Conflict-free AGV scheduling method for punctual material distribution in workshop

Abstract

The invention relates to a conflict-free AGV scheduling method for punctualized material distribution in a workshop, and the method comprises the following steps: obtaining a material distribution task, and dividing a decision period into a plurality of stages; an independent scheduling algorithm is used in each stage for AGV scheduling, tasks and departure time distributed by each AGV in each trip are optimized, a conflict-free optimal path is planned, an AGV scheduling scheme for minimizing time window penalty and distribution cost is realized, the scheduling algorithm comprises task allocation and path planning, in the task allocation, the path planning is performed, and in the path planning, the time window penalty and the distribution cost are reduced. According to the method, travel division is carried out based on departure time estimation, AGV assignment of each travel is carried out based on a heuristic rule, initial path planning is carried out by adopting an A * algorithm according to a task allocation result in path planning, conflict detection is carried out based on a path time window, and conflict resolution is carried out in combination with waiting and local re-planning. Compared with the prior art, cooperation of task allocation and path planning under the punctualization distribution requirement is achieved, and the calculation efficiency and the resource utilization rate can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of AGV scheduling, and in particular to a conflict-free AGV scheduling method for just-in-time material distribution in a workshop. Background Art

[0002] In the modern economy, logistics is the foundation of all industries. From the timely delivery of fresh food in daily life to the precise delivery of emergency supplies after disasters, the timeliness of delivery is emphasized. In the manufacturing industry, the adoption of the Just-in-Time (JIT) principle reflects the shift in the supply chain management paradigm, emphasizing the synchronization of production processes with customer needs to minimize inventory costs and improve response speed.

[0003] Material distribution in the workshop plays a unique role in the supply chain. It not only needs to consider the transportation and storage of materials, but also needs to be closely integrated with the production process. It has a crucial impact on production efficiency and cost, and is an important research topic for realizing intelligent manufacturing. Timely distribution of materials is the key to ensuring the smooth operation of the production line. Any delay or interruption in the material flow will affect the entire production process, resulting in idle machines and costly downtime. In addition, the cost of material distribution accounts for a high proportion of the total production cost, and some are even as high as 30%. With the development of manufacturing technology, small-batch and high-frequency distribution methods that can quickly respond to customers' personalized customization needs have become the mainstream, which also puts higher requirements on material distribution scheduling. Automated Guided Vehicles (AGV) is an intelligent logistics equipment that integrates multiple advanced technologies and can travel along a planned path. It has the characteristics of automation, high flexibility, high efficiency, and high reliability. It is widely used in material distribution in manufacturing workshops. The intelligent logistics system composed of mobile robots such as AGV can greatly reduce labor costs and improve the efficiency of material distribution.

[0004] The workshop material distribution problem is similar to the AGV scheduling problem, the vehicle routing problem (VRP) and the inventory routing problem (IRP), but it also exhibits different characteristics. Unlike the single-access constraint in the traditional vehicle routing problem, the production line usually involves parallel operations, and the carts need to make multiple round trips between the warehouse and the site, which increases the complexity of the problem. In addition, in the production of high-end equipment such as aircraft, due to the long production cycle, large amount of materials, and large differences in processing time of different processes, the material distribution tasks are unevenly distributed in time. If the distribution within the entire production cycle is arranged at one time, performance will be degraded.

[0005] Just-in-time material distribution methods can be divided into two types: milk-run distribution and point-to-point distribution. Milk-run distribution usually follows a fixed circular route to serve each workstation or warehouse in turn for periodic replenishment. However, milk-run distribution has disadvantages such as redundant routes and poor flexibility. Point-to-point distribution can directly transport materials from the warehouse to the target workstation, which can quickly respond to real-time or emergency needs and is more flexible.

[0006] In the point-to-point delivery mode, the system must first reasonably allocate multiple delivery tasks to one or more vehicles, which mainly involves the following decision-making issues: ① If the number of tasks exceeds the current vehicle's carrying capacity, it is necessary to deliver them in batches and multiple times, and select the tasks that the vehicle needs to deliver for this trip. ② If a vehicle needs to deliver multiple tasks in one trip, it is necessary to decide the delivery order of the tasks. ③ Decide on the departure time of each trip of the vehicle. In order to achieve on-time delivery, task allocation mainly focuses on how to reduce the penalty for early or overdue delivery of materials. Generally, there are three types of methods. The first type of method allows the vehicle to deliver continuously and reduces the time window penalty value by optimizing the delivery order. This type of method is mostly suitable for problem scenarios with short decision cycles and compact tasks. The second type of method reduces the vehicle's lead time by inserting idle time, and determines the departure time only based on the current task requirements. The third type of method reverses the departure time based on the material demand time window and the determined driving time. This type of method has the greatest potential because it is more flexible, but it needs to assume that the cars travel along the shortest path and have a certain travel time. It does not consider the time consumption caused by conflicts between cars, and it is impossible to arrange a suitable departure time.

