Intelligent optimization method and system for process parameters in whole process of enameled wire production

Abstract

The application discloses an intelligent optimization method and system for process parameters in the whole production process of enameled wire, and the method comprises the following steps: acquiring real-time operation data of each process of the production line; inputting the real-time operation data into a pre-built prediction model to predict future quality indexes, wherein the prediction model is a model containing physical coupling constraints between processes; comparing the predicted quality indexes with preset standards, and triggering online re-optimization when the prediction deviation exceeds a threshold; and based on a multi-objective optimization algorithm, solving optimal process parameter combinations in a feasible region meeting the coupling constraints and issuing the combinations for execution. The application solves the technical problems in the prior art that the optimization of enameled wire process parameters relies on manual experience, lacks modeling and global optimization of physical coupling across processes in the whole process, and cannot realize online real-time prediction and adaptive control, and realizes intelligent and predictive closed-loop optimization of the whole production process of enameled wire, and can be widely applied to the field of intelligent control of process parameters in enameled wire production enterprises.

Description

Technical Field

[0001] This invention belongs to the field of industrial intelligent manufacturing and process control technology, specifically relating to an intelligent optimization method and system for process parameters throughout the entire enameled wire production process. Background Technology

[0002] Enameled wire production involves multiple closely coupled processes, including drawing, annealing, coating, baking, cooling, and winding. The process parameters of each process (such as drawing speed, annealing temperature, coating viscosity, and baking temperature profile) interact and jointly determine the electrical, mechanical, and appearance quality of the product. With the manufacturing industry's transformation towards intelligent manufacturing, achieving precise control and optimization of process parameters through digital means is crucial for reducing trial-and-error costs, improving product consistency, increasing production efficiency, and reducing energy consumption. This represents a key technological direction for the enameled wire industry to achieve high-quality development.

[0003] However, existing technologies still have significant shortcomings in optimizing enameled wire process parameters. On the one hand, process parameter settings mainly rely on manual experience or local adjustments for single processes, lacking modeling and global optimization of the physical coupling relationships between all processes, resulting in low parameter matching between different processes and large fluctuations in product quality. On the other hand, existing automated adjustment schemes mostly adopt an offline iterative mode of "trial production, testing, and correction," which is essentially a post-production correction. It cannot perceive changes in operating conditions in real time during production and predict quality trends in advance. Moreover, each optimization is independent of the others, lacking knowledge accumulation and continuous learning mechanisms based on historical data, making it difficult to meet the needs of high-speed production lines for real-time, forward-looking control and adaptive optimization. Summary of the Invention

[0004] This invention provides an intelligent optimization method and system for process parameters throughout the entire enameled wire production process. It aims to solve the technical problems in existing enameled wire production process parameter optimization, such as reliance on manual experience or local adjustments, lack of physical coupling modeling and global optimization across all processes, and the fact that existing automated adjustment schemes mostly adopt a post-correction mode of "trial production, testing, and correction," which cannot provide online real-time prediction and adaptive control, and lack historical knowledge accumulation and continuous learning mechanisms.

[0005] To address the aforementioned technical problems, this invention provides an intelligent optimization method for process parameters throughout the entire enameled wire production process, characterized by the following steps:

[0006] Obtain real-time operating data for each process of the enameled wire production line, wherein each process includes at least wire drawing, annealing, coating, baking, cooling and winding processes;

[0007] The real-time operating data is input into a pre-built prediction model to predict future product quality indicators. The prediction model is a model that includes physical coupling constraints between processes.

[0008] The predicted quality indicators are compared with the preset quality standards. When the prediction deviation exceeds the set threshold, the online re-optimization process is triggered with the current working conditions as the initial condition.

[0009] The optimal combination of process parameters is solved within the feasible region that satisfies the physical coupling constraints between the processes based on the multi-objective optimization algorithm, and the optimal combination of process parameters is then distributed to the execution mechanisms of each process for adaptive adjustment.

