Knowledge-driven digital twin discrete workshop material distribution optimization method

Abstract

The application discloses a knowledge-driven digital twin discrete workshop material distribution optimization method, links multi-source heterogeneous data generated in a discrete manufacturing workshop material distribution process by using a knowledge graph to form a unified data type, realizes structured representation of data knowledge, further carries out data mining and relation acquisition on the unified data, can solve problems such as information islands in the data management process to discover hidden relations and rules among data, maps knowledge data in the knowledge graph to a digital twin workshop to form a knowledge layer, realizes the establishment of a high-fidelity virtual workshop corresponding to a physical workshop under the condition of obtaining massive multi-source heterogeneous workshop material distribution data, improves the path optimization algorithm of single-task AGVs and multi-task AGVs, and the improved algorithm is more suitable for the operation mechanism of the digital twin workshop when solving the material distribution path optimization problem, and thus the efficiency of the AGV material distribution process in the workshop is improved.

Description

Technical Field

[0001] This invention belongs to the field of material distribution path optimization technology in discrete manufacturing workshops, specifically involving a knowledge-driven digital twin discrete workshop material distribution optimization method. Background Technology

[0002] With the rapid development of new technologies such as the Industrial Internet of Things (IIoT), deep learning, and knowledge engineering, their integration into the manufacturing industry has become a new opportunity for upgrading. Developing intelligent manufacturing technology is one of the core objectives of the manufacturing industry's transformation, upgrading, and high-quality development strategy. This indicates that traditional manufacturing models must undergo intelligent transformation to improve the efficiency of intelligent manufacturing. According to statistics from the American Production and Inventory Control Society (APICS), in the product manufacturing process, the manufacturing process itself accounts for only about 5% of the entire process, with the rest entirely spent on logistics; moreover, over 40% of a company's total production costs are spent on logistics. Therefore, how to further improve the efficiency of logistics in manufacturing workshops is a common challenge faced by industrial enterprises in the manufacturing sector.

[0003] As a high-level development model of manufacturing, intelligent manufacturing has become a trend and a major breakthrough for improving the quality of manufacturing in my country. The intelligent transformation of production logistics in discrete manufacturing workshops is one of the important means to achieve this goal. Because digital twin workshop technology enables data interaction and transmission between virtual and physical workshops, managing workshop data becomes more effective. Therefore, the emergence of digital twin technology provides a new approach to improving the efficiency of production logistics in discrete manufacturing workshops. However, research on digital twin workshop technology is still in its early stages. On the one hand, the essence of digital twins is the process of simulating multi-source heterogeneous data. Therefore, a digital twin workshop is the process of digitally deconstructing multi-source heterogeneous data generated in the physical workshop and projecting it into the virtual workshop for digital simulation. However, with the increasing digitalization, networking, and intelligent transformation of discrete manufacturing workshops, hundreds of millions of multi-source heterogeneous historical data, such as delivery task information, transportation information, and location information, are generated during the material distribution and operation process in the production workshop. These data come from diverse sources and have inconsistent formats, leading to distortion problems such as duplication, missing data, and errors when projected into the virtual workshop. On the other hand, the data relationships generated during the material distribution operation in the digital twin workshop are complex. The lack of effective unified management of the material distribution-related data in the workshop can lead to information silos, which in turn makes it difficult to make decisions on the material distribution operation and reduces the efficiency of material distribution.

[0004] In summary, the current challenges facing production logistics in discrete manufacturing workshops can be categorized into three main points: First, how to reduce the time and costs spent on logistics in the manufacturing process and further improve production logistics efficiency; second, how to analyze and utilize the massive amounts of heterogeneous data generated during the material distribution process in digital twin physical workshops, and how to solve the problem of data type integration and unification to provide data support for the formation of high-fidelity virtual workshops; and third, the complex relationships among the distribution data generated during the workshop material distribution process, making it difficult to discover hidden relationships and patterns between data, and lacking collaborative interaction mechanisms between data. Summary of the Invention

[0005] The purpose of this invention is to provide a knowledge-driven digital twin discrete workshop material distribution optimization method, enabling production managers to perform forward-looking path planning and optimization of the workshop material distribution process in a dynamic environment based on the real-time interaction between the physical and virtual workshops, thereby improving the efficiency of material distribution in discrete manufacturing workshops.

[0006] The objective of this invention is achieved through the following technical means: a knowledge-driven digital twin discrete workshop material distribution optimization method, comprising the following steps: Step 1: Extract data about workshop material distribution through knowledge extraction and map it to a graph database to form a knowledge graph of workshop material distribution relationships; Step 2: Construct a digital twin physical workshop model. The real-time status of the AGV material delivery process is perceived through a sensor network. Then, the digital twin virtual workshop is modeled by combining knowledge graph technology and digital twin technology, thus forming a mirror virtual workshop synchronized with the physical workshop. The material types and processing types of the processing stations in the material delivery relationship are mapped one-to-one with the corresponding material locations and processing station locations in the workshop digital model. Based on the mirror virtual workshop, the logical model of material delivery events in the digital twin virtual workshop is carried out through virtual-physical interaction mapping method, providing a foundation for subsequent path optimization. Step 3: Based on the workshop material delivery information data in the knowledge graph, construct workshop material delivery task information, and based on the digital twin physical workshop model, construct a task allocation and scheduling model that takes into account both material delivery time and AGV delivery efficiency. For single-task AGV path optimization problems and multi-task AGV path optimization problems, propose corresponding algorithms to solve path optimization and obtain the optimal path.

