Differential precise current distribution control method in multi-workpiece batch electroplating process

Abstract

The invention discloses a differentiated precise current distribution control method in a multi-workpiece batch electroplating process. The method comprises the following steps: acquiring multi-workpiece feature data and preprocessing the multi-workpiece feature data; workpiece difference characteristic analysis and grouping are carried out; modeling and analyzing the electric field distribution of the electrolytic cell; plating layer growth dynamic monitoring and prediction are carried out; and generating a differential current distribution strategy. A multi-frequency differential resistance spectrum acquisition system is adopted to realize real-time monitoring of plating layer growth, an improved independent component analysis algorithm fused with spatial constraint is applied to perform workpiece group resistance spectrum decoupling, and micro-area accurate current modulation is realized through a self-adaptive waveform library of an electric field distortion area. The problems that dynamic characteristics of plating layers are difficult to monitor and differentiation control of workpieces with similar characteristics is insufficient in multi-workpiece electroplating are solved, and the quality consistency of the plating layers is improved.

Description

Technical Field

[0001] The present invention belongs to the field of electroplating processes and their control, and in particular, to a differential precise current distribution control method during the batch electroplating of multiple workpieces. Background Art

[0002] As an important branch of surface treatment technology, the electroplating process has a wide range of applications in the fields of electronics, aviation, automotive, etc. In a batch production environment, it is an inevitable choice to process multiple workpieces with different shapes, sizes or materials simultaneously to improve production efficiency, but this also poses higher requirements for the precise control of the electroplating process. Achieving differential precise current distribution control during the batch electroplating of multiple workpieces can not only improve the consistency of coating quality, but also reduce material consumption, energy loss, and improve production efficiency, which has important economic and technical value.

[0003] Current multi-workpiece electroplating control technologies are mainly divided into three categories: the batch electroplating method based on workpiece pre-classification, the electric field regulation method based on auxiliary shielding devices, and the current control method based on pulse parameter adjustment. The method based on workpiece pre-classification realizes differential control between batches by pre-measuring the geometric features of workpieces and dividing similar workpieces into the same batch for electroplating; the method based on auxiliary shielding devices realizes the protection of high current density regions by arranging auxiliary shielding objects around the workpieces to adjust the local electric field distribution; the method based on pulse parameter adjustment controls the current distribution of different types of workpieces in the time dimension by changing parameters such as pulse frequency and duty cycle. In addition, some researchers have tried to identify the shape of workpieces through simple computer vision technology and set basic electroplating parameters in combination with empirical models, but lack the precise perception and control ability of the microscopic differences between and within workpieces.

[0004] However, the existing technology has obvious deficiencies in the monitoring of the dynamic characteristics of coating growth and the differentiated control of groups of workpieces with similar characteristics. First, in terms of the dynamic monitoring of coating growth, the existing methods mainly rely on the initial workpiece characteristics for open-loop control or simple timing feedback adjustment, and lack real-time monitoring and accurate modeling of changes in surface geometric characteristics during coating growth. Especially in a multi-workpiece co-slot environment, the electric fields generated by each workpiece interfere with each other, resulting in complex changes in the electric field distribution, and the existing resistance spectrum monitoring technology cannot effectively distinguish such interference, and thus cannot accurately capture the dynamic characteristics of the coating growth of each workpiece. Secondly, in terms of the differentiated control of workpiece groups, the existing technology mostly adopts a rough grouping method based on specification classification, and lacks the ability to identify and respond to slight differences within the same specification workpiece group. When there are workpieces with similar shapes but different detailed structures in a batch of workpieces, these local differences will cause subtle changes in the electric field distribution, and traditional control methods are difficult to achieve accurate electroplating parameter adjustments for these slight differences, resulting in fluctuations in the coating quality of the same batch of workpieces. In addition, in terms of micro-area current control of complex-shaped workpieces, the existing pulse modulation method lacks consideration of the interaction between electric fields in each micro-area, and cannot solve the problem of mutual interference between micro-area current parameters, which limits the further improvement of electroplating quality. Summary of the invention

[0005] The purpose of the invention is to provide a differentiated and precise current distribution control method in the process of batch electroplating of multiple workpieces, in order to solve at least one technical problem existing in the prior art.

[0006] Technical solution, a differentiated and precise current distribution control method for batch electroplating of multiple workpieces, including: Acquire feature data of multiple workpieces and perform preprocessing to obtain a workpiece data set; Based on the artifact data set, analyze the artifact difference characteristics and group them, obtain the optimized artifact grouping and calculate the intra-group difference matrix; Based on the optimized workpiece grouping and workpiece data set, the electric field distribution of the electrolytic cell is modeled and analyzed to obtain the actual current density distribution in the micro area and the workpiece electric field interference matrix; Based on the workpiece electric field interference matrix and the intra-group difference matrix, the coating growth dynamic monitoring and prediction are carried out to obtain the group coating growth characteristics; The group coating growth characteristics are read, and the differentiated current distribution control strategy is generated by combining the actual current density distribution in the micro-area and optimizing the workpiece grouping.

[0007] Beneficial effect: The present invention solves the problems of difficulty in monitoring dynamic characteristics of coatings in electroplating of multiple workpieces and insufficient differentiated control of workpieces with similar characteristics, realizes accurate identification and control of minute differences within a group of workpieces with similar characteristics, and simultaneously realizes precise current regulation of complex micro-areas on the surface of workpieces, thereby improving the uniformity and quality consistency of the coating. Brief Description of the Drawings

[0008] Figure 1 It is a flowchart of the steps of a differential precise current distribution control method in the multi-workpiece batch electroplating process provided by an embodiment of the present application.

[0009] Figure 2 It is a flowchart of the steps of analyzing the differential characteristics of workpieces and grouping them provided by an embodiment of the present application.

[0010] Figure 3 It is a flowchart of the steps of constructing a multi-level workpiece similarity metric provided by an embodiment of the present application.

[0011] Figure 4 It is a flowchart of the steps of grouping workpieces and selecting representative workpieces provided by an embodiment of the present application. Detailed Description of the Embodiment

[0012] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0013] It should be specifically noted that, for clearly showing the step flow of the present application, serial numbers are marked for each step in the specification. These serial numbers are only for the convenience of description and do not limit the execution order of the steps. In actual operation, according to the technical requirements of the specific implementation scenario, each step can be executed in a different order from that shown in the specification, and in some cases, parallel processing between steps can also be achieved.

[0014] As Figure 1 shown, the differential precise current distribution control method in the multi-workpiece batch electroplating process includes the following steps: S1. Obtain the characteristic data of multiple workpieces and perform preprocessing to obtain a workpiece data set; Specifically, the characteristic data of multiple workpieces includes the three-dimensional point cloud data of the workpieces, the surface conductivity distribution data of each part of the workpieces, and the material composition data of the workpiece surfaces. Among them, the three-dimensional point cloud data describes the detailed information of the surface shape and spatial structure of the workpieces; the surface conductivity distribution data reflects the current conduction ability of different parts of the workpiece surfaces; the material composition data records the chemical components of the materials on the workpiece surfaces.

