Multi-agent collaborative control method and system for SMT production line

Abstract

The embodiment of the application provides a multi-agent collaborative control method and system for an SMT production line, a device agent state file is constructed through multi-device state acquisition and time reference alignment, optimal task allocation is realized by combining a bidding and tender mechanism and a collaborative decision network model, and closed-loop self-healing control is realized through an exception diagnosis and a knowledge base driven re-bidding mechanism, effectively solving the deficiencies of traditional technologies in aspects of multi-device state management, agent task allocation decision and exception response, and providing technical support for intelligent collaborative control and stable operation of the SMT production line.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a multi-agent collaborative control method and system for SMT production lines. Background Technology

[0002] Existing collaborative control methods for SMT (Surface Mount Technology) production lines have significant shortcomings. Traditional systems perform poorly in the acquisition and management of multi-device status data. They typically lack a unified time reference alignment and missing value filling mechanism for the operating parameters of heterogeneous equipment such as solder paste printers, pick-and-place machines, reflow ovens, and testing equipment. They also fail to effectively link the capacity parameters of each device to construct a device intelligent agent status profile, resulting in insufficient consistency and completeness of multi-device status information, making it difficult to support subsequent refined collaborative task scheduling.

[0003] Furthermore, existing technologies face bottlenecks in multi-agent task allocation decisions. Most systems lack the ability to drive agent bidding based on production order parsing, construct bidding parameter combinations by combining task completion time predictions and resource consumption predictions, and then feed them into a collaborative decision-making network model for fusion inference. Consequently, they cannot achieve optimal task allocation mapping through comprehensive scoring and descending order, making it difficult to effectively guarantee the rationality of production line resource scheduling and production efficiency.

[0004] Existing systems have technical shortcomings in responding to task execution anomalies. They lack a closed-loop self-healing mechanism that inputs execution status feedback records into an anomaly diagnosis model for execution status pattern matching, queries response strategies based on an anomaly response knowledge base, and drives the re-bidding of incomplete subtasks. This affects the continuous and stable operation of the production line under equipment malfunction scenarios. Solving these problems is crucial for improving the intelligence level of multi-agent collaborative control in SMT production lines. Summary of the Invention

[0005] To address the problems in existing technologies, this application provides a multi-agent collaborative control method and system for SMT production lines. It can effectively solve the shortcomings of traditional technologies in multi-device status management, agent task allocation decision-making, and anomaly response, and provide technical support for the intelligent collaborative control and stable operation of SMT production lines.

[0006] To solve at least one of the above problems, this application provides the following technical solution:

[0007] Firstly, this application provides a multi-agent cooperative control method for SMT production lines, including:

[0008] The original state data set is obtained by collecting equipment operating parameters and load status from solder paste printers, pick-and-place machines, reflow ovens and testing equipment in the SMT production line and adding timestamps. The original state data set is aligned and missing values ​​are filled according to a unified time base to obtain a standardized state sequence. The standardized state sequence is grouped according to equipment identifier and associated with the capacity parameters of each equipment to obtain the equipment intelligent agent state file.

[0009] Based on the production order parsing, a set of subtasks is obtained and a task bidding request is issued to each device agent. Each device agent reads the device agent status file and calculates the predicted task completion time and resource consumption to obtain a bidding parameter combination. The bidding parameter combination is sent to the collaborative decision network model to perform fusion inference to obtain a comprehensive bidding score. The comprehensive bidding scores are sorted in descending order and the first device is selected to obtain a task allocation mapping table.

[0010] Each device agent receives sub-tasks according to the task allocation mapping table and drives the equipment to perform production actions. During the execution process, the task progress and equipment status are periodically collected to obtain execution status feedback records. The execution status feedback records are sent to the anomaly diagnosis model for execution status pattern matching to obtain anomaly judgment results. Based on the anomaly judgment results, the anomaly response knowledge base is queried to obtain response strategies. According to the response strategies, the unfinished sub-tasks are re-tendered and updated control instructions are issued to the newly selected device agents.

[0011] Furthermore, it also includes: establishing data connections with the control systems of solder paste printers, pick-and-place machines, reflow ovens, and testing equipment respectively through industrial communication protocols, reading equipment operating parameters and load status from each control system, and adding the timestamp of the acquisition time and equipment identifier to obtain the original status data set;

[0012] The timestamps of each record in the original state data set are converted to a unified time base to obtain a time-aligned data set. Continuity detection is performed on the time-aligned data set according to a preset sampling period, and missing positions are filled by interpolation with adjacent valid values ​​to obtain a standardized state sequence.

[0013] Furthermore, it also includes: performing a grouping operation on the standardized state sequence according to the equipment identifier field to obtain a set of equipment group states, and reading the capacity parameters and process capability parameters corresponding to each equipment from the production line configuration library to obtain a set of equipment capability parameters;

[0014] The device intelligent agent status file is obtained by performing association matching between each group in the device group status set and the device capability parameter set according to the device identifier.

[0015] Furthermore, it also includes: parsing the product model, bill of materials, and process requirements in the production order to obtain task description information; decomposing the task description information according to workstation type and process sequence to obtain a set of sub-tasks; and encapsulating the set of sub-tasks into a task bidding request and publishing it to each equipment agent.

[0016] After receiving the task bidding request, each device agent reads the current load and capacity parameters in the device agent status file, calculates the task completion time prediction value based on the current load and capacity parameters and according to the preset time estimation rules, calculates the resource consumption prediction value based on the process requirements in the task bidding request and according to the preset resource consumption rules, and combines the task completion time prediction value and the resource consumption prediction value to obtain the bidding parameter combination.

[0017] Furthermore, it also includes: transforming the predicted task completion time and predicted resource consumption in the bidding parameter combination according to a preset normalization rule to obtain a normalized bidding feature vector, and sending the normalized bidding feature vector into the multi-layer fusion structure of the collaborative decision network model to perform feature weighting and nonlinear transformation to obtain a comprehensive bidding score;

[0018] The comprehensive bid scores are sorted in descending order of numerical value to obtain a score ranking sequence. The first device in the score ranking sequence is selected as the winning device for the corresponding sub-task, and a mapping relationship between the sub-task and the winning device is established to obtain a task allocation mapping table.

[0019] Furthermore, it also includes: each device agent queries the sub-task corresponding to its own device from the task allocation mapping table and converts the sub-task into a device control command and sends it to the device control system to drive the device to perform production actions. During the execution process, the device control system reads the number of completed workpieces, the number of remaining workpieces, and the device operating parameters according to a preset collection cycle to obtain execution status feedback records.

