A Workshop Machine Tool and AGV Scheduling Method Based on Improved Contract Network Protocol and Hybrid Simulation

Abstract

This invention discloses a machine tool and AGV scheduling method for a workshop based on an improved contract network protocol and hybrid simulation. Specifically, it employs discrete event simulation and multi-agent simulation methods to construct a production process sub-model controlling the workshop's product manufacturing process and an agent decision-making sub-model handling workshop decision-making problems. These two models are then integrated to construct a hybrid simulation model for integrated scheduling problems in flexible workshops. Based on this, a contract network protocol is used as the negotiation mechanism between agents. Based on an expanded task buffer pool and a bidding priority mechanism, and combined with the process of task execution, a machine tool and AGV selection strategy conforming to the characteristics of the integrated scheduling problem is formulated. This invention improves upon issues such as bidding broadcasting, malicious bidding, singular task authorization, and lack of feedback on task completion quality during the contract network negotiation process. It realizes the task allocation process of the hybrid simulation model based on the improved contract network, achieving integrated scheduling of machine tools and AGVs in flexible production workshops.

Description

Technical Field

[0001] This invention belongs to the field of AGV system applications, and in particular relates to a workshop machine tool and AGV scheduling method based on an improved contract network protocol and hybrid simulation. Background Technology

[0002] The integrated scheduling problem of machine tools and AGVs in flexible workshops considers AGV scheduling in addition to flexible workshop scheduling, making it more in line with actual production. It mainly includes research on machine tool selection, AGV selection, the execution sequence of workpieces on machine tools and AGVs, and AGV path selection. Currently, many scholars both domestically and internationally have studied this problem, but most still treat machine tool scheduling and AGV scheduling as two separate issues. In contrast, only a few scholars have emphasized the importance of integrated scheduling of machine tools and AGVs in flexible workshops.

[0003] There are three main methods for solving the integrated scheduling problem of machine tools and AGVs in flexible work workshops: mathematical programming, heuristic methods, and metaheuristic methods.

[0004] Mathematical programming methods include mixed integer programming and branch and bound methods. These methods establish a mathematical model by limiting some variables to integers and can use exact algorithms to obtain exact solutions to the problem. For example, Zhang et al. established an integer programming model with transportation resource constraints and variable processing time, aiming to minimize the maximum completion time and the waiting time of workpieces in the buffer. They used a hybrid algorithm combining genetic algorithm and tabu search to solve the problem. The genetic algorithm performs a global search to allocate machine tools and AGVs, while the tabu search performs a local search to improve the quality of the solution and obtain a better solution.

[0005] While mathematical programming methods can find the optimal solution to a problem, their computational complexity is high, especially when dealing with large-scale problems, where their drawbacks are easily exposed. Heuristic and metaheuristic methods, mainly rule-based allocation and intelligent optimization algorithms, can guarantee the optimal or near-optimal solution within an acceptable computational time even when the problem complexity is high. Therefore, many studies have adopted this method to integrate the scheduling of machine tools and AGVs, especially focusing on intelligent algorithms and their hybrid algorithms. For example, Wu Le et al. proposed an improved discrete differential evolution algorithm based on hybrid variable neighborhood search, established a corresponding mathematical model, and solved the problem through a three-layer encoding structure based on processes, machine tools, and AGVs. Chen Kui et al. established an optimization model for the integrated scheduling problem based on minimizing the maximum completion time objective, and then proposed a hybrid discrete particle swarm optimization algorithm combining heuristic initialization methods, competitive learning mechanisms, and random restart mechanisms to optimize the model. Zhou Xin introduced the simulated annealing algorithm into the genetic algorithm, realizing the machine tool selection and sorting problem of process tasks based on two encoding methods: machine tool and process. He added process transfer time during the decoding process, but essentially only implemented machine tool scheduling considering transportation time and did not incorporate the AGV scheduling problem. Homayouni et al. only encoded the process sequencing part and proposed a multi-starting-point biased random key genetic algorithm to solve the problem. They implemented machine tool selection, AGV selection, and process sequencing on machine tools by combining different heuristic rules. Zheng et al. used an adaptive genetic algorithm with dynamically adjustable crossover and mutation rates to solve this type of scheduling problem to improve the algorithm's performance in the search space. They used this genetic algorithm to optimize the workpiece machining route and used heuristic rules to assign tasks to idle AGVs.

[0006] Most scholars have focused on intelligent algorithms and their improvements when solving these problems. While intelligent algorithms can achieve good scheduling results, they often struggle to accurately establish mathematical models, especially when dealing with large-scale, complex manufacturing systems. The resulting mathematical models contain numerous constraints and limitations, making it difficult to accurately reflect the actual situation. Furthermore, intelligent algorithms are prone to getting trapped in local optima during the solution process, often requiring the use of other methods for improvement. Moreover, when dealing with large-scale problems, they may also face issues such as high computational load and long computation time.

[0007] Simulation methods are also an effective approach to solving the workshop scheduling problem. Compared with intelligent algorithms, simulation methods can accurately simulate the complex dynamic behavior of manufacturing systems, more realistically reflect the actual situation of manufacturing systems, and have greater advantages in modeling large-scale complex dynamic systems. However, most studies still use a single simulation method to solve the problem under study, which has the limitation of not being able to fully reflect the characteristics of the problem under study.

[0008] References:

[0009] 【1】Erol R,Sahin C,Baykasoglu A,et al.Amulti-agent based approach todynamic scheduling of machines and automated guided vehicles in manufacturingsystems[J].Applied Soft Computing,2012,12(6):1720-1732.

[0010] 【2】SAHIN C,DEMIRTAS M,EROL R,et al.Amulti-agent based approach todynamic scheduling with flexible processing capabilities[J].Journal ofIntelligent Manufacturing,2017,28(8):1827-1845.

[0011] 【3】BILGE ULUSOY G.ATime Window Approach to Simultaneous Schedulingof Machines and Material Handling System in an FMS[J].Operations Research,1995,43(6):1058-1070.

[0012] 【4】KACEM I,HAMMADI S,BORNE P.Pareto-optimality approach for flexiblejob-shop scheduling problems:hybridization of evolutionary algorithms andfuzzy logic[J].Mathematics and Computers in Simulation,2002,60(3):245-276.

[0013] 【5】KUMAR MVS,JANARDHANA R,RAO CS P.Simultaneous scheduling ofmachines and vehicles in an FMS environment with alternative routing[J].TheInternational Journal of Advanced Manufacturing Technology,2011,53(1-4):339-351. Summary of the Invention

[0014] To address the shortcomings of existing technologies, this invention provides a workshop machine tool and AGV scheduling method based on an improved contract network protocol and hybrid simulation.

[0015] The present invention provides a workshop machine tool and AGV scheduling method based on an improved contract network protocol and hybrid simulation, specifically as follows:

[0016] Step 1: Construct a hybrid simulation model for the integrated scheduling problem of machine tools and AGVs in a flexible workshop.

[0017] A hybrid simulation model is established by combining Discrete Event Simulation (DEMS) and Multi-Agent Simulation (ABMS). During modeling, the workshop is divided into a production process sub-model and an agent decision-making sub-model. The hybrid simulation model for this problem is built using the hybrid simulation method of DEMS combined with ABMS. The production process sub-model built by the DEMS method drives the product flow in the workshop, while the agent decision-making sub-model built by the ABMS method handles the decision-making problem in the workshop.

