Tungsten filament copper plating production line control system and method based on digital twinning

Abstract

The invention discloses a tungsten filament copper plating production line control system and method based on digital twinning, and relates to the technical field of photovoltaic material preparation and automatic control. The acquisition module is used for acquiring electrochemical parameters, hydrodynamic parameters, motion state parameters and visual morphological parameters of each line position in real time at millisecond-level high frequency; the real-time digital twinning engine is used for constructing a virtual twinning body comprising a physical geometric model, a multi-physics field coupling model, a real-time quality prediction model and a data driving compensation model; and the man-machine interaction and decision support interface is used for displaying a virtual-real mapping picture, a quality prediction curve, alarm information and an operation instruction recommended by the system in real time. The traditional'black box 'electrochemical process is converted into a visual three-dimensional scene through the digital twin engine, an operator can synchronously observe the plating growth state, anode consumption and bubble distribution of sixteen tungsten filaments on a human-computer interface, and the anomaly positioning time is remarkably shortened.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic material preparation and automation control technology, specifically to a control system and method for a tungsten wire copper plating production line based on digital twins. Background Technology

[0002] Driven by the need for cost reduction and efficiency improvement in the photovoltaic industry, diamond wire, using ultra-fine, high-strength tungsten wire as the substrate, is a key material for cutting silicon wafers. Tungsten wire requires precision copper plating pretreatment before diamond micropowder can be bonded. Sixteen-wire copper plating machines have become the mainstream production equipment due to their high efficiency, but their multi-line, high-speed, and continuous production characteristics also bring severe control challenges.

[0003] Currently, the existing control system of the 16-line copper plating machine has the following defects: "Black box" production and quality lag: Existing control systems can only monitor macroscopic electrical and motion parameters such as voltage, current, and speed, and cannot perceive key quality indicators such as coating thickness, uniformity, and crystallization status in real time. Quality inspection relies on post-production offline sampling, and by the time problems are discovered, a large number of scraps have already been generated, making it impossible to achieve pre-production warnings and in-process intervention; Fixed process parameters lack adaptability: Production parameters such as current density and line speed are usually set and fixed based on experience. However, when the performance of incoming tungsten wire batches fluctuates, the composition of the plating solution changes over time, or the equipment condition changes such as anode passivation, fixed parameters cannot guarantee the continuous production of optimal products. Frequent manual adjustments are required, which is highly dependent on the operator's experience. Summary of the Invention

[0004] The purpose of this invention is to provide a control system and method for a tungsten wire copper plating production line based on digital twins, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a control system for a tungsten wire copper plating production line based on digital twins, comprising: A multi-source sensing and data acquisition system is deployed on the physical production line of a 16-line copper plating machine to collect electrochemical parameters, fluid dynamic parameters, motion state parameters and visual morphological parameters of each line position in real time at a high frequency of milliseconds. A real-time digital twin engine is communicatively connected to a multi-source sensing and data acquisition system to construct a virtual twin that includes a physical geometric model, a multi-physics coupling model, a real-time quality prediction model, and a data-driven compensation model. The engine synchronizes the physical entity state at a speed of seconds, thereby achieving the technical effect of coating thickness prediction error of less than 0.1μm and thickness standard deviation of 16-line products of less than 0.1μm. An adaptive optimization controller, coupled with a real-time digital twin engine, is used to dynamically adjust the process parameters of each production line based on a model predictive control algorithm to eliminate the deviation between predicted quality and target quality. A multi-line collaborative optimization module, which works in conjunction with an adaptive optimization controller, is used to minimize the variance of the coating quality index of the sixteen-line product through a distributed constraint optimization method, thereby ensuring multi-line consistency. The human-computer interaction and decision support interface is used to display the virtual-real mapping screen, quality prediction curve, alarm information and system recommended operation instructions in real time.

[0006] Furthermore, the multi-source sensing and data acquisition system includes: An electrochemical sensor array is used to collect data on copper ion concentration, pH value, and temperature distribution in the plating bath. A current-voltage sensor array is used to independently monitor the current and voltage of each circuit in the sixteen-wire circuit. A fluid and motion sensor array is used to collect data on plating solution flow rate, various linear tensions, and linear velocity. The machine vision unit is used to acquire images of key areas of the plating tank at a rate of no less than 25fps to monitor bubble distribution and anode consumption status. The data from each sensor is transmitted to the real-time digital twin engine via an industrial Ethernet in a first cycle, wherein the first cycle is configured to be less than or equal to the sum of the sensor response time and the time constant of the electroplating process kinetics.