[0020] The device operating parameters in the execution status feedback record are extracted and organized into a status feature vector. The status feature vector is sent to the anomaly diagnosis model to perform a similarity calculation with the normal operation mode to obtain a mode deviation value. The mode deviation value is compared with a preset anomaly threshold condition to obtain an anomaly judgment result.

[0021] Furthermore, it also includes: extracting the anomaly type label and the anomaly occurrence device identifier from the anomaly determination result to obtain anomaly description information, and querying the standard processing flow stored in the anomaly response knowledge base according to the anomaly description information to obtain a response strategy.

[0022] Based on the response strategy, the unfinished subtasks corresponding to the abnormal devices are extracted from the original task allocation mapping table to obtain the subtasks to be reassigned. The subtasks to be reassigned are encapsulated into a task bidding request and published to the device intelligent agent with alternative execution capability. The device intelligent agent that wins the bid is selected according to the comprehensive bidding score, and the updated control command is sent to the device intelligent agent that wins the bid.

[0023] Secondly, this application provides a multi-agent collaborative control system for SMT production lines, comprising:

[0024] The equipment status monitoring module is used to collect equipment operating parameters and load status from solder paste printers, pick-and-place machines, reflow ovens and testing equipment in the SMT production line, and add timestamps to obtain a raw status data set. The raw status data set is aligned and missing values ​​are filled according to a unified time base to obtain a standardized status sequence. The standardized status sequence is grouped according to equipment identifier and associated with the capacity parameters of each equipment to obtain an equipment intelligent agent status file.

[0025] The equipment task allocation module is used to obtain a set of sub-tasks based on the parsing of production orders and to issue task bidding requests to each equipment agent. Each equipment agent reads the status file of the equipment agent and calculates the predicted value of task completion time and the predicted value of resource consumption to obtain a combination of bidding parameters. The combination of bidding parameters is sent to the collaborative decision network model to perform fusion inference to obtain a comprehensive bidding score. The comprehensive bidding scores are sorted in descending order and the first-ranked equipment is selected to obtain a task allocation mapping table.

[0026] The collaborative control module is used for each device agent to receive sub-tasks according to the task allocation mapping table and drive the equipment to perform production actions. During the execution process, the module periodically collects task progress and equipment status to obtain execution status feedback records. The execution status feedback records are sent to the anomaly diagnosis model for execution status pattern matching to obtain anomaly judgment results. Based on the anomaly judgment results, the module queries the anomaly response knowledge base to obtain response strategies. According to the response strategies, the unfinished sub-tasks are re-tendered and updated control instructions are issued to the newly selected device agents.

[0027] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the multi-agent cooperative control method for SMT production lines.

[0028] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-agent cooperative control method for SMT production lines.

[0029] Fifthly, this application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the multi-agent cooperative control method for SMT production lines.

[0030] As can be seen from the above technical solution, this application provides a multi-agent collaborative control method and system for SMT production lines. It constructs equipment agent status files by collecting multiple device status data and aligning them with a time base, achieves optimal task allocation by combining a bidding mechanism and a collaborative decision-making network model, and realizes closed-loop self-healing control through anomaly diagnosis and a knowledge base-driven re-bidding mechanism. This effectively solves the shortcomings of traditional technologies in multi-device status management, agent task allocation decision-making, and anomaly response, and provides technical support for the intelligent collaborative control and stable operation of SMT production lines. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart illustrating the multi-agent collaborative control method for SMT production lines in an embodiment of this application.

[0033] Figure 2 This is a structural diagram of the multi-agent collaborative control system for SMT production lines in the embodiments of this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0036] In view of the problems existing in the prior art, this application provides a multi-agent collaborative control method and system for SMT production lines. By collecting the status of multiple devices and aligning with the time base, a device agent status profile is constructed. Optimal task allocation is achieved by combining a bidding mechanism and a collaborative decision-making network model. Closed-loop self-healing control is achieved through anomaly diagnosis and a knowledge base-driven re-bidding mechanism. This effectively solves the shortcomings of traditional technologies in multi-device status management, agent task allocation decision-making, and anomaly response, providing technical support for the intelligent collaborative control and stable operation of SMT production lines.

[0037] To effectively address the shortcomings of traditional technologies in multi-device state management, agent task allocation decision-making, and anomaly response, and to provide technical support for the intelligent collaborative control and stable operation of SMT production lines, this application provides an embodiment of a multi-agent collaborative control method for SMT production lines. See [link to embodiment]. Figure 1 The multi-agent collaborative control method for SMT production lines specifically includes the following:

[0038] Step S101: Collect equipment operating parameters and load status from solder paste printers, pick-and-place machines, reflow ovens and testing equipment in the SMT production line and add timestamps to obtain a raw state data set. Align the raw state data set according to a unified time base and fill in missing values ​​to obtain a standardized state sequence. Group the standardized state sequence according to equipment identifier and associate it with the capacity parameters of each equipment to obtain an equipment intelligent agent state file.

[0039] This embodiment establishes data connections with the control systems of the solder paste printer, pick-and-place machine, reflow oven, and inspection equipment in the SMT production line via industrial communication protocols. After the connection is established, the operating parameters and load status of each device are periodically read from the control system. The operating parameters of the solder paste printer include printing pressure, squeegee speed, and stencil cleaning cycle; the operating parameters of the pick-and-place machine include the placement head position, nozzle vacuum, and feeder status; the operating parameters of the reflow oven include the measured temperature of each temperature zone and the conveyor belt speed; and the operating parameters of the inspection equipment include the inspection cycle time and defect count. This embodiment adds a timestamp and device identifier to each parameter reading, and aggregates all the collected records to form a raw state data set.

[0040] Based on the original state data set, this embodiment performs a unified time base conversion on the timestamps of each record. The conversion process reads the local clock offset of each device's control system, subtracts the corresponding offset from the timestamp of each record, and maps it to the production line's unified time base to obtain a time-aligned data set. For example, if the local clock of a pick-and-place machine's control system is several milliseconds ahead of the production line's reference clock, then all timestamps recorded by that device need to have this offset subtracted to complete the alignment.