[0007] Therefore, if we want to improve the on-time delivery efficiency of the third method, after clarifying the delivery task of the small vehicle, we also need to plan the optimal driving path between the starting point and the end point. Traditional path planning usually searches for the shortest path between the starting point and the end point in a static environment where the global map information is known. The problem of multi-vehicle path planning with conflicts caused by limited channels remains to be solved.

[0008] Another problem with task allocation is that all cars in the workshop are always enabled when assigning cars. Common car assignment rules include first arrive first serve (FAFS), minimum idle time (MIT), shortest rectilinear distance (SRD), etc. Existing research uses one or more priority rules to select among all cars, so that all cars are almost always in working state. However, the more cars there are in motion, the more path conflicts need to be handled, and when the number of tasks is small, it causes a waste of resources.

[0009] In summary, the existing AGV scheduling methods have the following main problems: 1) It is difficult to coordinate task allocation and path planning under the requirement of just-in-time delivery; 2) All vehicles in the workshop are always enabled during scheduling, resulting in a waste of resources; 3) When resolving conflicts, replanning the entire path may cause a deadlock, reducing the stability of the system. Summary of the invention

[0010] The purpose of the present invention is to provide a conflict-free AGV scheduling method for just-in-time material distribution in workshops. A phased scheduling scheme is proposed. In each phase, two sub-problems of task allocation and path planning are solved. In terms of task allocation, a trip division method is designed to solve the departure time decision problem considering path conflicts under JIT distribution requirements. At the same time, an AGV assignment method based on heuristic rules is proposed, which can flexibly adjust the number of AGVs used. In order to solve the path conflict problem, a hybrid path planning algorithm combining the A* algorithm with the heuristic method is proposed. The entire optimization process is iterated through a genetic algorithm, which can achieve efficient and conflict-free material distribution.

[0011] The purpose of the present invention can be achieved by the following technical solutions:

[0012] A conflict-free AGV scheduling method for on-time material distribution in a workshop, the method comprising the following steps:

[0013] Get material delivery tasks,

[0014] Divide the decision cycle into stages;

[0015] In each stage, an independent scheduling algorithm is used to schedule AGVs, optimize the tasks and departure times of each AGV for each trip, and plan the optimal conflict-free path to achieve an AGV scheduling solution that minimizes time window penalties and delivery costs. The scheduling algorithm includes task allocation and path planning. In the task allocation, the itinerary is divided based on the departure time estimate, and the AGV is assigned to each itinerary based on heuristic rules. In the path planning, the A* algorithm is used for initial path planning based on the result of task allocation, and conflict detection is performed based on the path time window, and conflict resolution is performed in combination with waiting and local replanning.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] (1) The present invention establishes a JIT material distribution model with a time window to minimize the time window penalty and distribution cost. In terms of task allocation, a trip division method based on departure time estimation and an AGV assignment method based on heuristic rules are designed. In terms of path planning, a hybrid path planning algorithm combining the A* algorithm and the heuristic algorithm is proposed, and a genetic algorithm is used to jointly optimize task allocation and path planning, thereby achieving the coordination of task allocation and path planning under the requirements of just-in-time distribution, ensuring timely delivery of materials without delays, and avoiding path conflicts, thus meeting actual needs.

[0018] (2) The present invention designs a trip division method based on departure time estimation to solve the departure time decision problem considering path conflicts, which can reasonably reserve delivery lead time and increase feasible solutions.

[0019] (3) A cart assignment method based on heuristic rules is designed to enable a reasonable number of carts, achieve timely delivery of materials with fewer AGVs, and maintain a small number of line-side material storage, thus avoiding resource waste, improving computing efficiency and resource utilization, and achieving a balance between performance and cost.

[0020] (4) The present invention designs a local replanning method to retain the previous path without conflict and minimize the changes of the new path compared to the original path, which can improve the quality and robustness of the solution and has good potential in terms of computational efficiency and scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the layout of a single warehouse workshop of the present invention;

[0022] Figure 2 It is a schematic diagram of the phase division of the present invention;

[0023] Figure 3 It is a staged scheduling flow chart of the present invention;

[0024] Figure 4 It is a flow chart of genetic algorithm in the stages of the present invention;

[0025] Figure 5 Schematic diagram of two types of path conflicts of the present invention, where (5a) is a same-direction conflict and (5b) is an opposite-direction conflict;

[0026] Figure 6 It is the time window feature of the conflicting relationship of the present invention;

[0027] Figure 7 is a specific example of a conflict of opposite directions;

[0028] Figure 81 is a comparison of path time windows before and after the opposite conflicts are resolved in an embodiment, wherein (8a) is the path time window before conflict 1 is resolved, (8b) is the path time window after conflict 1 is resolved, (8c) is the path time window before conflict 2 is resolved, and (8d) is the path time window after conflict 2 is resolved;

[0029] Fig. 9 It is the line side material storage quantity of each station of the second assembly line in one embodiment. DETAILED DESCRIPTION

[0030] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0031] This embodiment provides a conflict-free AGV scheduling method for on-time material delivery in a workshop, the method comprising the following steps:

[0032] S1, obtain material distribution tasks,

[0033] S2 divides the decision cycle into multiple stages;

[0034] S3, in each stage, an independent scheduling algorithm is used to schedule AGVs, the tasks and departure times of each AGV for each trip are optimized, and the optimal conflict-free path is planned to achieve an AGV scheduling solution that minimizes the time window penalty and delivery cost. The scheduling algorithm includes task allocation and path planning. In the task allocation, the itinerary is divided based on the departure time estimate, and the AGV is assigned to each itinerary based on heuristic rules. In the path planning, the A* algorithm is used for initial path planning based on the result of task allocation, and conflict detection is performed based on the path time window, and conflict resolution is performed in combination with waiting and local replanning.