[0015] S2. Based on the workpiece data set, analyze the differential characteristics of the workpieces and group them to obtain an optimized workpiece grouping and calculate the within-group differential matrix; Specifically, using the existing dataset, extract the unique attributes of each workpiece, which may affect the electroplating effect. For example, one workpiece has a rougher surface while another is smoother, and the current distribution during electroplating will be different. According to the analysis results, divide the workpieces into different groups. The characteristics of the workpieces in each group should be as similar as possible to ensure that the electroplating strategy for each group is more accurate. For example, divide the workpieces with smooth surfaces into one group and those with rough surfaces into another group. For each group, further analyze the subtle differences among the workpieces within the group and describe these differences using a mathematical matrix, which helps to further fine-tune the electroplating parameters and minimize the electroplating effect differences among the workpieces within the group.

[0016] S3. Based on the optimized workpiece grouping and workpiece dataset, conduct modeling and analysis of the electric field distribution in the electrolytic cell to obtain the actual current density distribution in the micro-region and the workpiece electric field interference matrix; Specifically, establish a virtual electrolytic cell model that can simulate how current flows in the electrolytic cell and how this current is distributed to the surfaces of different workpieces. Through modeling, calculate the actual current density borne by each small region (micro-region) on the workpiece surface. This is like analyzing how much current each piece of the workpiece surface "absorbs", reflecting the accuracy of the current distribution during the electroplating process. The electric fields between workpieces will affect each other. For example, one workpiece may "block" part of the current path of another workpiece. The electric field interference matrix is used to describe this complex relationship of electric field interaction between workpieces.

[0017] S4. Based on the workpiece electric field interference matrix and the within-group difference matrix, conduct dynamic monitoring and prediction of the coating growth to obtain the coating growth characteristics of the group; Specifically, use real-time data monitoring technology to observe the actual growth of the coating in each region on the workpiece surface. For example, the coating may grow too fast or too slow in some regions, and this dynamic information can help to identify problems. Combine the monitoring data and the previous analysis results to predict the coating growth pattern and overall characteristics of each group of workpieces, such as which region of the coating is most likely to be completed early or requires more current.

[0018] S5. Read the coating growth characteristics of the group, and combine the actual current density distribution in the micro-region and the optimized workpiece grouping to generate a differentiated current distribution control strategy.

[0019] Specifically, construct a differentiated current distribution plan to ensure that the current distribution during the electroplating process can meet the specific requirements of different workpieces and regions.

[0020] In this embodiment, by accurately analyzing the differential characteristics of each workpiece and the characteristics of the electric field distribution, a differential current distribution strategy is generated, making the electroplated layer more uniform, reducing defects, and improving the quality and performance of the coating. The intelligent grouping of workpieces and the current distribution control strategy can reduce the electroplating time, lower energy consumption, and improve production efficiency. Due to the use of dynamic monitoring and prediction technologies, the current distribution strategy can be adjusted in real time to adapt to the complex shapes and materials of different workpieces, realizing personalized electroplating. By reducing material waste during the electroplating process, lowering power consumption, and improving the efficiency of batch production, the production cost can be effectively saved. Based on comprehensive data analysis and modeling, scientific guidance can be provided for the electroplating process, reducing the dependence on traditional experimental and empirical methods.

[0021] According to one aspect of the present application, the steps of performing preprocessing include: S11. Collect multi-modal workpiece feature data: Use a three-dimensional laser scanner to perform high-precision scanning on the geometric features of the workpiece to obtain the three-dimensional point cloud data of the workpiece; Use an electrical conductivity sensor array to measure the surface electrical conductivity distribution data of each part of the workpiece; Detect the material composition data on the surface of the workpiece through an X-ray fluorescence analyzer.

[0022] S12. Extract the microscopic surface topography features of the workpiece: Apply a local curvature analysis algorithm to the three-dimensional point cloud data to calculate the curvature distribution features of the workpiece surface; Perform multi-scale analysis on the workpiece surface based on wavelet transform to extract the surface roughness features of the workpiece; Use the region growing method to identify the microscopic concave and convex features on the workpiece surface, including sharp corners, holes, and grooves, etc.

[0023] S13. Map electroplating key parameters and feature fusion: Establish a mapping relationship between the material composition data and the electrochemical activity to generate an electrochemical activity distribution map of the workpiece; Combine the surface electrical conductivity distribution data and the curvature distribution features to construct a current absorption tendency index of the workpiece; Fuse the surface roughness features and the microscopic concave and convex features to generate an electric field perturbation sensitivity map of the workpiece.

[0024] S14. Standardize feature data and detect anomalies: Perform unified dimension processing on all feature data and convert it into a standardized workpiece feature vector; Apply the Local Outlier Factor (LOF) algorithm to identify feature anomaly regions; Perform adaptive sampling enhancement on the feature anomaly regions to improve the data accuracy and reliability of the anomaly regions, and generate a workpiece data set containing normal and anomaly region marks.

[0025] As Figure 2 shown, according to one aspect of the present application, the steps of analyzing the differential characteristics of workpieces and grouping them include: S21. Read the workpiece dataset, perform dimensionality reduction and extract key features to obtain the dimensionality-reduced feature representation and feature importance weights; based on the dimensionality-reduced feature representation and feature importance weights, construct a multi-level workpiece similarity metric to obtain the comprehensive similarity. S22. Based on the comprehensive similarity, group the workpieces and select representative workpieces to obtain the optimized workpiece grouping and representative workpieces. S23. Based on the optimized workpiece grouping and representatives, perform intra-group difference quantification and analysis to obtain the intra-group difference matrix and the principal components of the intra-group differences.

[0026] As Figure 3 shown, according to one aspect of the present application, the steps of constructing a multi-level workpiece similarity metric to obtain the comprehensive similarity include: Extract data from the dimensionality-reduced feature representation and feature importance weights, and use a multi-level similarity metric function for the electroplating process to calculate the macroscopic shape similarity, microscopic structure similarity, and material property similarity. Based on the macroscopic shape similarity, microscopic structure similarity, and material property similarity, calculate the similarity scores between workpieces to obtain the workpiece similarity matrix. According to the distribution of each level of features in the workpiece similarity matrix, combined with the electroplating perturbation sensitivity weighting mechanism, dynamically adjust the level weight coefficients. Combine the workpiece similarity matrix, level weight coefficients, and feature importance weights to calculate the comprehensive similarity between workpieces.

[0027] As Figure 4 shown, according to one aspect of the present application, the steps of grouping the workpieces and selecting representative workpieces to obtain the optimized workpiece grouping and representative workpieces include: Based on the comprehensive similarity, preliminarily group the workpieces through a spectral clustering algorithm to obtain the initial workpiece grouping. Adopt a grouping evaluation index based on electroplating process sensitivity to evaluate the rationality of the initial workpiece grouping to obtain the grouping evaluation result. Adopt an adaptive grouping adjustment algorithm to optimize the grouping boundary according to the grouping evaluation result to obtain the optimized workpiece grouping. For each optimized workpiece grouping, perform centrality analysis and use the workpiece with the highest centrality as the representative workpiece.