[0041] Accordingly, this embodiment performs continuity detection on the time-aligned data set according to a preset sampling period. The detection process calculates the time interval between adjacent records based on the timestamp sequence of each parameter channel, compares the time interval with the preset sampling period, and determines that data is missing at that position when the time interval exceeds the preset sampling period. This embodiment fills the missing positions with interpolated adjacent valid values. The interpolation process reads the valid data records adjacent to the missing position and calculates the fill value according to linear interpolation rules. After filling, a standardized state sequence is obtained.

[0042] Based on the standardized state sequence, this embodiment performs a grouping operation according to the device identifier field. The grouping process traverses each record in the standardized state sequence, reads the device identifier field value of the record, and groups records with the same device identifier into the same group, resulting in a device group state set. Each group in the device group state set corresponds to all operating status records of a specific device on the production line.

[0043] Specifically, in this embodiment, the production capacity parameters and process capability parameters corresponding to each piece of equipment are read from the production line configuration library. The production capacity parameters include processing capacity per unit time and maximum number of parallel tasks, while the process capability parameters include the type of component packages that can be processed and their precision level. These parameters are aggregated to obtain a set of equipment capability parameters. In this embodiment, each group in the equipment group status set is associated and matched with the set of equipment capability parameters according to its equipment identifier. The matching process uses the equipment identifier as the association key to bind the operating status record of each group with the capability parameters of the corresponding equipment. After binding, a device agent status file is obtained. This device agent status file is used in subsequent step S201 for each device agent to read the current load and production capacity parameters and calculate the bidding parameter combination when receiving a task bidding request.

[0044] Step S102: Based on the production order parsing, a set of sub-tasks is obtained and a task bidding request is issued to each device agent. Each device agent reads the device agent status file and calculates the predicted task completion time and resource consumption to obtain a bidding parameter combination. The bidding parameter combination is sent to the collaborative decision network model to perform fusion inference to obtain a comprehensive bidding score. The comprehensive bidding score is sorted in descending order and the first device is selected to obtain a task allocation mapping table.

[0045] This embodiment parses the product model, bill of materials, and process requirements in the production order to obtain task description information. The parsing process reads the structured fields of the production order, extracting the model identifier of the product to be produced, the required component list, and the process parameter requirements for each workstation. This embodiment decomposes the task description information according to workstation type and process sequence. Based on the standard process flow of the SMT production line, the complete production task is broken down into solder paste printing sub-tasks, component mounting sub-tasks, reflow soldering sub-tasks, and quality inspection sub-tasks. Each sub-task carries a corresponding workstation type identifier and process parameter constraints, which are then aggregated to form a sub-task set.

[0046] Based on the aforementioned set of subtasks, this embodiment encapsulates each subtask into a task bidding request and publishes it to each device agent. The encapsulation process adds a task description, constraints, expected completion time, and evaluation metrics to each subtask, forming a bidding request message that conforms to the communication protocol specifications. Each device agent receives a task bidding request matching its own workstation type through a message subscription mechanism, preparing to perform bidding parameter calculations.

[0047] Accordingly, after receiving the task bidding request, each device agent reads the current load and capacity parameters from the device agent status file output in step S101. The reading process extracts the device's recent operating status record and the length of the assigned task queue from the device agent status file, obtaining the device's unit-time processing capacity and the types of component packages it can process. In this embodiment, based on the current load and capacity parameters and according to a preset time estimation rule, the predicted task completion time is calculated. The estimation rule divides the number of sub-tasks by the unit-time processing capacity and adds the estimated consumption time of the current task queue.

[0048] Based on the predicted task completion time, this embodiment calculates the predicted resource consumption value according to the process requirements in the task bidding request and a preset resource consumption rule. The resource consumption rule estimates the number of feeder switching times, nozzle replacement times, and energy consumption levels based on the number of components involved in the sub-task, packaging complexity, and other factors. This embodiment combines the predicted task completion time value and the predicted resource consumption value to obtain a bidding parameter combination, which reflects the expected cost for the device agent to execute the corresponding sub-task.

[0049] Specifically, in this embodiment, the predicted task completion time and resource consumption values ​​in the bidding parameter combination are numerically transformed according to a preset normalization rule to obtain a normalized bidding feature vector. The transformation process maps each predicted value to a unified numerical range to eliminate dimensional differences. In this embodiment, the normalized bidding feature vector is fed into the multi-layered fusion structure of the collaborative decision-making network model to perform feature weighting and nonlinear transformation. The collaborative decision-making network model comprehensively weighs time and resource factors and outputs a scalar-form comprehensive bidding score.

[0050] Based on the comprehensive bid evaluation, this embodiment sorts the comprehensive bid evaluations returned by each device agent for the same sub-task in descending order of numerical value to obtain a score ranking sequence. The ranking process uses the comprehensive bid evaluation as the sorting key, with higher scores listed first. This embodiment selects the first device in the score ranking sequence as the winning device for the corresponding sub-task, establishing a mapping relationship between sub-tasks and winning devices. All sub-task mapping relationships are aggregated to obtain a task allocation mapping table. This task allocation mapping table is used in subsequent step S103 for each device agent to query its corresponding sub-task and drive the device to execute production actions.

[0051] Step S103: Each device agent receives sub-tasks according to the task allocation mapping table and drives the device to perform production actions. During the execution process, the task progress and device status are periodically collected to obtain execution status feedback records. The execution status feedback records are sent to the anomaly diagnosis model for execution status pattern matching to obtain anomaly judgment results. Based on the anomaly judgment results, the anomaly response knowledge base is queried to obtain response strategies. According to the response strategies, the unfinished sub-tasks are re-tendered and updated control instructions are issued to the newly selected device agents.

[0052] As can be seen from the above description, the multi-agent collaborative control method for SMT production lines provided in this application can construct device agent status profiles by collecting multiple device statuses and aligning them with time bases, achieve optimal task allocation by combining a bidding mechanism and a collaborative decision-making network model, and achieve closed-loop self-healing control through anomaly diagnosis and a knowledge base-driven re-bidding mechanism. This effectively solves the shortcomings of traditional technologies in multi-device status management, agent task allocation decision-making, and anomaly response, providing technical support for the intelligent collaborative control and stable operation of SMT production lines.

[0053] In one embodiment of the multi-agent cooperative control method for SMT production lines in this application, the method may further include the following:

[0054] Step S201: Establish data connections with the control systems of the solder paste printer, pick and place machine, reflow oven, and testing equipment respectively through the industrial communication protocol, read the equipment operating parameters and load status from each control system, and add the timestamp of the acquisition time and equipment identifier to obtain the original status data set;

[0055] Step S202: Convert the timestamps of each record in the original state data set to a unified time base to obtain a time-aligned data set. Perform continuity detection on the time-aligned data set according to a preset sampling period and fill the missing positions with adjacent valid values ​​to obtain a standardized state sequence.