[0018] The production process sub-model is the main operating line of the hybrid simulation model, controlling and advancing the workpiece's technological process. When a decision event occurs, the decision problem is submitted to the intelligent agent decision sub-model composed of multiple agents for processing through an event triggering mechanism. The obtained decision result then drives the operation of the main line. The interaction between the two models is realized through the hybrid simulation model integration interface. The specific logic of the model operation is as follows:

[0019] 1) When a new order arrives, the system Agent generates the corresponding workpiece Agent and process Agent according to the order information and associates the two. Then, the generated workpiece Agent enters the system through the "order entry" object in the production process sub-model to start the production process.

[0020] 2) When the production process sub-model generates a new process to be executed, it needs to make decisions on the location of the AGV and the machine tool. That is, the workpiece agent enters the "negotiation" module, which is the connection module between the production process sub-model and the agent decision-making sub-model. When the entry event of the "negotiation" module is triggered, the "negotiation required" information is sent to the process agent of the workpiece through the port, thereby submitting the process task allocation problem to the agent decision-making sub-model. The corresponding agent negotiates the task allocation result according to the communication protocol, and reads and changes the variables of the workpiece agent through the active object penetration interface, and stores the returned result in the workpiece agent.

[0021] 3) The workpiece agent that receives the task allocation result is released from the "negotiation" module using the activity object penetration interface and enters the AGV transportation link of the production process sub-model. After being released by the "Queue 1" object according to the priority order, the workpiece agent executes the transportation process. Similarly, the path planning problem in the transportation process is decided by the agent. The workpiece agent enters the "Waiting and Executing Transportation" module, which calls the transportation AGV for it, and the AGV agent decides the path planning to transport the workpiece to the processing machine tool. Accordingly, the workpiece agent enters the processing module of the production process sub-model to perform processing until the processing is completed and a new process is generated. Repeat steps 2) and 3) until all processes are completed.

[0022] 4) The production process sub-model continues to operate until a dynamic disturbance event occurs in the system, and is identified by the "judgment AGV fault" and "judgment machine tool fault" objects. The affected workpiece Agent is returned to the "negotiation" module when the disturbance event occurs. Similarly, the disturbance response problem under dynamic disturbance is submitted to the agent decision sub-model for resolution.

[0023] Step 2: Improved contract network protocol negotiation based on expanded task buffer pool and tender priority.

[0024] 1) Bidding stage

[0025] Solution to the bidding broadcast problem under task buffer pool capacity limitation: In order to reduce communication and computation, the bidding broadcast problem is solved by limiting the number of receivers; when selecting a processing machine for a certain process, bidding information is only sent to machine tools that have the processing capacity for this process and whose buffer pool is not full; when selecting a handling equipment, since all AGVs in the system have handling capacity, bidding information is only sent to AGV Agents whose buffer pool is not full.

[0026] Solution to the problem of vicious bidding under bidding priority: In the hybrid simulation model, after a workpiece enters the system, it first enters the production process sub-model modeled by discrete events. When the system reaches the point where it needs to select a transport AGV and a processing machine for a certain process of the workpiece, the workpiece first flows into the "negotiation" module in the production process sub-model. This module notifies the corresponding workpiece Agent to activate its process Agent, and then the process Agent sends bidding information and AGVs. The Agent and Machine Tool Agent communicate and negotiate until the negotiation ends. Only then is the workpiece released from the "negotiation" process, and the "negotiation" module begins processing the negotiation process for the next Agent. When multiple processes require bidding simultaneously in the system, the workpiece Agents corresponding to these processes are assigned bidding priorities by the "queue" object before entering the negotiation module. Then, they initiate bidding in the negotiation module in order of priority to complete the negotiation. Furthermore, process bidding events generated simultaneously in the system are concurrent events, which are serialized by the simulation engine. That is, they are transformed into information sent in the "negotiation" at the same time according to the bidding priority. In terms of message processing mechanism, these simultaneously sent information are executed in the order they are sent. The system processes only one bidding information at a time until the bidding information is processed and the contract is signed before the processing of the next bidding information begins. After all messages are processed, the simulation clock remains unchanged.

[0027] 2) Bidding stage

[0028] When calculating the bid content of the participant Agent, the task information carried by the pre-booked tasks in the task buffer pool is used to predict the various information for processing the bidding task in the future, and the information is sent to the bid initiator as the bid content.

[0029] 3) Bidding evaluation stage

[0030] Based on the production process sub-model in the hybrid simulation model, the start and end times of the actual handling and processing of each step can be obtained. These actual times are compared with the estimated times, and corrections are made if the deviation is too large, so as to improve the accuracy of the calculation based on the reservation information. At the same time, the actual task completion status is compared with the contract information.

[0031] The expanded task buffer pool is specifically as follows:

[0032] 1) The number of task types that can be accommodated in the buffer pool has been expanded.

[0033] The machine tool agent task buffer pool can accommodate tasks that are being processed by the machine tool, tasks that have arrived at the machine tool buffer and are waiting to be processed, tasks that are being transported to the machine tool by AGV, and tasks that are waiting to be transported to the machine tool in other machine tool buffers; the AGV agent task buffer pool can accommodate tasks that are being transported and tasks that are waiting to be transported at the machine tool buffer or material loading and unloading point.

[0034] 2) Expand the function of utilizing task information in the task buffer pool.

[0035] Each task stored in the task buffer pool carries relevant information about its production process, including the process name, the name of the workpiece to which it belongs, the machine tool, the transport AGV, the start time of transport, the end time of transport, the start time of processing, and the end time of processing. This information is searchable and usable, and can be called up wherever this information is needed during system operation.

[0036] Step 3: Machine tool and AGV selection based on expanded task buffer pool.

[0037] 1) AGV Selection Strategy

[0038] The evaluation metric is minimizing the completion time.

[0039] Once a process is initiated, in order to process it as quickly as possible, the AGV with the earliest handling time should be selected for transfer. Therefore, when the AGV Agent receives the bidding information from the process Agent, the content of the tender document submitted to it is the estimated value of the earliest start time of the process. The calculation process of this value is shown in equations (1) to (2):

[0040]

[0041]

[0042] In the formula, The estimated time for the start of the j-th process of the i-th workpiece on the k-th AGV; The estimated waiting time for the j-th process of the i-th workpiece on the k-th AGV is given. This is the estimated end time of the last pending task in the task buffer pool of the k-th AGV. Let t be the no-load travel time of the k-th AGV from its current position to the j-th process position of the i-th workpiece; t is the current time when the system is running.

[0043] After receiving all the bids returned by the bidding AGVs, the process Agent in the bidding state enters the bid decision state. According to formula (3), it selects the AGV with the earliest transport time as the transport equipment and signs a contract with it. The contract time for starting transport is the earliest start time submitted by the AGV. Accordingly, the process is placed in the task buffer pool of the contracted AGV as the last task to be executed, and the information it carries is used for the next bid calculation of the contracted AGV.

[0044]

[0045] In the formula, A ij The contracted AGV is for the j-th process of the i-th workpiece.

[0046] The actual execution results are used to provide feedback on the estimated value, detect whether there is a deviation and correct the deviation; the deviation detection formula is shown in Equation (4). When the formula is not satisfied, the deviation accumulates to the upper limit, and the corresponding AGV Agent updates the estimated start time of all remaining tasks in its task buffer pool to correct the deviation.

[0047]

[0048] In the formula, Let be the actual end time of the j-th process of the i-th workpiece on the k-th AGV; The estimated time for the completion of the j-th process of the i-th workpiece on the k-th AGV; threshold is the allowable deviation threshold of the system.

[0049] 2) Machine tool selection strategy

[0050] After the process agent completes the selection of the transport AGV, it continues to send bidding information to the machine tools in the machine tool set. In accordance with the goal of minimizing the completion time, when selecting machine tools, it is necessary to select the machine tool with the earliest completion time. Therefore, the bidding document submitted by the machine tool agent is the estimated completion time of the bidding process, and its calculation process is shown in equations (5) to (8).