[0028] Specifically, perform non-linear feature dimensionality reduction and key feature extraction. Apply the t-SNE (t-Distributed Stochastic Neighbor Embedding) algorithm to the workpiece dataset to project the high-dimensional features into a low-dimensional space to obtain the dimensionality-reduced feature representation; use an improved adaptive feature importance evaluation algorithm to identify the key electroplating features affecting the electroplating quality from the dimensionality-reduced feature representation; based on feature sensitivity analysis, calculate the influence degree of each feature on the electroplating result to generate the feature importance weights.

[0029] Construct a multi-level workpiece similarity metric. Design a multi-level similarity metric function for the electroplating process, including macro shape similarity, microstructural similarity, and material property similarity; calculate the similarity scores of each pair of workpieces at each level to generate a workpiece similarity matrix; apply the information entropy weight method to dynamically adjust the level weight coefficients according to the distribution of each level's features; combine the feature importance weights and level weight coefficients to calculate the comprehensive similarity between workpieces.

[0030] Select adaptive workpiece grouping and representative workpieces. Based on the comprehensive similarity, apply the spectral clustering algorithm to preliminarily group the workpieces to obtain the initial workpiece grouping; design a grouping evaluation index based on the sensitivity of the electroplating process to evaluate the rationality of the initial workpiece grouping; adopt an adaptive grouping adjustment algorithm to optimize the grouping boundary according to the evaluation results to generate the optimized workpiece grouping; for each grouping, select representative workpieces that can represent the characteristics of the group based on centrality analysis.

[0031] Quantify and analyze the intra-group differences. For the workpieces within each optimized workpiece grouping, calculate the feature deviations between them and the representative workpieces to generate an intra-group difference matrix; apply principal component analysis (PCA) to identify the main patterns of intra-group differences and extract the intra-group difference principal components; based on the intra-group difference principal components, construct a difference distribution model for the workpieces within the group to provide a basis for subsequent differential current distribution.

[0032] In another embodiment of the present application, obtaining the level weight coefficients further includes: applying an improved information entropy weight method and introducing an electroplating perturbation sensitivity weighting mechanism, where regions with high curvature and high current density fluctuations obtain enhanced weights, and the level weight coefficients are dynamically adjusted.

[0033] Specifically, in an actual electroplating production line, when applying the improved information entropy weight method to 50 connector workpieces with different shapes, an electroplating perturbation sensitivity weighting mechanism is introduced. The specific implementation process is as follows: First, calculate the similarity between workpieces i and j at three levels of macro shape, microstructural, and material properties: Macro shape similarity: SS(i, j) = exp(-‖S_i - S_j‖ 2 / 0.2 2 ); Microstructural similarity: MS(i, j) = exp(-‖M_i - M_j‖ 2 / 0.15 2 ); Material property similarity: CS(i, j) = exp(-‖C_i - C_j‖ 2 / 0.25 2)。Then construct the workpiece similarity matrix SM, with a matrix size of 50×50, where each element represents the similarity score between two workpieces at each level. Next, apply the improved information entropy weight method, introducing a weighted mechanism for electroplating perturbation sensitivity. The traditional information entropy weight calculation formula is: w_k = (1 - E_k) / ∑(1 - E_j); where E_k and E_j are different information entropies; the improved calculation formula is: w_k = (1 - E_k) × D_k / ∑[(1 - E_j)× D_j], where D_k and D_j are electroplating perturbation sensitivity factors. For high-curvature regions (average curvature > 0.5mm -1 ) and high-current-density fluctuation regions (fluctuation > 15%), set enhanced weights: D_k = 1.0 + 0.5×(H_curve + H_current), where H_curve and H_current are binary indicators of curvature and current density fluctuation (present = 1, absent = 0). Through this improvement, when processing "U"-shaped small connectors, the weight at the microstructure level increases from 0.25 to 0.42, making the grouping focus more on the key factors of coating quality, and ultimately increasing the uniformity index of workpiece grouping by 18%.

[0034] According to one aspect of the present application, the steps of modeling and analyzing the electric field distribution in the electrolytic cell to obtain the actual current density distribution in the micro-region and the workpiece electric field interference matrix include: S31. Obtain the geometric structure of the electrolytic cell and the electrode arrangement, combine the optimized workpiece grouping and the workpiece dataset, model the reference electric field of the multi-workpiece electrolytic cell, and obtain the complete model of the multi-workpiece electrolytic cell and the ideal reference electric field distribution; S32. Based on the complete model of the multi-workpiece electrolytic cell and the ideal reference electric field distribution, perform electric field interference mapping and quantification between workpieces to obtain the workpiece electric field interference matrix and the mutual interference intensity coefficient; S33. Based on the workpiece electric field interference matrix and the mutual interference intensity coefficient, construct a dynamic electric field distribution prediction model; S34. Combine the workpiece feature vectors in the workpiece dataset, perform micro-region electric field characteristic analysis and current density mapping to obtain the actual current density distribution in the micro-region.

[0035] According to one aspect of the present application, the steps of performing electric field interference mapping and quantification between workpieces to obtain the workpiece electric field interference matrix and the mutual interference intensity coefficient include: Based on the complete model of the multi-workpiece electrolytic cell, simulate the electric field distribution when a single workpiece exists, and generate a series of single-workpiece electric field perturbation maps; Calculate the difference between the single-workpiece electric field perturbation map and the ideal reference electric field distribution, and construct the workpiece electric field interference matrix; Combine the preset electric field interference intensity quantification index and the workpiece electric field interference matrix, and calculate the mutual interference intensity coefficient between workpieces; Based on the mutual interference intensity coefficient, identify the main propagation paths of electric field interference and construct an interference propagation network model.

[0036] Specifically, conduct reference electric field modeling for a multi-workpiece electrolytic cell. Based on the geometric structure and electrode arrangement of the electrolytic cell, construct a basic geometric model of the electrolytic cell; combine the conductivity distribution and temperature distribution of the electrolyte to establish an electrolyte conductivity field; integrate the geometric feature data of the workpiece, place the workpiece in the basic geometric model of the electrolytic cell to generate a complete model of the multi-workpiece electrolytic cell; apply the finite element analysis method to calculate the ideal reference electric field distribution without workpiece interference.

[0037] Conduct electric field interference mapping and quantification between workpieces. Based on the complete model of the multi-workpiece electrolytic cell, simulate the electric field distribution when a single workpiece exists to generate a series of single-workpiece electric field perturbation maps; construct a workpiece electric field interference matrix by superimposing the differences between each single-workpiece electric field perturbation map and the ideal reference electric field distribution; design a quantization index for electric field interference intensity, calculate the mutual interference intensity coefficient between workpieces; identify the main propagation paths of electric field interference and construct an interference propagation network model.

[0038] Construct a dynamic electric field distribution prediction model. Based on the workpiece electric field interference matrix and the mutual interference intensity coefficient, construct a static electric field distribution model considering the mutual interaction of multiple workpieces; integrate the principles of fluid dynamics to simulate the influence of the flow of electroplating solution on the electric field distribution and generate a flow field influence factor; combine the static electric field distribution model and the flow field influence factor to construct a dynamic electric field distribution prediction model; apply the Monte Carlo method to simulate the influence of small changes in workpiece position on the electric field distribution and generate an uncertainty interval of the electric field distribution.

