Multi-industry mixed-line SMT production line scheduling optimization system based on digital twinning

Abstract

The application belongs to the technical field of intelligent scheduling, and particularly relates to a multi-industry mixed-line SMT production line scheduling optimization system based on digital twinning, which comprises a digital twinning model construction module, a multi-industry mixed-line production line digital twinning model is constructed based on a device information model and a product process knowledge base; a multi-agent reinforcement learning scheduling module, which comprises four agents of scheduling, line changing, material distribution and quality control, and realizes adaptive scheduling decision through a multi-objective composite reward function; a knowledge graph diagnosis module, which constructs a scheduling knowledge graph based on historical cases, realizes intelligent retrieval, adaptation and reuse of a scheduling scheme; and a central collaborative controller, which forms a perception-analysis-decision-execution-learning closed loop. The application realizes efficient adaptive scheduling of multi-industry mixed-line SMT production lines of communication, automobile electronics, medical electronics and the like through the fusion of digital twinning and multi-agent reinforcement learning.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology, and in particular to a multi-industry mixed-line SMT production line scheduling optimization system based on digital twins. Background Technology

[0002] Surface Mount Technology (SMT) production lines are a core production link in the electronics manufacturing industry, widely used in communication equipment, consumer electronics, automotive electronics, medical electronics, aerospace, and many other sectors. As electronic products evolve towards multi-variety, small-batch, and high-frequency changeover, SMT production lines are transforming from traditional single-product mass production to multi-industry mixed-line production—the same production line needs to switch between producing communication modules, automotive controllers, medical circuit boards, and other products with different industry and process requirements at different times. In mixed-line production, production line scheduling faces multiple challenges, including significant differences in product processes, tight changeover windows, a wide variety of materials, and dynamic changes in equipment status. Traditional scheduling methods based on fixed cycle times and manual experience are no longer sufficient to meet the demands of efficient production.

[0003] Currently, the scheduling optimization of SMT production lines mainly relies on the following technical solutions: First, static scheduling based on Manufacturing Execution System (MES) to prepare production plans in advance according to order delivery dates and standard working hours; second, using commercial Advanced Planning and Scheduling (APS) software to perform finite capacity scheduling based on constraint theory; third, some advanced production lines have introduced IoT-based equipment data acquisition systems to achieve real-time monitoring of production line status and rule-based event response scheduling; and fourth, a few studies have attempted to apply reinforcement learning to SMT production line scheduling, using intelligent agents to learn the optimal scheduling strategy.

[0004] However, the aforementioned existing technologies have the following shortcomings when practically applied to multi-industry mixed-line SMT production lines:

[0005] First, existing scheduling systems lack digital twin models to support multi-industry mixed-line scenarios; second, existing scheduling algorithms have a single optimization objective and are difficult to take into account the multi-dimensional performance indicators under multi-industry mixed-line scenarios; third, existing scheduling systems lack the ability to adaptively reschedule for production line emergencies; and fourth, existing scheduling knowledge and experience lack a structured accumulation and reuse mechanism. Summary of the Invention

[0006] 1. Technical problems to be solved

[0007] The purpose of this invention is to solve the problems in the prior art by proposing a multi-industry mixed-line SMT production line scheduling optimization system based on digital twins, which integrates multi-agent reinforcement learning and knowledge graph to achieve adaptive optimization scheduling.

[0008] 2. Technical Solution

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] Firstly, this application provides a multi-industry mixed-line SMT production line scheduling optimization system based on digital twins, including:

[0011] The digital twin model construction module is used to construct a digital twin model of a multi-industry mixed-line SMT production line based on the equipment information model and the product process knowledge base. The digital twin model includes a physical space, a virtual space, and a data interaction layer.

[0012] The multi-agent reinforcement learning scheduling module includes a scheduling agent, a line-changing agent, a material delivery agent, and a quality control agent, which are used to execute adaptive scheduling decisions based on the digital twin model.

[0013] The knowledge graph diagnostic module is used to construct a scheduling knowledge graph based on historical scheduling cases and process knowledge, and to realize the retrieval, adaptation and reuse of scheduling schemes through graph reasoning;

[0014] The central collaborative controller is used to coordinate the collaborative operation of the digital twin model construction module, the multi-agent reinforcement learning scheduling module, and the knowledge graph diagnostic module, and form a closed-loop management process of "perception-analysis-decision-execution-learning".