[0007] The data on workshop material distribution is data about the material distribution path in the workshop, including at least the type of material, the type of material processed at the processing station, and the production process.

[0008] After the data on workshop material distribution is extracted, it also includes semi-structured data and unstructured data. The semi-structured data is transformed into structured data through a packaging model, and the unstructured data is transformed into structured data through entity recognition and relation extraction.

[0009] The specific steps for transforming unstructured data into structured data are as follows: entity recognition is performed through a bidirectional encoder representation-bidirectional long short-term memory network-conditional random field model, and relation extraction is performed through a cascaded pointer network model.

[0010] Step 2 specifically involves, Step 2.1: Establish a digital twin physical workshop model and obtain real-time data on the material distribution process in the physical workshop, including the location information of logistics equipment and materials. Based on the location information of logistics equipment and materials and the workshop layout, establish the current layout of the workshop and construct a digital twin physical workshop model. Step 2.2: Based on the digital twin physical workshop model, a dynamic mapping relationship between the virtual workshop and the real workshop is constructed using workshop material delivery events. This involves mapping the material types and processing types at processing stations in the real workshop material delivery relationship to the corresponding material locations and processing station locations in the digital twin physical workshop model, thereby making the equipment in the actual physical workshop correspond to the equipment in the virtual workshop constructed by the digital twin physical workshop model. Step 2.3: Use complex event processing technology to describe the processing rules of material delivery events, formulate corresponding rules and constraints that are consistent with actual material delivery, and enable the virtual workshop to display the running logic of AGV path optimization process in material delivery events in the physical workshop. Step 2.4: Based on knowledge graph technology and digital twin physical workshop model, the material delivery operation status of the workshop is simulated and monitored in real time in the virtual workshop composed of the digital twin physical workshop model.

[0011] The mapping relationship between the digital twin physical workshop model and the current workshop layout and logistics equipment location information in step 2.1 is as follows:

[0012]

[0013] In the formula, This refers to an Internet of Things (IoT) network built using the Industrial Internet of Things (IIoT) for interconnecting material distribution resources in a digital twin physical workshop. The network nodes include two types of nodes: production workstations and AGV carts. This includes data information for AGV (Automated Guided Vehicle) logistics equipment, such as location, status, and speed. A logical channel for connecting production workstations and AGV vehicle nodes and supporting their data interaction and industrial communication.

[0014] In step 2.4, the relationship between the digital twin model virtual workshop and the material distribution operation status is as follows:

[0015]

[0016]

[0017]

[0018] In the formula, For virtual workshops; This refers to a collection of resources for material distribution in the workshop, including workstations, AGVs, etc. This represents a set of production processes in the workshop, including the production process and production process parameters. Indicates the material delivery route in the workshop; This indicates the relationship between material distribution resources, production processes, and distribution routes. Indicates the first element in the material distribution resource set. Delivery resources, with various attribute sets including delivery number. Delivery Name Delivery type Distribution geometry set Delivery status set Delivery Action Set Delivery dataset Delivery data interface set ,other ; Indicates the first step in the production process Classification of production processes, and various attribute sets including production process number. Production process name Production process type Production process standard set Production process set Production time requirements set Production constraint set ,other ; Indicates the first in the set of delivery routes Class path, various attribute sets including path number Path name Path type Path length Optimal path Available paths Two-dimensional geometric set of paths Data interface set ,other .

[0019] In step 2.2, the mapping relationship between the virtual workshop and the physical workshop in the digital twin workshop model is as follows:

[0020]

[0021]

[0022] In the formula, For virtual workshops; A collection of virtual workshop resources; A set of physical attributes for the virtual workshop; A collection of material delivery behaviors in a virtual workshop; A set of logical models for material distribution in a virtual workshop; Connect the "physical workshop model" to the "virtual workshop model"; For interconnection; For the virtual-real interaction mapping of digital twin workshops.

[0023] In step 2.3, the material delivery logic is as follows:

[0024]

[0025] In the formula, For the first The first type of distribution resource Delivery-related events, such as AGV delivery preparation, AGV path optimization, material arrival, and material departure events; For a time-constrained set of delivery events within a single AGV, an association matching matrix is ​​generated. Time constraints; The directed graph adjacency matrix for the delivery tasks. The delivery logic for material distribution; The expanded form is denoted as:

[0026] for , denoted as:

[0027] In the association matching matrix, The expanded form is denoted as:

[0028]

[0029] In the formula, For the first Material distribution resources; Starting point To the finish line The number of delivery tasks is determined by the priority of the AGV material delivery tasks. The workshop AGV will deliver materials according to the priority of the delivery tasks.