[0039] Conduct micro-region electric field characteristic analysis and current density mapping. Based on the dynamic electric field distribution prediction model, calculate the initial electric field intensity at each point on the workpiece surface; combine the microscopic topography data in the workpiece feature vector to identify the electric field hot spots with abnormal electric field distribution; apply the improved Butler-Volmer equation to establish the mapping relationship between the micro-region electric field intensity and the theoretical current density; consider the cathode polarization effect to correct the theoretical current density and obtain a more accurate micro-region actual current density distribution.

[0040] In another embodiment of the present application, use a modified Laplace equation considering the non-linear effect of ion migration to calculate the difference between the single-workpiece electric field perturbation map and the ideal reference electric field distribution, construct a workpiece electric field interference matrix; design a quantization index for electric field interference intensity, introduce a spatial gradient weighting factor, and calculate the mutual interference intensity coefficient between workpieces based on the workpiece electric field interference matrix.

[0041] Specifically, during the construction of the electroplating bath model, the modified Laplace equation considering the non-linear effect of ion migration is applied to map the electric field interference between workpieces. The specific implementation process is as follows: The traditional Laplace equation is: ▽·(σ▽φ)= 0; the modified equation considers the non-linear effect of ion migration: ▽·(σ▽φ) + ▽·(σμ|▽φ|▽φ) =0, where μ is the ion migration coefficient, and the non-linear effect is significant in the high current density region (>4A / dm 2 )). Where σ is the conductivity of the workpiece material; ▽ is the vector differential operator; φ is the electric potential.

[0042] During the implementation process, the electroplating bath is divided into finite element meshes. For the meshes close to the workpiece surface, fine meshes of 0.5mm are used, and meshes of 2-5mm are used in other regions. For the "L"-shaped connector with sharp corners, after separately simulating its electric field distribution, the difference from the ideal reference electric field is calculated, and the workpiece electric field interference matrix FEDM is constructed: FEDM(i, x, y, z) = SEDF(i, x, y, z) - REF(x, y, z); where (x, y, z) are the spatial coordinate points, SEDF represents the electric field interference distribution function, and REF is the ideal reference electric field function; when quantifying the electric field interference intensity, the spatial gradient weighting factor G(x, y, z) is introduced: G(x, y, z) =1 + λ×|▽φ(x, y, z)| 2 , λ is set to 0.2 to increase the interference weight in the high gradient region. Calculate the mutual interference intensity coefficient between workpieces i and j: MIFC(i, j) = ∫∫∫ |FEDM(i, x, y, z) × FEDM(j, x, y, z)| × G(x, y, z) dxdydz. For two adjacent "U"-shaped connectors, the interference coefficient calculated by the traditional method is 0.167, and the improved method obtains 0.231, which more accurately reflects the strong interference in the high current density region, and thus more effective compensation is carried out in the subsequent current distribution, resulting in a 15% improvement in the final electroplating thickness uniformity.

[0043] According to one aspect of the present application, the steps of performing dynamic monitoring and prediction of coating growth and obtaining the coating growth characteristics of the group include: S41. Obtain differential resistance spectrum data through a multi-frequency differential resistance spectrum acquisition system; S42. Based on the differential resistance spectrum data, obtain the coating growth state data through the mapping relationship between the resistance spectrum characteristics and the coating growth state; S43. Combine the workpiece electric field interference matrix, the differential resistance spectrum data, and the intra-group difference matrix to perform decoupling and interference elimination of the workpiece group resistance spectrum, and obtain the coating growth characteristics of the group; S44. Based on the coating growth state data and the group coating growth characteristics, perform simulations using the coating growth dynamic prediction model to obtain a prediction uncertainty assessment.

[0044] Specifically, map the resistance spectrum characteristics to the coating growth state. Extract the frequency-domain characteristics of the differential resistance spectrum data to identify the key frequency characteristics representing the coating growth state; based on the principle of electrochemical impedance spectroscopy (EIS), construct an equivalent circuit model to correlate the resistance spectrum characteristics with the physical parameters of the electroplating process; apply a non-linear regression algorithm to fit the parameters of the equivalent circuit model to generate electroplating process parameters; establish a mapping relationship between the electroplating process parameters and indicators such as the coating growth rate and uniformity to generate coating growth state data.

[0045] Construct a coating growth dynamic prediction model. Integrate the coating growth state data and the group coating growth characteristics to construct an initial coating growth history database; apply long short-term memory network (LSTM) to train the coating growth time series prediction model based on the coating growth history database; combine the workpiece feature vector and the actual current density distribution in the micro-region to develop a micro-region coating growth predictor considering micro-region characteristics; design a dynamic confidence interval estimation method to provide a prediction uncertainty assessment for the prediction results and guide the adjustment of subsequent control strategies.

[0046] According to one aspect of the present application, a multi-frequency differential resistance spectrum acquisition system includes: An alternating current impedance measurement circuit with frequency scanning capabilities that superimposes a weak alternating current signal without disturbing the normal electroplating process; A reference electrode array arranged in the electrolytic cell to form a multi-point sampling network; A partition modulation strategy module that develops an electrolytic cell partition modulation strategy to achieve partition resistance spectrum acquisition of different regions of the electrolytic cell by controlling the alternating current impedance measurement circuit and the reference electrode array; A time-series differential sampling module that performs differential processing on the resistance spectrum data at adjacent time points to generate differential resistance spectrum data.

[0047] According to one aspect of the present application, the steps of decoupling and eliminating interference from the resistance spectra of workpiece groups to obtain group coating growth characteristics include: Based on the workpiece electric field interference matrix, construct a resistance spectrum interference model in a multi-workpiece environment; Combined with the resistance spectrum interference model, use the independent component analysis algorithm (ICA) to separate the independent resistance spectrum components of each workpiece group from the differential resistance spectrum data; Use a method based on Kalman filtering to process the independent resistance spectrum components to eliminate the resistance spectrum fluctuations caused by non-coating growth factors and obtain purified resistance spectrum data; Combined with the within-group difference matrix, further refine the purified resistance spectrum data to obtain group coating growth characteristics.