[0007] Furthermore, the real-time digital twin engine includes: Physical geometry model, configured to create a 3D parametric model of the device, used to visualize the geometric state and spatial layout of the physical device; A multiphysics coupling model, integrating electric field, flow field, and electrochemical models, is used to calculate the local current density and mass transfer efficiency at each line position. The electric field model employs the Laplace equation. To describe the potential distribution, the flow field model uses the incompressible Navier-Stokes equation to describe the plating solution flow, and the electrochemical model uses the Butler-Volmer equation to describe the copper deposition kinetics. Among them, the current density calculated by the electric field model is added to the energy equation as a Joule heat source term, which affects the temperature distribution in the flow field model. The flow velocity calculated by the flow field model affects the mass transfer process on the electrode surface, thereby changing the limiting current density and effective ion concentration in the Butler-Volmer equation. The real-time quality prediction model includes a feature extraction layer, a machine learning computation layer, and an output layer. The feature extraction layer is configured to receive the current density distribution, flow velocity distribution, and process parameters such as linear velocity, plating bath temperature, and Cu output from the multi-physics coupling model.2+ The concentration and the coating thickness deviation sequence within the historical data window are used. The machine learning calculation layer uses a gradient boosting tree or neural network algorithm to perform non-linear mapping on the features. The output layer is configured to output the coating thickness prediction value and deposition rate prediction value for each of the sixteen line positions. A data-driven compensation model is used to correct systematic errors in multiphysics coupling models online using machine learning algorithms. A real-time rendering engine is used to visualize the status of the virtual production line.

[0008] Furthermore, the adaptive optimization controller, based on the prediction results of the digital twin engine, continuously optimizes the process parameters of each production line using a model predictive control algorithm, wherein the optimization objective function is: J=Σ_{i=1}^{N}||Y_pred(k+i)-Y_ref(k+i)||²_Q+Σ_{i=0}^{N_u-1}||ΔU(k+i)||²_R, The constraints include: U_min≤U(k+i)≤U_max, ΔU_min≤ΔU(k+i)≤ΔU_max, Y_min≤Y_pred(k+i)≤Y_max, Where Y_pred is the predicted coating thickness vector, Y_ref is the target thickness, U is the control variable vector, ΔU is the rate of change of the control variable, Q and R are weight matrices, N is the prediction time domain, and N_u is the control time domain.

[0009] Furthermore, the multi-line collaborative optimization module adopts a distributed model predictive control architecture, decomposing the global optimization problem into sixteen sub-problems for parallel solution through Lagrange dual decomposition, wherein the objective function of the sub-problem of the j-th line contains coupling terms. λ_j is a Lagrange multiplier. The central coordinator iteratively updates the global average control value. The convergence condition is given by λ_j until convergence. .

[0010] Furthermore, the human-computer interaction and decision support interface is further configured as follows: Displays a 3D virtual production line rendered by a physical geometry model, and overlays process parameters, predicted coating thickness, and early warning status for each line position on the virtual equipment in real time. A tiered warning is triggered when any predicted value deviates from the target thickness by more than a dynamic threshold. Record and store complete process parameter timing data for each of the sixteen lines to support quality traceability and the construction of a process knowledge base.

[0011] A control method for a tungsten wire copper plating production line based on digital twins, the method comprising the following steps: S1: Construct a digital twin model that includes a physical geometry model, a multiphysics coupling model, and a real-time quality prediction model. The multiphysics coupling model includes at least an electric field, a flow field, and an electrochemical deposition field. The model parameters are continuously corrected using historical production data and real-time sensor data. S2: Using the first cycle to fuse the multi-source sensor data of sixteen lines, generate a high-confidence state estimate for each line, the state estimate including the coating thickness distribution, copper ion concentration field and temperature gradient distribution; S3: Based on the real-time quality prediction model, predict the coating thickness change trend of each line position in the second cycle. The real-time quality prediction model takes the output of the multi-physics coupling model, the current process parameters and the historical production data window as input, and generates the predicted values ​​of coating thickness and deposition rate through machine learning algorithms. S4: The rolling optimization algorithm is used to generate the process parameter adjustment amount for each line position, and the control command is generated after the adjustment amount is distributed and coordinated with the goal of minimizing the consistency deviation of the sixteen-line coating thickness. S5: Execute the control command and collect the sensor data for the next first cycle to form a closed-loop feedback; S6: Compare the actual coating thickness measurement with the predicted value, update the digital twin model parameters online, and store the process scheme that meets the consistency requirements into the knowledge base.