[0056] Each device agent queries the task allocation mapping table output in step S102 for its corresponding subtask. The query process uses the device identifier as the search key to extract the subtask record associated with that device identifier from the task allocation mapping table, obtaining the subtask's process parameters, workpiece quantity, and constraints. In this embodiment, the subtask is converted into a device control command. The conversion process generates a corresponding control program number and parameter setting value based on the subtask's process parameters, and then sends the device control command to the device control system to drive the device to execute production actions.

[0057] After the equipment control command is issued, this embodiment reads the task progress and equipment status from the equipment control system according to a preset acquisition cycle during execution. The read content includes the number of completed workpieces, the number of remaining workpieces, and equipment operating parameters. The equipment operating parameters include the nozzle vacuum degree of the pick-and-place machine, the feeding status of the feeder, the measured temperature of the reflow oven, the conveyor belt speed, the inspection pass rate of the inspection equipment, and the defect distribution. This embodiment adds timestamps to the data acquired in each acquisition cycle and aggregates them to form an execution status feedback record.

[0058] Accordingly, this embodiment extracts and organizes the device operating parameters from the execution status feedback record into a state feature vector. The organization process arranges the values ​​of each parameter channel in a preset order to form a fixed-length vector structure for subsequent model input. In this embodiment, the state feature vector is fed into an anomaly diagnosis model to perform similarity calculations with the normal operating mode. The anomaly diagnosis model performs pattern matching on the current state feature vector based on the normal operating state distribution features learned during the training phase, and outputs the pattern deviation value between the current state and the normal mode.

[0059] Based on the pattern deviation value, this embodiment compares it with a preset anomaly threshold to obtain an anomaly determination result. The comparison process determines whether the pattern deviation value exceeds the preset threshold range. If it does, it is determined to be an abnormal state and labeled with an anomaly type; otherwise, it is determined to be a normal state. For example, if the pattern deviation value corresponding to the feeding status parameter of a pick-and-place machine's feeder exceeds the threshold range, the anomaly determination result is labeled as a feeding anomaly type.

[0060] Specifically, in this embodiment, anomaly type tags and the identifier of the device where the anomaly occurred are extracted from the anomaly determination results to obtain anomaly description information. Based on this anomaly description information, this embodiment queries the standard processing flow stored in the anomaly response knowledge base according to preset index rules. The anomaly response knowledge base stores the response action sequences and task transfer rules corresponding to various anomalies using the anomaly type tags as index keys. The query yields a response strategy. The response strategy indicates whether task reassignment needs to be performed and the scope and priority constraints of the reassignment.

[0061] Based on the aforementioned response strategy, this embodiment extracts the unfinished subtasks corresponding to the abnormal equipment from the original task allocation mapping table to obtain the subtasks to be reassigned. The extraction process locates the record of the abnormal equipment in the task allocation mapping table based on its identifier, reads the remaining workpiece quantity and process parameters of that record to form the subtasks to be reassigned. This embodiment encapsulates the subtasks to be reassigned as a task bidding request and publishes it to a device agent with alternative execution capabilities. Each candidate device agent returns a combination of bidding parameters according to the bidding process in step S102, and the comprehensive bidding score is calculated by the collaborative decision-making network model. This embodiment selects the winning device agent based on the comprehensive bidding score and issues updated control commands to the winning device agent to execute the remaining production tasks.

[0062] In one embodiment of the multi-agent cooperative control method for SMT production lines in this application, the method may further include the following:

[0063] Step S301: Perform a grouping operation on the standardized state sequence according to the equipment identifier field to obtain the equipment group state set, and read the capacity parameters and process capability parameters corresponding to each equipment from the production line configuration library to obtain the equipment capability parameter set;

[0064] Step S302: Perform association matching between each group in the device group status set and the device capability parameter set according to the device identifier to obtain the device intelligent agent status file.

[0065] In this embodiment, the standardized state sequence output in step S101 is grouped according to the device identifier field. The grouping process traverses each record in the standardized state sequence, reads the device identifier field value carried by each record, and groups records with the same device identifier into the same group. After the grouping is completed, this embodiment obtains a device group state set, in which each group corresponds to all operating status records of a specific device in the SMT production line during the acquisition period.

[0066] Based on the aforementioned equipment grouping status set, this embodiment reads the capacity parameters and process capability parameters corresponding to each piece of equipment from the production line configuration library. The production line configuration library stores the basic equipment information and capability specifications entered during the production line deployment phase. This embodiment retrieves the corresponding configuration records one by one according to the equipment identifier list involved in the equipment grouping status set. Capacity parameters include processing capacity per unit time, maximum number of parallel tasks, and maximum continuous running time. Process capability parameters include the types of component packages that can be processed, the mounting accuracy level, and the temperature control range. This embodiment aggregates the retrieved equipment configuration records to form a set of equipment capability parameters.

[0067] Accordingly, this embodiment performs association matching between each group in the device group status set and the device capability parameter set based on the device identifier. The matching process uses the device identifier as the association key, searching for configuration records with the same device identifier in the device capability parameter set for each group in the device group status set, and binding the operating status record sequence, corresponding capacity parameters, and process capability parameters of that group. For example, if the device identifier of a high-speed pick-and-place machine is a specific number, then all status records in the corresponding group of that device will be associated with the placement speed parameter and the processable package type parameter corresponding to that number.

[0068] Specifically, in this embodiment, after completing the association matching of all groups, a device agent status file is obtained. The device agent status file is organized with the device identifier as the index, and each file record contains two parts: the historical operating status sequence of the device and static capability parameters. The operating status sequence reflects the dynamic load changes and parameter fluctuations of the device in the recent data collection period, while the static capability parameters reflect the inherent processing capabilities and process applicability range of the device.

[0069] Based on the device agent status file, this embodiment stores it in the agent shared data area for subsequent steps to access. The device agent status file is used in subsequent step S102 for each device agent to read the current load status and capacity parameters when receiving a task bidding request, and to calculate the predicted task completion time and resource consumption to form a bidding parameter combination.

[0070] In one embodiment of the multi-agent cooperative control method for SMT production lines in this application, the method may further include the following:

[0071] Step S401: Parse the product model, bill of materials, and process requirements in the production order to obtain task description information. Decompose the task description information according to workstation type and process sequence to obtain a set of sub-tasks. Encapsulate the set of sub-tasks into a task bidding request and publish it to each equipment agent.