[0048] Specifically, the hybrid resistance spectrum signal is combined with the resistance spectrum interference model (ESIM), and the ICA algorithm is applied for signal separation. Specifically: Construct a priori information matrix: Use ESIM to construct the a priori mixing matrix A_prior, where each element A_prior(i, j) represents the theoretical influence intensity of workpiece j on measurement point i, and A_prior(i, j) = α × exp(-d(i, j) / λ)× FEDM_norm(i, j), where d(i, j) is the distance from measurement point i to workpiece j, λ is the attenuation coefficient (set to 15 mm), FEDM_norm(i, j) is the normalized electric field interference intensity, and α is the proportionality coefficient. The traditional ICA algorithm usually uses a randomly initialized mixing matrix, and here A_prior is used as the initial estimate. Set the mixing matrix A_initial of ICA = A_prior + ε, where ε is a small random perturbation (to prevent local optimality). Expand the standard ICA objective function and add a penalty term for the a priori knowledge of ESIM: J(W) = J_standard(W) + β × ||W·A_prior - I||_F, where W is the separation matrix, J_standard is the standard ICA objective function, ||·||_F is the Frobenius norm, and β is the penalty weight (set to 0.3). In each ICA iteration, both the statistical independence of the signal and the consistency with the theoretical model are considered; an alternating optimization strategy is used: first update the separation matrix W based on statistical independence, and then fine-tune the separation matrix W based on the ESIM constraint; the gradient update formula: W_new = W_old - η × [▽J_standard(W) + β × ▽(||W·A_prior - I||_F)], where η is the learning rate and W_old is the separation matrix before update. As the iteration progresses, the value of β is dynamically adjusted according to the convergence situation. In the initial stage, β is larger (0.5) to emphasize a priori knowledge; in the later stage, β decreases (down to 0.1) to enhance the adaptability to actual data.

[0049] In another embodiment of the present application, in the partition modulation strategy module, an electrolytic cell partition modulation strategy is developed, and a spatial phase encoding technology is introduced. By injecting phase-orthogonal excitation signals in different regions, spatial separation of multiple workpiece interference sources is achieved, and partitioned resistance spectrum acquisition of different regions of the electrolytic cell is realized based on the AC impedance measurement circuit and the reference electrode array. In the time series differential sampling module, combined with the spatial phase decoding algorithm, the resistance spectrum data at adjacent time points are differentially processed to generate differential resistance spectrum data.

[0050] Specifically, when implementing the spatial phase encoding differential resistance spectroscopy technology on an actual electroplating production line, the specific operations are as follows: Design an AC impedance measurement circuit with a frequency scanning range of 100 Hz - 100 kHz, and the signal amplitude is 10 mV, which is much lower than the normal electroplating voltage (2 - 3 V). Arrange 16 platinum reference electrodes in the electrolytic cell to form a 4×4 grid sampling array, and each electrode maintains a distance of 10 mm from the nearest workpiece. The traditional partition acquisition method injects and acquires signals sequentially for 9 regions (3×3 grid), but this method cannot distinguish the contributions of different workpieces in a multi-workpiece environment. Introduce the spatial phase encoding technology to inject excitation signals with the same frequency but different phases simultaneously into 9 regions: s_i(t) = A_i × sin(ωt + θ_i), where A_i is the signal amplitude factor of region i; ω is the angular frequency of the signal; t is the time; θ_i is the phase of region i, and an orthogonal design is adopted: θ_i = 2π×(i - 1) / 9 to ensure a phase interval of 40°. At the signal acquisition end, use the spatial phase decoding algorithm: r_i = ∑ R(t)·sin(ωt + θ_i) dt; where r_i is the result signal after spatial phase decoding of region i; R(t) is the mixed received signal, and the signals of each region are separated through orthogonal correlation with the specific phases of each region. The traditional time series differential sampling is: DERS(f, t) = ERS(f, t) - ERS(f, t - Δt); after improvement, phase decoding is added: DERS(f, t, i)= ERS_i(f, t) - ERS_i(f, t - Δt), where ERS_i represents the decoded signal of region i, and Δt is the time difference. After applying this technology, when testing the simultaneous electroplating of 7 workpieces, the signal interference suppression rate is increased from the original 12 dB to 26 dB, and the accuracy of extracting the resistance spectrum characteristics is increased by 32%, providing a more reliable data basis for subsequent coating growth monitoring.

[0051] In another embodiment of the present application, an improved independent component analysis algorithm integrating spatial constraints is applied. By introducing the prior knowledge of the spatial positions of the workpieces and the constraints of the electroplating solution flow pattern, the independent resistance spectrum components of each workpiece group are separated from the differential resistance spectrum data. Design a signal processing method based on adaptive Kalman filtering. By adjusting the process noise covariance matrix in real time, the low-frequency noise caused by the electroplating solution flow is selectively suppressed, and the independent resistance spectrum components are processed to eliminate the resistance spectrum fluctuations caused by non-coating growth factors, obtaining purified resistance spectrum data.

[0052] Specifically, when processing the resistance spectrum signals during the simultaneous electroplating of 50 workpieces, the specific process of implementing the improved independent component analysis algorithm with fused spatial constraints and adaptive Kalman filtering is as follows: First, based on the workpiece electric field disturbance matrix FEDM, a resistance spectrum interference model ESIM is constructed: ESIM(i, j) = ∑ FEDM(i, x, y, z) × FEDM(j, x, y, z)dxdydz; The separation process of the traditional ICA algorithm is: S = W × X, where S is the separated signal source matrix, X is the mixed observation signal matrix, and W is the separation matrix. The improved ICA algorithm incorporates the workpiece spatial position constraint: F(W) = J(W) + λ × ∑∑ C(i, j) × |W_i · W_j|, where J(W) is the original ICA objective function, C(i, j) is the spatial proximity of workpieces i and j, and λ = 0.3 is the balance parameter. The workpiece spatial position constraint term penalizes the solutions that violate the spatial distribution law, making the separation result more physically meaningful.

[0053] In addition, the electroplating solution flow pattern constraint is also incorporated. Through hydrodynamic simulation, the electroplating solution velocity field v(x, y, z) is obtained, the signal propagation delay matrix T(i, j) is calculated, and the ICA solution is corrected in time series. In the signal processing stage, the traditional Kalman filter uses a fixed process noise covariance matrix Q. The improved adaptive Kalman filter dynamically adjusts Q by real-time estimating the low-frequency noise characteristics caused by the electroplating solution flow: Q(t) = Q_base + β × A_flow(t) × A_flow(t) T , where A_flow(t) is the flow noise direction vector, β = 0.4 is the weight coefficient, T is the transpose, and Q_base is the basic process noise covariance matrix. Finally, combining the intra-group difference matrix IDM, the group coating growth feature GDGF is calculated: GDGF(g) = ∑ [w_i × PERS(i) × (1 + γ × IDM(i, RW))], where w_i is the weight of workpiece i within the group, γ = 0.2 is the difference adjustment coefficient, IDM(i, RW) is the difference between workpiece i and the representative workpiece RW, and PERS(i) is the purified resistance spectrum data. Applying this technology to process an electroplating batch containing 7 groups of workpieces, the signal source separation accuracy is increased from 72% to 91%, the low-frequency flow noise suppression rate is increased by 18 dB, and the coating growth monitoring accuracy is increased by 25%.

[0054] According to one aspect of the present application, the steps of generating a differentiated current distribution control strategy include: S51. Construct a multi-objective electroplating quality evaluation index, including the quality-parameter mapping relationship; S52. Based on the optimized workpiece grouping and group plating growth characteristics, combined with the quality-parameter mapping relationship, construct an optimization strategy for group-level current parameters to obtain anti-interference group current parameters; S53. Based on the actual current density distribution in the micro-region, formulate a differential current modulation scheme for the micro-region to obtain the workpiece micro-region current modulation scheme; S54. Integrate the anti-interference group current parameters and the workpiece micro-region current modulation scheme, and integrate a multi-level current distribution control strategy to obtain a complete current distribution control strategy.