[0015] In one possible implementation, the digital twin model building module includes a device-level information model unit, which is used for standardized access of different types and brands of devices. The device-level information model unit achieves standardized and unified device connection, data collection, parameter distribution and event reporting through differentiated adapters. The device-level information model includes static attribute information, dynamic operating status information and maintenance information of the device.

[0016] In one possible implementation, the product process knowledge base includes product process knowledge from at least two of the communication equipment, automotive electronics, medical electronics, and consumer electronics industries. The process knowledge includes product mounting process parameters, quality inspection standards, special process requirements, and standard process times. The digital twin model is constructed based on the coupling mapping relationship between the equipment information model and the product process knowledge base.

[0017] In one possible implementation, the multiple intelligent agents share information through a collaborative communication protocol, which defines a state synchronization mechanism, an action negotiation mechanism, and conflict resolution rules among the intelligent agents. The scheduling intelligent agent is responsible for the overall production plan scheduling of the production line, the line change intelligent agent is responsible for the selection of line change timing and optimization of line change sequence, the material delivery intelligent agent is responsible for AGV scheduling and material delivery decisions, and the quality control intelligent agent is responsible for quality early warning and defect root cause localization.

[0018] In one possible implementation, the reward function of the multi-agent reinforcement learning scheduling module is a multi-objective composite function, including output reward, quality reward, delivery reward, changeover efficiency reward, and violation penalty; the weight of each reward item is adaptively adjusted according to the current production target; the multi-agent reinforcement learning scheduling module performs simulation training based on the digital twin model in the offline stage, and performs distributed decision-making based on the real-time production line status in the online stage.

[0019] In one possible implementation, when a device malfunction, emergency order insertion, or material shortage event is detected, the central collaborative controller triggers a rescheduling process: the digital twin model performs rescheduling simulation based on the current real-time status of the production line, evaluates the expected effects of multiple candidate rescheduling schemes, and selects the scheme with the best global optimization objective for execution; the effectiveness of the rescheduling scheme is evaluated through simulation confidence level, and a manual review process is triggered when the confidence level is lower than a preset threshold.

[0020] In one possible implementation, the knowledge graph diagnostic module extracts entities, relationships, and attributes from historical scheduling cases to construct a structured scheduling knowledge graph; the entity types include products, equipment, materials, process parameters, scheduling rules, anomaly types, and handling plans; the relationship types include the applicability relationship between products and equipment, the configuration relationship between products and process parameters, and the association relationship between anomalies and handling plans.

[0021] In one possible implementation, the knowledge graph diagnostic module retrieves historical scheduling schemes similar to the new product using a graph similarity matching algorithm. The graph similarity matching algorithm includes entity similarity calculation, relation path matching, and structural similarity scoring, and the three are weighted and summed to obtain a similarity score. Based on the candidate basic schemes, a graph neural network is used to adapt and optimize the schemes to generate customized scheduling schemes.

[0022] In one possible implementation, the digital twin model includes a production line configuration-process capability matrix, which includes the mapping relationship between equipment and product type, the matching relationship between equipment capability and process requirements, and the mapping relationship between material delivery path and mounting station. When equipment is added or removed from the production line, new products are introduced, or process parameters are adjusted, the digital twin model automatically triggers a calibration update process.

[0023] In one possible implementation, the central collaborative controller further includes a knowledge closed-loop update unit, which is used to collect actual operation result data after the scheduling scheme is executed, and feed back the quantitative score and deviation analysis of the scheme execution effect to the knowledge graph diagnosis module; when the execution effect of the scheduling scheme reaches the preset standard, the new scheme knowledge is stored in the knowledge graph; when the execution effect does not reach the preset standard, the failure reason and deviation analysis result are recorded.

[0024] 3. Beneficial effects

[0025] Compared with the prior art, the advantages of this invention are:

[0026] (1) In this application, a digital twin model of mixed production line that can cover products from multiple industries such as communications, automotive electronics, and medical electronics is constructed through a standardized access system for equipment information models and a multi-industry product process knowledge base, to support scheduling decisions in complex mixed production line scenarios.