[0035] The above method is specifically implemented through the following scheme.

[0036] (1) Problem description

[0037] (11) Assumptions related to workshop layout

[0038] a. There is V in the workshop L There are V identical assembly lines, each with S Stations, each station has V A Warehouses, lineside storage areas, and intersections are all abstracted as nodes where AGVs can pass through or stay, such as Figure 1 shown.

[0039] b. A central warehouse is responsible for the materials of all assembly lines in the workshop, assuming that there are sufficient materials in the warehouse.

[0040] c. The size of a station is L S ×W S , L S is the length of a station, W S The logistics channels in the workshop are all two-way single lanes. When multiple AGVs occupy the same section between two adjacent nodes at the same time, path conflicts will occur.

[0041] (12) Assumptions related to delivery tasks

[0042] a. The assembly time of a product at a station is T A , the time to move between two stations is T M The workshop status is T M +T A It is a periodic cycle, which is set as the decision cycle, and the starting time of each cycle is called time 0.

[0043] b. There are J material delivery tasks in the decision cycle. The material demand time t of any material delivery task j∈{1,2,...,J} Dem,j ∈[T M ,T M +T A ), demand quantity q j and the target node N j Known. There is no temporary insertion of new tasks.

[0044] c. The materials required for assembly work are mainly divided into three categories: ① Large structural parts with small quantity and irregular shape, such as engines, are usually delivered directly by large logistics equipment such as cranes. ② Common standard parts with large demand and small volume, such as screws and nuts, are usually stored in separate areas along the line and replenished regularly. ③ Assemblies with many types and medium volume, such as instruments. Due to the correlation between assembly parts, the assembly parts required for the same process have been pre-packed into one or more standard material boxes through kitting. This article mainly considers the supply of assembly parts, and the material demand q j Corresponding to the number of standard material boxes.

[0045] (13) Assumptions related to the distribution process

[0046] a. There are K identical AGVs in the workshop. All vehicles are available at the initial moment. The speed v of each vehicle is always constant, and the maximum number of standard bins that can be loaded is Q.

[0047] b. In one trip, the AGV starts from the warehouse, delivers the loaded tasks (can be one or more) to their target nodes, and then returns to the warehouse. Each AGV can reach any node on the map along the logistics channel, but it does not need to pass through all nodes in one trip.

[0048] c. Material requirement q for any task j j are all less than the maximum carrying capacity of the car, Q. A task cannot be divided, that is, each task can only be delivered once by an AGV, while an AGV can deliver multiple tasks in one trip.

[0049] d. A trip is performed by only one AGV, and the total amount of materials in one trip does not exceed the maximum load capacity Q of the vehicle.

[0050] e. Factors such as loading and unloading time, AGV failure, AGV power, and lineside area capacity are not considered.

[0051] (2) Scheduling Model

[0052] (21) Symbol Definition

[0053] A collection of task numbers.

[0054] AGV serial number collection.

[0055] A set of discrete time sequence numbers.

[0056] A set of path node numbers.

[0057] The set of program numbers of car k, R k is the total number of trips of car k.

[0058] From the starting point To the end The path of Indicates the node number. represents the road segment between adjacent nodes, Represents the Manhattan distance between nodes.

[0059] [t Dem,j -T W ,t Dem,j ]: Material demand time window of task j, where T W is the time window size.

[0060] t j : The delivery time of task j.

[0061] α,β: penalty factors for early or late delivery.

[0062] N j : The target node of task j.

[0063] q j : Material requirement for task j.

[0064] Q: What is the maximum load capacity of an AGV?

[0065] f k : The fixed cost of each start of the k-th AGV.

[0066] s k : The cost per unit distance traveled by the k-th AGV.

[0067] d kr : The total distance of the k-th AGV’s r-th trip.

[0068] (22) Decision variables

[0069] If the delivery task j is delivered by the kth AGV in its rth trip, then x jkr =1, otherwise x jkr = 0. If multiple tasks are delivered in one trip, the material demand time t Dem Sort in ascending order and deliver in sequence.

[0070] If the kth AGV's rth trip starts at time m, then y krd =1, otherwise y krd =0.

[0071] If the kth AGV passes through path node n at time m, then z kmn =1, otherwise z kmn =0.