[0012] Furthermore, the physical geometry model in the digital twin model described in S1 is configured as follows: Create parametric 3D models of each component of the equipment, including the anode shape, the cathode tungsten wire routing path, the plating tank geometry, and the fluid domain boundary; Provide spatial constraints to limit the computational domain of multiphysics coupling models; Output a 3D visualization data stream to the human-computer interaction interface to support virtual-real synchronous mapping; The multiphysics coupling model described in S1 is established in the following way: The electric field model is used to solve for the potential distribution, and the conductivity σ is a function of temperature and Cu. 2+ A function of concentration; The flow field model is used to solve the velocity vector field of the plating solution, and the velocity distribution affects the mass transfer boundary layer on the electrode surface. The electrochemical deposition model calculates the local current density based on the Butler-Volmer equation and is bidirectionally coupled with the electric field and the flow field.

[0013] Furthermore, in S2, the fusion of multi-source sensor data employs a state estimation algorithm, which is configured to eliminate measurement noise and outliers, and output an estimated coating thickness and its uncertainty quantification index. The uncertainty quantification index is characterized by the trace or the largest eigenvalue of the estimated covariance matrix.

[0014] Furthermore, the objective function of the rolling optimization algorithm described in S4 is configured as follows: Minimize the cumulative squared deviation between the predicted coating thickness and the target thickness, while simultaneously minimizing the cumulative change in the control quantity; The weight matrix for predicting coating thickness is configured to be significantly larger than the weight matrix for control quantity changes, in order to prioritize ensuring product consistency across sixteen lines.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. By using a digital twin engine, the traditional "black box" electrochemical process is transformed into a visualized three-dimensional scene. Operators can simultaneously observe the coating growth status, anode consumption, and bubble distribution of the sixteen tungsten wires on the human-machine interface, making the production process clear at a glance and significantly shortening the anomaly location time. 2. By adopting a multi-line collaborative optimization control strategy, the coating thickness distribution of each line position tends to be consistent, which solves the product quality fluctuation problem caused by inter-line differences in traditional 16-line equipment and significantly enhances the consistency of multi-line parallel production. 3. The system has adaptive adjustment capabilities, which can automatically optimize process parameters when the performance of incoming materials, the state of plating solution or the operating conditions of equipment change, reducing reliance on human experience and lowering the frequency of human intervention and the difficulty of operation. 4. By integrating multi-source sensor data and intelligent algorithms, the system achieves fusion perception of multi-dimensional information such as electrochemistry, fluid dynamics and visual morphology, which significantly enhances the equipment's ability to respond to environmental disturbances. 5. The control method is based on model prediction and rolling optimization, which can realize the early correction of quality deviations during the production process, avoid batch quality anomalies, and improve the overall yield and resource utilization rate. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a schematic diagram of the system modules of the present invention; Figure 2 This is a schematic diagram of the modules included in the real-time digital twin engine of the present invention; Figure 3 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0019] like Figure 1 and Figure 2 As shown, the present invention provides a control system for a tungsten wire copper plating production line based on digital twins, comprising: Multi-source sensing and data acquisition system: The multi-source sensing and data acquisition system is deployed on the physical production line of the 16-line copper plating machine to collect electrochemical parameters, fluid dynamic parameters, motion state parameters and visual morphological parameters of each line position in real time at millisecond-level high frequency.

[0020] Multi-source sensing and data acquisition systems specifically include: An electrochemical sensor array is used to collect data on copper ion concentration, pH value, and temperature distribution in the plating bath. A current-voltage sensor array is used to independently monitor the current and voltage of each circuit in the sixteen-wire circuit. A fluid and motion sensor array is used to collect data on plating solution flow rate, various linear tensions, and linear velocity. The machine vision unit is used to acquire images of key areas of the plating tank at a rate of no less than 25fps to monitor bubble distribution and anode consumption status. The data from each sensor is transmitted to the real-time digital twin engine via an industrial Ethernet in a first cycle, wherein the first cycle is configured to be less than or equal to the sum of the sensor response time and the time constant of the electroplating process kinetics.