[0027] (2) In this application, through the distributed collaboration of four intelligent agents—scheduling, line change, material distribution, and quality control—and combined with a multi-objective composite reward function, intelligent scheduling optimization of mixed production lines under high-frequency line change and dynamic disturbances is realized, effectively improving the flexibility and operating efficiency of the production line.

[0028] (3) In this application, by structuring and accumulating historical scheduling cases and process knowledge, the scheduling scheme is generated quickly by using graph similarity matching and neural network adaptation, which significantly shortens the scheduling compilation cycle for new product introduction.

[0029] (4) In this application, the scheduling strategy is continuously optimized and knowledge is updated through the closed-loop feedback of digital twin simulation and multi-agent reinforcement learning, so as to support the continuous evolution of production line scheduling capabilities as business evolves. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the multi-industry mixed-line SMT production line scheduling optimization system based on digital twin proposed in this invention.

[0031] Figure 2 This is a flowchart illustrating the digital twin model construction method proposed in this invention;

[0032] Figure 3This is a flowchart illustrating the adaptive scheduling decision-making method based on multi-agent reinforcement learning proposed in this invention.

[0033] Figure 4 This is a flowchart illustrating the scheduling knowledge reuse and continuous optimization method proposed in this invention. Detailed Implementation

[0034] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0035] Reference Figure 1 A multi-industry mixed-line SMT production line scheduling and optimization system based on digital twins, including:

[0036] The digital twin model building module is used to construct digital twin models of multi-industry mixed-line SMT production lines based on equipment information models and product process knowledge bases. The digital twin model includes physical space, virtual space, and data interaction layer. The digital twin model building module includes equipment-level information model units, which are used for standardized access of different types and brands of equipment. Standardized and unified equipment connection, data acquisition, parameter distribution, and event reporting are achieved through differentiated adapters. The equipment-level information model includes equipment static attribute information, dynamic operating status information, and maintenance information.

[0037] The product process knowledge base contains product process knowledge from at least two of the following industries: communication equipment, automotive electronics, medical electronics, and consumer electronics. The process knowledge includes product mounting process parameters, quality inspection standards, special process requirements, and standard process time. The digital twin model is constructed based on the coupling mapping relationship between the equipment information model and the product process knowledge base.

[0038] The multi-agent reinforcement learning scheduling module includes a scheduling agent, a line change agent, a material delivery agent, and a quality control agent, which are used to execute adaptive scheduling decisions based on a digital twin model. The multiple agents share information through a cooperative communication protocol, which defines the state synchronization mechanism, action negotiation mechanism, and conflict resolution rules among the agents. The scheduling agent is responsible for the overall production plan scheduling of the production line, the line change agent is responsible for the selection of line change timing and optimization of line change sequence, the material delivery agent is responsible for AGV scheduling and material delivery decisions, and the quality control agent is responsible for quality early warning and defect root cause localization.

[0039] The reward function of the multi-agent reinforcement learning scheduling module is a multi-objective composite function, including output reward, quality reward, delivery reward, changeover efficiency reward, and violation penalty. The weight of each reward item is adaptively adjusted according to the current production target. The multi-agent reinforcement learning scheduling module is trained in simulation based on a digital twin model in the offline stage, and performs distributed decision-making based on the real-time production line status in the online stage.

[0040] The knowledge graph diagnostic module is used to construct a scheduling knowledge graph based on historical scheduling cases and process knowledge, and to realize the retrieval, adaptation and reuse of scheduling schemes through graph reasoning. The knowledge graph diagnostic module retrieves historical scheduling schemes similar to new products through a graph similarity matching algorithm. The graph similarity matching algorithm includes entity similarity calculation, relation path matching and structural similarity scoring, and the similarity score is obtained by weighted summation of the three. Based on the candidate basic schemes, the scheme adaptation and optimization are performed through graph neural networks to generate customized scheduling schemes.

