A multi-source data driven manufacturing enterprise multi-agent collaboration system

Abstract

The application relates to the technical field of intelligent manufacturing and production scheduling, in particular to a multi-source data driven manufacturing enterprise multi-agent collaborative system, which comprises a data acquisition and knowledge graph construction module, which acquires multi-source data and constructs a process knowledge graph; a scheduling task execution and feedback optimization module, which calculates the process readiness of a current target process through a process preposition constraint readiness submodule of a scheduling agent; when the process readiness reaches the standard, a task is issued to a candidate equipment pool; a device agent in an equipment-process performance quantification submodule responds to the task, calculates equipment-process performance indexes and feeds back to the scheduling agent of a comprehensive scheduling bid decision submodule, calculates a comprehensive allocation score after receiving the feedback, and awards the task to the candidate equipment with the highest score. The application aims to realize real-time sensing of dynamic equipment performance and a multi-agent collaborative scheduling method strictly following process timing constraints.

Description

Technical Field

[0001] This application relates to the field of intelligent manufacturing and production scheduling technology, specifically to a multi-agent collaborative system for manufacturing enterprises driven by multi-source data. Background Technology

[0002] With the deepening development of intelligent manufacturing and the Industrial Internet, manufacturing enterprises are facing flexible production demands characterized by diverse varieties, variable batch sizes, and short delivery times. Against this backdrop, traditional centralized production scheduling systems are gradually revealing problems such as slow response times, poor adaptability, and insufficient utilization of multi-source heterogeneous data. Multi-agent systems, due to their distributed, self-organizing, and collaborative decision-making characteristics, have become one of the effective technological approaches to solving these problems.

[0003] However, in a flexible, multi-product manufacturing environment, multiple candidate devices are typically available for the same process, and these devices vary significantly in terms of processing efficiency, processing quality, current occupancy status, changeover time, and energy consumption. Traditional scheduling methods often prioritize minimizing completion time as the sole optimization objective, or simply weight the processes in multi-objective optimization, leading to efficiency losses such as idle equipment and work-in-process inventory buildup during actual execution. Furthermore, the coupling and conflicts between multiple performance indicators such as processing efficiency, yield rate, changeover time loss, and energy consumption lack a systematic and comprehensive scoring mechanism to support collaborative task allocation decisions among multiple agents. Summary of the Invention

[0004] In view of the above, it is necessary to provide a multi-agent collaborative system for manufacturing enterprises driven by multi-source data to solve the above problems.

[0005] One embodiment of this application provides a multi-source data-driven multi-agent collaborative system for manufacturing enterprises, the system comprising: Data Acquisition and Knowledge Graph Construction Module: Collects multi-source data from manufacturing enterprises, extracts process features from unstructured data, and constructs a process knowledge graph that includes hard constraints on process topology and soft constraints on resource attributes; The pre-process constraint readiness submodule configures a scheduling agent to monitor the process status. Based on the real-time processing queue and efficiency of the equipment where the pre-process of the current target task is located, it dynamically predicts the remaining completion time of the pre-process and calculates the process readiness of the current target task. When the process readiness reaches the target, it publishes a task message to the candidate equipment pool. The equipment-process performance quantification submodule configures equipment intelligent agents to respond to the task messages, extracts parameters from historical process data and process knowledge graphs, analyzes the processing efficiency, product processing quality, changeover time loss, energy consumption and load of each candidate equipment, calculates equipment-process performance indicators and feeds them back to the comprehensive scheduling bidding decision submodule. The integrated scheduling bidding decision submodule configures a scheduling agent to collect feedback from all device agents, calculates an integrated allocation score based on the process readiness, assigns the task to the candidate device with the highest integrated allocation score, and updates the queue status. The device agent synchronizes the updated queue status to the scheduling agent to trigger the next stage of dynamic scheduling.

[0006] Preferably, the dynamic prediction of the remaining completion time of the preceding process includes the following steps: Based on the process knowledge graph, identify all the preceding processes of the current target task; for each preceding process, obtain the real-time status of the equipment assigned to it, and obtain the remaining completion time based on the completion time of the equipment from the current processing process to the preceding process of the current target task and the processing efficiency of the equipment in the same historical target task process.