[0051]

[0052]

[0053]

[0054]

[0055] In the formula, This is the estimated time for the completion of the j-th process of the i-th workpiece on the m-th machine tool; This is the estimated start time for the j-th process of the i-th workpiece on the m-th machine tool; Let be the processing time of the j-th process for the i-th workpiece on the m-th machine tool; The estimated time for the j-th process of the i-th workpiece to wait for processing on the m-th machine tool; This is the estimated end time of the last pending task in the task buffer pool of the m-th machine tool; Let be the load travel time of the winning AGV from the current position of the bidding process to the machine tool in the j-th process of the i-th workpiece.

[0056] Before calculating the value, the machine tool agent needs to verify whether the machine tool in the previous process of the bidding process is the same machine tool as the bidding process. If it is the same machine tool, then there is no AGV handling step in the execution of that process. In this case, the calculation process... All are 0.

[0057] After receiving all the tenders returned by the tendered machine tools, the process agent in the bidding state enters the tender decision state. According to formula (9), it selects the machine tool with the earliest end processing time as the processing equipment and signs a contract with it. The contract end processing time is the earliest end processing time submitted by the machine tool. Accordingly, the process is placed in the task buffer pool of the contracted machine tool as the last task to be executed. The information it carries is used for the next bidding calculation. Finally, the process agent also needs to check whether the winning machine tool is the processing machine tool of the previous process of the related workpiece agent. If the processing machine tools of the two processes are the same, there is no need for AGV transportation. The previous winning AGV agent needs to be notified to delete this process from its task buffer pool.

[0058]

[0059] In the formula, M ij The contracted machine tool is the machine tool for the j-th process of the i-th workpiece.

[0060] Similarly, deviation detection and correction are required. The deviation detection method is shown in Equation (10). When the equation is not satisfied, the deviation accumulates to the upper limit. The corresponding machine tool agent updates the estimated end time of all remaining tasks in its task buffer pool to correct the deviation.

[0061]

[0062] In the formula, Let be the actual completion time of the j-th process of the i-th workpiece on the m-th machine tool.

[0063] Step 4: Task allocation based on the hybrid simulation model of the improved contract network.

[0064] 1) Coordination of overall goals

[0065] Global goals are coordinated by prioritizing them, including the following two priorities:

[0066] Priority of issuance

[0067] Each process is assigned a priority for the bidding order. This priority is based on the total number of remaining processes for the workpiece. The more remaining processes a workpiece has, the higher its priority and the earlier its negotiation order.

[0068] AGV and Machine Tool Agent Task Execution Priority

[0069] The contract execution time for each process task won by the AGV Agent and the machine tool Agent through bidding is calculated based on the existing task information in their task buffer pool. Then, newly won tasks are placed into the buffer pool to participate in the bidding for the next task. In other words, the execution order of these tasks is their queuing order in the buffer pool. Correspondingly, the process agent carries the priority that determines its execution order. When selecting the next task to execute, the AGV and the machine tool only need to select from their task buffer pools in sequence.

[0070] 2) Task allocation process during hybrid simulation model runtime

[0071] When an order arrives, the system agent generates corresponding workpiece agents and process agents based on the order information and associates them. The generated workpiece agent enters the system through the "Order Entry" object in the production process sub-model of the hybrid simulation model, initiating the production process. When the system reaches the point where task allocation for a specific process of a workpiece is required, the workpiece agent flows into the "Negotiation" object, submitting the task allocation decision problem to the intelligent agent decision sub-model for processing, and the corresponding process agent is activated. The activated process agent initiates the improved contract network protocol negotiation process according to the bidding priority assigned to it in the "queue," determining its handling based on the established AGV and machine tool selection strategy. AGVs and machine tools are stored in the task buffer pools of the contracted AGV and machine tool according to the task execution priority, waiting for execution. When the contracted AGV is idle, the workpiece agent flows into "Get AGV" to obtain the contracted AGV. The "AGV Transport" and "AGV Release" objects control the AGV to transport the workpiece to the machine tool and release the AGV. Similarly, when the contracted machine tool is idle, the workpiece flows into "Get Machine Tool", "Process", and "Release Machine Tool" in sequence to complete the processing operation. When the process is completed, the corresponding AGV and machine tool correct the information in the task buffer pool according to the actual execution result of the task. At the same time, the workpiece agent returns from "Judgment New Process" to start the next process flow.

[0072] The beneficial technical effects of this invention are as follows:

[0073] This invention employs discrete event simulation and multi-agent simulation methods to construct a production process sub-model controlling the product manufacturing process in the workshop and an agent decision-making sub-model handling workshop decision-making problems. These two models are then integrated to construct a hybrid simulation model for the integrated scheduling problem in flexible workshops. Based on this, a contract network protocol is used as the negotiation mechanism between agents. By expanding the task buffer pool and implementing a bidding priority mechanism, issues such as bidding broadcasting, vicious bidding, singular task authorization, and lack of feedback on task completion quality are addressed in the contract network negotiation process. Furthermore, considering the process of task execution, a machine tool and AGV selection strategy that conforms to the characteristics of the integrated scheduling problem is formulated. This enables the task allocation process based on the improved contract network in the hybrid simulation model, achieving integrated scheduling of machine tools and AGVs in flexible production workshops. Finally, simulation results demonstrate the effectiveness and superiority of the proposed method. Attached Figure Description

[0074] Figure 1 This invention presents a hybrid simulation model integrating DEMS and ABMS.

[0075] Figure 2 This is a schematic diagram of the task buffer pool and its expansion structure.

[0076] Figure 3 To simplify the contract network protocol process.

[0077] Figure 4 This invention improves the contract network protocol negotiation process.

[0078] Figure 5 This describes the process of executing the steps in this invention.

[0079] Figure 6 This invention improves the task allocation process of hybrid simulation models under contract networks.

[0080] Figure 7 This is a comparison of simulation results when t / p > 0.25 in this invention.

[0081] Figure 8 This is a comparison of simulation results when t / p < 0.25 in this invention.

[0082] Figure 9 This is a 2D layout diagram of the workshop model for an example. Detailed Implementation

[0083] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0084] The present invention provides a workshop machine tool and AGV scheduling method based on an improved contract network protocol and hybrid simulation, specifically as follows:

[0085] Step 1: Construct a hybrid simulation model for the integrated scheduling problem of machine tools and AGVs in a flexible workshop.

[0086] By combining Discrete Event Simulation (DEMS) and Multi-Intelligent Simulation (ABMS), a hybrid simulation model is established, which is used to construct a hybrid simulation model for the integrated scheduling problem of machine tools and AGVs in flexible workshops. Figure 1 As shown.

[0087] Since solving the integrated scheduling problem of machine tools and AGVs involves two levels of issues: driving the workshop production process and solving decision-making problems, the workshop is divided into a production process sub-model and an agent decision-making sub-model during modeling. A hybrid simulation model for this problem is built using a hybrid simulation method combining DEMS and ABMS. The production process sub-model built by the DEMS method drives the flow of products in the workshop, while the agent decision-making sub-model built by the ABMS method handles the decision-making problems in the workshop.

[0088] The production process sub-model is the main operating line of the hybrid simulation model, controlling and advancing the workpiece's technological process. When a decision event occurs, the decision problem is submitted to the intelligent agent decision sub-model composed of multiple agents for processing through an event triggering mechanism. The obtained decision result then drives the operation of the main line. The interaction between the two models is realized through the hybrid simulation model integration interface. The specific logic of the model operation is as follows:

[0089] 1) When a new order arrives, the system Agent generates the corresponding workpiece Agent and process Agent according to the order information and associates the two. Then, the generated workpiece Agent enters the system through the "order entry" object in the production process sub-model to start the production process.