[0041] The central collaborative controller coordinates the collaborative operation of the digital twin model construction module, the multi-agent reinforcement learning scheduling module, and the knowledge graph diagnostic module, forming a closed-loop management process of "perception-analysis-decision-execution-learning." The central collaborative controller also includes a knowledge closed-loop update unit, which collects actual operation result data after executing the scheduling plan, and feeds back the quantitative score and deviation analysis of the plan's execution effect to the knowledge graph diagnostic module. When the execution effect of the scheduling plan meets the preset standard, the new plan knowledge is stored in the knowledge graph; when the execution effect does not meet the preset standard, the reasons for failure and deviation analysis results are recorded.

[0042] Example 1:

[0043] Reference Figure 2 A method for constructing a digital twin model for multi-industry mixed-line SMT production lines, achieving high-fidelity mapping from physical production lines to virtual space, includes the following steps:

[0044] Step 1: Construct a standardized access system for production line equipment information models.

[0045] The SMT production line includes equipment from multiple brands and types, including but not limited to: printers, pick-and-place machines, reflow ovens, automated optical inspection equipment, X-ray inspection equipment, and automated guided vehicles. For different types and brands of equipment, based on the electronic assembly equipment interaction information model standard, differentiated adapter services are developed to achieve standardized and unified equipment connection, data acquisition, parameter distribution, and event reporting, thus constructing an equipment-level information model.

[0046] The equipment-level information model includes: equipment static attribute information (model, specifications, design capacity, supported component types), dynamic operating status information (current speed, rejection rate, nozzle status, temperature profile), and maintenance information (cumulative running time, last maintenance time, and predicted lifespan of vulnerable parts).

[0047] Step 2: Construct a multi-industry product process knowledge base and process model.

[0048] For SMT mounting products in different industries such as communication equipment, automotive electronics, medical electronics, and consumer electronics, product process knowledge bases are constructed. The process knowledge bases include: product identification information (part number, version, industry category), mounting process parameters (solder paste type, printing parameters, reflow soldering temperature profile, mounting sequence constraints), quality inspection standards (AOI detection threshold, X-ray detection standards), and special process requirements (traceability requirements for automotive electronics, cleanliness requirements for medical electronics).

[0049] Based on the aforementioned process knowledge base, process models for each product are constructed. These process models include standard working hours, material requirements lists, equipment resource requirements, and quality control points for each process.

[0050] Step 3: Establish the coupling mapping relationship between production line equipment and product processes.

[0051] The equipment-level information model is coupled and mapped with the product process model to construct a production line configuration-process capability matrix. The coupling mapping includes: the mapping relationship between equipment and product types (the range of product types that each pick-and-place machine can mount), the matching relationship between equipment capabilities and process requirements (whether the mounting speed meets the cycle time requirement, whether the mounting accuracy meets the product tolerance requirement), and the mapping relationship between material delivery paths and mounting stations (AGV feeding routes, feeder station allocation).

[0052] Step 4: Construct a digital twin model of a multi-industry mixed production line.

[0053] Based on the aforementioned equipment information model, product process knowledge base, and coupling mapping relationship, a digital twin model of a multi-industry mixed-line SMT production line is constructed. The digital twin model includes a physical space, a virtual space, and a data interaction layer: Physical space: the physical entities of the production line equipment and their operating status data, collected in real-time via IoT communication protocols.

[0054] Virtual space: the geometric model, behavioral model, and rule model of the production line. The geometric model includes the three-dimensional layout of the equipment and the relative positions between the equipment. The behavioral model includes the operating rules and performance degradation rules of the equipment under different working conditions. The rule model includes the constraint rules for the operation of the production line (such as line changeover rules, material delivery rules, and quality control rules).

[0055] Data Interaction Layer: Enables real-time data mapping from physical space to virtual space, and facilitates the transmission and execution of simulation decisions from virtual space to physical space. Step 5: Perform online calibration and synchronous updates of the digital twin model.

[0056] When equipment is added or removed from the production line, equipment performance degrades, new products are introduced, or process parameters are adjusted, the digital twin model automatically triggers a calibration and update process. The update includes: correcting equipment performance model parameters based on real-time equipment operating data; expanding the product model library based on the new product process knowledge base; and updating the geometric model based on production line layout adjustment information.