[0030] Step 3 specifically involves, Step 3.1: Based on the objectives of AGV material delivery task scheduling, a preliminary single-task delivery model is constructed, followed by the establishment of a multi-task model. Step 3.2: Based on the task characteristics of a single-task AGV, use the improved A... The algorithm solves for the shortest path between the two task points separately, specifically including the following steps: Using improved A The algorithm starts from the starting point and searches for the destination, using a cost function. After selecting expansion points, a loop mode is initiated, and the process is repeated continuously until the target point is reached, ultimately generating a driving path, denoted as:

[0031] In the formula, At the current node, when searching for the target point The value is the estimated cost; From the starting node to the current node The actual cost function; For the current node The estimated cost function for the target point; For the newly added turning cost function, improve A A turning cost function is added to the algorithm to reduce the number of turns the AGV makes during material delivery, thereby improving the AGV's delivery efficiency. Recorded as:

[0032]

[0033]

[0034] In the formula, As a turning factor, the AGV single-task material delivery path optimization process is based on This determines whether the turning cost function should be calculated. For path length, The turning coefficient, in the estimated cost function, represents the movement cost between nodes, denoted as:

[0035] In the formula, For nodes x-coordinate For nodes The ordinate; Step 3.3: Based on the task characteristics of multi-task AGVs, the multi-task material delivery path optimization problem is abstracted into a traveling salesman problem. Using an improved ant colony algorithm, with the objective of minimizing the AGV's travel distance, the optimal path for the AGV to pass through all task delivery points is calculated. This specifically includes the following steps: The multi-task material delivery route optimization problem is abstracted as a traveling salesman problem: if tasks... It is a task The subsequent tasks, then the subsequent task variables The value is 1 in all cases and 0 in all other cases, therefore The expression is denoted as:

[0036] The multi-task AGV material delivery path optimization model is denoted as:

[0037]

[0038] In the formula, The objective function for the multi-task AGV material delivery path optimization model is... These are constraints; Let the number of nodes be The number of ants is , Represents a node With nodes The distance between them express and Between The pheromone concentration at time , At any moment, ants The selection of the next node is influenced by the pheromone concentration, so it chooses from the node... Transfer to node The probability is , denoted as:

[0039] In the formula, For distance heuristic function, As a factor of pheromone importance, The pheromone, as the importance factor of the heuristic function, is continuously updated iteratively in the ant colony algorithm, denoted as:

[0040] In the formula, The pheromone volatility coefficient. For the first Only ants at the node With nodes The pheromones released between them have a value of , Indicates the total amount of pheromones. Ants The total distance traveled The total amount of pheromones released by ants during the search. The beneficial effects of this invention are: linking multi-source heterogeneous data generated during material distribution in discrete manufacturing workshops using knowledge graphs to form a unified data type, achieving a structured representation of data knowledge; further data mining and relationship acquisition of the unified data can solve problems such as information silos in data management, enabling the discovery of hidden relationships and patterns between data; mapping knowledge data in the knowledge graph to form a knowledge layer in the digital twin workshop, enabling the establishment of a high-fidelity virtual workshop corresponding to the physical workshop while acquiring massive amounts of multi-source heterogeneous workshop material distribution data; considering the incompatibility between traditional workshop AGV material distribution path optimization methods and the operating mechanism of digital twin workshops, the algorithm for path optimization of single-task AGVs and multi-task AGVs is improved. The improved algorithm is more adaptable to the operating mechanism of digital twin workshops when solving material distribution path optimization problems, thereby improving the efficiency of workshop AGV material distribution processes. Attached Figure Description

[0041] Figure 1 This is the overall flow of the method of the present invention; Figure 2 This invention relates to the knowledge graph construction process and the knowledge-driven digital twin model construction process of the present invention. Figure 3 This invention relates to the material distribution path optimization process based on a workshop digital twin model.

[0042] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Detailed Implementation

[0043] A knowledge-driven digital twin discrete workshop material distribution optimization method includes the following steps: Step 1: Extract data about workshop material distribution through knowledge extraction and map it to a graph database to form a knowledge graph of workshop material distribution relationships; The data on workshop material distribution is data about the material distribution path in the workshop, including at least the type of material, the type of material processed at the processing station, and the production process.

[0044] After the data on workshop material distribution is extracted, in addition to structured data, it also includes semi-structured data and unstructured data. The semi-structured data is transformed into structured data through the packaging model, and the unstructured data is transformed into structured data through entity recognition and relation extraction.

[0045] The specific steps for transforming unstructured data into structured data are as follows: entity recognition is performed through a bidirectional encoder representation-bidirectional long short-term memory network-conditional random field model, and relation extraction is performed through a cascaded pointer network model.

[0046] Step 1.1: Equip unmanned transport vehicles, materials, and processing stations with sensors to collect material delivery data in the workshop. Then, classify the collected delivery data into structured, semi-structured, and unstructured data, with unstructured text data accounting for the largest proportion.

[0047] Step 1.2: Based on the collected data of various types, use knowledge extraction technology to transform the semi-structured and unstructured data in the material distribution process into structured data, forming a unified data type and realizing the structured representation of data knowledge.

[0048] Step 1.2.1: Use a wrapper model to extract semi-structured data from files in formats such as XML and JSON and convert them into structured data.

[0049] Step 1.2.2: Use a bidirectional encoder representation-bidirectional long short-term memory network-conditional random field model and a cascaded pointer network model to transform unstructured data from text data into structured data.