[0007] Preferably, the calculation of the process readiness of the current target task specifically involves: Take the maximum value from the remaining completion time of all preceding processes to determine the shortest time that the target task still needs to wait. Multiply the shortest time with a preset attenuation coefficient and use the negative correlation mapping result as the process readiness of the current target task.

[0008] Preferably, the processing efficiency of the candidate equipment is specifically the ratio of the standard working time of the target process in the process knowledge graph to the average processing time of the candidate equipment in the history of processing the target process.

[0009] Preferably, the product processing quality is obtained through the yield rate, and the specific process is as follows: The standard values ​​and tolerance ranges of quality parameters for the target process are extracted from the process knowledge graph. Combined with the historical measured data of the candidate equipment, the normalized tolerance deviation of each key parameter in each historical sample is calculated. The maximum normalized tolerance deviation among all key parameters in each historical sample is obtained. The average value of the maximum normalized tolerance deviation obtained from all historical samples is obtained. The negative exponential function mapping result after multiplying the average value with the preset adjustment coefficient is used as the yield rate.

[0010] Preferably, the specific process for obtaining the changeover time loss is as follows: Extract the processing attributes of process nodes from the process knowledge graph, including geometric features, parameter features, resource features and environmental features, and construct a high-dimensional feature vector; The similarity between the high-dimensional feature vectors of the current process and the target process of each candidate device is analyzed, and the difference between the value 1 and the similarity is taken as the deviation. The changeover time loss is obtained by accumulating the product of the complexity coefficient and the deviation, combined with the preset basic changeover time of each candidate device. The complexity coefficient is preset by performing linear regression fitting on the historical changeover data of each candidate device.

[0011] Preferably, the performance index of the computing device-process is positively correlated with the processing efficiency and the product processing quality, and negatively correlated with the changeover time loss, energy consumption and load.

[0012] Preferably, the comprehensive allocation score is obtained by weighted summation of the process readiness degree and the normalized results of the equipment-process performance index, with the weights being preset values.

[0013] Preferably, if there are multiple highest scores in the comprehensive allocation, the following indicators are compared: the earliest expected start time, the average energy consumption, and the current allocated queue length. The smaller the value, the higher the priority.

[0014] Preferably, if the scheduling agent does not receive a confirmation response from the selected device within a preset time, the grant will be automatically revoked and the task will be reassigned to the candidate device with the second-highest score.

[0015] This application has at least the following beneficial effects: This application, through its data acquisition and knowledge graph construction module, not only constructs a process knowledge graph containing entity nodes and semantic relationship edges, providing a clear hard constraint foundation for process topology for scheduling, but also makes implicit experience knowledge explicit through quantified relationships (such as the calculation of changeover time loss based on process vector similarity), significantly improving the scheduling system's understanding depth and response speed of complex process logic.

[0016] Building upon this foundation, this application utilizes an equipment-process performance quantification submodule to deeply consider the efficiency and yield of specific processes processed by the equipment in the past, and integrates changeover time loss, normalized energy consumption, and current queuing load calculated based on cosine similarity. This mechanism effectively overcomes the shortcomings of traditional scheduling that relies solely on standard working hours, making task allocation decisions more closely aligned with the actual availability and health status of the equipment, thereby achieving high-precision matching of production resources.

[0017] Furthermore, this application calculates the estimated completion time of the set of preceding processes through a process pre-constraint readiness submodule and introduces an exponential decay function to quantify the maturity of the current process allocation (process readiness). By weightedly fusing process readiness with equipment performance indicators to generate a comprehensive allocation score, it effectively avoids logical deadlocks and equipment idleness caused by "subsequent processes being allocated before preceding processes". This mechanism ensures the global temporal consistency of the comprehensive scheduling bidding decision submodule in the multi-agent distributed decision-making process, greatly reducing the risks of queuing congestion and work-in-process inventory backlog on the manufacturing floor. Attached Figure Description

[0018] Figure 1 A block diagram of a multi-agent collaborative system for manufacturing enterprises driven by multi-source data is provided in this application. Figure 2 The specific flowchart of the multi-agent closed-loop cooperative control provided in this application is shown. Detailed Implementation

[0019] In the description of the embodiments of this application, the words "exemplary," "or," and "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary," "or," and "for example" is intended to present the relevant concepts in a specific manner.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this application's specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0021] It should also be noted that the terms "first" and "second" in this application and its accompanying drawings are used to distinguish similar objects, rather than to describe a specific order or sequence. The methods disclosed in the embodiments of this application or the methods shown in the flowcharts include one or more steps for implementing the method. Without departing from the scope of protection of this application, the execution order of multiple steps can be interchanged, and some steps can also be deleted.