[0072] (23) Optimization goals

[0073] In order to achieve JIT delivery, a time window constraint is set to allow early arrival but not late arrival. j (See formula (1)) is not in the time window [t Dem,j -T W ,t Dem,j The penalty value within ] is shown in formula (2). The objective function f that minimizes the time window penalty t As shown in formula (3). The distribution cost includes the fixed cost and the travel distance cost of each trip of the car. Formula (4) extracts the path node of the kth AGV's rth trip and uses the function Calculate the total distance traveled. For example, when The objective function f that minimizes the total delivery cost d See formula (5).

[0074]

[0075] d kr = Dist({n|z kmn y krm' y k(r+1)m” =1,m'≤m≤m”}) (4)

[0076]

[0077] (24) Constraints

[0078]

[0079]

[0080] Among them, formula (6) indicates that a task can only be delivered by one AGV; formula (7) indicates that an AGV can only set out once in one trip; formula (8) indicates that the number of cars in use at any time does not exceed the total number of cars; formula (9) indicates that a car can only pass through one path node at the same time; formula (10) indicates that a path node can only be occupied by one car at the same time; formula (11) indicates that the total amount of materials carried by an AGV in one trip does not exceed its maximum carrying capacity.

[0081] (3) Scheduling model solution

[0082] (31) Phased scheduling process

[0083] Due to the long assembly time, large number of materials, and uneven distribution of tasks in time, the decision cycle is divided into multiple small stages. Specifically: All material distribution tasks are dispatched according to the demand time t Dem Divided into H+1 task sets, denoted as like Figure 2 As shown. The number of tasks in stage h is J h , the stage length is L; the task set Contains only t Dem =T M mission, Contains only t Dem ∈(T M ,T M +L] task, Contains only t Dem ∈(T M +T A -L,T M +T A ] task, where T Ais the assembly time of the product at a station, T M is the time it takes to move between two stations.

[0084] Each stage has an independent scheduling algorithm. For stage h (h ≥ 1), enter the latest available time set of all AGVs after completing stage h-1 and task collection Output AGV scheduling plan renew And input the next stage, t Avail,h-1,k represents the latest available time of AGV k in stage h-1, represents the set of AGV sequence numbers, K is the number of AGVs; tasks in stage h will not be advanced to stage h-1 or delayed to stage h+1 for processing.

[0085] The phased scheduling process is as follows Figure 3 Considering that the scheduling problem studied in this embodiment is an NP-hard problem, a heuristic algorithm combined with a genetic evolution operator is designed in the stage.

[0086] (32) Intra-stage scheduling algorithm flow

[0087] The independent scheduling algorithm uses a genetic algorithm. The specific steps of using the genetic algorithm for AGV scheduling are as follows:

[0088] A chromosome is represented as is the task number, t Ld is the maximum lead time for material delivery, that is, the earliest time for task j to arrive at the target node is t Dem,j -t Ld , t Dem,j is the demand time of task j; materials are allowed to arrive in advance, and the line-side storage capacity is used to ensure that the vehicle completes the delivery plan, and the lead time does not exceed t Ld , reducing the storage pressure on the line side. At the same time, it also reserves margin for subsequent conflict handling operations.

[0089] Due to the different number of tasks in different stages, the global search capability of genetic algorithm is used to set the [t LL ,t LU ] Select the appropriate t Ld , repeat the following operations to generate the initial population: assign all tasks in the stage to the required time t Dem Sort in ascending order. If the demand time is the same, sort in ascending order according to the distance from the target node to the warehouse. Divide the sorted task sequence into several segments, and randomly sort the tasks in each segment to form the first J of the chromosome. h bit; finally in [t LL ,t LU] Randomly select an integer in the interval as the last digit t Ld ;

[0090] Decoding and fitness calculation: Decode each individual in the population and calculate the fitness value. The decoding part includes task allocation and path planning. The task allocation algorithm divides the trip under the premise of satisfying the vehicle capacity constraint and delivery time constraint, assigns an AGV vehicle to each trip and decides its departure time. In order to avoid conflicts, the starting point, end point and departure time of the vehicle obtained by the task allocation algorithm are input into the path planning algorithm to obtain the optimal path without conflict. When calculating the individual fitness, the weighted addition of different target values ​​is converted into a single-objective problem, as shown in formula (12).

[0091] f=w1f t +w2f d (12)

[0092] Among them, w1 and w2 are weights, f t is the objective function that minimizes the time window penalty, f d The objective function is to minimize the total delivery cost;

[0093] Evolution: Evolve a new generation of population through selection, crossover and mutation, and continuously iterate to approach the optimal solution. This part belongs to the general operation of genetic algorithm, and this embodiment will not be repeated here. The algorithm flow is as follows Figure 4 shown.

[0094] (33) Task Allocation Algorithm

[0095] Step 1) Estimated departure time:

[0096] Assume that in a trip the set of transport tasks According to the material demand time t of the task Dem , the shortest driving distance d, the car speed v and the maximum advance time t without considering path conflicts Ld Estimated time interval for the car to depart from the warehouse [t DL ,t DU ], specifically including the following steps:

[0097] Step 11) Calculate the number of tasks x to be delivered in order i Total distance travelled;

[0098] Step 12) Calculate the departure time interval

[0099] Step 13) Check Does it intersect with the departure time interval of the previous task? If not, then task x i If it is not suitable for this delivery, the algorithm terminates; otherwise, it continues to loop and outputs the final interval [tDL ,t DU ].