[0010] Furthermore, the prediction model is a hybrid digital twin model, and the construction process of the hybrid digital twin model includes:

[0011] Process mechanism models were constructed based on the physicochemical principles of each process, and data-driven models were constructed based on historical production data.

[0012] A residual modeling strategy is adopted to couple the process mechanism model with the data-driven model: the process mechanism model calculates the preliminary predicted value, the data-driven model learns the residual between the preliminary predicted value and the actual value, and the sum of the preliminary predicted value and the residual is used as the final prediction output.

[0013] Furthermore, the physical coupling constraints between processes include at least one of the following:

[0014] Temporal coupling constraint: characterizes the matching relationship between wire drawing speed and annealing time;

[0015] Thermo-coupling constraint: characterizing the matching relationship between the annealing exit temperature and the coating inlet temperature;

[0016] Rheological coupling constraint: characterizes the synergistic relationship between painting speed, paint viscosity and baking temperature, so as to constrain the deviation between the actual paint film thickness and the target paint film thickness within a preset range;

[0017] Mechanical coupling constraint: characterizes the matching relationship between the state of the paint film after baking and the winding tension.

[0018] Furthermore, the multi-objective optimization algorithm employs an improved NSGA-II algorithm, which includes:

[0019] Constraint Repair Operator: When a generated individual violates the physical coupling constraint between processes, the individual is repaired to the feasible domain boundary through local search;

[0020] Process grouping cross operator: The decision variables are grouped according to each process, and the cross operation is performed on a process group basis to maintain the overall coordination of parameters within the same process group.

[0021] Furthermore, when the online re-optimization process is triggered, the system retrieves the historical optimization case with the highest similarity to the current operating condition from the preset process parameter optimization knowledge base, and performs local fine-tuning using the optimal parameters of the historical optimization case as the initial solution to generate the optimal process parameter combination.

[0022] Furthermore, the method also includes establishing a continuous learning mechanism:

[0023] Transfer learning is used to transfer process knowledge from existing operating conditions to new operating conditions;

[0024] Incremental learning is employed to continuously update the hybrid digital twin model and the process parameter optimization knowledge base using newly generated production data.

[0025] Furthermore, the predicted time step is dynamically and adaptively determined based on the production line operating speed and the physical length of each process.

[0026] Another aspect of the present invention provides an intelligent optimization system for process parameters of the entire enameled wire production process, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in any one of the above embodiments.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] Compared with existing technologies, the intelligent optimization method and system for the entire process parameters of enameled wire production provided by this invention constructs a predictive model that includes physical coupling constraints between processes and inputs real-time operating data into this model to predict future product quality indicators. This allows for the prediction of trends before quality deviations actually occur, unlike the "post-production correction" model of existing technologies that rely on trial production sample testing data. This represents a paradigm shift from "offline iteration" to "online prevention," fundamentally avoiding the generation of defective products. When the prediction deviation exceeds a set threshold, an online re-optimization process is triggered using the current operating conditions as the initial condition. Based on a multi-objective optimization algorithm, this process satisfies the physical constraints between processes. Solving for the optimal combination of process parameters within the feasible region of coupled constraints can overcome the coupling conflict problem caused by the isolated setting of parameters for each process, ensuring that the parameters of all processes such as wire drawing, annealing, painting, baking, cooling, and winding can be physically coordinated, thereby obtaining a globally optimal solution that takes into account electrical performance, mechanical performance, energy consumption, and production efficiency. By automatically distributing the optimal combination of process parameters obtained from the solution to the actuators of each process for adaptive adjustment, closed-loop automatic control of the production process can be achieved. It can quickly respond to changes in operating conditions without human intervention, significantly reducing the dependence on operator experience, reducing product quality fluctuations caused by human factors, and improving the overall efficiency and product consistency of the production line. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the intelligent optimization method for process parameters throughout the entire enameled wire production process of this invention.

[0030] Figure 2 This is a flowchart illustrating the intelligent optimization method for process parameters throughout the entire enameled wire production process of this invention.