[0090] 2) When the production process sub-model generates a new process to be executed, it needs to make decisions on the location of the AGV and the machine tool. That is, the workpiece agent enters the "negotiation" module, which is the connection module between the production process sub-model and the agent decision-making sub-model. When the entry event of the "negotiation" module is triggered, the "negotiation required" information is sent to the process agent of the workpiece through the port, thereby submitting the process task allocation problem to the agent decision-making sub-model. The corresponding agent negotiates the task allocation result according to the communication protocol, and reads and changes the variables of the workpiece agent through the active object penetration interface, and stores the returned result in the workpiece agent.

[0091] 3) The workpiece agent that receives the task allocation result is released from the "negotiation" module using the activity object penetration interface and enters the AGV transportation link of the production process sub-model. After being released by the "Queue 1" object according to the priority order, the workpiece agent executes the transportation process. Similarly, the path planning problem in the transportation process is decided by the agent. The workpiece agent enters the "Waiting and Executing Transportation" module, which calls the transportation AGV for it, and the AGV agent decides the path planning to transport the workpiece to the processing machine tool. Accordingly, the workpiece agent enters the processing module of the production process sub-model to perform processing until the processing is completed and a new process is generated. Repeat steps 2) and 3) until all processes are completed.

[0092] 4) The production process sub-model continues to operate until a dynamic disturbance event occurs in the system, and is identified by the "judgment AGV fault" and "judgment machine tool fault" objects. The affected workpiece Agent is returned to the "negotiation" module when the disturbance event occurs. Similarly, the disturbance response problem under dynamic disturbance is submitted to the agent decision sub-model for resolution.

[0093] Step 2: Improved contract network protocol negotiation based on expanded task buffer pool and tender priority.

[0094] To address issues such as broadcast bidding, malicious bidding, limited task authorization, and lack of feedback on task completion quality during the contract network negotiation process, an improved contract network protocol is proposed based on expanding the task buffer pool and prioritizing bidding.

[0095] (1) Task buffer pool mechanism

[0096] Considering that traditional contract network protocols require devices to complete their current tasks before participating in bidding, leading to significant meaningless communication and the inability to obtain optimal allocation schemes due to devices currently executing tasks failing to participate in bidding, this paper introduces a task buffer pool for the participant agent. This allows the agent to accept multiple task authorizations, increasing the likelihood of assigning tasks to the best executor. The task buffer pool, essentially a task queue, ensures that after accepting one task authorization, the participant agent can continue to accept other task authorizations. New tasks are compared with existing tasks in the buffer pool, and the new task is stored in the appropriate position. Once the current task is completed, tasks in the buffer pool are executed sequentially.

[0097] When Zhu Yunfei et al. used the Contract Network protocol to solve the workshop scheduling problem, they increased the opportunities for workstations to bid and accept task authorization by adding a task buffer pool to the machine tool agent. However, their proposed task buffer pool only contained the tasks to be processed in the input buffer. Wang Shuang, in scheduling machine tools and AGVs in a smart workshop, only set up a task buffer pool for the AGV agent and did not specify which tasks were included in the task buffer pool. Furthermore, none of the above literature utilized the task information stored in the task buffer pool.

[0098] (2) Expand the task buffer pool mechanism

[0099] When addressing the integrated scheduling problem of machine tools and AGVs in flexible work workshops, independent task buffer pool modules were added for both machine tools and AGV agents. The functionality of the task buffer pools was expanded in the following two aspects:

[0100] 1) The number of task types that can be accommodated in the buffer pool has been expanded.

[0101] The machine tool agent task buffer pool can hold tasks such as Figure 2 The large elliptical frame shows tasks being processed by the machine tool, tasks that have arrived at the machine tool buffer and are waiting to be processed, tasks being transported to the machine tool by the AGV, and tasks waiting to be transported to the machine tool in other machine tool buffers. The AGV Agent task buffer pool can hold tasks such as... Figure 2 The small and medium-sized ovals indicate tasks that are being moved and tasks that are waiting to be moved at machine tool buffer zones or material loading / unloading points.

[0102] 2) Expand the function of utilizing task information in the task buffer pool.

[0103] Each task stored in the task buffer pool carries relevant information about its production process, including the process name, the name of the workpiece to which it belongs, the machine tool, the transport AGV, the start time of transport, the end time of transport, the start time of processing, and the end time of processing. This information is searchable and usable, and can be called up wherever this information is needed during system operation.

[0104] Among them, M m Let A represent the m-th machine tool. a P represents the a-th AGV; i (M m Pe represents the i-th task being processed on the m-th machine tool; i (M m A a Ps represents the i-th task waiting to be transported in the output buffer of the m-th machine tool, and this task is transported by the a-th AGV; i (Mm ) represents the i-th task waiting to be processed in the input buffer of the m-th machine tool; T ai (→M m ) indicates that the a-th AGV is carrying the i-th task, and the machine tool for processing the task is the m-th machine tool.

[0105] (3) Contract network protocol improvement method based on extended task buffer pool and tendering priority

[0106] In the basic contract network and its extended forms, the signing process involves multiple negotiation rounds. To reduce communication volume, this paper adopts the simplified contract network protocol proposed by Sang Zelei (hereinafter referred to as JCNP), such as... Figure 3 As shown, the number of communications during the negotiation process is reduced to three. In the context of the integrated scheduling problem of machine tools and AGVs in a flexible workshop, the initiator in the contract network corresponds to the process agent in this paper, and the participants correspond to the machine tool agent and the AGV agent.

[0107] To address the issues existing in the basic contract net protocol, and considering the characteristics of the integrated scheduling problem, an improved contract net protocol process is proposed based on a hybrid simulation model and an extended task buffer pool, as follows: Figure 4 As shown

[0108] 1) Bidding stage

[0109] Solution to the bidding broadcast problem under task buffer pool capacity limitation: In order to reduce communication and computation, the bidding broadcast problem is solved by limiting the number of receivers; when selecting a processing machine for a certain process, bidding information is only sent to machine tools that have the processing capacity for this process and whose buffer pool is not full; when selecting a handling equipment, since all AGVs in the system have handling capacity, bidding information is only sent to AGV Agents whose buffer pool is not full.

[0110] Solution to the problem of vicious bidding under bidding priority: In the hybrid simulation model, after a workpiece enters the system, it first enters the production process sub-model modeled by discrete events. When the system reaches the point where it needs to select a transport AGV and a processing machine for a certain process of the workpiece, the workpiece first flows into the "negotiation" module in the production process sub-model. This module notifies the corresponding workpiece Agent to activate its process Agent, and then the process Agent sends bidding information and AGVs. The Agent and Machine Tool Agent communicate and negotiate until the negotiation ends. Only then is the workpiece released from the "negotiation" process, and the "negotiation" module begins processing the negotiation for the next Agent. When multiple processes requiring bidding occur simultaneously in the system, the workpiece Agents corresponding to these processes are assigned bidding priorities by a "queue" object before entering the negotiation module. They then initiate bidding sequentially in the negotiation module according to their priorities. Furthermore, process bidding events generated simultaneously within the system are concurrent events, processed serially by the simulation engine. This transforms them into information sent sequentially in the "negotiation" process according to their bidding priorities at the same time. In terms of message processing, these simultaneously sent messages are executed in the order they were sent. The system processes only one bidding message at a time until that message is processed and a contract is signed before starting processing the next one. The simulation clock remains unchanged after all messages are processed. This avoids the vicious bidding situation where multiple bid initiators exist simultaneously in the system, participants bid to multiple initiators, and ultimately win multiple tasks simultaneously, resulting in the inability to execute tasks on time.