[0100] The pseudo code of the algorithm is shown in Table 1, where Represents a warehouse node.

[0101] Table 1

[0102]

[0103]

[0104] Step 2) Itinerary division:

[0105] Considering the capacity constraints of the trolley, the delivery task is divided into multiple trips, and the definition is the task set for the rth trip, S T,r is the departure time interval of the rth trip, and the trip division includes the following steps:

[0106] Step 21) Set task x i With current itinerary The tasks in the group are combined to form a temporary group

[0107] Step 22) Determine whether the vehicle capacity constraint is satisfied. If not, assign task x i Assign to the next trip and jump to step 21); if satisfied, execute the next step;

[0108] Step 23) Temporary group Press t to complete all tasks in Dem Sort in ascending order. According to the estimated departure time interval in step 1), determine whether there is a feasible departure time interval. If not, assign task x i Assign to the next trip and jump to step 21); if yes, execute the next step;

[0109] Step 24) Set task x i Add to current trip.

[0110] The pseudo code of the algorithm is shown in Table 2.

[0111] Table 2

[0112]

[0113] Step 3) AGV Assignment

[0114] Assume that chromosome Divided into R trips, which is the task set for the rth trip Assign cars and decide their departure time t D,r, the earliest available car is selected from the cars that have been started in the stage. If there is no car that meets the time constraint, the earliest available car is selected from all cars. In this way, the number of cars used can be reasonably arranged in different stages. The pseudo code of the algorithm is shown in Table 3, where the definition It is the collection of cars that have been started in the stage, AGV cars Available time t Avail,k Initialized to the completion time of the previous stage

[0115] Table 3

[0116]

[0117] (34) Path planning algorithm

[0118] The A* algorithm is used to plan the initial optimal path for the car, calculate the path time window and detect whether there is a conflict, and different resolution strategies are adopted for different conflict types. The A* algorithm will not be described in detail, and this embodiment focuses on the conflict detection method and conflict resolution strategy.

[0119] Calculating the path time window is the main method for detecting path conflicts in a static environment. As long as the departure time, running speed, starting point, end point and driving route of the car are determined, the time period from the car entering to leaving a certain road section, that is, the path time window, can be calculated. Two types of path conflicts can be summarized:

[0120] 1. Conflict in the same direction: Two cars drive into the same road section in the same direction, and their path time windows overlap, as shown in Figure (5a). In this case, one car is ordered to wait for the other car to pass before entering the conflicting road section.

[0121] 2. Conflict: When two cars are traveling towards each other, they will enter each other's current or previous road segment, as shown in Figure (5b). The path time windows interact, such as Figure 6 The part within the dotted box is shown. Figure 7 A specific case is shown. The starting point of AGV1 is node 5 and the end point is node 1; the starting point of AGV2 is node 1 and the end point is node 6. Both vehicles start from the starting point at the same time. If the waiting strategy is still adopted, AGV2 needs to wait until AGV1 reaches node 1 before it can start. The waiting time is long and may cause delivery delays. Therefore, the path of AGV2 needs to be replanned to bypass the conflicting section.

[0122] Combining waiting and local replanning to resolve conflicts is as follows:

[0123] Use the A* algorithm to plan the initial optimal path between the starting point and the end point Start traversing the collection from i=1 Nodes within Calculate the route The path time window is checked to see if there is a conflict. If there is a conflict in the same direction, the previous section is Wait; if there is a conflict, perform local replanning and update The specific steps of the local replanning are as follows:

[0124] Step 41) Traverse forward from l=i Nodes within Find the nearest intersection node If no intersection node is found when l=1, jump to step 43);

[0125] Step 42) do not consider and Judging from the intersection Is there any other traversable direction? If so, use A* algorithm to find it. If there are other paths between the end points, the algorithm ends; otherwise, let i = c and jump back to step 41);

[0126] Step 43) do not consider Judging from the starting point Check whether there are other traversable directions. If so, use the A* algorithm to find other paths between the starting point and the end point, and the algorithm ends; otherwise, let the car wait at the starting point for a preset time, and then use the A* algorithm to re-plan the optimal path between the starting point and the end point, and the algorithm ends.

[0127] Repeat the above operation until there is no path conflict. The pseudo code of path planning is shown in Table 4.

[0128] Table 4

[0129]

[0130]

[0131]

[0132] (4) Experimental results and performance verification

[0133] In this embodiment, a series of numerical experiments are conducted to verify the performance of the algorithm proposed in the present invention in solving the material distribution problem of the aircraft pulse assembly line. All algorithms are coded in MATLAB R2022a and run on a PC equipped with an Intel Core i9 processor and 16GB RAM. The process data of an actual aircraft pulse assembly line is taken as the research object, and there are 298 assembly processes within 6 hours. Given the target node and material demand time of each process, the material demand in the range of {1, 2, 3, 4, 5} and the line edge position in the range of {1, 2, 3} are randomly generated for each process to form a material distribution task. Assuming that there are 3 identical pulse assembly lines in the workshop, there are 298 × 3 = 894 tasks in total. Parameter settings are detailed in Table 5. Table 6 captures 4 task information in a certain stage as an example. In Table 6, j is the task number; AO Code is the assembly process number corresponding to the task; L / S / LS is the assembly line / station / line edge storage area, and the location of the assembly process.