[0050] ①: Entity recognition process The entity recognition process mainly uses a bidirectional encoder representation-bidirectional long short-term memory network-conditional random field model. The model is based on the bidirectional long short-term memory network model, which can learn the context information of words or phrases in the discrete manufacturing workshop material distribution path optimization text dataset. Adding a bidirectional encoder representation can improve the accuracy of entity recognition in the logistics field, and adding a conditional random field model can take into account the relationship between words in the dataset, thus improving the logic of natural language text.

[0051] Combining the advantages of the three models, the main workflow of entity recognition is as follows: First, the training set of text corpus on material distribution path optimization in discrete manufacturing workshops is input into the bidirectional encoder model. The input natural language text is converted into word vector form. Within the same sentence, the model calculates the degree of association between each word and the other words. Based on the calculation results, the weight coefficient matrix is ​​continuously adjusted to obtain the feature vector expression of the word, denoted as:

[0052] In the formula, , , It is a vector matrix of text on optimizing the material distribution path in the workshop. This is the Embedding dimension, and we need to avoid " "The score increases indefinitely with the increase of vector dimension, utilizing..." The weights are fractionalized, the dimensionality is considered, and only the features inherent in the words themselves are considered last. Then, the word vectors are input into a bidirectional long short-term memory (LSTM) network model for training to extract contextual information from the text. Information from consecutive moments at the same time is concatenated and merged according to position. Afterwards, the hidden layers of the bidirectional LSM network model will output... ,in The matrix is In the matrix The number of words in the statement representing the material delivery in the workshop. The number of labeled tags represents the number of tags in a predicted sequence. ,in for The predicted label index at time step is denoted as:

[0053] In the formula, The transition probability matrix is ​​the key component for training the Conditional Random Field (CRF) model. CRF models address the problem of inconsistent labeling between words, such as labeling "AGV" as "workstation" and "workstation" as "material." Here, we train the transition probability matrix to locate the label positions and relationships, denoted as:

[0054] In the formula, In the sequence Generate sequences under the conditions The probability of a correct sequence During training, it is necessary to maximize the log probability of the sequence, denoted as:

[0055] In the formula, During training, we need to maximize the log probability of the sequence. During prediction, we select the prediction result that maximizes the prediction probability, denoted as:

[0056] In the formula, To maximize the prediction probability, the Conditional Random Field (CRF) model is used to annotate the input labeled sequence with entities. The CRF model receives the output from the previous layer and outputs the most likely labeled sequence, ultimately completing the entity recognition task.

[0057] Step 1.3: Map the extracted structured data to the Neo4j high-performance graph database to form the corresponding knowledge graph, and use a graph database management system to manage the formed knowledge graph.

[0058] ② Relation extraction process The relation extraction process primarily utilizes a layered pointer network model. This model involves two steps: First, the input text is encoded in the encoding layer, where head entities are identified in the head entity recognition layer. Second, based on the identified head entities, the corresponding tail entities are identified in the tail entity-relation recognition layer.

[0059] The layered pointer network model uses a bidirectional encoder model as the first encoding layer, denoted as:

[0060]

[0061] In the formula, the input sentence is represented as a vector matrix. , A matrix representing each embedding. Indicates the position in the word sequence , It is the hidden state vector. Indicates the context of the sentence. The number of Transformer structures.

[0062] In the head entity recognition layer of the stacked pointer network model, the start and end positions (0 or 1) of the head entity are marked to determine the head entity position, denoted as:

[0063]

[0064] In the formula, and They represent the first The probability (0 or 1) of the start and end positions of the word vector head entity. It is the first in the input sequence Word vectors at each position, Indicates training weights, Indicates deviation, It is the sigmoid activation function.

[0065] Then, the cascaded pointer network model determines the sentence through optimized computation. The range is denoted as:

[0066] In the formula, Indicates the length of the head entity, if If true, then Otherwise, it is 0. It is the binary representation of the header position of the entity. It is indicated by the tail position. These are parameters that require practice.

[0067] In the tail entity-relationship identification layer of the cascaded pointer network model, different relations are used to label the tail entity, denoted as:

[0068]

[0069] In the formula, and This represents the probability of the tail entity's location, while This indicates the first step in the head entity detection process. Vector form of an object.

[0070] For relationships The annotation process for the tail entity annotation tool is denoted as:

[0071] In the formula, Indicates the starting position of the tail entity. This indicates the end position of the tail entity. For "null" objects... , and All are 0. It is a relationship The parameters of the tail entity tagger are then determined. The obtained tail entities need to be matched with the relation recognition layer. Finally, the relation extraction process is completed, and the corresponding structured data is obtained.

[0072] like Figure 1 and Figure 2As shown, by collecting information such as the physical workshop layout and the location of each piece of equipment, and using RFID sensors (Radio Frequency Identification) to identify material information in real time, track material location and link inventory, etc., the entire material tracking process is completed; and LiDAR sensors are used for real-time obstacle avoidance and path planning, etc., for AGV path navigation and obstacle avoidance.

[0073] And various logistics systems, such as Warehouse Management System (WMS) and Transportation Management System (TMS). This information is collected to obtain various structured, semi-structured, and unstructured data. Entity identification, relation extraction, and knowledge mapping are then performed on this data to form a knowledge graph about logistics and transportation.