[0022] The following description, in conjunction with the accompanying drawings, details a specific solution for a multi-source data-driven multi-agent collaborative system for manufacturing enterprises provided in this application.

[0023] Please see Figure 1The diagram illustrates a block diagram of a multi-source data-driven multi-agent collaborative system for manufacturing enterprises, provided in one embodiment of this application. The system includes: a data acquisition and knowledge graph construction module, and a scheduling task execution and feedback optimization module. The scheduling task execution and feedback optimization module includes a process pre-constraint readiness submodule, an equipment-process performance quantification submodule, and a comprehensive scheduling bidding decision submodule.

[0024] This application first proposes a multi-source data-driven multi-agent collaborative system for manufacturing enterprises, applied in the field of intelligent manufacturing and production scheduling technology. The system includes: Data Acquisition and Knowledge Graph Construction Module: Collects multi-source data from manufacturing enterprises, extracts process features from unstructured data, and constructs a process knowledge graph that includes hard constraints on process topology and soft constraints on resource attributes.

[0025] This application connects to ERP or e-commerce platforms in real time via API to obtain key information such as order quantity, product type, and delivery time.

[0026] Unstructured data such as process documents and operation logs are integrated and transformed into a structured format using OCR and NLP technologies. This results in a process knowledge graph containing both hard constraints (process sequence, equipment compatibility) and soft constraints (minimizing waiting time, maximizing equipment utilization). The process knowledge graph carries the hard constraints of process logic through the topological relationships between entities and the soft constraint parameters required for scheduling optimization through node attributes. The process knowledge graph acquisition process is as follows: First, define the entity node types of the process knowledge graph, including: product nodes, process nodes, equipment nodes, material nodes, and quality characteristic nodes.

[0027] Secondly, define the semantic relationship edge types between entity nodes, including: "Product_Contains_Process", "Process_Predecessor_Process", "Process_Successor_Process", "Process_Executable on_Equipment", "Process_Consumes_Materials", "Process_Constrains_Quality Characteristics", and "Equipment_Switches to_Equipment".

[0028] For each entity node and its corresponding relationships, the relevant attribute fields are extracted and populated. During the structured graph storage and graph construction process, structured data (such as product models in ERP, process definitions in MES, and equipment ledgers) is directly populated into the corresponding node attributes through data mapping rules. Unstructured data (such as process documents and operation logs) uses OCR and NLP technologies to extract key information: process documents extract process names, process sequence descriptions, equipment requirement descriptions, and tolerance parameters; operation logs extract actual equipment changeover records to calculate the average changeover time. Finally, all nodes, relationships, and attributes are imported into the graph database to form a process knowledge graph that can be queried and reasoned about in real time.

[0029] All collected data is stored in a unified data lake (a data lake is a storage system used to store large amounts of structured, semi-structured and unstructured data in their raw format).

[0030] At this point, the collection of relevant data and the acquisition of the process knowledge graph have been completed.

[0031] The scheduling task execution and feedback optimization module: The scheduling agent monitors the process status through the process pre-constraint readiness submodule, calculates the process readiness of the current target process based on the estimated completion time of the preceding processes, and issues the task to the candidate equipment pool when the process readiness reaches the target. The equipment agent in the equipment-process performance quantification submodule responds to the task, calculates the equipment-process performance index by combining local historical processing data and knowledge graph parameters, and provides feedback to the scheduling agent in the comprehensive scheduling bidding decision submodule. After collecting all feedback, the scheduling agent calculates the comprehensive allocation score based on the process readiness and assigns the task to the candidate equipment with the highest score.

[0032] Before scheduling the agent to broadcast the process task, the state information of the previous process is first obtained and the process readiness of the current process is calculated. .like Furthermore, the preceding processes are not yet completed, and the task remains in the waiting pool; only when The task message broadcast is triggered only when all preceding processes are completed. The device agent only returns the raw performance indicators in the response message and does not calculate the final comprehensive allocation score itself.