[0111] 2) Bidding stage

[0112] The existence of a task buffer pool ensures that tasks are negotiated as soon as they are generated, reducing task waiting time. When calculating the bid content of participating agents, task information (such as estimated end time of handling and estimated end time of processing) from existing pre-booked tasks in the task buffer pool is used to predict various information for processing the bidding task in the future, and this information is sent to the bid initiator as part of the bid content. Furthermore, the specific content of the bids exchanged during the negotiation process should be reasonably set according to the actual problem, tailored to the specific issues.

[0113] 3) Bidding evaluation stage

[0114] Because the benchmark calculation for competitive processes is based on the pre-booked information in the task buffer pool—essentially a forecast of future execution—the actual time information for each process may deviate from the contract time, requiring correction. Using the production process sub-model within the hybrid simulation model, the start and end times of actual handling and processing for each step can be obtained. Comparing these actual times with the estimated times and correcting for significant deviations improves the accuracy of benchmark calculations based on pre-booked information. Simultaneously, comparing the actual task completion status with contract information provides feedback on task completion quality.

[0115] Step 3: Machine tool and AGV selection based on expanded task buffer pool.

[0116] 1) Process Execution Analysis

[0117] When addressing integrated scheduling problems, most scholars have only developed machine tool selection strategies that consider AGV load transportation time, neglecting the impact of AGV idle transportation time, and simply selecting AGVs based on their idle status. Furthermore, the results obtained through formulas are estimated values, lacking feedback information from actual execution. Therefore, it is necessary to develop more reasonable negotiation content for contract network protocols in the tender documents, taking into account the characteristics of integrated scheduling problems.

[0118] Based on the workpiece-centric production process sub-model described above, it can be seen that regardless of when or where a process occurs, it must first be transported by an AGV (except for cases where the processing machine is the same) before it can enter the processing machine for processing, until the processing is completed, thus ending the current process and generating a new process. Figure 5 As shown, this is a continuous process. In this process, machine tool selection and AGV selection are interdependent. For example, different AGV selections will result in different start transport times, thus affecting the start processing time of the workpiece on the machine tool. Conversely, different machine tool selections will result in different AGV destinations, thus resulting in different end transport times. In the entire process, AGV selection has the earliest impact; therefore, the transport AGV for this process should be determined first before machine tool selection. Furthermore, if machine tools are selected first, when formulating the machine tool selection strategy, since the transport AGV for the process is not yet determined, the idle travel time of the transport AGV cannot be determined; only the load transport time from the workpiece position to the machine tool position can be considered.

[0119] 2) AGV Selection Strategy

[0120] Task allocation strategies serve the goals pursued by the entire manufacturing system. The evaluation metric is minimizing completion time. When formulating AGV and machine tool selection strategies for processes and arranging their execution sequence on the two resources, these strategies must also serve this goal.

[0121] Once a process is initiated, in order to process it as quickly as possible, the AGV with the earliest handling time should be selected for transfer. Therefore, when the AGV Agent receives the bidding information from the process Agent, the content of the tender document submitted to it is the estimated value of the earliest start time of the process. The calculation process of this value is shown in equations (1) to (2):

[0122]

[0123]

[0124] In the formula, The estimated time for the start of the j-th process of the i-th workpiece on the k-th AGV; The estimated waiting time for the j-th process of the i-th workpiece on the k-th AGV is given. This is the estimated end time of the last pending task in the task buffer pool of the k-th AGV. Let t be the no-load travel time of the k-th AGV from its current position to the j-th process position of the i-th workpiece; t is the current time when the system is running.

[0125] After receiving all the bids returned by the bidding AGVs, the process Agent in the bidding state enters the bid decision state. According to formula (3), it selects the AGV with the earliest transport time as the transport equipment and signs a contract with it. The contract time for starting transport is the earliest start transport time submitted by the AGV. Accordingly, the process is placed in the task buffer pool of the contracted AGV as the last task to be executed. The information it carries (start transport time, end transport time, machine tool position, etc.) is used for the next bid calculation of the contracted AGV.

[0126]

[0127] In the formula, A ij The contracted AGV is for the j-th process of the i-th workpiece.

[0128] Before a process is actually executed, the results calculated by the formula are all predictions for the future, which may deviate from the actual execution results. Furthermore, the calculation based on the values ​​of the pre-booked task information in the task buffer pool will accumulate deviations with each new pre-booked task added to the task buffer pool, and the obtained prediction results will gradually lose their reference value. Therefore, it is necessary to use the actual execution results to provide feedback on the estimated value, detect whether the deviation exists, and correct the deviation. The deviation detection formula is shown in Equation (4). When the formula is not satisfied, the deviation accumulates to the upper limit, and the corresponding AGV Agent updates the estimated start time of all remaining tasks in its task buffer pool to correct the deviation.

[0129]

[0130] In the formula, Let be the actual end time of the j-th process of the i-th workpiece on the k-th AGV; The estimated time for the completion of the j-th process of the i-th workpiece on the k-th AGV; threshold is the allowable deviation threshold of the system.

[0131] 2) Machine tool selection strategy

[0132] After the process agent completes the selection of the transport AGV, it continues to send bidding information to the machine tools in the machine tool set. In accordance with the goal of minimizing the completion time, when selecting machine tools, it is necessary to select the machine tool with the earliest completion time. Therefore, the bidding document submitted by the machine tool agent is the estimated completion time of the bidding process, and its calculation process is shown in equations (5) to (8).

[0133]

[0134]

[0135]

[0136]

[0137] In the formula, This is the estimated time for the completion of the j-th process of the i-th workpiece on the m-th machine tool; This is the estimated start time for the j-th process of the i-th workpiece on the m-th machine tool; Let be the processing time of the j-th process for the i-th workpiece on the m-th machine tool; The estimated time for the j-th process of the i-th workpiece to wait for processing on the m-th machine tool; This is the estimated end time of the last pending task in the task buffer pool of the m-th machine tool; Let be the load travel time of the winning AGV from the current position of the bidding process to the machine tool in the j-th process of the i-th workpiece.

[0138] As can be seen from equations (5) to (7), the determination of the machine tool's value is influenced by the AGV selected in the previous step, and only in this step can the end transport time of the contracted AGV in this process be determined. Specifically, before calculating the value, the machine tool agent needs to verify whether the machine tool in the previous process of the bidding process is the same machine tool. If it is, then there is no AGV transport step in the execution of this process, and the calculation process... All are 0.

[0139] After receiving all the tenders returned by the tendered machine tools, the process agent in the bidding state enters the tender decision state. According to formula (9), it selects the machine tool with the earliest end processing time as the processing equipment and signs a contract with it. The contract end processing time is the earliest end processing time submitted by the machine tool. Accordingly, the process is placed in the task buffer pool of the contracted machine tool as the last task to be executed. The information it carries (start processing time, end processing time, etc.) is used for the next bidding calculation. Finally, the process agent also needs to check whether the winning machine tool is the processing machine tool of the previous process of the related workpiece agent. If the processing machine tools of the two processes are the same, there is no need for AGV transportation. It is necessary to notify the previous winning AGV agent to delete this process from its task buffer pool.

[0140]

[0141] In the formula, M ij The contracted machine tool is the machine tool for the j-th process of the i-th workpiece.

[0142] Similarly, the tender content and the final contract signing time obtained by using the above formula are estimated results, which may deviate from the actual execution time of the process. Deviation detection and correction are required. The deviation detection method is shown in Equation (10). When the formula is not satisfied, the deviation accumulates to the upper limit. The corresponding machine tool agent updates the estimated end time of all remaining tasks in its task buffer pool to correct the deviation.