[0021] Real-time digital twin engine: The real-time digital twin engine communicates with the multi-source sensing and data acquisition system to construct a virtual twin that includes a physical geometric model, a multi-physics coupling model, a real-time quality prediction model, and a data-driven compensation model. It synchronizes the physical entity state at a speed of seconds, thereby achieving the technical effect of coating thickness prediction error of less than 0.1μm and thickness standard deviation of less than 0.1μm for 16-line products.

[0022] The real-time digital twin engine specifically includes a physical geometry model, a multiphysics coupling model, a real-time quality prediction model, a data-driven compensation model, and a real-time rendering engine.

[0023] Among them, the physical geometry model is configured to create a three-dimensional parametric model of the device, which is used to visualize the geometric state and spatial layout of the physical device.

[0024] A multi-physics coupling model, integrating electric field model, flow field model and electrochemical model, is used to calculate the local current density and mass transfer efficiency at each line position. The electric field model uses the Laplace equation to describe the potential distribution in the electrolyte. ; σ: The conductivity of the electrolyte, a function of temperature and concentration: σ = f(T, [Cu] 2+ ]); φ: electric potential; The flow field model uses the incompressible Navier-Stokes equations to describe the plating solution flow:

[0025] ρ: plating solution density; u: flow velocity vector; p: pressure; μ: dynamic viscosity; F: volume force; The electrochemical model uses the Butler-Volmer equation to describe the electrode reaction kinetics of copper deposition:

[0026] i: Local current density; Exchange current density is related to electrode surface state and ion concentration; α_c: Anode and cathode transfer coefficients; F: Faraday constant; R: Ideal gas constant; T: Absolute temperature; η: Overpotential, η = φ_solid - φ_electrolyte - E_eq, where E_eq is the equilibrium potential.

[0027] The current density calculated by the electric field model is added to the energy equation as a Joule heat source term, affecting the temperature distribution in the flow field model; the flow velocity calculated by the flow field model affects the mass transfer process on the electrode surface, thereby changing the limiting current density and effective ion concentration in the Butler-Volmer equation.

[0028] The real-time quality prediction model includes a feature extraction layer, a machine learning computation layer, and an output layer. The feature extraction layer is configured to receive the current density distribution, flow velocity distribution, and process parameters such as linear velocity, plating bath temperature, and Cu output from the multi-physics coupling model. 2+ The concentration and the coating thickness deviation sequence within the historical data window are used. The machine learning calculation layer uses a gradient boosting tree or neural network algorithm to perform nonlinear mapping on the features. The output layer is configured to output the coating thickness prediction value and deposition rate prediction value for each of the sixteen lines.

[0029] A data-driven compensation model is used to correct systematic errors in multiphysics coupling models online using machine learning algorithms.

[0030] A real-time rendering engine is used to visualize the status of the virtual production line.

[0031] Adaptive optimization controller: The adaptive optimization controller is coupled with a real-time digital twin engine to dynamically adjust the process parameters of each production line based on the model predictive control algorithm, thereby eliminating the deviation between the predicted quality and the target quality. The adaptive optimization controller, based on the prediction results of the digital twin engine, continuously optimizes the process parameters of each production line using a model predictive control algorithm. The objective function for optimization is: J=Σ_{i=1}^{N}||Y_pred(k+i)-Y_ref(k+i)||²_Q+Σ_{i=0}^{N_u-1}||ΔU(k+i)||²_R, The constraints include: U_min≤U(k+i)≤U_max, control quantity constraints, such as upper and lower limits of current; ΔU_min≤ΔU(k+i)≤ΔU_max, constraining the rate of change of the control variable to prevent sudden changes; Y_min≤Y_pred(k+i)≤Y_max, output quantity constraints, such as thickness tolerance; Where Y_pred is the predicted coating thickness vector, Y_ref is the target thickness, U is the control variable vector, ΔU is the rate of change of the control variable, Q and R are weight matrices, N is the prediction time domain, and N_u is the control time domain.

[0032] Multi-line collaborative optimization module: The multi-line collaborative optimization module works in conjunction with the adaptive optimization controller to minimize the variance of the coating quality index of the 16-line product through a distributed constraint optimization method, ensuring multi-line consistency.