[0074] This establishes a mapping relationship between the actual physical workshop and the virtual workshop to simulate the physical workshop. Based on the material distribution event information extracted from the knowledge graph, various information actually collected is mapped to the virtual workshop, and the logistics path is automatically optimized according to the distribution process of logistics resources.

[0075] For example, first, based on the layout of the physical workshop, construct an identical virtual workshop; Furthermore, based on the process flow, for example, if the process of processing station A requires transporting materials from point B to point C after processing, various sensors monitor the status of the material shelves at point B, replenish the stock, monitor whether the unmanned transport vehicle (AGV) transports the goods to point A, and monitor whether the processing station has completed the transportation. Once completed, the AGV is dispatched to transport the processed goods to point C. Through sensors, the status of each link in the material transportation process is monitored in real time, thereby realizing automatic handling according to the specific process flow.

[0076] Step 2: Construct a digital twin physical workshop model. The real-time status of the AGV material delivery process is perceived through a sensor network. Then, the digital twin virtual workshop is modeled by combining knowledge graph technology and digital twin technology, thus forming a mirror virtual workshop synchronized with the physical workshop. The material types and processing types of the processing stations in the material delivery relationship are mapped one-to-one with the corresponding material locations and processing station locations in the workshop digital model. Based on the mirror virtual workshop, the logical model of material delivery events in the digital twin virtual workshop is carried out through virtual-physical interaction mapping method, providing a foundation for subsequent path optimization. Step 3: Based on the workshop material delivery information data in the knowledge graph, construct workshop material delivery task information, and based on the digital twin physical workshop model, construct a task allocation and scheduling model that takes into account both material delivery time and AGV delivery efficiency. For single-task AGV path optimization problems and multi-task AGV path optimization problems, propose corresponding algorithms to solve path optimization and obtain the optimal path.

[0077] Step 2 specifically involves, Step 2.1: Establish a digital twin physical workshop model and obtain real-time data on the material distribution process in the physical workshop, including the location information of logistics equipment and materials. Based on the location information of logistics equipment and materials and the workshop layout, establish the current layout of the workshop and construct a digital twin physical workshop model. The mapping relationship between the digital twin physical workshop model and the current workshop layout and logistics equipment location information in step 2.1 is as follows:

[0078]

[0079] In the formula, This refers to an Internet of Things (IoT) network built using the Industrial Internet of Things (IIoT) for interconnecting material distribution resources in a digital twin physical workshop. The network nodes include two types of nodes: production workstations and AGV carts. This includes data information for AGV (Automated Guided Vehicle) logistics equipment, such as location, status, and speed. A logical channel for connecting production workstations and AGV vehicle nodes and supporting their data interaction and industrial communication.

[0080] Step 2.2: Based on the digital twin physical workshop model, a dynamic mapping relationship between the virtual workshop and the real workshop is constructed using workshop material delivery events. This involves mapping the material types and processing types at processing stations in the real workshop material delivery relationship to the corresponding material locations and processing station locations in the digital twin physical workshop model, thereby making the equipment in the actual physical workshop correspond to the equipment in the virtual workshop constructed by the digital twin physical workshop model. Step 2.3: Use complex event processing technology to describe the processing rules of material delivery events, formulate corresponding rules and constraints that are consistent with actual material delivery, and enable the virtual workshop to display the running logic of AGV path optimization process in material delivery events in the physical workshop. In step 2.2, the mapping relationship between the virtual workshop and the physical workshop in the digital twin workshop model is as follows:

[0081]

[0082]

[0083] In the formula, For virtual workshops; A collection of virtual workshop resources; A set of physical attributes for the virtual workshop; A collection of material delivery behaviors in a virtual workshop; A set of logical models for material distribution in a virtual workshop; Connect the "physical workshop model" to the "virtual workshop model"; For interconnection; For the virtual-real interaction mapping of digital twin workshops.

[0084] Having established an initial connection between the virtual and physical workshops, the next step is to achieve a deeper connection and build a dynamic mapping relationship between them. This connection will be driven by workshop material delivery events, denoted as:

[0085] In the formula, It is a set of virtual workshop attributes; It is a delivery event pattern; if these two match, then the action behavior... occur.

[0086] In step 2.3, the material delivery logic is as follows:

[0087]

[0088] In the formula, For the first The first type of distribution resource Delivery-related events, such as AGV delivery preparation, AGV path optimization, material arrival, and material departure events; For a time-constrained set of delivery events within a single AGV, an association matching matrix is ​​generated. Time constraints; The directed graph adjacency matrix for the delivery tasks. The delivery logic for material distribution; The expanded form is denoted as:

[0089] for , denoted as:

[0090] In the association matching matrix, The expanded form is denoted as:

[0091]

[0092] In the formula, For the first Material distribution resources; Starting point To the finish line The number of delivery tasks is determined by the priority of the AGV material delivery tasks. The workshop AGV will deliver materials according to the priority of the delivery tasks.

[0093] Step 2.4: Based on knowledge graph technology and digital twin physical workshop model, the material delivery operation status of the workshop is simulated and monitored in real time in the virtual workshop composed of the digital twin physical workshop model.