[0055] Specifically, construct a multi-objective electroplating quality evaluation index. Define an electroplating quality index system including plating thickness uniformity, surface roughness, bonding strength, etc.; based on the electroplating process requirements, assign index weights to each quality index; combined with the workpiece dataset, set differential target quality parameters for different workpiece groups; construct an association model between electroplating quality and current parameters to form a quality-parameter mapping relationship.

[0056] Construct an optimization strategy for group-level current parameters. Based on the optimized workpiece grouping and group plating growth characteristics, determine the basic group current parameters for each workpiece group; apply a multi-objective optimization algorithm to optimize the group current parameters according to the quality-parameter mapping relationship and the target quality parameters; consider the mutual interference intensity coefficient, and perform interference compensation on the optimized parameters to obtain anti-interference group current parameters; combined with the prediction uncertainty assessment, design a parameter tolerance interval for each workpiece group to improve the robustness of the control strategy.

[0057] Construct a differential current modulation scheme for micro-regions within the workpiece. Based on the actual current density distribution in the micro-region and the output of the micro-region plating growth predictor, identify the micro-region control targets that require differential treatment; design a dedicated pulse waveform template library for different types of micro-regions, including various waveforms such as square waves, trapezoidal waves, and exponential waves; select the most suitable waveform template for each micro-region control target, and optimize the waveform parameters according to the micro-region characteristics to generate a micro-region pulse parameter set; ensure the coordination between the micro-region pulse parameters to avoid mutual interference, and form a complete workpiece micro-region current modulation scheme.

[0058] Integrate a multi-level current distribution control strategy. Integrate the anti-interference group current parameters and the workpiece micro-region current modulation scheme to form a preliminary multi-level current distribution strategy; apply electroplating process simulation technology to verify the effectiveness of the multi-level current distribution strategy and optimize it according to the simulation results; design a parameter adjustment strategy for different stages of the electroplating process to form a complete timing control strategy; formulate alternative solutions for abnormal situations to enhance the adaptability of the control strategy, and finally generate a complete current distribution control strategy.

[0059] According to one aspect of the present application, the steps of formulating a differential current modulation scheme for the micro-region to obtain the workpiece micro-region current modulation scheme include: Based on the actual current density distribution in the micro-region and the output of the micro-region plating growth predictor, identify the micro-region regulation targets that require differential treatment; Design an adaptive waveform library for the electric field distortion region, including special waveforms such as variable-amplitude exponential waves for high-curvature sharp corners, feedback-modulated trapezoidal waves for deep holes, and double-peak composite waves for slits; Select the most suitable waveform template for each micro-region regulation target, and dynamically optimize the waveform parameters through the electric field gradient response algorithm to generate a micro-region pulse parameter set; Adopt waveform interference pre-compensation technology to predict and eliminate the waveform crosstalk effect between adjacent micro-regions, ensure the coordination between the pulse parameters of each micro-region, and form a complete workpiece micro-region current modulation scheme.

[0060] Specifically, during the electroplating process of precision connectors, the specific process of implementing differential current modulation for the complex micro-regions on the workpiece surface is as follows: First, identify the micro-region regulation targets that require special treatment. Divide the surface of the "U"-shaped connector into 30 micro-regions. When the predicted local plating growth rate deviation exceeds ±15%, mark it as a regulation target. In the example, 8 key micro-regions are identified, including 3 high-curvature sharp corner regions, 2 deep hole regions, and 3 slit regions. For the electric field distortion region, design an adaptive waveform library, including three types of special waveforms: Variable-amplitude exponential wave (for high-curvature sharp corners): i(t) = i_base × [1 + α × exp(-t / τ1)] × rect(t / T, DC), where α is the variable-amplitude coefficient (range 0.2 - 0.8), τ1 is the time constant (1 - 10 ms), rect is the rectangular wave function, DC is the duty cycle, i_base is the reference value of the current, and T is the period of the waveform.

[0061] Feedback-modulated trapezoidal wave (for deep holes): i(t) = i_base × [1 + β × (1 - exp(-t / τ2))] × trap(t / T, RT, FT), where β is the modulation coefficient (0.1 - 0.5), τ2 is the deep hole charging time constant (5 - 20 ms), trap is the trapezoidal wave function, and RT and FT are the rising and falling time ratios.

[0062] Double-peak composite wave (for slits): i(t) = i_base × [rect(t / T, DC1) + γ × rect((t - Δt) / T, DC2)], where γ is the secondary peak amplitude ratio (0.3 - 0.7), Δt is the peak-to-peak delay (T / 4 - T / 2), and DC1 and DC2 are the duty cycles of the two rectangular waves.

[0063] Waveform parameter optimization adopts the electric field gradient response algorithm: Params = Params_init + η × ▽E_local × (E_target / E_local - 1), where ▽E_local is the local electric field gradient, η is the learning rate (0.05 - 0.2), E_target is the target electric field strength, Params_init is the initial parameter value in the waveform optimization algorithm, and E_local is the local electric field strength.

[0064] For adjacent micro-regions, the waveform interference pre-compensation technique is applied: i_adj(t) = i_orig(t) - ∑ κ_j ×i_j(t-δ_j), where κ_j is the interference coefficient of adjacent micro-region j (usually <0.15), δ_j is the interference delay time (0 - 2ms), i_orig(t) is the original uncorrected modulation current signal, and i_j is the modulation current signal of adjacent micro-region j. This embodiment improves the plating quality of the complex micro-regions of the "U" - shaped connector: the thickness uniformity in the sharp - corner area is increased by 29%, the deep - hole coverage rate is increased by 35%, the bonding strength in the slit area is increased by 22%, and the plating transition between different micro-regions is smoother, and the surface quality score is increased from 72 points to 91 points.

[0065] This embodiment is applied to an electroplating production line for manufacturing precision electronic connectors, which needs to gold - plate copper - alloy connectors with different shapes and sizes simultaneously. The production batch contains 50 workpieces, including three basic shapes: long - strip, "L" - shaped, and "U" - shaped. There are differences in size and fine structure for workpieces of each shape. The specific implementation steps are as follows: Step 1: Acquisition and pre - processing of workpiece feature data.

[0066] Use a KEYENCE VK - X1000 three - dimensional laser scanner to scan 50 workpieces, and the resolution is set to 0.5μm. The original three - dimensional point - cloud data obtained by scanning is denoted as P_raw. The point - cloud of each workpiece contains about 1 million points, and each point contains spatial coordinate (x, y, z) information. At the same time, use a Fischer SIGMASCOPE SMP350 conductivity measuring instrument to uniformly select 20 measurement points on the surface of each workpiece to obtain the surface conductivity distribution data C_raw. The measurement conditions are: frequency 2MHz, probe diameter 2mm. Use a Bruker M4 TORNADO X - ray fluorescence analyzer to detect the material composition on the surface of the workpiece, and 5 representative points are selected for measurement for each workpiece to obtain the material composition data M_raw.