[0143]

[0144] In the formula, Let be the actual completion time of the j-th process of the i-th workpiece on the m-th machine tool.

[0145] Step 4: Task allocation based on the hybrid simulation model of the improved contract network.

[0146] 1) Coordination of overall goals

[0147] By formulating a selection strategy for AGVs and machine tools, the optimal transport AGV and processing machine tool for each process are determined. However, in this process, the process agent is selfish, aiming to complete its processing as early as possible. This achieves a local optimum for a single agent's goal, which may not necessarily lead to the optimal solution for the overall system operation. Other methods are needed to coordinate global goals to ensure a better overall system operation result. This invention coordinates global goals by prioritizing them, including the following two priority levels:

[0148] Priority of issuance

[0149] Due to the serialization function of the simulation engine, bidding actions generated at the same time are executed sequentially. However, the execution order of these bidding actions affects the minimum completion time target pursued by the system. To better achieve the system goal, each action is assigned a bidding priority, which is based on the total number of remaining actions for the workpiece. Workpieces with more remaining actions have higher priority and are negotiated earlier. This avoids workpieces with a large number of actions being executed last, where the unfinished actions of the remaining workpiece are still being processed after all other workpieces in the system have been completed, thus severely impacting the system's completion time.

[0150] AGV and Machine Tool Agent Task Execution Priority

[0151] The contract execution time for each process task won by the AGV Agent and the machine tool Agent through bidding is calculated based on the existing task information in their task buffer pool. Then, newly won tasks are placed into the buffer pool to participate in the bidding for the next task. In other words, the execution order of these tasks is their queuing order in the buffer pool. Correspondingly, the process agent carries the priority that determines its execution order. When selecting the next task to execute, the AGV and the machine tool only need to select from their task buffer pools in sequence.

[0152] 2) Task allocation process during hybrid simulation model runtime

[0153] The task allocation process in the hybrid simulation model based on the improved contract net protocol is as follows: Figure 6As shown, when an order arrives, the system agent generates a corresponding workpiece agent and a process agent based on the order information and associates them. The generated workpiece agent enters the system through the "Order Entry" object in the production process sub-model of the hybrid simulation model, initiating the production process. When the system reaches the point where task allocation for a specific process of a workpiece is required, the workpiece agent flows into the "Negotiation" object, submitting the task allocation decision problem to the intelligent agent decision sub-model for processing, and the corresponding process agent is activated. The activated process agent initiates the improved contract network protocol negotiation process according to the bidding priority assigned to it in the "queue," determining its AGV and machine tool selection strategy. The AGV and machine tool are transported together and stored in the task buffer pools of the contracted AGV and machine tool according to the task execution priority, awaiting execution. When the contracted AGV is idle, the workpiece Agent flows into "Get AGV" to obtain the contracted AGV. The "AGV Transport" and "AGV Release" objects control the AGV to transport the workpiece to the machine tool and release the AGV. Similarly, when the contracted machine tool is idle, the workpiece flows into "Get Machine Tool", "Process", and "Release Machine Tool" in sequence to complete the processing operation. When the process is completed, the corresponding AGV and machine tool correct the information in the task buffer pool according to the actual execution result of the task. At the same time, the workpiece Agent returns from "Judgment New Process" to start the next process flow.

[0154] Example:

[0155] To test the effectiveness of the proposed method, simulation verification of the method was performed using literature examples, and the results were compared and analyzed with those of other algorithms. During the experiment, a hybrid simulation model of the experimental cases was built using the multi-method modeling and simulation software AnyLogic. Based on this, the task allocation method program based on the contract network protocol and the improved contract network protocol in this chapter was implemented. Finally, the results obtained were compared and analyzed with those of other algorithms. The experimental cases included the Bilge[3] example and the Kacem[4] example.

[0156] (1) Simulation and Result Analysis of the Bilge Case

[0157] 1) Experimental Case

[0158] The test case comes from the research of Bilge[3] and Kumar[5]. This case was first defined by Bilge and is independent of machine tool flexibility. Kumar et al. extended the problem by increasing machine tool flexibility to test the effectiveness of their differential evolution algorithm in solving the machine tool and AGV integrated scheduling problem. This invention uses their dataset to analyze the performance of the method in this chapter.

[0159] 2) Results Comparison and Analysis

[0160] To verify the effectiveness of the hybrid simulation model method and the improved contract net protocol based on hybrid simulation in terms of solution quality, the machine tool and AGV task allocation method in the FMAS system developed by Cenk et al.[2] was first adopted to realize the task allocation based on the hybrid simulation model, thereby verifying the effectiveness of the hybrid simulation model. For ease of representation, this paper names it the HS-CNP method. Secondly, the improved contract net protocol method in this chapter was used to simulate and test the selected dataset under the hybrid simulation model to verify the improvement effect of the contract net. For ease of representation, this paper names it the HS-pCNP method.

[0161] During the experiment, the input simulation variables were four different layout scenarios, 10 types of workpiece sets, and different t / p ratios (the ratio of transportation time to processing time) in the test case. The total production time (maximum completion time) was selected as the evaluation index, and the optimization results of Kumar, Erol[1], and Cent et al. were used for comparison for evaluation. The comparison results are shown in Tables 1 and 2. In the table, EX in the experiment column represents an experiment, the last number represents the layout, and the middle number represents the workpiece set. For example, EX11 indicates that the experiment was conducted with the first type of workpiece set input in the first layout scenario. The extra 0 or 1 at the end of the experiment column in Table 2 indicates that the processing time was increased by 2 times or 3 times to obtain different t / p ratios. For example, EX110 indicates that the experiment was conducted with the first type of workpiece set input with the processing time increased by 2 times in the first layout scenario.

[0162] Table 1 Comparison of simulation results under different scenarios when t / p > 0.25

[0163]

[0164]

[0165] Table 1 shows the simulation results for different scenarios when t / p > 0.25. It can be seen that 95% of the data obtained by the HS-CNP method is superior to FMS. Since both methods use the same approach at the task allocation level, the only difference is that the HS-CNP method is based on a hybrid simulation model, while FMAS uses a pure MAS structure. This proves the feasibility of the hybrid simulation model method in solving the flexible job shop integrated scheduling problem, and that the results obtained are superior. Furthermore, 52.5% of the data obtained by the HS-CNP method is superior to FDE, compared to only 12.5% ​​of the data obtained by FMS, representing an optimization performance improvement of approximately four times.

[0166] Furthermore, the data obtained using the improved contract net method (HS-pCNP) is superior to HS-CNP in 67.5% of cases, demonstrating the effectiveness of the improved contract net. Additionally, it is superior to FDE in 57.5% of cases, representing a further improvement over HS-CNP and further proving the effectiveness of the improved contract net. Similarly, HS-pCNP is superior to FMS in 95% of cases. Compared to MAS, both methods are nearly 100% superior.

[0167] Table 2 Comparison of simulation results under different scenarios when t / p < 0.25

[0168]

[0169]

[0170] Table 2 shows the simulation results for different scenarios when t / p < 0.25. It can be observed that the data obtained by the HS-CNP method is only superior to FMS in 50% of cases and to FDE in 45.2% of cases, while HS-pCNP achieves 73.8% and 69% respectively. The improved contract net method is significantly better than the unimproved state, which is directly reflected in the 81% data where the HS-pCNP method outperforms HS-CNP, demonstrating the effectiveness of the HS-pCNP contract net protocol improvement method. When t / p < 0.25, the proportion of transportation time in the entire workpiece processing is relatively small, and the improvement effect of the HS-CNP and HS-pCNP methods is lower compared to when t / p > 0.25. This also proves that the method proposed in this invention is more advantageous in solving integrated scheduling problems where transportation time cannot be ignored.