[0134] Table 5 Parameter settings

[0135]

[0136]

[0137] Table 6 Material distribution task examples

[0138] j AO Code L / S / LS <![CDATA[t Dem,j (s)]]> <![CDATA[q j ]]> 15 25 1-1-1 24696 4 60 865 1-2-2 23400 4 358 865 2-2-2 23400 4 809 2936 3-5-2 25200 1

[0139] (41) Validation of path planning algorithm

[0140] A target node is randomly generated for each car. The car starts from the warehouse, reaches the target node, and then returns to the warehouse. Figure 8 The before-and-after comparison of the time windows of two conflicting paths resolved during the planning process is shown. It can be seen that the final result does not have conflicting features such as overlapping or interactive time windows, which proves the effectiveness of the conflict resolution strategy.

[0141] (42) Task Allocation Algorithm Evaluation

[0142] The decision cycle T M +T A It is divided into 12 stages from 0 to 11, and the length of the hth stage is L. When h = 1, L = 1h; when h > 1, L = 0.5h. Since the task of stage 0 does not require just-in-time delivery, it can be delivered before the assembly starts, so it is not considered in this section.

[0143] (1) Time window constraint validity verification

[0144] Since the time window constraint is set not only to ensure timely delivery but also to reduce the accumulation of materials on the line side, the material storage quantity of each station on the assembly line is observed to verify its effectiveness. A simple priority rule method is used for comparison. This method only considers the AGV capacity constraint to divide the itinerary, artificially inserts idle time, and uses the FAFS rule for AGV assignment. The total time window penalty f of the present invention and the priority rule method is t They are 310.2 and 5891.2 respectively. Although both algorithms can deliver the materials before the end of the time window, the priority rule method will result in a higher penalty. Fig. 9 It can be seen that the priority rule method will cause a large amount of materials to accumulate in the line edge area, while the present invention can always maintain a small amount of line edge materials, thereby achieving the purpose of reducing the accumulation of line edge materials.

[0145] (2) Cart Assignment Method

[0146] A common method of assigning cars is to assign the earliest trip to the earliest available car, which is called the First Arrive First Serve (FAFS) rule. The car assignment part adopts the FAFS rule and the method proposed in this article respectively, and the other parts of the task allocation algorithm remain unchanged. As can be seen from Table 7, the present invention realizes the function of adjusting the number of cars used in different stages. As can be seen from Table 8, the FAFS rule can achieve a better objective function value due to the use of more cars, but at the same time it also makes the waiting time of vehicles caused by conflicts longer, local replanning more frequent, and the algorithm running time longer. Although the present invention is slightly inferior to the method based on the FAFS rule in terms of objective function value, it improves resource utilization and algorithm efficiency, and achieves a balance between performance and cost.

[0147] Table 7 Experimental results of each stage under different car assignment methods

[0148]

[0149] Table 8 Total experimental results under different car assignment methods

[0150]

[0151]

[0152] (3) Itinerary division and departure time decision method

[0153] The proposed algorithm is compared with two other algorithms, TD-I and TD-II. TD-I and TD-II only consider the AGV capacity constraint when dividing the trip, and divide it into a group if it does not exceed the maximum capacity. Different idle time insertion methods are used when deciding the departure time. TD-I inserts idle time between stages, and all cars have an initial available time t in stage 1. Avail,0,k = 2h, k∈{1,...,6}. The initial available time t of the car k that was not started in the h-1th stage is Avail,h-1,k =min(max(T Avail,h-1 ),t Avail,h-1,k +L). The AGV starts the next trip immediately after returning to the warehouse in the stage. The TD-II algorithm inserts idle time between trips so that the delivery time of the first task of the trip is exactly the start time of the time window. Between different trips, the AGV can wait in the warehouse for a period of time. The other parts are the same as this algorithm. Table 9 shows that the proposed algorithm significantly reduces the time window penalty, while the delivery cost does not change significantly.

[0154] Table 9 Experimental results under different trip division methods

[0155] algorithm <![CDATA[f d ]]> contrast <![CDATA[f t ]]> contrast TD-I 2454.0 -6.3％ 879.4 -64.7％ TD-II 2270.4 +1.0％ 425.0 -27.0％ The present invention 2300.3 / 310.2 /

[0156] (43) Performance at different production scales

[0157] By adjusting the number of assembly lines and AGVs in the workshop, the effectiveness of the present invention in large-scale production scenarios is verified. Table 10 shows the experimental results under two larger scales. The model and algorithm proposed in the present invention have the ability to handle such large-scale problems and show the application potential of adapting to different problem scales.