[0033] The multi-line collaborative optimization module adopts a distributed model predictive control architecture. It decomposes the global optimization problem into sixteen sub-problems through Lagrange duality decomposition, solving them in parallel. The objective function of the sub-problem on the j-th line contains coupling terms. λ_j is a Lagrange multiplier. The central coordinator iteratively updates the global average control value. The convergence condition is given by λ_j until convergence. ; Subproblem objective function: .

[0034] Human-computer interaction and decision support interface: The human-computer interaction and decision support interface is used to display the virtual-real mapping screen, quality prediction curve, alarm information and system-recommended operation instructions in real time.

[0035] The human-computer interaction and decision support interface is further configured as follows: Displays a 3D virtual production line rendered by a physical geometry model, and overlays process parameters, predicted coating thickness, and early warning status for each line position on the virtual equipment in real time. A tiered warning is triggered when any predicted value deviates from the target thickness by more than a dynamic threshold. Record and store complete process parameter timing data for each of the sixteen lines to support quality traceability and the construction of a process knowledge base.

[0036] The hardware components of this system mainly include the following: Multi-source sensor network: Electrochemical sensors are arranged in each electroplating tank unit (measurement range: Cu). 2+ The system includes a 10-50 g / L sensor (accuracy ±0.5 g / L), a temperature sensor (measurement range: 20-80℃, accuracy ±0.5℃), a current sensor (measurement range: 0-200A, accuracy ±0.5A), and a machine vision camera (resolution 2048×2048, acquisition rate 30fps); sensor data is transmitted to the central controller via industrial Ethernet. Central Processing Unit: It adopts a high-performance industrial computer (CPU: Intel i7-12700H, RAM: 32GB DDR4) paired with edge computing devices (NVIDIA Jetson Xavier) to run digital twin models and optimization algorithms; all hardware meets the IP67 protection level and is suitable for the high temperature and high humidity environment of the electroplating workshop. Actuators include a programmable power supply (0-200A output, accuracy ±0.1A), a metering pump (flow range 0-500mL / min, accuracy ±1%), a thermoelectric cooler (temperature control range 20-80℃, accuracy ±0.5℃), and a servo drive system (linear speed control range 0-30m / min, accuracy ±0.1m / min).

[0037] Workflow: Taking copper plating with a 32μm tungsten wire (target plating thickness 2.5μm) as an example, the system workflow is as follows: 1. Initialization: After the system starts, the digital twin engine loads the initial process parameters (current density: 120A, temperature: 50℃, line speed: 20m / min) and historical production data, and initializes the multiphysics model.

[0038] 2. Data Acquisition and Fusion: Multi-source sensors acquire process data from each production line in real time. The data fusion module generates high-confidence state estimates through a Kalman filter algorithm, including the plating thickness, copper ion concentration, and temperature distribution at each production line.

[0039] 3. Status Prediction and Optimization: The digital twin engine predicts the coating thickness change trend of each line position within the next 30 seconds; the optimization control module calculates the optimal process parameter adjustment amount based on the prediction results; the collaborative optimization module coordinates the resource allocation of each line position and generates control commands.

[0040] 4. Command execution: The actuator adjusts the process parameters of each line position according to the control command (such as reducing the current of line position 3 from 126.2A to 125.0A, and turning on the replenishment pump of line position 8 for 5 seconds to replenish the plating solution).

[0041] 5. Learning and Adaptation: The system compares the actual coating thickness with the predicted value, automatically adjusts the parameters of the digital twin model, and improves the accuracy of the next prediction. Simultaneously, successful process solutions are recorded in the knowledge base for use in subsequent similar production scenarios.

[0042] To verify the effectiveness of the system of this invention, a comparative experiment was designed with traditional fixed parameter control systems and simple PID feedback control systems: Experiment 1: Comparison of Multi-line Consistency Control Effects Objective: To verify the ability of the collaborative optimization algorithm to maintain product consistency across 16 lines while overcoming common disturbances (such as main tank concentration fluctuations).

[0043] Method: During an 8-hour production process, a simulated Cu plating solution was used. 2+ The concentration decreased slowly by 5%. After completion, the coating thickness of all tungsten wire products was measured.

[0044]

[0045] Conclusion: The system of this invention improves product consistency (standard deviation) by more than 62%, and the process capability index CpK far exceeds the industry benchmark of 1.33, proving that it has six sigma-level quality control capabilities.