[0033] This application analyzes orders based on the product's technological process. The process knowledge graph, as the core knowledge base of intelligent manufacturing, stores complete information about the product's technological processes in a structured manner. This typically includes process nodes (detailed information on each production step), process relationships (sequence, parallelism, and dependency relationships between processes), resource constraints (equipment, materials, manpower, and other resources required for each process), time parameters (standard working hours, setup time, changeover time, etc.), and quality requirements (quality control points and standards for each process). After obtaining an order, based on the product model in the order, the application retrieves the corresponding process route from the process knowledge graph to obtain the complete process sequence and process parameters.

[0034] For each product's process sequence, there are often multiple pieces of equipment capable of completing the task for each process. However, these different pieces of equipment vary in processing capacity, current occupancy status, changeover time, energy consumption, and maintenance schedule. Therefore, each process corresponds to a dedicated pool of candidate equipment. This application extracts historical processing data for similar processes from each candidate equipment and quantifies and generates performance indicators for its execution of the current process. The specific process is as follows: For the j-th process of order i The scheduling agent obtains the set of candidate equipment for the process based on the process knowledge graph. ,in, This indicates that the scheduling agent generates a process task message, which includes the following fields: process ID, order ID, product model, standard working hours for the process, list of materials required for the process and reserve quantity, process constraints, key parameters of quality requirements, and release time. The scheduling agent broadcasts the message to the device monitoring agent corresponding to the candidate device set through Agent Communication Language (ACL). After receiving the task message, the monitoring agent corresponding to each candidate device calculates its own device-process performance index and process readiness for that process based on local real-time status data (current processing queue, remaining processing time, equipment health status, energy consumption curve, etc.). After the bidding window closes, the scheduling agent performs global max-min normalization on all equipment data under the target task, then calculates the comprehensive allocation score, and returns a bidding response message containing the comprehensive allocation score, the earliest expected start time (based on the currently allocated process queue of the device, the remaining processing time of all processes in the queue is accumulated, and the total time is the earliest time when the device can process the new task), the expected processing time (in this embodiment, the standard working time of the process is obtained from the process knowledge graph and corrected by the processing efficiency E of the candidate device, i.e.: expected processing time = standard working time / E), and the expected energy consumption (calculated based on the product of the rated power of the equipment in the graph and the expected processing time) to the scheduling agent. Among them, the process readiness, device-process performance index, and comprehensive allocation score are obtained by the following sub-modules.

[0035] The pre-process constraint readiness submodule configures the scheduling agent to monitor the process status. Based on the real-time processing queue and efficiency of the equipment where the pre-process of the current target task is located, it dynamically predicts the remaining completion time of the pre-process and calculates the process readiness of the current target task. When the process readiness reaches the target, it publishes a task message to the candidate equipment pool.

[0036] In multi-agent scheduling, if task allocation is based solely on the single-point performance of equipment, hard constraints on the process path may be ignored (e.g., if the preceding "milling" process is not yet completed, assigning the "heat treatment" task to an idle equipment will result in subsequent equipment waiting and work-in-process inventory buildup). Therefore, in this embodiment, before broadcasting tasks and bidding for performance, the scheduling agent first introduces a pre-constraint satisfaction evaluation system. Only when the preceding process of the current target process has been completed or is about to be completed is it allowed to be released to the candidate equipment pool for subsequent performance index bidding and allocation, thereby achieving collaborative decision-making between process constraints and equipment capabilities. Therefore, this application further introduces a pre-constraint satisfaction index when allocating tasks, and makes collaborative decisions with performance indicators, so that subsequent processes are only allowed to be allocated when the preceding process has been completed (or is about to be completed), otherwise their allocation priority is reduced.

[0037] Among them, for orders The Steps Define its set of preceding processes. Each preceding process The remaining completion time is ,in The processing time can be dynamically estimated based on the current queue of equipment already allocated to the preceding process and the processing time. For example, this application illustrates that the equipment processes according to the First-Come, First-Served (FCFS) principle. The assigned processes are ( , , ),in, For orders The Steps The preceding process. The current process performed by the equipment is as follows: Then, the process steps are obtained from the process knowledge graph. , , Standard working hours and obtain process During the processed period, For (process) Standard working hours / equipment historical processes Processing efficiency - process Processing time + process Standard working hours / equipment historical processes Processing efficiency + procedures Standard working hours / equipment historical processes The processing efficiency. If all preceding processes have been completed (i.e. ),but Otherwise, there are incomplete prerequisite processes. Define the latest prerequisite completion time: The remaining waiting time for the current process is: The unit is minutes. At this moment: ,in, The decay coefficient having the reciprocal dimension of time (e.g.) To control the sensitivity of waiting for punishment, this application selects... ). Values ​​in The longer the waiting time, The closer it is to 0. Here, exp() is an exponential function with the natural constant e as its base.