[0057] The synchronization accuracy of the digital twin model is evaluated by the model-physical deviation index. When the deviation exceeds a preset threshold, a model correction process is triggered to ensure the consistency between the virtual model and the physical production line.

[0058] Example 2:

[0059] Reference Figure 3 This embodiment provides an adaptive scheduling decision-making method based on multi-agent reinforcement learning to achieve intelligent scheduling optimization of multi-industry mixed-line SMT production lines under high-frequency line changes and dynamic disturbances, including the following steps:

[0060] Step 1: Construct a multi-agent collaborative decision-making framework.

[0061] The scheduling decision of the SMT production line is decomposed into collaborative decision-making by multiple intelligent agents, including: a scheduling intelligent agent: responsible for the overall production planning and scheduling of the production line, and determining the processing sequence and start time of each order on the production line;

[0062] Line switching agent: responsible for making line switching strategy decisions when switching between different products, including selecting the timing of line switching, optimizing the line switching sequence, and quickly configuring debugging parameters;

[0063] Material delivery intelligent agent: responsible for AGV scheduling and material delivery decisions, delivering materials in advance based on the material status of the pick-and-place machine and the line changeover plan;

[0064] The quality control agent is responsible for quality early warning, root cause localization of quality defects, and decision-making on appropriate actions. Information is shared among the agents via a collaborative communication protocol, which defines the state synchronization mechanism, action negotiation mechanism, and conflict resolution rules between the agents.

[0065] Step 2: Define the state space and action space for multi-agent reinforcement learning.

[0066] For each agent, define its observed state space and executable action space.

[0067] The state space of the scheduling agent includes: the current work-in-process queue of the production line, the priority and delivery date of each order, and the current load and health status of each device; the action space includes: selecting the next order to process and determining the processing order of each order.

[0068] The state space of the line-changing agent includes: the current product type being produced, the next product type to be produced, the time window required for line changing, and the current parameter configuration of each device; the action space includes: selecting the line-changing timing (immediate line changing / delayed line changing / preparation before line changing).

[0069] The state space of the material delivery agent includes: the remaining material quantity of each pick-and-place machine, the inventory status of the material warehouse, and the current task and position of the AGV; the action space includes: selecting a delivery task, assigning an AGV, and determining the delivery path.

[0070] The state space of the quality control agent includes: SPI / AOI detection data, product yield trend, and equipment parameter drift; the action space includes: triggering quality warnings, adjusting process parameter thresholds, and generating defect root cause diagnosis reports.

[0071] Step 3: Construct a multi-objective composite reward function.

[0072] The reward function of the multi-agent reinforcement learning is designed as a composite function containing multiple optimization objectives to take into account multi-dimensional performance indicators in multi-industry mixed-line scenarios:

[0073]

[0074] in For production bonuses, the ratio of actual production output to theoretical capacity of the production line is calculated. Quality rewards are calculated based on the product's first-pass yield and quality warning interception rate. Delivery bonuses are calculated based on the on-time delivery rate of orders. The bonus for line switching efficiency is calculated based on the ratio of actual line switching time to standard line switching time; As a penalty for violations, it is used to restrain actions that violate production line constraints (such as running out of materials, exceeding capacity, or changing lines after the allotted time). to The adaptive weights for each reward item are dynamically adjusted based on current production targets (e.g., focusing on output during peak seasons and cost during off-seasons).

[0075] Step 4: Perform offline training and online decision-making based on digital twins.

[0076] In the offline phase, a simulation environment is built based on the digital twin model. Multiple agent reinforcement learning is conducted through simulation of various production scenarios (including normal production, emergency order insertion, equipment failure, material shortage, line switching, etc.) to obtain the initial scheduling strategy.

[0077] The simulation environment samples are generated through Monte Carlo sampling of the digital twin model, which supports the reproduction of the historical state of the actual production line and the forward-looking extrapolation of hypothetical scenarios.

[0078] During the online phase, the trained strategy is deployed to the actual production line. Each agent executes distributed decisions based on real-time collected production line status data and achieves millisecond-level response through edge computing nodes.

[0079] Step 5: Perform online rescheduling optimization under emergency events.