[0031] Figure 3 This is the execution flowchart of the improved NSGA-II multi-objective optimization algorithm in this invention. Detailed Implementation

[0032] The technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0033] In one embodiment of this application, an intelligent optimization method for process parameters throughout the entire enameled wire production process is provided. The method is deeply integrated with the Manufacturing Execution System (MES) and Distributed Control System (DCS) within the workshop via Industrial Ethernet or Fieldbus, covering all process stages including wire drawing, annealing, coating, baking, cooling, and winding.

[0034] This embodiment constructs a hybrid digital twin model that includes physical coupling constraints between processes, and achieves advanced prediction and multi-objective global optimization based on the hybrid digital twin model. Specifically, it first acquires real-time operating data of each process, and uses the hybrid digital twin model to predict future product quality indicators. When the prediction deviation exceeds a set threshold, an online re-optimization process is triggered. Within the feasible domain that satisfies physical coupling constraints such as timing, thermodynamics, rheology, and mechanics, it solves for the optimal combination of process parameters that balances quality, energy consumption, and efficiency, and automatically sends the results to the actuators of each process for adaptive adjustment. This achieves a paradigm shift from "post-event correction" to "pre-event prevention," significantly improving product quality consistency and production efficiency.

[0035] The method steps of this application will be described in detail below with reference to specific embodiments.

[0036] S1. Obtain real-time operating data of each process in the enameled wire production line, wherein each process includes at least wire drawing, annealing, coating, baking, cooling and winding.

[0037] In some implementations, communication connections are established with the programmable logic controllers (PLCs) of each process and the sensors and actuators deployed on the production site via industrial Ethernet or fieldbus (such as Profinet, EtherCAT, etc.) to collect the operating data of each process in real time. The sensors include, but are not limited to: speed sensors and tension sensors installed on the wire drawing machine to collect wire drawing speed and wire tension; thermocouple temperature sensors installed in the annealing furnace to collect annealing temperature; viscometers and laser diameter gauges installed in the painting process to collect paint viscosity and wet film diameter; K-type thermocouple arrays installed in each section of the baking oven to collect temperature distribution in each temperature zone of the oven; thermocouple anemometers installed in the cooling section to collect cooling wind speed and wind temperature; and tension sensors and meter counters installed on the take-up machine to collect take-up tension and production length.

[0038] The collected data dimensions cover equipment operating parameters, environmental parameters, raw material parameters, and product quality inspection data. Among them, equipment operating parameters include no fewer than twenty-eight dimensions of process parameters such as wire drawing speed, annealing temperature, coating speed, temperature of each zone of the oven, cooling wind speed, and take-up tension; environmental parameters include workshop temperature and humidity; raw material parameters include copper rod grade, paint batch viscosity, and solid content; and product quality inspection data includes ten quality indicator dimensions such as breakdown voltage, paint film continuity, pinhole count, adhesion, thermal shock, and scratch resistance.

[0039] After acquiring the raw data, the data is cleaned and standardized. The preprocessing operations include: removing outliers caused by sensor failure or signal interference, using linear interpolation or nearest neighbor mean imputation to handle missing values, and normalizing or standardizing data of different dimensions to eliminate the impact of dimension differences on subsequent model training and optimization. The preprocessed data is stored in a real-time database.

[0040] S2. Input the real-time operating data into a pre-built prediction model to predict future product quality indicators. The prediction model is a model that includes physical coupling constraints between processes.

[0041] In some implementations, the prediction model employs a hybrid digital twin model architecture, which deeply couples a process mechanism model based on physicochemical principles with a data-driven model based on historical data through a residual modeling strategy. This hybrid modeling approach aims to address the problems that pure mechanism models are difficult to accurately describe complex nonlinear relationships and that pure data-driven models lack physical interpretability and extrapolation capabilities.