[0067] Apply the local curvature analysis algorithm to the three-dimensional point cloud data P_raw to calculate the curvature distribution feature K of the workpiece surface. For each point p_i in the point cloud, select the point set N_i within its neighborhood radius r = 0.8 mm, and use the least squares method to fit the local quadratic surface: z = ax 2 + by 2 + cxy + dx + ey + f, where a, b, c, d, e, f are fitting parameters. Calculate the principal curvatures κ_1 and κ_2 through these parameters: κ_1 = a + b + sqrt((a - b) 2 + c 2 ); κ_2 = a + b - sqrt((a - b) 2 + c 2 ), and obtain the mean curvature K_i = (κ_1 + κ_2) / 2 and Gaussian curvature G_i = κ_1 × κ_2 of the point p_i. Statistically analyze the curvature data of all points to generate the curvature distribution feature K of the workpiece, including the mean curvature distribution map and Gaussian curvature distribution map.

[0068] Apply the Daubechies wavelet transform to perform a 4-level decomposition on the workpiece surface to extract the surface roughness feature R. After meshing the point cloud data, perform wavelet decomposition along the x and y directions to extract the surface detail information at different scales, and generate multi-scale roughness indices, including Ra (arithmetic mean roughness), Rq (root mean square roughness), and Rz (ten-point mean roughness). Use the region growing method to identify the microscopic concave and convex features F on the workpiece surface. The algorithm starts from the curvature extreme points and performs region growing based on the curvature similarity criterion (the curvature difference between adjacent points < 5%) to identify the feature regions such as sharp corners, holes, and grooves, and record their positions, areas, and depth / height information.

[0069] Based on the material composition data M_raw, establish the mapping relationship with the electrochemical activity. For the gold plating process, mainly examine the copper content (Cu%), zinc content (Zn%) and impurity element content, and calculate the electrochemical activity index EA: EA = 0.85×Cu% - 0.2×Zn% - 1.5×Impurity%, where Impurity% is the total content of impurity elements. Map the EA value to the workpiece surface to generate the electrochemical activity distribution map. Combine the surface conductivity distribution data C_raw and the curvature distribution feature K to construct the current absorption tendency index CAI of the workpiece: CAI = w1×C_norm + w2×K_norm, where C_norm is the normalized conductivity value, K_norm is the absolute value of the normalized average curvature, and w1 = 0.7 and w2 = 0.3 are the weight coefficients (determined based on experience). Integrate the surface roughness feature R and the micro-convexity and concavity feature F to generate the electric field perturbation sensitivity map EDS of the workpiece: EDS = α×R_norm + β×F_intensity + γ×F_density, where R_norm is the normalized roughness value, F_intensity represents the intensity (depth / height) of the convexity and concavity feature, F_density represents the density of the convexity and concavity features per unit area, and α = 0.4, β = 0.35 and γ = 0.25 are the weight coefficients.

[0070] Perform standardization processing on all feature data, using the min-max normalization method: X_norm = (X - X_min) / (X_max - X_min), where X represents the original feature value, and X_min and X_max are the minimum and maximum values of this feature respectively. The standardized feature vector FV includes: geometric dimension parameters, curvature distribution feature K, surface roughness feature R, micro-convexity and concavity feature F, electrochemical activity index EA, current absorption tendency index CAI, and electric field perturbation sensitivity EDS. Apply the improved Local Outlier Factor (LOF) algorithm to identify the feature anomaly regions. Compared with the traditional LOF algorithm, this method introduces feature weights and assigns higher weights to the features that have a greater impact on the electroplating quality. For each sample point p, calculate the average distance ratio_p of its k nearest neighbors (k = 5), and the average distance ratio_N(p) of the k nearest neighbor points themselves. The LOF value is calculated as: LOF(p) = ratio_N(p) / ratio_p; when the LOF value is greater than 1.8, mark this region as a feature anomaly region. Perform adaptive sampling enhancement on the feature anomaly regions, increase the sampling density in the anomaly regions, and improve the data accuracy. Generate the final workpiece dataset FD.

[0071] Step 2: Analyze and group the differential features of the workpieces.

[0072] Apply the t-SNE algorithm to the workpiece dataset FD for non-linear dimensionality reduction. Different from traditional t-SNE, this method uses an adaptive perception parameter σ, which is dynamically adjusted according to the local density: σ_i = σ_base × (1 + 0.5×log(ρ_i / ρ_avg)), where σ_base = 30 is the base parameter value, ρ_i is the local density around point i, and ρ_avg is the average density. The initial dimension is 27 dimensions (including all features), and a 3D representation is obtained after dimensionality reduction, generating a dimensionality-reduced feature representation RF. An improved adaptive feature importance evaluation algorithm is used to identify key electroplating features from the dimensionality-reduced feature representation RF. Combining permutation importance and partial dependence analysis, calculate the contribution degree of each feature to the electroplating quality. The feature importance is calculated as: FI(j) = w1×PI(j) + w2×PD(j), where PI(j) is the permutation importance score of feature j, PD(j) is the partial dependence score of feature j, and w1 = 0.6 and w2 = 0.4 are weight coefficients. According to the calculation results, the top 5 most important features are: Current Absorption Tendency Index (CAI), Microscopic Concavity-Convex Feature Density, Electrochemical Activity Index (EA), Surface Mean Curvature, and Gaussian Curvature Extreme Value Distribution. Assign importance weights FIW to these features, and the weight values are proportional to their importance scores.

[0073] Design a multi-level similarity metric function for the electroplating process, including macro shape similarity, micro-structure similarity, and material property similarity, and integrate the dimensionality-reduced feature representation RF and the feature importance weight FIW. The calculation method of macro shape similarity SS: SS(i, j) = exp(-‖S_i - S_j‖ 2 / σ_s 2 ), where S_i and S_j represent the macro shape features of workpiece i and j respectively, which are extracted from the dimensionality-reduced feature representation RF and weighted according to the shape-related weights in the feature importance weight FIW. σ_s is the kernel width parameter (set to 0.2).

[0074] The calculation method of micro-structure similarity MS: MS(i, j) = exp(-‖M_i - M_j‖ 2 / σ_m 2 ), where M_i and M_j represent the micro-structure features of workpiece i and j (including surface roughness and microscopic concavity-convex features), which are also extracted from RF and weighted according to the relevant weights in FIW. σ_m is the kernel width parameter (set to 0.15).

[0075] The calculation method of material property similarity MTS: MTS(i, j) = exp(-‖C_i - C_j‖ 2 / σ_c 2) where \(C_i\) and \(C_j\) represent the material properties of workpieces \(i\) and \(j\) (including the electro-chemical activity index and current absorption tendency), which are extracted from RF and weighted according to the relevant weights in FIW. \(\sigma_c\) is the kernel width parameter (set to 0.25).