[0072] Step S402: After receiving the task bidding request, each device agent reads the current load and capacity parameters in the device agent status file, calculates the task completion time prediction value based on the current load and capacity parameters and according to the preset time estimation rules, calculates the resource consumption prediction value based on the process requirements in the task bidding request and according to the preset resource consumption rules, and combines the task completion time prediction value and the resource consumption prediction value to obtain the bidding parameter combination.

[0073] This embodiment parses the product model, bill of materials, and process requirements from the production order to obtain task description information. The parsing process reads the structured data fields of the production order, extracting the model identifier and batch quantity of the product to be produced, the model, package type, and required quantity of each component in the bill of materials, and the parameter settings and quality standards for each workstation in the process requirements. This embodiment organizes the extracted content according to a unified data structure to form task description information.

[0074] Based on the task description information, this embodiment decomposes it into a set of subtasks according to workstation type and process sequence. The decomposition process follows the standard process flow of an SMT production line, breaking down the complete production task into four types of workstation-level subtasks: solder paste printing subtask, component placement subtask, reflow soldering subtask, and quality inspection subtask. Each subtask inherits the process parameters and quality standards related to its workstation from the task description information, and adds a workstation type identifier and process sequence number. For example, the component placement subtask carries the placement point information and package type constraints for all components in the bill of materials.

[0075] Accordingly, this embodiment encapsulates the set of subtasks into task bidding requests and publishes them to each device agent. The encapsulation process constructs a bidding request message conforming to the communication protocol specification for each subtask. The message content includes a task description, constraints, expected completion time, and evaluation metrics. This embodiment sends each task bidding request to the message channel through a message publishing mechanism. Each device agent receives the task bidding request matching its subscribed workstation type identifier.

[0076] After the task bidding request is issued, each device agent receives the task bidding request and reads the current load and capacity parameters from the device agent status file output in step S302. The reading process extracts the operating status record of the device's most recent acquisition cycle from the device agent status file to obtain the current task queue length and device occupancy rate, and extracts the capacity parameters to obtain the processing capacity per unit time and the maximum number of parallel tasks.

[0077] Specifically, each device agent calculates the predicted task completion time based on the current load and the production capacity parameters, according to a preset time estimation rule. The time estimation rule divides the number of workpieces in the subtask by the unit time processing capacity of the device to obtain the basic processing time, adds up the estimated time of each task to be executed in the current task queue to obtain the queue waiting time, and adds the basic processing time and the queue waiting time to obtain the predicted task completion time.

[0078] Based on the predicted task completion time, each device agent calculates the predicted resource consumption value according to the process requirements in the task bidding request and a preset resource consumption rule. The resource consumption rule estimates the number of feeder changes based on the number of component package types involved in the sub-task, the number of nozzle changes based on the component size range, and the energy consumption level based on processing time and equipment power. These consumption indicators are then weighted and summarized according to preset weights to obtain the predicted resource consumption value. In this embodiment, the predicted task completion time and the predicted resource consumption value are combined to obtain a bidding parameter combination. This bidding parameter combination is then used by the collaborative decision-making network model in subsequent step S501 to perform fusion inference to calculate the comprehensive bidding score.

[0079] In one embodiment of the multi-agent cooperative control method for SMT production lines in this application, the method may further include the following:

[0080] Step S501: The predicted task completion time and the predicted resource consumption in the bidding parameter combination are numerically transformed according to the preset normalization rule to obtain the normalized bidding feature vector. The normalized bidding feature vector is then sent to the multi-layer fusion structure of the collaborative decision network model to perform feature weighting and nonlinear transformation to obtain the comprehensive bidding score.

[0081] Step S502: Sort the comprehensive bid scores in descending order of numerical value to obtain a score sorting sequence. Select the first device in the score sorting sequence as the winning device for the corresponding sub-task and establish a mapping relationship between the sub-task and the winning device to obtain a task allocation mapping table.

[0082] In this embodiment, the predicted task completion time and predicted resource consumption values ​​in the bidding parameter combinations output in step S402 are transformed according to a preset normalization rule. The transformation process, for each device agent's submitted bidding parameter combinations, reads the predicted task completion time and maps it to a zero-to-one interval based on the statistical distribution range of historical task completion times; it also reads the predicted resource consumption and maps it to a zero-to-one interval based on the statistical distribution range of historical resource consumption. This embodiment arranges the transformed values ​​in a preset order to form a normalized bidding feature vector, which eliminates scale differences between parameters of different dimensions.

[0083] Based on the normalized bid feature vector, this embodiment feeds it into the multi-layer fusion structure of the collaborative decision-making network model to perform feature weighting and nonlinear transformation. The first layer of the collaborative decision-making network model receives the normalized bid feature vector as input and performs weighted fusion of time and resource factors through a weight matrix. The intermediate layers introduce a nonlinear activation function to capture the interaction between time and resource factors. The output layer maps the fused high-level features into scalar values ​​to obtain the comprehensive bid score.

[0084] Accordingly, this embodiment explains the calculation logic of the comprehensive bid score. During the training phase, the collaborative decision-making network model learns the weights of time and resource factors on task execution quality based on historical task allocation records and execution effect feedback data. During the inference phase, the model assigns higher weights to the time component in the normalized bid feature vector to reflect the importance of delivery deadlines, and assigns appropriate weights to the resource component to balance equipment load balancing requirements. A higher comprehensive bid score indicates a higher overall suitability of the device agent for executing the corresponding sub-task.

[0085] Based on the comprehensive bid evaluation, this embodiment sorts the comprehensive bid evaluations returned by each device agent for the same subtask in descending order of numerical value to obtain a score ranking sequence. The sorting process uses the comprehensive bid evaluation as the sorting key, arranging the score values ​​from highest to lowest to form a priority sequence for candidate devices. For example, if a component placement subtask receives bid responses from three high-speed pick-and-place machines, the comprehensive bid evaluations of the three machines are sorted in descending order to form the score ranking sequence for that subtask.

[0086] Specifically, in this embodiment, the first device in the scoring ranking sequence is selected as the winning device for the corresponding sub-task. The selection process reads the device identifier ranked first in the scoring ranking sequence and identifies that device as the winning bidder for the corresponding sub-task. This embodiment establishes a mapping relationship between sub-tasks and winning devices; the mapping record includes the sub-task identifier, the winning device identifier, and the comprehensive bid score. This embodiment repeats the above winning selection process for all sub-tasks, aggregating the mapping records of each sub-task to obtain a task allocation mapping table. This task allocation mapping table is used in subsequent step S601 for each device agent to query the sub-task corresponding to its device and drive the device to execute production actions.