[0158] Table 10 Experimental results under different production scales

[0159]

[0160]

[0161] In view of the characteristics of long production cycle, large number of tasks and uneven distribution of tasks over time, the present invention designs a phased scheduling method to achieve the coordination of task allocation and path planning under JIT requirements. For task allocation, a trip division method based on departure time estimation and a car assignment method based on heuristic rules are designed to decide the departure time and adjust the number of cars; for path planning, a hybrid path planning algorithm combining the A* algorithm with the heuristic method is designed to effectively avoid path conflicts. In order to prove the effectiveness of the proposed model and algorithm, an actual aircraft pulse assembly line is taken as the research object. The experimental results show that: 1) the proposed algorithm can ensure timely delivery of materials without delays, and avoid path conflicts, meeting actual needs; 2) the algorithm can improve computing efficiency and resource utilization, and achieve a balance between performance and cost; 3) the algorithm has good scalability and good engineering application potential.

[0162] In future research, more complexities, such as multiple warehouses or different types of AGVs, can be explored. In addition, the uncertainty inherent in real production introduces dynamics in the material demand stage and the material delivery stage, which requires research on dynamic scheduling.

[0163] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A conflict-free AGV scheduling method for on-time material delivery in a workshop, characterized in that: The method comprises the following steps: Get material delivery tasks, Divide the decision cycle into stages; In each stage, an independent scheduling algorithm is used to schedule AGVs, optimize the tasks and departure times of each AGV for each trip, and plan the optimal conflict-free path to achieve an AGV scheduling solution that minimizes time window penalties and delivery costs. The scheduling algorithm includes task allocation and path planning. In the task allocation, the itinerary is divided based on the departure time estimate, and the AGV is assigned to each itinerary based on heuristic rules. In the path planning, the A* algorithm is used for initial path planning based on the result of task allocation, and conflict detection is performed based on the path time window, and conflict resolution is performed in combination with waiting and local replanning.

2. The conflict-free AGV scheduling method for on-time material delivery in a workshop according to claim 1 is characterized in that: The decision cycle is divided into multiple stages: all material distribution tasks are divided into multiple stages according to the demand time t Dem Divided into H+1 task sets, denoted as The number of tasks in stage h is J h , the stage length is L; the task set Contains only t Dem =T M mission, Contains only t Dem ∈(T M ,T M +L] task, Contains only t Dem ∈(T M +T A -L,T M +T A ] task, where T A is the assembly time of the product at a station, T M is the time to move between two stations; For stage h (h ≥ 1), enter the latest available time set of all AGVs after completing stage h-1 and task collection Output AGV scheduling plan renew And input the next stage, t Avail,h-1,k represents the latest available time of AGV k in stage h-1, represents the set of AGV sequence numbers, K is the number of AGVs; tasks in stage h will not be advanced to stage h-1 or delayed to stage h+1 for processing.

3. The conflict-free AGV scheduling method for on-time material delivery in a workshop according to claim 1 is characterized in that: The independent scheduling algorithm adopts a genetic algorithm, and the specific steps of using the genetic algorithm to schedule AGVs are as follows: A chromosome is represented as is the task number, t Ld is the maximum lead time for material delivery, that is, the earliest time for task j to arrive at the target node is t Dem,j -t Ld , t Dem,j is the demand time of task j; materials are allowed to arrive in advance, and the line-side storage capacity is used to ensure that the vehicle completes the delivery plan, and the lead time does not exceed t Ld ; Using the global search capability of genetic algorithms in the artificially set [t LL ,t LU ] Select the appropriate t Ld , the process is as follows: Generate the initial population: All tasks in the stage are assigned according to the required time t Dem Sort in ascending order. If the demand time is the same, sort in ascending order according to the distance from the target node to the warehouse. Divide the sorted task sequence into several segments, and randomly sort the tasks in each segment to form the first J of the chromosome. h bit; finally in [t LL ,t LU ] Randomly select an integer in the interval as the last digit t Ld ; Decoding and fitness calculation: Decode each individual in the population and calculate the fitness value. The decoding part includes task allocation and path planning. The task allocation algorithm divides the trip under the premise of satisfying the vehicle capacity constraint and delivery time constraint, assigns an AGV vehicle to each trip and decides its departure time. The starting point, end point and departure time of the vehicle obtained by the task allocation algorithm are input into the path planning algorithm to obtain the optimal path without conflict. When calculating the individual fitness, the weighted addition of different target values ​​is converted into a single-objective problem. The individual fitness is expressed as: f=w1f t +w2f d Among them, w1 and w2 are weights, f t is the objective function that minimizes the time window penalty, f d The objective function is to minimize the total delivery cost; Evolution: Evolve a new generation of population through selection, crossover and mutation, and continuously iterate to approach the optimal solution.

4. The conflict-free AGV scheduling method for on-time material delivery in a workshop according to claim 3 is characterized in that: The objective function of minimizing the time window penalty is expressed as: Among them, p j is the delivery time t j Not in the time window [t Dem,j -T W ,t Dem,j ] within the penalty value, p j =α(t Dem,j -T W -t j )+β(t j -t Dem,j ) Among them, T w is the time window size, α, β are penalty factors for advance or delay, t j is the delivery time of task j, N j is the target node of task j; Represents the AGV serial number set; represents the set of program numbers of car k; Represents a set of discrete time sequence numbers; The objective function of minimizing the total delivery cost is expressed as: Where K is the number of AGVs, f k is the fixed cost of each start of the kth AGV, R k is the total number of trips of car k, s k is the cost per unit distance traveled by the kth AGV, d kr is the total distance of the kth AGV’s rth trip.