[0042] Specifically, the construction of the process mechanism model is based on the physicochemical mechanisms of each process in enameled wire production. For example, in the baking process, a mechanism model based on heat and mass transfer equations is established to describe the quantitative relationship between temperature, cooling wind speed, linear velocity and curing degree of the enamel film in each zone of the oven. In the drawing process, a mechanism model based on plastic deformation theory is established to describe the quantitative relationship between drawing speed, die geometry parameters and wire mechanical properties. The process mechanism model receives real-time collected equipment operating parameters (such as temperature, speed, viscosity, etc.) as input and calculates preliminary predicted values ​​according to preset physical formulas. However, due to unmodeled dynamic factors such as equipment aging, environmental fluctuations and slight differences in raw materials in the actual production process, relying solely on the mechanism model often results in systematic biases.

[0043] To correct the aforementioned deviations, this embodiment further constructs a data-driven model. This data-driven model employs a deep neural network structure, such as a multilayer perceptron (MLP) or a long short-term memory (LSTM) network. During the model training phase, historical production data is collected, including process parameters, environmental parameters, raw material parameters, and corresponding actual product quality inspection data for each process step. The data-driven model specifically learns the residual between the predicted values ​​of the process mechanism model and the actual inspection values; that is, the residual is defined. ,in For actual quality indicators, The values ​​are predicted by the mechanistic model; the deep neural network takes process parameters as input and residuals as input. The output target is trained to capture nonlinear perturbations and complex coupling effects that are not covered by the mechanistic model.

[0044] In the online prediction phase, the reasoning process of the hybrid digital twin model is as follows: First, the real-time collected operation data of each process is input into the process mechanism model to calculate the preliminary prediction value; second, the same real-time operation data is input into the trained deep neural network data-driven model to obtain the residual prediction value; finally, the preliminary prediction value and the residual prediction value are added together to obtain the final quality index prediction value.

[0045] Through this residual modeling strategy, the process mechanism model provides a basic prediction framework that conforms to physical laws, ensuring the basic rationality and interpretability of the model under unseen operating conditions; while the data-driven model focuses on correcting local deviations, significantly improving prediction accuracy; tests show that under the same production conditions, the average absolute percentage error of the hybrid digital twin model can be controlled within 3.8%, which is better than the single mechanism model or the single data-driven model.

[0046] In some implementations, the predicted time step is dynamically and adaptively determined based on the production line speed and the physical length of each process; specifically, the real-time line speed of the current production line is first obtained. And the physical distance between the current monitoring point and the final quality inspection point. Based on formula Calculate the theoretical transmission time and add a certain safety margin to it. This determines the final prediction time step. Among them, safety margin This is used to cover the response delay of the actuator and the network transmission delay; thus, it ensures that there is enough time to complete the issuance and execution of parameter adjustment instructions before the actual occurrence of quality deviation, thereby achieving true proactive prediction and control.

[0047] The physical coupling constraints between processes include at least one of temporal coupling constraints, thermodynamic coupling constraints, rheological coupling constraints, and mechanical coupling constraints.

[0048] Regarding the temporal coupling constraint, it is mainly used to characterize the matching relationship between wire drawing speed and annealing time. The temporal coupling constraint is limited by the following logical relationship: the annealing time is equal to the effective heating zone length of the annealing furnace divided by the wire drawing speed; to ensure the adequacy of the annealing process and avoid over-annealing, the deviation between the actual calculated value of the annealing time and the theoretical target value shall not exceed a preset ratio (e.g., ±5%); if the wire drawing speed generated by the optimization algorithm changes, the operating parameters of the annealing furnace must be adjusted synchronously or the resulting change in annealing time must be accepted.

[0049] Thermo-coupling constraints are mainly used to characterize the matching relationship between the annealing exit temperature and the coating inlet temperature. Thermo-coupling constraints are defined as follows: the absolute value of the difference between the conductor temperature at the annealing process exit and the conductor temperature at the coating process inlet does not exceed a preset threshold (e.g., 30°C). This constraint ensures that the conductor is in a suitable thermodynamic state when it enters the coating die.