[0087] In one embodiment of the multi-agent cooperative control method for SMT production lines in this application, the method may further include the following:

[0088] Step S601: Each device agent queries the sub-task corresponding to its own device from the task allocation mapping table and converts the sub-task into a device control command and sends it to the device control system to drive the device to execute production actions. During the execution process, the device control system reads the number of completed workpieces, the number of remaining workpieces, and the device operating parameters according to the preset collection cycle to obtain the execution status feedback record.

[0089] Step S602: Extract and organize the device operating parameters in the execution status feedback record into a status feature vector, send the status feature vector into the anomaly diagnosis model to perform a similarity calculation with the normal operation mode to obtain a mode deviation value, and compare the mode deviation value with the preset anomaly threshold condition to obtain an anomaly judgment result.

[0090] Each device agent queries the task allocation mapping table output in step S502 for its corresponding subtask. The query process uses the device identifier as the search key, traverses each mapping record in the task allocation mapping table, and extracts the subtask record whose identifier matches the device identifier. In this embodiment, the task identifier, process parameters, workpiece quantity, and constraints are read from the extracted subtask record to prepare for the conversion of the subtask into device control instructions.

[0091] Based on the subtask records, this embodiment converts the subtasks into equipment control instructions. The conversion process involves querying the equipment control program library based on the process parameters carried by the subtask, matching the corresponding control program number with the parameter setting template, and filling the parameter setting template with the specific process values ​​of the subtask to form a complete equipment control instruction. In this embodiment, the equipment control instruction is sent to the equipment control system through the equipment driver interface. After receiving the instruction, the equipment control system drives the equipment to execute production actions.

[0092] Accordingly, this embodiment reads the task progress and equipment status from the equipment control system according to a preset acquisition cycle during execution. The reading process periodically accesses the status register of the equipment control system through the equipment driver interface to obtain the number of completed and remaining workpieces to reflect the task execution progress, and to obtain equipment operating parameters to reflect the current working status of the equipment. The equipment operating parameters include the placement head position of the pick-and-place machine, the nozzle vacuum level, the feeder's material count, the measured temperature of each temperature zone of the reflow oven, the conveyor belt speed, and the inspection cycle and defect detection count of the inspection equipment. This embodiment adds timestamps to the data acquired in each acquisition cycle and aggregates them to form an execution status feedback record.

[0093] Based on the execution status feedback record, this embodiment extracts and organizes the equipment operating parameters into a status feature vector. The extraction process separates the equipment operating parameter fields from the execution status feedback record, and the organization process arranges the values ​​of each parameter channel in a preset order to form a fixed-length vector structure. For example, if the operating parameters of a pick-and-place machine include three channels: nozzle vacuum, feeder status, and placement pressure, then the status feature vector consists of three components in a fixed order.

[0094] Specifically, in this embodiment, the state feature vector is fed into the anomaly diagnosis model to perform similarity calculation with the normal operation mode. During the training phase, the anomaly diagnosis model learns the distribution characteristics of the normal operation mode based on historical state feature vectors collected during the device's normal operation. During the inference phase, the anomaly diagnosis model receives the current state feature vector, calculates the distance metric between it and the distribution center of the normal operation mode, and outputs a mode deviation value. A larger mode deviation value indicates a more significant difference between the current state and the normal operation mode.

[0095] Based on the stated pattern deviation value, this embodiment compares it with a preset anomaly threshold condition to obtain an anomaly determination result. The comparison process determines whether the pattern deviation value exceeds the preset threshold range. If it does, the anomaly determination result is marked as an anomaly and labeled with an anomaly type tag according to the deviation direction and parameter channel affixation. If it does not exceed the threshold, the anomaly determination result is marked as a normal state. The anomaly determination result is used in subsequent step S701 to query the anomaly response knowledge base to obtain the corresponding response strategy.

[0096] In one embodiment of the multi-agent cooperative control method for SMT production lines in this application, the method may further include the following:

[0097] Step S701: Extract the anomaly type label and the anomaly occurrence device identifier from the anomaly determination result to obtain anomaly description information. Based on the anomaly description information, query the standard processing flow stored in the anomaly response knowledge base according to the preset index rules to obtain the response strategy.

[0098] Step S702: Based on the response strategy, extract the unfinished sub-tasks corresponding to the abnormal devices from the original task allocation mapping table to obtain the sub-tasks to be reassigned. Encapsulate the sub-tasks to be reassigned into a task bidding request and publish it to the device intelligent agent with alternative execution capability. Select the re-winning device intelligent agent according to the comprehensive bidding score and send the updated control command to the re-winning device intelligent agent.

[0099] This embodiment extracts the anomaly type tag and the identifier of the device that caused the anomaly from the anomaly determination result output in step S602 to obtain anomaly description information. The extraction process reads records marked as abnormal states from the anomaly determination result and retrieves the anomaly type tag and the identifier of the device that caused the anomaly from these records. The anomaly type tag reflects the specific category of the anomaly, including categories such as feeding anomaly, vacuum anomaly, temperature anomaly, and detection timeout anomaly. This embodiment combines the anomaly type tag and the identifier of the device that caused the anomaly to form anomaly description information.

[0100] Based on the anomaly description information, this embodiment queries the standard processing procedures stored in the anomaly response knowledge base according to preset indexing rules. The anomaly response knowledge base uses anomaly type tags as the primary index key, storing standard processing procedures and historical processing experience corresponding to various anomalies. The query process retrieves the anomaly response knowledge base using the anomaly type tags in the anomaly description information, matching the corresponding standard processing procedure records. For example, if the anomaly type tag is "feeding anomaly," the retrieved standard processing procedure includes a sequence of actions such as pausing the mounting of the corresponding component on the current feeder, triggering task transfer, and notifying material replenishment. This embodiment uses the retrieved standard processing procedure records as a response strategy.

[0101] Accordingly, this embodiment extracts the unfinished subtasks corresponding to the abnormal device from the original task allocation mapping table based on the response strategy. The extraction process locates the subtask record corresponding to the abnormal device in the task allocation mapping table based on the device identifier in the abnormal description information, and reads the remaining number of workpieces and process parameters for that subtask. This embodiment encapsulates the remaining number of unfinished workpieces and the corresponding process parameters to form a subtask to be reassigned. This subtask to be reassigned inherits the process constraints and quality standards of the atomic task.