[0038] Equipment-Process Performance Quantification Submodule: Configures equipment intelligent agents to respond to the task messages, extracts parameters from historical process data and process knowledge graphs, analyzes the processing efficiency, product processing quality, changeover time loss, energy consumption and load of each candidate equipment, calculates equipment-process performance indicators and feeds them back to the comprehensive scheduling bidding decision submodule.

[0039] First, candidate devices Q (using Q as an example) are extracted from the unified data lake for the target process. (Order The Historical processing data for each process step. Based on time-series data collected by the SCADA system and completion records from the MES system, equipment-process performance indicators are constructed. .

[0040] in, This indicates the processing efficiency of the target process in the history of the equipment. The calculation method is as follows: extract the standard working time of the process and the historical average processing time of the target process in the system, and take the ratio of the standard working time to the historical average processing time as... It should be noted that if the target equipment has no historical records for this process, the historical average processing time will be the general average of this process within the system; if the system has no records for this process, E will be the default value of 1.

[0041] Y represents the yield rate of products processed by the equipment. The higher the yield rate, the higher the performance index of the candidate equipment. The yield rate is obtained as follows: the "quality requirement" node of the target process is extracted from the process knowledge graph, and the set of N key dimensions or characteristic parameters that need to be controlled in this process is obtained. and its standard value With tolerance upper and lower limits Extract the historical records of the target processing steps for candidate equipment Q from the SCADA / MES system, and obtain the corresponding data for each batch / piece of product. measured value To eliminate the influence of different dimensions, the normalized tolerance deviation is calculated using the relative tolerance percentage deviation. The specific formula is as follows: It should be noted that if it is a one-sided tolerance, then use ,in, This represents the one-sided tolerance value; further, the maximum normalized tolerance deviation among N key parameters in a single historical sample is obtained, and then the average of the maximum normalized tolerance deviations obtained from all historical samples is calculated, denoted as . The yield rate can be calculated using the following formula: In the formula, exp() is an adjustment coefficient used to control the penalty sensitivity of quality deviation (in this embodiment, the value is 5, and the implementer can preset it according to the precision level of the process), and exp() is an exponential function with the natural constant e as the base.

[0042] C represents the equipment changeover time loss. If the product model currently being produced by the equipment is inconsistent with the product model of order i, the equipment needs to be changed (i.e., the same equipment can often handle multiple types of processes; for example, replacing process A of the currently produced product with process GX required by the product model of order i will result in a corresponding changeover time loss). The larger the changeover time loss, the lower the performance index of the candidate equipment. The changeover time loss is obtained as follows: extract the processing attributes of the "process" node from the process knowledge graph and construct a high-dimensional feature vector. The feature dimensions include: geometric features: machining part type, size range, and geometric tolerance grade; parametric features: spindle speed range, feed rate range, and depth of cut range (continuous value normalization); resource features: tool / fixture encoding embedding vector; and environmental features: temperature requirements. It should be noted that categorical features (such as fixture type and machining part) are encoded using One-Hot encoding, while continuous value features (such as spindle speed and geometric tolerance grade) are range-scaled and normalized, and then concatenated to form a high-dimensional feature vector.

[0043] Assuming the current production process is A, and the target order's production process is GX, obtain the vector. and . Deviation: ,in For vectors and Cosine similarity measures the degree of similarity between two vectors. Transformation time loss. Typically determined by the basic fixed replacement time. With complexity increment time constitute: ; In the formula, This indicates the basic fixed changeover time, which is the average basic time for the equipment to run and changeover without load (such as program recall and tool movement time) obtained from the equipment manual or MES historical changeover records. Indicates the incremental time complexity; The complexity coefficient is obtained through linear regression fitting of historical changeover data. The independent variable is the deviation between the feature vectors of the current process and the target process, and the dependent variable is the difference between the actual total changeover time and the basic fixed changeover time. The slope of the calculated value is the complexity coefficient. For example, if historically the changeover from process A to process B took 30 minutes, the calculated value would be... ,like ,but If historical fitting data is lacking, it can be used directly. As a normalized replacement time loss factor, it participates in subsequent scheduling decisions, that is... (Between 0 and 1), the smaller the value, the better it is to reduce waiting time.