[0050] Regarding rheological coupling constraints, they are mainly used to characterize the synergistic relationship between painting speed, paint viscosity, and baking temperature, so as to constrain the deviation between the actual paint film thickness and the target paint film thickness within a preset range. The rheological coupling constraints are based on the paint film thickness prediction model, in which the paint film thickness is directly proportional to the painting speed, paint viscosity, and coating coefficient, and inversely proportional to the baking temperature and cooling wind speed. The system requires that the temperature curves of each section of the baking oven be dynamically matched according to the real-time collected paint viscosity and the set painting speed, so that the deviation between the predicted actual paint film thickness and the target paint film thickness does not exceed the preset range (e.g., ±8%).

[0051] Regarding the mechanical coupling constraint, it is mainly used to characterize the matching relationship between the state of the paint film after baking and the take-up tension. The mechanical coupling constraint is limited by a safety factor: the product of the take-up tension and the paint film adhesion coefficient must not exceed the tensile strength of the paint film, and a certain safety margin must be reserved (e.g., safety factor ≥ 1.2). The system dynamically calculates the maximum allowable take-up tension based on the paint film curing degree and adhesion level predicted by the digital twin model, and limits the take-up tension below this safety threshold during the optimization process.

[0052] By introducing the above four types of physical coupling constraints between processes, this embodiment can eliminate parameter combinations that, although mathematically satisfying a single objective function, are physically infeasible or may cause process conflicts during subsequent multi-objective optimization solutions, thereby ensuring that the issued optimal process parameter combinations have high engineering feasibility and production stability.

[0053] S3. Compare the predicted quality indicators with the preset quality standards. When the prediction deviation exceeds the set threshold, trigger the online re-optimization process with the current working conditions as the initial condition.

[0054] In some implementations, the predicted quality indicators output by the hybrid digital twin model are compared with preset target quality indicators or allowable fluctuation ranges; the set threshold can be configured according to different product specifications, customer quality requirements or historical production data; for example, for high value-added products, a stricter threshold can be set to intervene and adjust in advance; for conventional products, the threshold can be appropriately relaxed to reduce unnecessary frequent adjustments.

[0055] The set threshold is a pre-set upper limit of allowable deviation, and its value is determined according to the strictness of the quality standard of enameled wire products; the preset quality standard is the qualified standard value of the product; the preset quality standard is an absolute value standard, and the set threshold is a relative value tolerance.

[0056] Once the online re-optimization process is triggered, the system will use the current real-time operating conditions as the initial conditions to start the multi-objective optimization solution module. To ensure the real-time performance of the re-optimization and avoid the high computational time caused by starting a global search from scratch, the system first retrieves the historical optimization cases with the highest similarity to the current operating conditions from the pre-set process parameter optimization knowledge base. By calculating the Euclidean distance or cosine similarity between the current operating condition vector and the operating condition vectors of historical cases in the knowledge base, the system selects several sets of the closest historical optimization records. Subsequently, the optimal process parameter combinations in these historical cases are used as the initial solutions for local fine-tuning search, thereby generating the optimal process parameter combinations that satisfy the physical coupling constraints between processes within seconds.

[0057] S4. Based on the multi-objective optimization algorithm, the optimal combination of process parameters is solved within the feasible region that satisfies the physical coupling constraints between the processes, and the optimal combination of process parameters is sent to the execution mechanism of each process for adaptive adjustment.

[0058] In some implementations, an optimization objective function with multiple dimensions is first constructed; the optimization objectives include maximizing the overall product quality index, minimizing unit product energy consumption, and maximizing production efficiency; wherein, the overall product quality index can be further subdivided into a weighted combination of electrical performance (such as breakdown voltage), mechanical performance (such as elongation and springback angle), and appearance quality (such as pinhole count and coating continuity); unit product energy consumption refers to the total electrical and thermal energy consumed in producing a unit length or unit weight of enameled wire; and production efficiency is characterized by the production line operating speed and effective output rate.