[0102] Based on the subtasks to be reassigned, this embodiment encapsulates them into task bidding requests and issues them to device agents with alternative execution capabilities. The encapsulation process constructs a bidding request message conforming to the communication protocol specifications for the subtasks to be reassigned. The message content includes a task description, constraints, and urgency level identifier. This embodiment filters candidate device agents with the same process capabilities based on the workstation type of the subtasks to be reassigned, excludes the original equipment in an abnormal state, and issues task bidding requests to the filtered candidate device agents.

[0103] Specifically, after receiving the task bidding request, each candidate device agent calculates the combination of bidding parameters according to the bidding process in steps S402 and S501, and infers the comprehensive bidding score through the collaborative decision-making network model. In this embodiment, the comprehensive bidding scores returned by each candidate device agent are collected, sorted in descending order of numerical value, and the top-ranked device is selected as the re-winning device agent. This embodiment generates updated control instructions based on the process parameters of the sub-task to be reassigned, and sends the updated control instructions to the re-winning device agent, driving it to continue executing the remaining production tasks. After receiving the control instructions, the re-winning device agent drives the equipment to complete the production actions of the sub-task to be reassigned according to the execution process in step S601.

[0104] To effectively address the shortcomings of traditional technologies in multi-device state management, agent task allocation decision-making, and anomaly response, and to provide technical support for the intelligent collaborative control and stable operation of SMT production lines, this application provides an embodiment of a multi-agent collaborative control system for SMT production lines, which implements all or part of the aforementioned multi-agent collaborative control method. See [link to embodiment]. Figure 2 The multi-agent collaborative control system for SMT production lines specifically includes the following:

[0105] Equipment status monitoring module 10 is used to collect equipment operating parameters and load status from solder paste printers, pick and place machines, reflow ovens and testing equipment in the SMT production line and add timestamps to obtain a raw status data set. The raw status data set is aligned and missing values ​​are filled according to a unified time base to obtain a standardized status sequence. The standardized status sequence is grouped according to equipment identifier and associated with the capacity parameters of each equipment to obtain an equipment intelligent agent status file.

[0106] The equipment task allocation module 20 is used to obtain a set of sub-tasks based on the production order parsing and to issue task bidding requests to each equipment agent. Each equipment agent reads the equipment agent status file and calculates the predicted value of task completion time and the predicted value of resource consumption to obtain a combination of bidding parameters. The combination of bidding parameters is sent to the collaborative decision network model to perform fusion inference to obtain a comprehensive bidding score. The comprehensive bidding scores are sorted in descending order and the first-ranked equipment is selected to obtain a task allocation mapping table.

[0107] The collaborative control module 30 is used for each device agent to receive sub-tasks according to the task allocation mapping table and drive the equipment to perform production actions. During the execution process, the module periodically collects task progress and equipment status to obtain execution status feedback records. The module sends the execution status feedback records to the anomaly diagnosis model for execution status pattern matching to obtain anomaly judgment results. Based on the anomaly judgment results, the module queries the anomaly response knowledge base to obtain response strategies. According to the response strategies, the module re-bids for the unfinished sub-tasks and issues updated control instructions to the newly awarded device agents.

[0108] As can be seen from the above description, the multi-agent collaborative control system for SMT production lines provided in this application can construct device agent status profiles by collecting multiple device status data and aligning them with time bases. It can achieve optimal task allocation by combining a bidding mechanism and a collaborative decision-making network model, and achieve closed-loop self-healing control through anomaly diagnosis and a knowledge base-driven re-bidding mechanism. This effectively solves the shortcomings of traditional technologies in multi-device status management, agent task allocation decision-making, and anomaly response, and provides technical support for the intelligent collaborative control and stable operation of SMT production lines.

[0109] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the multi-agent cooperative control method for SMT production lines.

[0110] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned multi-agent collaborative control method for SMT production lines.

[0111] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the aforementioned multi-agent collaborative control method for SMT production lines.

[0112] In this embodiment of the invention, a device intelligent agent status profile is constructed by collecting multiple device statuses and aligning them with a time base. Optimal task allocation is achieved by combining a bidding mechanism with a collaborative decision-making network model. Closed-loop self-healing control is realized through anomaly diagnosis and a knowledge-based re-bidding mechanism. This effectively solves the shortcomings of traditional technologies in multi-device status management, intelligent agent task allocation decision-making, and anomaly response, providing technical support for the intelligent collaborative control and stable operation of SMT production lines.

[0113] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0114] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0115] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0116] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0117] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-agent collaborative control method for SMT production lines, characterized in that, The method includes: The original state data set is obtained by collecting equipment operating parameters and load status from solder paste printers, pick-and-place machines, reflow ovens and testing equipment in the SMT production line and adding timestamps. The original state data set is aligned and missing values ​​are filled according to a unified time base to obtain a standardized state sequence. The standardized state sequence is grouped according to equipment identifier and associated with the capacity parameters of each equipment to obtain the equipment intelligent agent state file. Based on the production order parsing, a set of subtasks is obtained and a task bidding request is issued to each device agent. Each device agent reads the device agent status file and calculates the predicted task completion time and resource consumption to obtain a bidding parameter combination. The bidding parameter combination is sent to the collaborative decision network model to perform fusion inference to obtain a comprehensive bidding score. The comprehensive bidding scores are sorted in descending order and the first device is selected to obtain a task allocation mapping table. Each device agent receives sub-tasks according to the task allocation mapping table and drives the equipment to perform production actions. During the execution process, the task progress and equipment status are periodically collected to obtain execution status feedback records. The execution status feedback records are sent to the anomaly diagnosis model for execution status pattern matching to obtain anomaly judgment results. Based on the anomaly judgment results, the anomaly response knowledge base is queried to obtain response strategies. According to the response strategies, the unfinished sub-tasks are re-tendered and updated control instructions are issued to the newly selected device agents.