[0171] To more intuitively illustrate the differences in results between different methods, 41 sets of data under two t / p ratios were extracted for graphical comparison, such as... Figures 7-8 As shown, it can be observed that the HS-CNP and HS-pCNP methods yielded better experimental results.

[0172] (2) Kacem Case Simulation and Result Analysis

[0173] 1) Experimental Case

[0174] Since the experimental layout in the Bilge case is relatively simple (4 machine tools, 2 AGVs), in order to further verify the performance of the method in this chapter in a larger scenario, the standard instances proposed by kacem[4] are used: kacem8X8, kacem10X10, kacem15X10 (i.e., nxm, where n represents the number of workpieces and m represents the number of machine tools) to continue the simulation experiment. Referring to the layout of scenario 1 in the Bilge case, the workshop layout and AGV transportation schedule of the kacem case are constructed. The 2D layout of the workshop model built using AnyLogic is as follows: Figure 9 As shown in Table 3, the AGV transportation time is as follows. For the 8x8 scenario, only the following needs to be removed. Figure 9 You only need to find M5, M10 and their connection paths.

[0175] Table 3 Transportation Timetable

[0176]

[0177] 2) Results Comparison and Analysis

[0178] During the experiment, the number of AGVs was set to 5 in the 8x8 scenario and 7 in the 10x10 and 15x10 scenarios. The simulation results for the three scenarios are compared in Table 4. It can be seen that the results obtained by the HS-PCNP method are all better than those of HS-CNP, with an average improvement of 27% in the completion time index. This proves the effectiveness of the method of the present invention in larger experimental scenarios and the advantages of the improved contract net method.

[0179] Table 4 Comparison of HS-CNP and HS-PCNP results in different scenarios

[0180]

[0181]

[0182] This invention addresses the static integrated scheduling problem of machine tools and AGVs in flexible workshops, proposing a hybrid simulation model solution method based on real-time allocation of process tasks using a contract network protocol. First, discrete event simulation and multi-agent simulation methods are employed to construct a production process sub-model controlling the workshop's product manufacturing process and an agent decision-making sub-model handling workshop decision-making problems. These two models are then integrated to construct a hybrid simulation model for the integrated scheduling problem in flexible workshops. Based on this, a contract network protocol is used as the negotiation mechanism between agents. By expanding the task buffer pool and implementing a bidding priority mechanism, issues such as bidding broadcasting, vicious bidding, singular task authorization, and lack of feedback on task completion quality are addressed during the contract network negotiation process. Furthermore, considering the process task execution, a machine tool and AGV selection strategy tailored to the characteristics of the integrated scheduling problem is formulated, realizing the task allocation process based on the improved contract network in the hybrid simulation model, thus achieving integrated scheduling of machine tools and AGVs in the flexible production workshop. Finally, two simulation examples are used to verify the effectiveness and superiority of the proposed method.