[0046] Experiment 2: Verification of Quality Prediction Accuracy and Adaptive Capability Objective: To verify the accuracy of the digital twin quality prediction model and the system's adaptive capability in response to fluctuations in incoming materials.

[0047] Method: Different batches of tungsten wire were used (with varying surface activities), and the system operated automatically without manual intervention. Offline measurement (SEM) was used to verify the difference between the predicted and actual thickness.

[0048] Results: Comparison Table of Quality Prediction Errors

[0049] Conclusion: The prediction model of this invention improves accuracy by approximately 70%, providing a reliable basis for optimized control. When dealing with incoming material fluctuations, traditional systems require several hours of manual adjustment, while the system of this invention automatically re-stabilizes all 16-line thicknesses within the target tolerance zone within 30 minutes.

[0050] Experiment 3: Fault Diagnosis and Overall Energy Consumption Performance Objective: To verify the system's diagnostic capabilities under abnormal conditions and its long-term energy efficiency.

[0051] Method: Simulate a localized passivation fault in the anode and record the system response. Analyze the overall energy consumption and yield rate over one month of production.

[0052]

[0053] Conclusion: Through precise real-time optimization and forward-looking control, this invention significantly improves quality and reliability while achieving considerable energy savings and virtually eliminating downtime caused by quality incidents.

[0054] like Figure 3 As shown, a control method for a tungsten wire copper plating production line based on digital twins includes the following steps: S1: Construct a digital twin model that includes a physical geometry model, a multiphysics coupling model, and a real-time quality prediction model. The multiphysics coupling model includes at least an electric field, a flow field, and an electrochemical deposition field. The model parameters are continuously corrected using historical production data and real-time sensor data. S2: Using the first cycle to fuse the multi-source sensor data of sixteen lines, generate a high-confidence state estimate for each line, the state estimate including the coating thickness distribution, copper ion concentration field and temperature gradient distribution; S3: Based on the real-time quality prediction model, predict the coating thickness change trend of each line position in the second cycle. The real-time quality prediction model takes the output of the multi-physics coupling model, the current process parameters and the historical production data window as input, and generates the predicted values ​​of coating thickness and deposition rate through machine learning algorithms. S4: The rolling optimization algorithm is used to generate the process parameter adjustment amount for each line position, and the control command is generated after the adjustment amount is distributed and coordinated with the goal of minimizing the consistency deviation of the sixteen-line coating thickness. S5: Execute the control command and collect the sensor data for the next first cycle to form a closed-loop feedback; S6: Compare the actual coating thickness measurement with the predicted value, update the digital twin model parameters online, and store the process scheme that meets the consistency requirements into the knowledge base.

[0055] The physical geometry model in the digital twin model described in S1 is configured as follows: Create parametric 3D models of each component of the equipment, including the anode shape, the cathode tungsten wire routing path, the plating tank geometry, and the fluid domain boundary; Provide spatial constraints to limit the computational domain of multiphysics coupling models; Output a 3D visualization data stream to the human-computer interaction interface to support virtual-real synchronous mapping.

[0056] The multiphysics coupling model described in S1 is established in the following way: The electric field model is used to solve for the potential distribution, and the conductivity σ is a function of temperature and Cu. 2+ A function of concentration; The flow field model is used to solve the velocity vector field of the plating solution, and the velocity distribution affects the mass transfer boundary layer on the electrode surface. The electrochemical deposition model calculates the local current density based on the Butler-Volmer equation and is bidirectionally coupled with the electric field and the flow field.

[0057] The fusion of multi-source sensor data described in S2 employs a state estimation algorithm. This state estimation algorithm is configured to eliminate measurement noise and outliers, and output an estimated coating thickness and its uncertainty quantification index. The uncertainty quantification index is characterized by estimating the trace or the maximum eigenvalue of the covariance matrix.

[0058] The objective function of the rolling optimization algorithm described in S4 is configured as follows: Minimize the cumulative squared deviation between the predicted coating thickness and the target thickness, while simultaneously minimizing the cumulative change in the control quantity; The weight matrix for predicting coating thickness is configured to be significantly larger than the weight matrix for control quantity changes, in order to prioritize ensuring product consistency across sixteen lines.