[0059] In order to solve the above multi-objective optimization problem while satisfying the physical coupling constraints between processes, this embodiment adopts an improved non-dominated sorting genetic algorithm (NSGA-II). Compared with the traditional NSGA-II algorithm, the improvements in this embodiment are mainly reflected in the constraint handling mechanism and the design of the crossover operator.

[0060] Specifically, the improved NSGA-II algorithm introduces a constraint repair operator. During the algorithm iteration process, when a randomly generated individual (i.e., a set of candidate process parameters) violates the preset inter-process physical coupling constraints, a local search mechanism is activated to repair the individual. For example, if the wire drawing speed set by an individual results in insufficient annealing time, violating the temporal coupling constraint, the constraint repair operator will automatically adjust the annealing temperature of the individual or correct the range of the wire drawing speed according to the physical length and thermodynamic model of the annealing furnace, mapping it to the boundary or interior of the feasible region. This repair strategy avoids the generation of a large number of invalid solutions and improves the convergence speed and search efficiency of the algorithm.

[0061] Furthermore, the improved NSGA-II algorithm also employs a process grouping crossover operator. When performing crossover operations, the decision variables are logically grouped according to six processes: wire drawing, annealing, painting, baking, cooling, and wire winding. The crossover operation is performed on a process group basis, that is, two parent individuals are randomly selected, and the parameter sets of one or more complete process groups are exchanged, rather than randomly crossing individual parameter bits. This method maintains the overall coordination of parameters within the same process group and avoids disrupting the process balance within the process due to randomly shuffling parameter combinations.

[0062] To ensure the real-time performance of online re-optimization and meet the second-level response requirements of high-speed production lines, this embodiment incorporates a knowledge-based rapid retrieval strategy. When the online re-optimization process is triggered, the system does not perform a global blind search from scratch, but first searches a pre-set process parameter optimization knowledge base. This knowledge base stores optimization cases from historical production processes. Each case includes a working condition feature vector (such as raw material batch, ambient temperature and humidity, product specifications, etc.) and its corresponding optimal process parameter combination. The system calculates the similarity between the current real-time working condition and historical cases in the knowledge base, and selects the top N historical optimization cases with the highest similarity. Subsequently, the optimal parameters of these historical cases are used as the initial solution population and substituted into the improved NSGA-II algorithm for local fine-tuning. Since the initial solution is already very close to the global optimum, the algorithm only needs to perform a fine search within a small neighborhood to find the optimal solution that satisfies the constraints of the current working condition. Combined with edge computing power support and a lightweight digital twin model, this process can compress the re-optimization computation time to the second level.

[0063] The optimal combination of process parameters obtained from the solution is sent to the actuators of each process for adaptive adjustment. Specifically, this includes: encapsulating the optimal combination of process parameters into a control command data packet via industrial Ethernet or fieldbus protocol, and sending it to the programmable logic controller (PLC) corresponding to each process of the enameled wire production line; wherein, the control command data packet contains specific setting values ​​for each process such as wire drawing, annealing, coating, baking, cooling and winding; after receiving the control command data packet, the PLC of each process parses the corresponding target parameters and drives the corresponding actuator to perform actions to achieve real-time adjustment of process parameters.

[0064] Specifically, for the wire drawing process, the PLC sends a frequency adjustment command to the wire drawing machine's frequency converter based on the wire drawing speed setpoint in the optimal parameters, adjusting the motor speed to change the wire drawing speed; simultaneously, it adjusts the unwinding tension control module based on the tension setpoint to ensure stable wire tension entering subsequent processes; for the annealing process, the PLC sends a power adjustment command to the annealing furnace temperature controller based on the annealing temperature setpoint in the optimal parameters, adjusting the output power of the heating elements to quickly bring the temperature inside the annealing furnace to the target temperature range; for the coating process, the PLC sends a power adjustment command to the coating machine based on the coating speed and paint viscosity correlation parameters in the optimal parameters. The metering pump sends pulse frequency commands to precisely control the amount of paint supplied per unit time, thereby regulating the wet film thickness. For the baking process, the PLC sends adjustment commands to the temperature controllers and frequency converters of the cooling fans in each temperature zone according to the optimal parameters of the temperature and cooling wind speed settings of each temperature zone in the oven. For the winding process, the PLC sends torque or position control commands to the winding machine servo driver according to the winding tension setting in the optimal parameters to dynamically adjust the winding tension. Thus, through the closed-loop control link from the optimization algorithm to the actuator, the automation and fine adjustment of the process parameters of the entire enameled wire production process are realized.