[0080] When unexpected events such as equipment failure, emergency order insertion, or material shortage are detected, the digital twin model performs rescheduling simulation based on the current real-time status of the production line, evaluates the expected effects of multiple candidate rescheduling schemes, and selects the scheme with the best global optimization objective for execution.

[0081] The effectiveness of the rescheduling scheme is evaluated through the confidence level of simulation. When the confidence level is lower than a preset threshold, a manual review process is triggered.

[0082] Example 3:

[0083] Reference Figure 4 This embodiment provides a method for scheduling knowledge reuse and continuous optimization based on knowledge graphs, which realizes the structured accumulation and automated reuse of scheduling experience and process knowledge, including the following steps.

[0084] Step 1: Construct a knowledge graph for SMT production line scheduling across multiple industries.

[0085] A scheduling knowledge graph is constructed by extracting entities, relationships, and attributes from historical scheduling cases, process documents, and expert experience from SMT production lines across multiple industries. The entity types in the knowledge graph include: products, equipment, materials, process parameters, scheduling rules, exception types, and handling solutions. The relationship types include: applicability relationships between products and equipment, configuration relationships between products and process parameters, association relationships between exceptions and handling solutions, and conflict and priority relationships between scheduling rules.

[0086] Step 2: Perform knowledge graph-based scheduling scheme retrieval and adaptation.

[0087] When a scheduling scheme needs to be developed for a new product, the system retrieves historical scheduling schemes for similar products (based on characteristics such as industry, type of mounted components, and process complexity) through a knowledge graph, calculates a similarity score using a graph similarity matching algorithm, and selects the historical scheme with the highest score as a candidate basic scheme.

[0088] The calculation process of the graph similarity matching algorithm includes: entity similarity calculation, relation path matching, and structural similarity scoring, and the similarity score is obtained by weighted summation of the three.

[0089] Step 3: Perform intelligent adaptation and optimization of the scheduling scheme.

[0090] Based on the candidate basic schemes, and combined with the current production line's equipment status, material inventory, and personnel configuration information, a graph neural network is used to adapt and optimize the schemes, generating a customized scheduling scheme for the current scenario.

[0091] The graph neural network takes the subgraph of the candidate basic scheme as input, learns the key constraints and optimization patterns of the scheduling scheme through multiple rounds of message passing, and outputs the adapted scheduling decision parameters.

[0092] Step 4: Perform a closed-loop update of the scheduling knowledge.

[0093] After executing the scheduling plan, the system collects actual operational result data (including actual output, switchover time, quality data, etc.) and feeds the plan's effectiveness back to the knowledge graph. This feedback includes a quantitative score of the plan's execution effectiveness and a deviation analysis between the scheduling plan and the actual results.

[0094] When the execution effect of the scheduling plan meets the preset success criteria, the complete knowledge of the new plan is stored in the knowledge graph, forming new knowledge nodes and relationship edges; when the execution effect does not meet the preset criteria, the reasons for failure and the results of deviation analysis are recorded to avoid similar errors from happening again.

[0095] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-industry mixed-line SMT production line scheduling and optimization system based on digital twins, characterized in that, include: The digital twin model construction module is used to construct a digital twin model of a multi-industry mixed-line SMT production line based on the equipment information model and the product process knowledge base. The digital twin model includes a physical space, a virtual space, and a data interaction layer. The multi-agent reinforcement learning scheduling module includes a scheduling agent, a line-changing agent, a material delivery agent, and a quality control agent, which are used to execute adaptive scheduling decisions based on the digital twin model. The knowledge graph diagnostic module is used to construct a scheduling knowledge graph based on historical scheduling cases and process knowledge, and to realize the retrieval, adaptation and reuse of scheduling schemes through graph reasoning; The central collaborative controller is used to coordinate the collaborative operation of the digital twin model construction module, the multi-agent reinforcement learning scheduling module, and the knowledge graph diagnostic module, and form a closed-loop management process of "perception-analysis-decision-execution-learning".

2. The multi-industry mixed-line SMT production line scheduling optimization system based on digital twins according to claim 1, characterized in that, The digital twin model construction module includes a device-level information model unit, which is used for standardized access of different types and brands of devices. It achieves standardized and unified device connection, data collection, parameter distribution and event reporting through differentiated adapters. The device-level information model includes static attribute information, dynamic operating status information and maintenance information of the device.