[0059] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A control system for a tungsten wire copper plating production line based on digital twins, characterized in that, include: A multi-source sensing and data acquisition system is deployed on the physical production line of a 16-line copper plating machine to collect electrochemical parameters, fluid dynamic parameters, motion state parameters and visual morphological parameters of each line position in real time at a high frequency of milliseconds. A real-time digital twin engine is communicatively connected to a multi-source sensing and data acquisition system to construct a virtual twin that includes a physical geometric model, a multi-physics coupling model, a real-time quality prediction model, and a data-driven compensation model. The engine synchronizes the physical entity state at a speed of seconds, thereby achieving the technical effect of coating thickness prediction error of less than 0.1μm and thickness standard deviation of 16-line products of less than 0.1μm. An adaptive optimization controller, coupled with a real-time digital twin engine, is used to dynamically adjust the process parameters of each production line based on a model predictive control algorithm to eliminate the deviation between predicted quality and target quality. A multi-line collaborative optimization module, which works in conjunction with an adaptive optimization controller, is used to minimize the variance of the coating quality index of the sixteen-line product through a distributed constraint optimization method, thereby ensuring multi-line consistency. The human-computer interaction and decision support interface is used to display the virtual-real mapping screen, quality prediction curve, alarm information and system recommended operation instructions in real time.

2. The control system for a tungsten wire copper plating production line based on digital twins according to claim 1, characterized in that, The multi-source sensing and data acquisition system includes: An electrochemical sensor array is used to collect data on copper ion concentration, pH value, and temperature distribution in the plating bath. A current-voltage sensor array is used to independently monitor the current and voltage of each circuit in the sixteen-wire circuit. A fluid and motion sensor array is used to collect data on plating solution flow rate, various linear tensions, and linear velocity. The machine vision unit is used to acquire images of key areas of the plating tank at a rate of no less than 25fps to monitor bubble distribution and anode consumption status. The data from each sensor is transmitted to the real-time digital twin engine via an industrial Ethernet in a first cycle, wherein the first cycle is configured to be less than or equal to the sum of the sensor response time and the time constant of the electroplating process kinetics.

3. The control system for a tungsten wire copper plating production line based on digital twins according to claim 1, characterized in that: The real-time digital twin engine includes: Physical geometry model, configured to create a 3D parametric model of the device, used to visualize the geometric state and spatial layout of the physical device; A multiphysics coupling model, integrating electric field, flow field, and electrochemical models, is used to calculate the local current density and mass transfer efficiency at each line position. The electric field model employs the Laplace equation. To describe the potential distribution, the flow field model uses the incompressible Navier-Stokes equation to describe the plating solution flow, and the electrochemical model uses the Butler-Volmer equation to describe the copper deposition kinetics. Among them, the current density calculated by the electric field model is added to the energy equation as a Joule heat source term, which affects the temperature distribution in the flow field model. The flow velocity calculated by the flow field model affects the mass transfer process on the electrode surface, thereby changing the limiting current density and effective ion concentration in the Butler-Volmer equation. The real-time quality prediction model includes a feature extraction layer, a machine learning computation layer, and an output layer. The feature extraction layer is configured to receive the current density distribution, flow velocity distribution, and process parameters such as linear velocity, plating bath temperature, and Cu output from the multi-physics coupling model. 2+ The concentration and the coating thickness deviation sequence within the historical data window are used. The machine learning calculation layer uses a gradient boosting tree or neural network algorithm to perform non-linear mapping on the features. The output layer is configured to output the coating thickness prediction value and deposition rate prediction value for each of the sixteen line positions. A data-driven compensation model is used to correct systematic errors in multiphysics coupling models online using machine learning algorithms. A real-time rendering engine is used to visualize the status of the virtual production line.

4. The control system for a tungsten wire copper plating production line based on digital twins according to claim 1, characterized in that: The adaptive optimization controller, based on the prediction results of the digital twin engine, continuously optimizes the process parameters of each production line using a model predictive control algorithm, wherein the objective function is: J=Σ_{i=1}^{N}||Y_pred(k+i)-Y_ref(k+i)||²_Q+Σ_{i=0}^{N_u-1}||ΔU(k+i)||²_R, The constraints include: U_min≤U(k+i)≤U_max, ΔU_min≤ΔU(k+i)≤ΔU_max, Y_min≤Y_pred(k+i)≤Y_max, Where Y_pred is the predicted coating thickness vector, Y_ref is the target thickness, U is the control variable vector, ΔU is the rate of change of the control variable, Q and R are weight matrices, N is the prediction time domain, and N_u is the control time domain.