[0065] In some implementations, the method further includes establishing a continuous learning mechanism; using transfer learning to transfer process knowledge from existing operating conditions to new operating conditions; and using incremental learning to continuously update the prediction model and the process parameter optimization knowledge base using newly generated production data.

[0066] Specifically, the system has a process parameter optimization knowledge base to store historical optimization cases. After each online re-optimization or offline global optimization, the system stores the current production condition data, the optimal process parameter combination obtained from the solution, and the product quality results subsequently detected as a complete optimization case in the knowledge base. The production condition data includes, but is not limited to, raw material batch information (such as copper rod grade and paint batch number), environmental parameters (workshop temperature and humidity), and initial equipment state parameters. The optimal process parameter combination includes key control parameters for the entire process, such as wire drawing speed, annealing temperature, paint viscosity, baking temperature curve, cooling air velocity, and take-up tension. The product quality results include key quality indicators such as breakdown voltage, pinhole count, and paint film adhesion.

[0067] When production conditions change, such as changing raw material batches or switching product specifications, the system uses a transfer learning mechanism to transfer existing knowledge to the new operating conditions. Specifically, the system first extracts the feature vector of the current new operating conditions and retrieves several historical optimization cases with the highest feature similarity to the new operating conditions from the process parameter optimization knowledge base. Since different batches of raw materials or similar specifications of products have commonalities in physicochemical mechanisms, the system uses a pre-trained source domain model (a hybrid digital twin model trained based on a large amount of historical data) as a foundation, freezes the parameters of the model's bottom-level feature extraction network, and only fine-tunes the parameters of the top-level output layer or some hidden layers. By introducing a small amount of trial production data or early production data under the new operating conditions as target domain samples, the system calculates the loss function between the predicted value and the actual value, and uses the backpropagation algorithm to update the weights of the fine-tuned layers. This approach enables the model to quickly adapt to the new operating conditions and significantly reduces the dependence on a large amount of labeled data under the new operating conditions.

[0068] During normal production, the system employs an incremental learning mechanism to continuously update the hybrid digital twin model and the process parameter optimization knowledge base. As the production line continues to operate, the system collects new production data and corresponding quality inspection data in real time. When the amount of newly accumulated data reaches a preset threshold, or when a systematic drift trend in model prediction error is detected, the incremental learning process is triggered. During incremental learning, the system retains the main structure of the original model and uses newly generated production data to update the model online. To avoid the "catastrophic forgetting" phenomenon, the system introduces a regularization term in the loss function to constrain the change in model parameters during the update process, ensuring that the model maintains its predictive ability for historical typical operating conditions while adapting to new distributed data. Simultaneously, new process parameter optimization cases are written into the knowledge base in real time, and the case index is updated.

[0069] In one embodiment of this application, an intelligent optimization system for process parameters of the entire enameled wire production process is also provided. The system includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the intelligent optimization method for process parameters of the entire enameled wire production process described in any of the above embodiments.