3. The multi-industry mixed-line SMT production line scheduling and optimization system based on digital twins according to claim 1, characterized in that, The product process knowledge base contains product process knowledge from at least two of the communication equipment, automotive electronics, medical electronics and consumer electronics industries. The process knowledge includes product mounting process parameters, quality inspection standards, special process requirements and standard process time. The digital twin model is constructed based on the coupling mapping relationship between the equipment information model and the product process knowledge base.

4. The multi-industry mixed-line SMT production line scheduling and optimization system based on digital twins according to claim 1, characterized in that, The multiple intelligent agents share information through a collaborative communication protocol, which defines the state synchronization mechanism, action negotiation mechanism, and conflict resolution rules among the intelligent agents. The scheduling intelligent agent is responsible for the overall production plan scheduling of the production line, the line change intelligent agent is responsible for the selection of line change timing and optimization of line change sequence, the material delivery intelligent agent is responsible for AGV scheduling and material delivery decisions, and the quality control intelligent agent is responsible for quality early warning and defect root cause location.

5. The multi-industry mixed-line SMT production line scheduling optimization system based on digital twins according to claim 1, characterized in that, The reward function of the multi-agent reinforcement learning scheduling module is a multi-objective composite function, including output reward, quality reward, delivery reward, line changeover efficiency reward, and violation penalty; the weight of each reward item is adaptively adjusted according to the current production target. The multi-agent reinforcement learning scheduling module performs simulation training based on the digital twin model in the offline phase, and executes distributed decisions based on the real-time production line status in the online phase.

6. The multi-industry mixed-line SMT production line scheduling and optimization system based on digital twins according to claim 1, characterized in that, When a device malfunction, emergency order insertion, or material shortage event is detected, the central collaborative controller triggers a rescheduling process: the digital twin model performs rescheduling simulation based on the current real-time status of the production line, evaluates the expected effects of multiple candidate rescheduling schemes, and selects the scheme with the best global optimization objective for execution; the effectiveness of the rescheduling scheme is evaluated through simulation confidence level, and a manual review process is triggered when the confidence level is lower than a preset threshold.

7. The multi-industry mixed-line SMT production line scheduling and optimization system based on digital twins according to claim 1, characterized in that, The knowledge graph diagnostic module extracts entities, relationships, and attributes from historical scheduling cases to construct a structured scheduling knowledge graph. The entity types include products, equipment, materials, process parameters, scheduling rules, anomaly types, and handling plans. The relationship types include the applicability relationship between products and equipment, the configuration relationship between products and process parameters, and the association relationship between anomalies and handling plans.

8. The multi-industry mixed-line SMT production line scheduling optimization system based on digital twins according to claim 1, characterized in that, The knowledge graph diagnostic module retrieves historical scheduling schemes similar to the new product using a graph similarity matching algorithm. The graph similarity matching algorithm includes entity similarity calculation, relation path matching, and structural similarity scoring, and the three are weighted and summed to obtain a similarity score. Based on the candidate basic schemes, a graph neural network is used to adapt and optimize the schemes to generate customized scheduling schemes.

9. The multi-industry mixed-line SMT production line scheduling and optimization system based on digital twins according to claim 1, characterized in that, The digital twin model includes a production line configuration-process capability matrix, which includes the mapping relationship between equipment and product type, the matching relationship between equipment capability and process requirements, and the mapping relationship between material delivery path and mounting station. When equipment is added or removed from the production line, new products are introduced, or process parameters are adjusted, the digital twin model automatically triggers a calibration and update process.

10. The multi-industry mixed-line SMT production line scheduling and optimization system based on digital twins according to claim 1, characterized in that, The central collaborative controller also includes a knowledge closed-loop update unit, which is used to collect actual operation result data after the scheduling scheme is executed, and feed back the quantitative score and deviation analysis of the scheme execution effect to the knowledge graph diagnosis module. When the execution effect of the scheduling plan meets the preset standard, the new plan knowledge is stored in the knowledge graph; when the execution effect does not meet the preset standard, the reasons for failure and the results of deviation analysis are recorded.

Images