5. The control system for a tungsten wire copper plating production line based on digital twins according to claim 1, characterized in that: The multi-line collaborative optimization module adopts a distributed model predictive control architecture. It decomposes the global optimization problem into sixteen sub-problems through Lagrange dual decomposition, solving them in parallel. The objective function of the sub-problem on the j-th line contains coupling terms. λ_j is a Lagrange multiplier. The central coordinator iteratively updates the global average control value. The convergence condition is given by λ_j until convergence. .

6. The control system for a tungsten wire copper plating production line based on digital twins according to claim 1, characterized in that: The human-computer interaction and decision support interface is further configured as follows: Displays a 3D virtual production line rendered by a physical geometry model, and overlays process parameters, predicted coating thickness, and early warning status for each line position on the virtual equipment in real time. A tiered warning is triggered when any predicted value deviates from the target thickness by more than a dynamic threshold. Record and store complete process parameter timing data for each of the sixteen lines to support quality traceability and the construction of a process knowledge base.

7. A control method for a tungsten wire copper plating production line based on digital twins, applied to the system described in claim 1, characterized in that, The method includes the following steps: S1: Construct a digital twin model that includes a physical geometry model, a multiphysics coupling model, and a real-time quality prediction model. The multiphysics coupling model includes at least an electric field, a flow field, and an electrochemical deposition field. The model parameters are continuously corrected using historical production data and real-time sensor data. S2: Using the first cycle to fuse the multi-source sensor data of sixteen lines, generate a high-confidence state estimate for each line, the state estimate including the coating thickness distribution, copper ion concentration field and temperature gradient distribution; S3: Based on the real-time quality prediction model, predict the coating thickness change trend of each line position in the second future cycle. The real-time quality prediction model takes the output of the multi-physics coupling model, the current process parameters and the historical production data window as input, and generates the predicted values ​​of coating thickness and deposition rate through machine learning algorithms. S4: The rolling optimization algorithm is used to generate the process parameter adjustment amount for each line position, and the control command is generated after the adjustment amount is distributed and coordinated with the goal of minimizing the consistency deviation of the sixteen-line coating thickness. S5: Execute the control command and collect the sensor data for the next first cycle to form a closed-loop feedback; S6: Compare the actual coating thickness measurement with the predicted value, update the digital twin model parameters online, and store the process scheme that meets the consistency requirements into the knowledge base.

8. The control method for a tungsten wire copper plating production line based on digital twins according to claim 7, characterized in that: The physical geometry model in the digital twin model described in S1 is configured as follows: Create parametric 3D models of each component of the equipment, including the anode shape, the cathode tungsten wire routing path, the plating tank geometry, and the fluid domain boundary; Provide spatial constraints to limit the computational domain of multiphysics coupling models; Output a 3D visualization data stream to the human-computer interaction interface to support virtual-real synchronous mapping; The multiphysics coupling model described in S1 is established in the following way: The electric field model is used to solve for the potential distribution, and the conductivity σ is a function of temperature and Cu. 2+ A function of concentration; The flow field model is used to solve the velocity vector field of the plating solution, and the velocity distribution affects the mass transfer boundary layer on the electrode surface. The electrochemical deposition model calculates the local current density based on the Butler-Volmer equation and is bidirectionally coupled with the electric field and the flow field.

9. The control method for a tungsten wire copper plating production line based on digital twins according to claim 7, characterized in that: The state estimation algorithm used in S2 to fuse multi-source sensor data is configured to eliminate measurement noise and outliers, and output an estimated coating thickness and its uncertainty quantification index. The uncertainty quantification index is characterized by the trace or the largest eigenvalue of the estimated covariance matrix.

10. The control method for a tungsten wire copper plating production line based on digital twin according to claim 7, characterized in that: The objective function of the rolling optimization algorithm described in S4 is configured as follows: Minimize the cumulative squared deviation between the predicted coating thickness and the target thickness, while simultaneously minimizing the cumulative change in the control quantity; The weight matrix for predicting coating thickness is configured to be significantly larger than the weight matrix for control quantity changes, in order to prioritize ensuring product consistency across sixteen lines.

Images