2. The multi-agent cooperative control method for SMT production lines according to claim 1, characterized in that, The process involves collecting equipment operating parameters and load status from solder paste printers, pick-and-place machines, reflow ovens, and testing equipment in the SMT production line, and adding timestamps to obtain a raw state data set. This raw state data set is then aligned and filled with missing values ​​according to a unified time base to obtain a standardized state sequence, including: Data connections are established with the control systems of solder paste printers, pick-and-place machines, reflow ovens, and testing equipment through industrial communication protocols. The original status data set is obtained by reading the equipment operating parameters and load status from each control system and adding the timestamp of the acquisition time and the equipment identifier. The timestamps of each record in the original state data set are converted to a unified time base to obtain a time-aligned data set. Continuity detection is performed on the time-aligned data set according to a preset sampling period, and missing positions are filled by interpolation with adjacent valid values ​​to obtain a standardized state sequence.

3. The multi-agent cooperative control method for SMT production lines according to claim 1, characterized in that, The step of grouping the standardized state sequence by device identifier and associating it with the production capacity parameters of each device to obtain the device intelligent agent state file includes: The standardized state sequence is grouped according to the equipment identifier field to obtain the equipment group state set. The capacity parameters and process capability parameters corresponding to each equipment are read from the production line configuration library to obtain the equipment capability parameter set. The device intelligent agent status file is obtained by performing association matching between each group in the device group status set and the device capability parameter set according to the device identifier.

4. The multi-agent cooperative control method for SMT production lines according to claim 1, characterized in that, The process involves parsing production orders to obtain a set of subtasks and issuing task bidding requests to each device agent. Each device agent reads the device agent status file and calculates the predicted task completion time and resource consumption to obtain a combination of bidding parameters, including: The product model, bill of materials, and process requirements in the production order are parsed to obtain task description information. The task description information is then decomposed according to workstation type and process sequence to obtain a set of sub-tasks. The set of sub-tasks is then encapsulated into a task bidding request and published to each equipment agent. After receiving the task bidding request, each device agent reads the current load and capacity parameters in the device agent status file, calculates the task completion time prediction value based on the current load and capacity parameters and according to the preset time estimation rules, calculates the resource consumption prediction value based on the process requirements in the task bidding request and according to the preset resource consumption rules, and combines the task completion time prediction value and the resource consumption prediction value to obtain the bidding parameter combination.

5. The multi-agent cooperative control method for SMT production lines according to claim 1, characterized in that, The process of feeding the combined bidding parameters into a collaborative decision-making network model to perform fusion inference and obtain a comprehensive bidding score, and then sorting the comprehensive bidding scores in descending order and selecting the first-ranked device to obtain a task allocation mapping table, includes: The predicted values ​​of task completion time and resource consumption in the bidding parameter combination are numerically transformed according to the preset normalization rules to obtain the normalized bidding feature vector. The normalized bidding feature vector is then fed into the multi-layer fusion structure of the collaborative decision network model to perform feature weighting and nonlinear transformation to obtain the comprehensive bidding score. The comprehensive bid scores are sorted in descending order of numerical value to obtain a score ranking sequence. The first device in the score ranking sequence is selected as the winning device for the corresponding sub-task, and a mapping relationship between the sub-task and the winning device is established to obtain a task allocation mapping table.

6. The multi-agent cooperative control method for SMT production lines according to claim 1, characterized in that, Each device agent receives sub-tasks according to the task allocation mapping table and drives the equipment to execute production actions. During execution, task progress and equipment status are periodically collected to obtain execution status feedback records. These execution status feedback records are then sent to an anomaly diagnosis model for execution status pattern matching to obtain anomaly determination results, including: Each intelligent device queries the sub-task corresponding to its own device from the task allocation mapping table and converts the sub-task into a device control command, which is then sent to the device control system to drive the device to perform production actions. During the execution process, the device control system reads the number of completed workpieces, the number of remaining workpieces, and the device operating parameters according to a preset collection cycle to obtain execution status feedback records. The device operating parameters in the execution status feedback record are extracted and organized into a status feature vector. The status feature vector is sent to the anomaly diagnosis model to perform a similarity calculation with the normal operation mode to obtain a mode deviation value. The mode deviation value is compared with a preset anomaly threshold condition to obtain an anomaly judgment result.

7. The multi-agent cooperative control method for SMT production lines according to claim 1, characterized in that, The step of querying the anomaly response knowledge base based on the anomaly determination result to obtain a response strategy, and then re-tendering the unfinished sub-tasks according to the response strategy and issuing updated control instructions to the newly awarded device agent, includes: An anomaly description information is obtained by extracting anomaly type tags and anomaly occurrence device identifiers from anomaly determination results. Based on the anomaly description information, a response strategy is obtained by querying the standard processing flow stored in the anomaly response knowledge base according to preset index rules. Based on the response strategy, the unfinished subtasks corresponding to the abnormal devices are extracted from the original task allocation mapping table to obtain the subtasks to be reassigned. The subtasks to be reassigned are encapsulated into a task bidding request and published to the device intelligent agent with alternative execution capability. The device intelligent agent that wins the bid is selected according to the comprehensive bidding score, and the updated control command is sent to the device intelligent agent that wins the bid.

8. A multi-agent collaborative control system for SMT production lines, characterized in that, The system includes: The equipment status monitoring module is used to collect equipment operating parameters and load status from solder paste printers, pick-and-place machines, reflow ovens and testing equipment in the SMT production line, and add timestamps to obtain a raw status data set. The raw status data set is aligned and missing values ​​are filled according to a unified time base to obtain a standardized status sequence. The standardized status sequence is grouped according to equipment identifier and associated with the capacity parameters of each equipment to obtain an equipment intelligent agent status file. The equipment task allocation module is used to obtain a set of sub-tasks based on the parsing of production orders and to issue task bidding requests to each equipment agent. Each equipment agent reads the status file of the equipment agent and calculates the predicted value of task completion time and the predicted value of resource consumption to obtain a combination of bidding parameters. The combination of bidding parameters is sent to the collaborative decision network model to perform fusion inference to obtain a comprehensive bidding score. The comprehensive bidding scores are sorted in descending order and the first-ranked equipment is selected to obtain a task allocation mapping table. The collaborative control module is used for each device agent to receive sub-tasks according to the task allocation mapping table and drive the equipment to perform production actions. During the execution process, the module periodically collects task progress and equipment status to obtain execution status feedback records. The execution status feedback records are sent to the anomaly diagnosis model for execution status pattern matching to obtain anomaly judgment results. Based on the anomaly judgment results, the module queries the anomaly response knowledge base to obtain response strategies. According to the response strategies, the unfinished sub-tasks are re-tendered and updated control instructions are issued to the newly selected device agents.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the multi-agent cooperative control method for SMT production lines as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-agent cooperative control method for SMT production lines as described in any one of claims 1 to 7.

Images