[0094] In step 2.4, the relationship between the digital twin model virtual workshop and the material distribution operation status is as follows:

[0095]

[0096]

[0097]

[0098] In the formula, For virtual workshops; This refers to a collection of resources for material distribution in the workshop, including workstations, AGVs, etc. This represents a set of production processes in the workshop, including the production process and production process parameters. Indicates the material delivery route in the workshop; This indicates the relationship between material distribution resources, production processes, and distribution routes. Indicates the first element in the material distribution resource set. Delivery resources, with various attribute sets including delivery number. Delivery Name Delivery type Distribution geometry set Delivery status set Delivery Action Set Delivery dataset Delivery data interface set ,other ; Indicates the first step in the production process Classification of production processes, and various attribute sets including production process number. Production process name Production process type Production process standard set Production process set Production time requirements set Production constraint set ,other ; Indicates the first in the set of delivery routes Class path, various attribute sets including path number Path name Path type Path length Optimal path Available paths Two-dimensional geometric set of paths Data interface set ,other .

[0099] Step 3 specifically involves, Step 3.1: To achieve the goal of AGV material delivery task scheduling, a relatively simple single-task delivery model is initially constructed, followed by a more complex multi-task model. The specific delivery task scheduling model is represented as follows: After receiving a material delivery task, the AGV will enter three stages: loading, transporting, and unloading. Let the task number be... =(1,2,…, n The car's number is =(1,2,…, m Then the car is performing the task. At that time, no-load duration Recorded as:

[0100] Carrying time Recorded as:

[0101] Total time for a single task Recorded as:

[0102] Overall efficiency of AGV single-task delivery Recorded as:

[0103] Task completion time Recorded as:

[0104] Complete the task within the specified time. Recorded as:

[0105] In the formula, This refers to the delivery time period for the materials; The difference between material delivery time and standard time Recorded as:

[0106] The objective function for delivery task scheduling is denoted as:

[0107] The correspondence between delivery tasks and AGVs is denoted as:

[0108] The relevant variables for the AGV loading status are denoted as:

[0109] The relevant variables for the AGV unloading status are denoted as:

[0110] The status of each task node is recorded as follows:

[0111] The constraints for delivery task scheduling are as follows:

[0112] In the formula, This means that an AGV completes only one delivery task.

[0113]

[0114] In the formula, This indicates that the number of loads and unloads for each task remains consistent.

[0115]

[0116] In the formula, This means that at any given moment, each node on the map will not be occupied by two or more AGVs simultaneously; at most, only one AGV will occupy it.

[0117] In the formula, This means that any AGV will not carry more than 5 pallets, and the weight it carries will meet the predetermined requirements by default.

[0118] Step 3.2: Based on the task characteristics of a single-task AGV, use the improved A... The algorithm solves for the shortest path between the two task points separately, specifically including the following steps: Using improved A The algorithm starts from the starting point and searches for the destination, using a cost function. After selecting expansion points, a loop mode is initiated, and the process is repeated continuously until the target point is reached, ultimately generating a driving path, denoted as:

[0119] In the formula, At the current node, when searching for the target point The value is the estimated cost; From the starting node to the current node The actual cost function; For the current node The estimated cost function for the target point; For the newly added turning cost function, improve A A turning cost function is added to the algorithm to reduce the number of turns the AGV makes during material delivery, thereby improving the AGV's delivery efficiency. Recorded as:

[0120]

[0121]

[0122] In the formula, As a turning factor, the AGV single-task material delivery path optimization process is based on This determines whether the turning cost function should be calculated. For path length, The turning coefficient, in the estimated cost function, represents the movement cost between nodes, denoted as:

[0123] In the formula, For nodes x-coordinate For nodes The ordinate.

[0124] Step 3.3: Based on the task characteristics of multi-task AGVs, the multi-task material delivery path optimization problem is abstracted into a traveling salesman problem. Using an improved ant colony algorithm, with the objective of minimizing the AGV's travel distance, the optimal path for the AGV to pass through all task delivery points is calculated. This specifically includes the following steps: The multi-task material delivery route optimization problem is abstracted as a traveling salesman problem: if tasks... It is a task The subsequent tasks, then the subsequent task variables The value is 1 in all cases and 0 in all other cases, therefore The expression is denoted as:

[0125] The multi-task AGV material delivery path optimization model is denoted as:

[0126]

[0127] In the formula, The objective function for the multi-task AGV material delivery path optimization model is... These are constraints; Let the number of nodes be The number of ants is , Represents a node With nodes The distance between them express and Between The pheromone concentration at time , At any moment, ants The selection of the next node is influenced by the pheromone concentration, so it chooses from the node... Transfer to node The probability is , denoted as:

[0128] In the formula, For distance heuristic function, As a factor of pheromone importance, The pheromone, as the importance factor of the heuristic function, is continuously updated iteratively in the ant colony algorithm, denoted as:

[0129] In the formula, The pheromone volatility coefficient. For the first Only ants at the node With nodes The pheromones released between them generally take a value of , Indicates the total amount of pheromones. Ants The total distance traveled. This represents the total amount of pheromones released by ants during the search.

