-driven intelligent flexible job shop scheduling,with the advantages of deep reinforcement learning with high learning ability and easy generalization application,this paper proposes an end-to-end deep reinforcement learning method for multi-resource collaborative scheduling decision of flexible job shop to improve the intelligent decision-making ability of intelligent factory.Different from the current deep reinforcement learning method based on dispatching rules,belong to the priority scheduling rules that focus on local optimization,in essence,the method presented in this paper makes scheduling decisions directly and quickly according to the current environment status to achieve global optimal.Specifically,firstly,according to the disjunctive graph model in the FJSPT mechanism modeling above(1),the original features of operation,machine and AGV nodes are designed.Secondly,according to the learning mechanism of encoder network in the Transformer framework,the feature extraction network within deep reinforcement learning method for FJSPT based on self-attention mechanism is designed to extract the advanced features of operation,machine and AGV nodes in the disjunctive graph.On this basis,an agent based on Actor-Critic network structure is constructed.The agent is trained by PPO method to obtain the best scheduling strategy.The experimental results show that the trained scheduling agent can effectively solve large-scale examples that it has never seen,and can quickly obtain relatively high-quality scheduling results when it is employed to solve public benchmark instance.In summary,this study aims to explore the coupling relationship between production and transportation activities in intelligent flexible job shops and propose a multi-resource collaborative scheduling method.The proposed method can be applied to advanced production scheduling systems within practical job shop,quickly generating scheduling plans for operations,machines,and AGVs,as well as material transport routes.It continuously updates and adjusts the scheduling scheme based on real-time production status,ensuring efficient and reliable operation of the job shop.The research findings will enrich the theory about production scheduling and provide a reference for job shop management in intelligent flexible manufacturing enterprises.