Claims

1. A workshop machine tool and AGV scheduling method based on an improved contract network protocol and hybrid simulation, characterized in that, Specifically: Step 1: Construct a hybrid simulation model for the integrated scheduling problem of machine tools and AGVs in a flexible workshop; A hybrid simulation model is established by combining Discrete Event Simulation (DEMS) and Multi-Agent Simulation (ABMS). During modeling, the workshop is divided into a production process sub-model and an agent decision-making sub-model. The hybrid simulation model for this problem is built using the DEMS-ABMS hybrid simulation method. The production process sub-model built by the DEMS method drives the product flow in the workshop, while the agent decision-making sub-model built by the ABMS method handles the decision-making problem in the workshop. The production process sub-model is the main operating line of the hybrid simulation model, controlling and advancing the workpiece's technological process. When a decision event occurs, the decision problem is submitted to the intelligent agent decision sub-model composed of multiple agents for processing through an event triggering mechanism. The obtained decision result then drives the operation of the main line. The interaction between the two models is realized through the hybrid simulation model integration interface. The specific logic of the model operation is as follows: 1) When a new order arrives, the system Agent generates the corresponding workpiece Agent and process Agent according to the order information and associates the two. Then, the generated workpiece Agent enters the system through the "order entry" object in the production process sub-model to start the production process. 2) When the production process sub-model generates a new process to be executed, it needs to make decisions on the location of the AGV and the machine tool. That is, the workpiece agent enters the "negotiation" module, which is the connection module between the production process sub-model and the agent decision-making sub-model. When the entry event of the "negotiation" module is triggered, the "negotiation required" information is sent to the process agent of the workpiece through the port, thereby submitting the process task allocation problem to the agent decision-making sub-model. The corresponding agent negotiates the task allocation result according to the communication protocol, and reads and changes the variables of the workpiece agent through the active object penetration interface, and stores the returned result in the workpiece agent. 3) The workpiece agent that receives the task allocation result is released from the "negotiation" module using the activity object penetration interface and enters the AGV transportation link of the production process sub-model. After being released by the "Queue 1" object according to priority, the workpiece agent executes the transportation process. Similarly, the path planning problem in the transportation process is decided by the agent. The workpiece agent enters the "Waiting and Executing Transportation" module, which calls the transportation AGV for it, and the AGV agent decides the path planning to transport the workpiece to the processing machine tool. Accordingly, the workpiece agent enters the processing module of the production process sub-model to perform processing until the processing is completed and a new process is generated. Repeat steps 2) and 3) until all processes are completed. 4) The production process sub-model continues to operate until a dynamic disturbance event occurs in the system, and is identified by the "judgment of AGV failure" and "judgment of machine tool failure" objects. The workpiece agents affected by the disturbance event are returned to the "negotiation" module. Similarly, the disturbance response problem under dynamic disturbance is submitted to the agent decision sub-model for resolution. Step 2: Improved contract network protocol negotiation based on expanded task buffer pool and tender issuance priority; 1) Bidding stage Solution to the bidding broadcast problem under task buffer pool capacity limitation: In order to reduce communication and computation, the bidding broadcast problem is solved by limiting the number of receivers; when selecting a processing machine for a certain process, bidding information is only sent to machine tools that have the processing capacity for this process and whose buffer pool is not full; when selecting handling equipment, since all AGVs in the system have handling capacity, bidding information is only sent to AGV Agents whose buffer pool is not full. Solution to the problem of vicious bidding under bidding priority: In the hybrid simulation model, after a workpiece enters the system, it first enters the production process sub-model modeled by discrete events. When the system reaches the point where it needs to select a transport AGV and a processing machine for a certain process of the workpiece, the workpiece first flows into the "negotiation" module in the production process sub-model. This module notifies the corresponding workpiece agent to activate its process agent, and then the process agent sends bidding information and AGVs. The Agent and Machine Tool Agent communicate and negotiate until the negotiation ends. Only then is the workpiece released from the "negotiation" and the "negotiation" module begins processing the negotiation process for the next Agent. When multiple processes require bidding simultaneously in the system, the workpiece Agents corresponding to these processes are assigned bidding priorities by the "queue" object before entering the negotiation module. Then, they initiate bidding in the negotiation module in order of priority to complete the negotiation. Furthermore, process bidding events generated simultaneously in the system are concurrent events, which are serialized by the simulation engine. That is, they are transformed into information sent in the "negotiation" at the same time according to the bidding priority. In terms of message processing mechanism, these simultaneously sent information are executed in the order they are sent. The system processes only one bidding information at a time until the bidding information is processed and the contract is signed before the processing of the next bidding information begins. After all messages are processed, the simulation clock remains unchanged. 2) Bidding stage When calculating the bid content of the participant Agent, the task information carried by the pre-booked tasks in the task buffer pool is used to predict the various information for processing the bidding task in the future, and send it as the bid content to the bid initiator. 3) Bid Evaluation Stage Based on the production process sub-model in the hybrid simulation model, the start and end times of the actual handling and processing of each step can be obtained. These actual times are compared with the estimated times, and corrections are made if the deviation is too large, so as to improve the accuracy of the calculation based on the reservation information. At the same time, the actual task completion status is compared with the contract information. The expanded task buffer pool is specifically as follows: 1) The number of task types that can be accommodated in the buffer pool has been expanded; The machine tool agent task buffer pool can accommodate tasks that are being processed by the machine tool, tasks that have arrived at the machine tool buffer and are waiting to be processed, tasks that are being transported to the machine tool by the AGV, and tasks that are waiting to be transported to the machine tool in other machine tool buffers; the AGV agent task buffer pool can accommodate tasks that are being transported and tasks that are waiting to be transported at the machine tool buffer or material loading and unloading point. 2) Expand the functionality of task information utilization in the task buffer pool; Each task stored in the task buffer pool carries relevant information about its production process, including the process name, the name of the workpiece to which it belongs, the machine tool, the transport AGV, the start time of transport, the end time of transport, the start time of processing, and the end time of processing. This information is searchable and usable, and can be called up wherever it is needed during system operation. Step 3: Machine tool and AGV selection based on expanded task buffer pool; 1) AGV Selection Strategy The evaluation metric is minimizing the completion time; Once a process is initiated, in order to process it as quickly as possible, the AGV with the earliest handling time should be selected for transfer. Therefore, when the AGVAgent receives the bidding information from the process Agent, the bid submitted to it contains the estimated value of the earliest start time of the process. The calculation process of this value is shown in equations (1) to (2): In the formula, The estimated time for the start of the j-th process of the i-th workpiece on the k-th AGV; The estimated waiting time for the j-th process of the i-th workpiece on the k-th AGV is given. This is the estimated end time of the last pending task in the task buffer pool of the k-th AGV. Let t be the no-load travel time of the k-th AGV from its current position to the j-th process position of the i-th workpiece; t is the current running time of the system. After receiving all the bids returned by the bidding AGVs, the process Agent in the bidding state enters the bid decision state. According to formula (3), it selects the AGV with the earliest transport time as the transport equipment and signs a contract with it. The contract time for starting transport is the earliest start time submitted by the AGV. Accordingly, the process is placed in the task buffer pool of the contracted AGV as the last task to be executed. The information it carries is used for the next bid calculation of the contracted AGV. In the formula, A ij The contracted AGV for the j-th process of the i-th workpiece; The actual execution results are used to provide feedback on the estimated value, detect whether there is a deviation and correct the deviation; the deviation detection formula is shown in equation (4). When the equation is not satisfied, the deviation accumulates to the upper limit, and the corresponding AGV Agent updates the estimated start time of all remaining tasks in its task buffer pool to correct the deviation. Where, Let be the actual end time of the j-th process of the i-th workpiece on the k-th AGV; The estimated time for the end of the j-th process of the i-th workpiece on the k-th AGV; threshold is the allowable deviation threshold of the system. 2) Machine tool selection strategy After the process agent completes the selection of the transport AGV, it continues to send bidding information to the machine tools in the machine tool collection. In accordance with the goal of minimizing the completion time, when selecting machine tools, it is necessary to select the machine tool with the earliest completion time. Therefore, the bidding document submitted by the machine tool agent is the estimated completion time of the bidding process, and its calculation process is shown in equations (5) to (8). In the formula, This is the estimated time for the completion of the j-th process of the i-th workpiece on the m-th machine tool; This is the estimated start time for the j-th process of the i-th workpiece on the m-th machine tool; Let be the processing time of the j-th process for the i-th workpiece on the m-th machine tool; The estimated time for the j-th process of the i-th workpiece to wait for processing on the m-th machine tool; This is the estimated end time of the last pending task in the task buffer pool of the m-th machine tool; The load travel time of the winning AGV from the current position of the bidding process to the machine tool in the j-th process of the i-th workpiece; Before calculating the value, the machine tool agent needs to verify whether the machine tool in the previous process of the bidding process is the same machine tool as the bidding process. If it is the same machine tool, then there is no AGV handling step in the execution of that process. In this case, the calculation process... All are 0; After receiving all the tenders returned by the tendered machine tools, the process agent in the bidding state enters the tender decision state. According to formula (9), it selects the machine tool with the earliest end processing time as the processing equipment and signs a contract with it. The contract end processing time is the earliest end processing time submitted by the machine tool. Accordingly, the process is placed in the task buffer pool of the contracted machine tool as the last task to be executed. The information it carries is used for the next bidding calculation. Finally, the process agent also needs to check whether the winning machine tool is the processing machine tool of the previous process of the related workpiece agent. If the processing machine tools of the two processes are the same, there is no need for AGV transportation. The previous winning AGV agent needs to be notified to delete this process from its task buffer pool. In the formula, M ij The contracted machine tool for the j-th process of the i-th workpiece; Similarly, deviation detection and correction are required. The deviation detection method is shown in Equation (10). When the equation is not satisfied, the deviation accumulates to the upper limit. The corresponding machine tool agent updates the estimated end time of all remaining tasks in its task buffer pool to correct the deviation. In the formula, Let be the actual completion time of the j-th process of the i-th workpiece on the m-th machine tool; Step 4: Task allocation based on the hybrid simulation model of the improved contract network; 1) Coordination of overall goals Global goals are coordinated by prioritizing them, including the following two priorities: Priority of issuance Assign a priority to each process in the bidding order. This priority is based on the total number of remaining processes for the workpiece. The more remaining processes a workpiece has, the higher its priority and the earlier its negotiation order. AGV and Machine Tool Agent Task Execution Priority The contract execution time for each process task won by the AGV Agent and the machine tool Agent through bidding is calculated based on the existing task information in their task buffer pool. Then, newly won tasks are placed into the buffer pool to participate in the bidding for the next task. In other words, the execution order of these tasks is their queuing order in the buffer pool. Correspondingly, the process agent carries the priority that determines its execution order. When selecting the next task to execute, the AGV and the machine tool only need to select from their task buffer pools in turn. 2) Task allocation process during hybrid simulation model runtime When an order arrives, the system agent generates corresponding workpiece agents and process agents based on the order information and associates them. The generated workpiece agent enters the system through the "Order Entry" object in the production process sub-model of the hybrid simulation model, initiating the production process. When the system reaches the point where task allocation for a specific process of a workpiece is required, the workpiece agent flows into the "Negotiation" object, submitting the task allocation decision problem to the intelligent agent decision sub-model for processing, and the corresponding process agent is activated. The activated process agent initiates the improved contract network protocol negotiation process according to the bidding priority assigned to it in the "queue," determining its handling based on the established AGV and machine tool selection strategy. AGVs and machine tools are stored in the task buffer pools of the contracted AGV and machine tool according to the task execution priority, waiting for execution. When the contracted AGV is idle, the workpiece agent flows into "Get AGV" to obtain the contracted AGV. The "AGV Transport" and "AGV Release" objects control the AGV to transport the workpiece to the machine tool and release the AGV. Similarly, when the contracted machine tool is idle, the workpiece flows into "Get Machine Tool", "Process", and "Release Machine Tool" in sequence to complete the processing operation. When the process is completed, the corresponding AGV and machine tool correct the information in the task buffer pool according to the actual execution result of the task. At the same time, the workpiece agent returns from "Judgment New Process" to start the next process flow.

Images