[0044] H represents the average energy consumption of the candidate equipment in the historical processing of the target process (in this embodiment, maximum and minimum normalization is performed, and the normalization range is the average energy consumption of all candidate equipment in the processing of the target process). The higher the energy consumption, the lower the performance index of the candidate equipment.

[0045] K represents the number of operations in the assigned operation queue of the candidate equipment.

[0046] Adding 1 to the denominator is to prevent the denominator from being 0, which would result in a meaningless fraction.

[0047] Before substituting the variables into the above formula, it is necessary to consider the changeover time loss. and the number of processes in the allocated process queue of candidate equipment. Standardized operations will be implemented. Specifically, based on the datasets returned by all candidate devices currently participating in the bidding, their respective... Value and The values ​​are respectively subjected to min-max normalization to ensure that their values ​​are uniformly mapped to... After the interval is calculated, the summation operation of the denominator terms is performed.

[0048] The integrated scheduling bidding decision submodule configures the scheduling agent to collect feedback from all device agents, calculates the integrated allocation score based on the process readiness, and assigns the task to the candidate device with the highest integrated allocation score, updating the queue status. The device agent synchronizes the updated queue status to the scheduling agent to trigger the next stage of dynamic scheduling.

[0049] The process readiness level at the time of process allocation can be obtained by following the steps above. And by taking the equipment-process performance index S of the candidate equipment, the corresponding comprehensive allocation score can be obtained. ,in and All are weighting factors. and Both are greater than 0 and less than 1, satisfying This application sets The implementer can make adjustments. Since performance metrics may be much larger than process readiness, the equipment-process performance metrics of all candidate devices are normalized to maximum and minimum values ​​here.

[0050] The scheduling agent collects responses from all candidate device agents within a set bidding waiting window (200ms in this embodiment). If no response is received from a device within the timeout period, it is considered not to participate in the bidding. For each bidding candidate device, they are sorted from highest to lowest based on their comprehensive allocation score. The device with the highest comprehensive allocation score is assigned a task. If multiple devices have the same highest score, the following secondary indicators are further compared: earliest expected start time (earlier, higher priority); average energy consumption (lower, higher priority); and current allocated queue length (shorter, higher priority). The scheduling agent sends a task assignment confirmation message to the winning device agent and a bidding termination message to the remaining candidate devices. If the scheduling agent does not receive a "confirmation response" from the selected device within a preset time, it automatically cancels the assignment and reassigns the task to the candidate device with the second-highest score. The device agent assigned the task will then proceed with the work process. Insert at the end of its own processing queue, and update the queue length and the estimated completion time (ETC) of each process in the queue.

[0051] After the assigned equipment agent updates its queue, it feeds back the processing completion signal and measured quality data to the scheduling agent and the data lake. The scheduling agent changes the order status to "next process pending assignment" and repeatedly executes the steps of assessing the readiness of the target process, broadcasting the task message, and bidding for candidate equipment until all processes within the order are completed and stored. This completes the multi-source data-driven, multi-agent closed-loop collaborative control.

[0052] The specific flowchart of multi-agent closed-loop collaborative control is as follows: Figure 2 As shown.