5. The conflict-free AGV scheduling method for on-time material delivery in a workshop according to claim 3 is characterized in that: The decision variables for AGV scheduling using genetic algorithm include: If the delivery task j is delivered by the kth AGV in its rth trip, then x jkr =1, otherwise x jkr =0; if multiple tasks are delivered in one trip, the material demand time t Dem Sort in ascending order and deliver in sequence; If the kth AGV's rth trip starts at time m, then y krd =1, otherwise y krd =0; If the kth AGV passes through path node n at time m, then z kmn =1, otherwise z kmn =0; in, Represents a set of task numbers; Represents the AGV serial number set; represents the set of program numbers of car k, R k is the total number of trips of car k; Represents a set of discrete time sequence numbers; Represents a set of path node numbers.

6. The conflict-free AGV scheduling method for on-time material delivery in a workshop according to claim 5, characterized in that: The constraints for AGV scheduling using genetic algorithms include:

7. The conflict-free AGV scheduling method for on-time material delivery in a workshop according to claim 3 is characterized in that: The trip division based on the departure time estimate specifically includes the following steps: Step 1) Estimated departure time: Assume that in a trip the set of transport tasks According to the material demand time t of the task Dem , the shortest driving distance d, the car speed v and the maximum advance time t without considering path conflicts Ld Estimated time interval for the car to depart from the warehouse [t DL ,t DU ]: Step 11) Calculate the number of tasks x to be delivered in order i Total distance travelled; Step 12) Calculate the departure time interval Step 13) Check Does it intersect with the departure time interval of the previous task? If not, then task x i If it is not suitable for this delivery, the algorithm terminates; otherwise, it continues to loop and outputs the final interval [t DL ,t DU ]; Step 2) Itinerary division: Considering the capacity constraints of the trolley, the delivery task is divided into multiple trips, and the definition is the task set for the rth trip, S T,r is the departure time interval of the rth trip, and the trip division includes the following steps: Step 21) Set task x i With current itinerary The tasks in the group are combined to form a temporary group Step 22) Determine whether the vehicle capacity constraint is satisfied. If not, assign task x i Assign to the next trip and jump to step 21); if satisfied, execute the next step; Step 23) Temporary group Press t to complete all tasks in Dem Sort in ascending order. According to the estimated departure time interval in step 1), determine whether there is a feasible departure time interval. If not, assign task x i Assign to the next trip and jump to step 21); if yes, execute the next step; Step 24) Set task x i Add to current trip.

8. The conflict-free AGV scheduling method for on-time material delivery in a workshop according to claim 3 is characterized in that: The specific AGV assignment for each trip based on heuristic rules is as follows: Assume that the chromosome Divided into R trips, which is the task set for the rth trip Assign cars and decide their departure time t D,r , prioritize the earliest available car among the cars that have been started in the stage. If there is no car that meets the time constraint, then select the earliest available car among all the cars, among which the AGV car Available time t Avail,k Initialized to the completion time of the previous stage 9. The conflict-free AGV scheduling method for on-time material delivery in a workshop according to claim 1, characterized in that: The conflict detection based on the path time window is specifically as follows: Determine the departure time, running speed, starting point, end point and driving route of the car, calculate the time period from the car entering to leaving a certain road section, that is, the path time window, and determine the path conflict type according to the path time window. The path conflict type includes: Same-direction conflict: Two cars drive into the same road section in the same direction, and their path time windows overlap; Opposite-direction conflict: Two cars are traveling towards each other and will enter each other's current or previous road segment, and the path time windows will interact.

10. The conflict-free AGV scheduling method for on-time material delivery in a workshop according to claim 9, characterized in that: The conflict resolution by combining waiting and local replanning is specifically as follows: Use the A* algorithm to plan the initial optimal path between the starting point and the end point Start traversing the collection from i=1 Nodes within Calculate the route The path time window is checked to see if there is a conflict. If there is a conflict in the same direction, the previous section is Wait; if there is a conflict, perform local replanning and update The specific steps of the local replanning are as follows: Step 41) Traverse forward from l=i Nodes within Find the nearest intersection node If no intersection node is found when l=1, jump to step 43); Step 42) do not consider and Judging from the intersection Is there any other traversable direction? If so, use A* algorithm to find it. If there are other paths between the end points, the algorithm ends; otherwise, let i = c and jump back to step 41); Step 43) do not consider Judging from the starting point Check whether there are other traversable directions. If so, use the A* algorithm to find other paths between the starting point and the end point, and the algorithm ends; otherwise, let the car wait at the starting point for a preset time, and then use the A* algorithm to re-plan the optimal path between the starting point and the end point, and the algorithm ends.

Images