[0070] In some implementations, the processor is configured to execute an improved multi-objective optimization algorithm and the inference logic of a hybrid digital twin model. Specifically, the processor receives real-time operational data and inputs it into a hybrid digital twin model deployed in memory to obtain advanced predicted values ​​of quality indicators. When the prediction deviation exceeds a set threshold, the processor invokes an improved NSGA-II algorithm stored in memory, combines it with inter-process physical coupling constraints, searches for the Pareto optimal solution set within the feasible region, and generates the optimal combination of process parameters. The processor connects to various sensors and actuators on the enameled wire production line via industrial Ethernet or fieldbus to form a closed-loop control system. It automatically sends control commands to the programmable logic controllers (PLCs) of each process via OPC UA or Modbus TCP / IP standard industrial communication protocols to drive the corresponding actuators and achieve adaptive adjustment of process parameters.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for intelligent optimization of process parameters throughout the entire enameled wire production process, characterized in that, Includes the following steps: Obtain real-time operating data for each process of the enameled wire production line, wherein each process includes at least wire drawing, annealing, coating, baking, cooling and winding processes; The real-time operating data is input into a pre-built prediction model to predict future product quality indicators. The prediction model is a model that includes physical coupling constraints between processes. The predicted quality indicators are compared with the preset quality standards. When the prediction deviation exceeds the set threshold, the online re-optimization process is triggered with the current working conditions as the initial condition. The optimal combination of process parameters is solved within the feasible region that satisfies the physical coupling constraints between the processes based on the multi-objective optimization algorithm, and the optimal combination of process parameters is then distributed to the execution mechanisms of each process for adaptive adjustment.

2. The intelligent optimization method for process parameters throughout the entire enameled wire production process according to claim 1, characterized in that, The prediction model is a hybrid digital twin model, and the construction process of the hybrid digital twin model includes: Process mechanism models were constructed based on the physicochemical principles of each process, and data-driven models were constructed based on historical production data. A residual modeling strategy is adopted to couple the process mechanism model with the data-driven model: the process mechanism model calculates the preliminary predicted value, the data-driven model learns the residual between the preliminary predicted value and the actual value, and the sum of the preliminary predicted value and the residual is used as the final prediction output.

3. The intelligent optimization method for process parameters throughout the entire enameled wire production process according to claim 1, characterized in that, The physical coupling constraints between processes include at least one of the following: Temporal coupling constraint: characterizes the matching relationship between wire drawing speed and annealing time; Thermo-coupling constraint: characterizing the matching relationship between the annealing exit temperature and the coating inlet temperature; Rheological coupling constraint: characterizes the synergistic relationship between painting speed, paint viscosity and baking temperature, so as to constrain the deviation between the actual paint film thickness and the target paint film thickness within a preset range; Mechanical coupling constraint: characterizes the matching relationship between the state of the paint film after baking and the winding tension.

4. The intelligent optimization method for process parameters throughout the entire enameled wire production process according to claim 1, characterized in that, The multi-objective optimization algorithm employs an improved NSGA-II algorithm, which includes: Constraint Repair Operator: When a generated individual violates the physical coupling constraint between processes, the individual is repaired to the feasible domain boundary through local search; Process grouping cross operator: The decision variables are grouped according to each process, and the cross operation is performed on a process group basis to maintain the overall coordination of parameters within the same process group.

5. The intelligent optimization method for process parameters throughout the entire enameled wire production process according to claim 2, characterized in that, When the online re-optimization process is triggered, the system retrieves the historical optimization case with the highest similarity to the current operating condition from the preset process parameter optimization knowledge base, and performs local fine-tuning using the optimal parameters of the historical optimization case as the initial solution to generate the optimal process parameter combination.

6. The intelligent optimization method for process parameters throughout the entire enameled wire production process according to claim 5, characterized in that, The method also includes establishing a continuous learning mechanism: Transfer learning is used to transfer process knowledge from existing operating conditions to new operating conditions; Incremental learning is employed to continuously update the hybrid digital twin model and the process parameter optimization knowledge base using newly generated production data.

7. The intelligent optimization method for process parameters throughout the entire enameled wire production process according to claim 1, characterized in that, The predicted time step is dynamically and adaptively determined based on the production line speed and the physical length of each process.

8. An intelligent optimization system for process parameters throughout the entire enameled wire production process, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 7.

Images