[0130] like Figure 1 and 2 As shown, once the virtual workshop is constructed, the mapping relationship between the virtual and physical workshops is established, and the operational logic and requirements for material delivery within the virtual workshop are set, the material delivery path is optimized based on sensor data from the physical workshop. Specifically, as follows... Figure 3 As shown, First, obtain the task information for material delivery, and then optimize using different methods depending on whether it involves multiple tasks or a single task. In single-task mode, read environmental information and determine the coordinates of the starting point and the target point; Define reached list point A and unreached list point B, both initialized to empty; The starting point enters Table A, and obstacles enter Table B; Using improved A The algorithm starts from point A and searches for the destination, using a cost function. Filter expansion points; Find the one with the smallest For nodes with values, add them to table B, and add their adjacent nodes to table A as parent nodes; If table B contains the target node (i.e., the destination), then the final cost is calculated to obtain the final path; otherwise, the improved A is reused. The algorithm starts from point A and searches for the destination, using a cost function. Filter expansion points; If a single logistics delivery task is completed, then check if all other tasks are completed. If not, return to the beginning to retrieve task information and continue to optimize other tasks. If completed, the process ends. If the single-task delivery process is not completed, the route optimization will be performed again. When multiple tasks are performed, the environmental information is also read to determine the coordinates of the task points; The simulation environment of the virtual workshop is updated based on sensor data; Using an improved ant colony algorithm to solve point-to-point paths; Simulation verifies the effectiveness of the path, executes the task according to the simulation results, and perceives the operating environment of the unmanned transport vehicle (AGV) in real time during task execution; If a dangerous state is detected, the virtual simulation environment will be updated and the path will be replanned. If there is no danger, continue executing the simulation results until the multitasking is completed. If the multitasking is not completed, continue executing the tasks according to the simulation results.

[0131] Once all tasks are completed, the transportation ends; otherwise, a new task is obtained and the route planning continues.

Claims

1. A knowledge-driven digital twin discrete workshop material distribution optimization method, characterized in that... This includes the following steps: Step 1: Extract data about workshop material distribution through knowledge extraction and map it to a graph database to form a knowledge graph of workshop material distribution relationships; Step 2: Construct a digital twin physical workshop model. The real-time status of the AGV material delivery process is perceived through a sensor network. Then, the digital twin virtual workshop is modeled by combining knowledge graph technology and digital twin technology, thus forming a mirror virtual workshop synchronized with the physical workshop. The material types and processing types of the processing stations in the material delivery relationship are mapped one-to-one with the corresponding material locations and processing station locations in the workshop digital model. Based on the mirror virtual workshop, the logical model of material delivery events in the digital twin virtual workshop is carried out through virtual-physical interaction mapping method, providing a foundation for subsequent path optimization. Step 3: Based on the workshop material delivery information data in the knowledge graph, construct workshop material delivery task information, and based on the digital twin physical workshop model, construct a task allocation and scheduling model that takes into account both material delivery time and AGV delivery efficiency. For single-task AGV path optimization problems and multi-task AGV path optimization problems, propose corresponding algorithms to solve path optimization and obtain the optimal path.

2. The knowledge-driven digital twin discrete workshop material distribution optimization method according to claim 1, characterized in that: The data on workshop material distribution is data about the material distribution path in the workshop, including at least the type of material, the type of material processed at the processing station, and the production process.

3. The knowledge-driven digital twin discrete workshop material distribution optimization method according to claim 1, characterized in that: After the data on workshop material distribution is extracted, it also includes semi-structured data and unstructured data. The semi-structured data is transformed into structured data through a packaging model, and the unstructured data is transformed into structured data through entity recognition and relation extraction.

4. The knowledge-driven digital twin discrete workshop material distribution optimization method according to claim 3, characterized in that: The specific steps for transforming unstructured data into structured data are as follows: entity recognition is performed through a bidirectional encoder representation-bidirectional long short-term memory network-conditional random field model, and relation extraction is performed through a cascaded pointer network model.

5. The knowledge-driven digital twin discrete workshop material distribution optimization method according to claim 1, characterized in that: Step 2 specifically involves, Step 2.1: Establish a digital twin physical workshop model and obtain real-time data on the material distribution process in the physical workshop, including the location information of logistics equipment and materials. Based on the location information of logistics equipment and materials and the workshop layout, establish the current layout of the workshop and construct a digital twin physical workshop model. Step 2.2: Based on the digital twin physical workshop model, a dynamic mapping relationship between the virtual workshop and the real workshop is constructed using workshop material delivery events. This involves mapping the material types and processing types at processing stations in the real workshop material delivery relationship to the corresponding material locations and processing station locations in the digital twin physical workshop model, thereby making the equipment in the actual physical workshop correspond to the equipment in the virtual workshop constructed by the digital twin physical workshop model. Step 2.3: Use complex event processing technology to describe the processing rules of material delivery events, formulate corresponding rules and constraints that are consistent with actual material delivery, and enable the virtual workshop to display the running logic of AGV path optimization process in material delivery events in the physical workshop. Step 2.4: Based on knowledge graph technology and digital twin physical workshop model, the material delivery operation status of the workshop is simulated and monitored in real time in the virtual workshop composed of the digital twin physical workshop model.

6. The knowledge-driven digital twin discrete workshop material distribution optimization method according to claim 1, characterized in that: The mapping relationship between the digital twin physical workshop model and the current workshop layout and logistics equipment location information in step 2.1 is as follows: In the formula, This refers to an Internet of Things (IoT) network built using the Industrial Internet of Things (IIoT) for interconnecting material distribution resources in a digital twin physical workshop. The network nodes include two types of nodes: production workstations and AGV carts. This includes data information for AGV (Automated Guided Vehicle) logistics equipment, such as location, status, and speed. A logical channel for connecting production workstations and AGV vehicle nodes and supporting their data interaction and industrial communication.