[0053] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description; sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0054] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A multi-source data driven manufacturing enterprise multi-agent collaborative system, characterized in that, The system includes: Data Acquisition and Knowledge Graph Construction Module: Collects multi-source data from manufacturing enterprises, extracts process features from unstructured data, and constructs a process knowledge graph that includes hard constraints on process topology and soft constraints on resource attributes; The pre-process constraint readiness submodule configures a scheduling agent to monitor the process status. Based on the real-time processing queue and efficiency of the equipment where the pre-process of the current target task is located, it dynamically predicts the remaining completion time of the pre-process and calculates the process readiness of the current target task. When the process readiness reaches the target, it publishes a task message to the candidate equipment pool. The equipment-process performance quantification submodule configures equipment intelligent agents to respond to the task messages, extracts parameters from historical process data and process knowledge graphs, analyzes the processing efficiency, product processing quality, changeover time loss, energy consumption and load of each candidate equipment, calculates equipment-process performance indicators and feeds them back to the comprehensive scheduling bidding decision submodule. The integrated scheduling bidding decision submodule configures a scheduling agent to collect feedback from all device agents, calculates an integrated allocation score based on the process readiness, assigns the task to the candidate device with the highest integrated allocation score, and updates the queue status. The device agent synchronizes the updated queue status to the scheduling agent to trigger the next stage of dynamic scheduling.

2. The multi-source data driven manufacturing enterprise multi-agent collaboration system of claim 1, wherein, The specific steps for dynamically predicting the remaining completion time of the preceding process include: Based on the process knowledge graph, identify all the preceding processes of the current target task; for each preceding process, obtain the real-time status of the equipment assigned to it, and obtain the remaining completion time based on the completion time of the equipment from the current processing process to the preceding process of the current target task and the processing efficiency of the equipment in the same historical target task process.

3. The multi-source data driven manufacturing enterprise multi-agent collaboration system of claim 1, wherein, The calculation of the process readiness of the current target task is specifically as follows: Take the maximum value from the remaining completion time of all preceding processes to determine the shortest time that the target task still needs to wait. Multiply the shortest time with a preset attenuation coefficient and use the negative correlation mapping result as the process readiness of the current target task.

4. The multi-source data driven manufacturing enterprise multi-agent collaboration system of claim 1, wherein, The processing efficiency of the candidate equipment is specifically the ratio of the standard working time of the target process in the process knowledge graph to the average processing time of the candidate equipment in the history of processing the target process.

5. The multi-source data-driven multi-agent collaborative system for manufacturing enterprises as described in claim 1, characterized in that, The product processing quality is obtained through the yield rate, and the specific process is as follows: The standard values ​​and tolerance ranges of quality parameters for the target process are extracted from the process knowledge graph. Combined with the historical measured data of the candidate equipment, the normalized tolerance deviation of each key parameter in each historical sample is calculated. The maximum normalized tolerance deviation among all key parameters in each historical sample is obtained. The average value of the maximum normalized tolerance deviation obtained from all historical samples is obtained. The negative exponential function mapping result after multiplying the average value with the preset adjustment coefficient is used as the yield rate.

6. The multi-source data-driven multi-agent collaborative system for manufacturing enterprises as described in claim 1, characterized in that, The specific process for obtaining the changeover time loss is as follows: Extract the processing attributes of process nodes from the process knowledge graph, including geometric features, parameter features, resource features and environmental features, and construct a high-dimensional feature vector; The similarity between the high-dimensional feature vectors of the current process and the target process of each candidate device is analyzed, and the difference between the value 1 and the similarity is taken as the deviation. The changeover time loss is obtained by accumulating the product of the complexity coefficient and the deviation, combined with the preset basic changeover time of each candidate device. The complexity coefficient is preset by performing linear regression fitting on the historical changeover data of each candidate device.

7. The multi-source data-driven multi-agent collaborative system for manufacturing enterprises as described in claim 1, characterized in that, The performance indicators of the computing device-process are positively correlated with the processing efficiency and the product processing quality, and negatively correlated with the changeover time loss, energy consumption and load.

8. The multi-source data-driven multi-agent collaborative system for manufacturing enterprises as described in claim 1, characterized in that, The comprehensive allocation score is specifically obtained by weighted summation of the process readiness degree and the normalized results of the equipment-process performance index, with the weights being preset values.

9. A multi-source data-driven multi-agent collaborative system for manufacturing enterprises as described in claim 1, characterized in that, If there are multiple highest scores in the comprehensive allocation, compare the following indicators: the earliest expected start time, the average energy consumption, and the current length of the allocated queue. The smaller the value, the higher the priority.

10. A multi-source data-driven multi-agent collaborative system for manufacturing enterprises as described in claim 1, characterized in that, If the scheduling agent does not receive a confirmation response from the selected device within a preset time, the grant will be automatically revoked and the task will be reassigned to the candidate device with the second-highest score.

Images