7. The knowledge-driven digital twin discrete workshop material distribution optimization method according to claim 1, characterized in that: In step 2.4, the relationship between the digital twin model virtual workshop and the material distribution operation status is as follows: In the formula, For virtual workshops; This refers to a collection of resources for material distribution in the workshop, including workstations, AGVs, etc. This represents a set of production processes in the workshop, including the production process and production process parameters. Indicates the material delivery route in the workshop; This indicates the relationship between material distribution resources, production processes, and distribution routes. Indicates the first element in the material distribution resource set. Delivery resources, with various attribute sets including delivery number. Delivery Name Delivery type Distribution geometry set Delivery status set Delivery Action Set Delivery dataset Delivery data interface set ,other ; Indicates the first step in the production process Classification of production processes, and various attribute sets including production process number. Production process name Production process type Production process standard set Production process set Production time requirements set Production constraint set ,other ; Indicates the first in the set of delivery routes Class path, various attribute sets including path number Path name Path type Path length Optimal path Available paths Two-dimensional geometric set of paths Data interface set ,other .

8. The knowledge-driven digital twin discrete workshop material distribution optimization method according to claim 1, characterized in that: In step 2.2, the mapping relationship between the virtual workshop and the physical workshop in the digital twin workshop model is as follows: In the formula, For virtual workshops; A collection of virtual workshop resources; A set of physical attributes for the virtual workshop; A collection of material delivery behaviors in a virtual workshop; A set of logical models for material distribution in a virtual workshop; Connect the "physical workshop model" to the "virtual workshop model"; For interconnection; For the virtual-real interaction mapping of digital twin workshops.

9. The knowledge-driven digital twin discrete workshop material distribution optimization method according to claim 1, characterized in that: In step 2.3, the material delivery logic is as follows: In the formula, For the first The first type of distribution resource Delivery-related events, such as AGV delivery preparation, AGV path optimization, material arrival, and material departure events; For a time-constrained set of delivery events within a single AGV, an association matching matrix is ​​generated. Time constraints; The directed graph adjacency matrix for the delivery tasks. The delivery logic for material distribution; The expanded form is denoted as: for , denoted as: In the association matching matrix, The expanded form is denoted as: In the formula, For the first Material distribution resources; Starting point To the finish line The number of delivery tasks is determined by the priority of the AGV material delivery tasks. The workshop AGV will deliver materials according to the priority of the delivery tasks.

10. The knowledge-driven digital twin discrete workshop material distribution optimization method according to claim 1, characterized in that: Step 3 specifically involves, Step 3.1: Based on the objectives of AGV material delivery task scheduling, a preliminary single-task delivery model is constructed, followed by the establishment of a multi-task model. Step 3.2: Based on the task characteristics of a single-task AGV, use the improved A... The algorithm solves for the shortest path between the two task points separately, specifically including the following steps: Using improved A The algorithm starts from the starting point and searches for the destination, using a cost function. After selecting expansion points, a loop mode is initiated, and the process is repeated continuously until the target point is reached, ultimately generating a driving path, denoted as: In the formula, At the current node, when searching for the target point The value is the estimated cost; From the starting node to the current node The actual cost function; For the current node The estimated cost function for the target point; For the newly added turning cost function, improve A A turning cost function is added to the algorithm to reduce the number of turns the AGV makes during material delivery, thereby improving the AGV's delivery efficiency. Recorded as: In the formula, As a turning factor, the AGV single-task material delivery path optimization process is based on This determines whether the turning cost function should be calculated. For path length, The turning coefficient, in the estimated cost function, represents the movement cost between nodes, denoted as: In the formula, For nodes x-coordinate For nodes The ordinate; Step 3.3: Based on the task characteristics of multi-task AGVs, the multi-task material delivery path optimization problem is abstracted into a traveling salesman problem. Using an improved ant colony algorithm, with the objective of minimizing the AGV's travel distance, the optimal path for the AGV to pass through all task delivery points is calculated. This specifically includes the following steps: The multi-task material delivery route optimization problem is abstracted as a traveling salesman problem: if tasks... It is a task The subsequent tasks, then the subsequent task variables The value is 1 in all cases and 0 in all other cases, therefore The expression is denoted as: The multi-task AGV material delivery path optimization model is denoted as: In the formula, The objective function for the multi-task AGV material delivery path optimization model is... These are constraints; Let the number of nodes be The number of ants is , Represents a node With nodes The distance between them express and Between The pheromone concentration at time , At any moment, ants The selection of the next node is influenced by the pheromone concentration, so it chooses from the node... Transfer to node The probability is , denoted as: In the formula, For distance heuristic function, As a factor of pheromone importance, The pheromone, as the importance factor of the heuristic function, is continuously updated iteratively in the ant colony algorithm, denoted as: In the formula, The pheromone volatility coefficient. For the first Only ants at the node With nodes The pheromones released between them have a value of , Indicates the total amount of pheromones. Ants The total distance traveled This represents the total amount of pheromones released by ants during the search.

Images