[0076] Calculate the above three types of similarity scores for each pair of workpieces to generate a workpiece similarity matrix SM: SM(i, j) = [SS(i, j), MS(i, j), MTS(i, j)]; Apply the information entropy weight method to dynamically adjust the hierarchical weight coefficient LW according to the distribution of each hierarchical feature. For the similarity metric \(k\), calculate its information entropy: \(E_k = -\sum p_{ki} \times \ln(p_{ki}) / \ln(n)\), where \(p_{ki}\) is the normalized value of feature \(k\) on the workpiece pair \((i, j)\) and \(n\) is the total number of workpiece pairs. The hierarchical weight is calculated as: \(LW_k = (1 - E_k) / \sum(1 - E_j)\). Combine the feature importance weight FIW and the hierarchical weight coefficient LW to calculate the comprehensive similarity CS between workpieces \(i\) and \(j\): CS(i, j) = \(LW_1\times SS(i, j)+LW_2\times MS(i, j)+LW_3\times MTS(i, j)\). Through the above calculations, a comprehensive similarity matrix CS is generated for all possible pairings of 50 workpieces, and this matrix will be used as the basis for subsequent workpiece grouping.

[0077] Based on the comprehensive similarity CS, apply the spectral clustering algorithm to initially group the 50 workpieces. Different from traditional spectral clustering, this method uses an adaptive similarity threshold to automatically determine the optimal threshold by analyzing the similarity distribution. Construct the similarity matrix W, calculate the Laplacian matrix \(L = D - W\), where \(D\) is the degree matrix (the diagonal elements are the sums of the corresponding rows). Solve the generalized eigenvalue problem \(Lv=\lambda Dv\), select the eigenvectors corresponding to the first \(k\) smallest non-zero eigenvalues to form the eigenmatrix V. Perform K-means clustering on the rows of the eigenmatrix V to obtain the initial workpiece grouping IG.

[0078] Design a grouping evaluation index based on the sensitivity of the electroplating process, including intra-group similarity, inter-group difference, and electroplating parameter sensitivity. The comprehensive score is: Score = 0.4×IntraSim + 0.3×InterDiff + 0.3×ParamSens; Use an adaptive grouping adjustment algorithm to optimize the grouping boundary through simulated annealing to generate an optimized workpiece grouping OG. Finally, the 50 workpieces are divided into 7 groups, namely long and large (8), long and medium (7), long and small (6), "L"-shaped large (9), "L"-shaped small (7), "U"-shaped large (6), and "U"-shaped small (7).

[0079] For each group, representative workpieces are selected based on centrality analysis. Calculate the centrality index of workpiece i: Centrality(i) = ∑ CS(i, j) / |G|; where |G| is the number of workpieces in the group. The workpiece with the highest centrality is selected as the representative workpiece RW of the group.

[0080] For the workpieces within each optimized workpiece group, calculate the feature deviation between them and the representative workpiece, and generate the within-group difference matrix IDM: IDM(i) = [FV(i) - FV(RW)] / FV(RW) × 100%; where FV(i) is the feature vector of workpiece i and FV(RW) is the feature vector of the representative workpiece. Apply principal component analysis (PCA) to identify the main patterns of within-group differences and extract the within-group difference principal components IDP. Retain the principal components that explain 95% of the variance, usually 3 - 5. Based on the within-group difference principal components IDP, construct the within-group workpiece difference distribution model IDM, and use the multivariate Gaussian distribution to describe the distribution of within-group workpiece features.

[0081] Step 3: Modeling and analysis of the electric field distribution in the electrolytic cell.

[0082] Based on the geometric structure and electrode arrangement of the electrolytic cell, construct the basic geometric model BGM of the electrolytic cell. The electroplating cell used is a rectangular cell with dimensions of 1200mm × 800mm × 600mm, and the anode is a titanium iridium oxide-coated titanium plate vertically placed on both sides with dimensions of 1000mm × 500mm. Measure the conductivity distribution of the electrolyte (potassium gold cyanide solution) at different positions, with the temperature set at 65 ± 2°C and the pH value at 10.5 ± 0.2. Construct the electrolyte conductivity field ECF, and generate the conductivity distribution model of the entire electroplating cell through cubic spline interpolation. Integrate the workpiece geometric feature data, place the workpieces in the basic geometric model of the electrolytic cell, and generate the complete model FCM of the multi-workpiece electrolytic cell. The workpieces are placed on a dedicated workpiece rack in 7 groups, and the workpiece spacing is not less than 30mm. Apply the finite element analysis method, use tetrahedral meshing for the electrolytic cell model, with the mesh size set to 0.5mm near the workpiece surface and 2 - 5mm in other areas. Solve the Laplace equation: ▽·(σ▽φ) = 0, where σ is the conductivity and φ is the electric potential. The boundary conditions are set as the anode electric potential of 1V and the cathode (workpiece) electric potential of 0V. Calculate the ideal reference electric field distribution REF without workpiece interference.

[0083] Based on the complete model of the multi-workpiece electrolytic cell FCM, the electric field distribution when a single workpiece exists is simulated in sequence to generate a series of single-workpiece electric field disturbance maps SEDF. During the simulation, other conditions are kept unchanged, and only one workpiece is placed. Calculate the difference between each single-workpiece electric field disturbance map and the ideal reference electric field distribution, and construct the workpiece electric field interference matrix FEDM: FEDM(i, x, y, z) = SEDF(i, x, y, z) - REF(x, y, z), where (x, y, z) represents the spatial coordinate points.

[0084] Design an index for quantifying the electric field interference intensity, and calculate the mutual interference intensity coefficient MIFC between workpieces. For workpieces i and j, the interference coefficient is calculated as: MIFC(i, j) = ∫∫∫ |FEDM(i, x, y, z) × FEDM(j, x, y, z)| dxdydz, and the integration range is the intersection of the influence regions of the two workpieces. Identify the main propagation paths of the electric field interference and construct the interference propagation network model IPNM. Construct a directed graph according to the MIFC value, with the nodes being workpieces, the weight of the edge being the interference coefficient, and the direction pointing from the interference source to the interfered workpiece.

[0085] Based on the workpiece electric field interference matrix FEDM and the mutual interference intensity coefficient MIFC, construct a static electric field distribution model SEFM considering the interaction of multiple workpieces. Adopt a modified linear superposition method: SEFM(x, y, z) = REF(x, y, z) + ∑ αi × FEDM(i, x, y, z), where αi is the influence coefficient of workpiece i, which is determined by iterative optimization with an initial value set to 1.0. Use the computational fluid dynamics method to simulate the influence of the electroplating solution flowing in a cycle at a flow rate of 200 L / min on the electric field distribution, and generate the flow field influence factor FIF: FIF(x, y, z) = 1 + β × v(x, y, z) / v_max × cos(θ), where v(x, y, z) is the flow velocity at point (x, y, z), v_max is the maximum flow velocity, θ is the angle between the flow direction and the electric field direction, and β = 0.15 is the influence coefficient. Combine the static electric field distribution model SEFM and the flow field influence factor FIF to construct a dynamic electric field distribution prediction model DEFM: DEFM(x, y, z, t) = SEFM(x, y, z) × FIF(x, y, z, t), where t represents the time variable. Apply the Monte Carlo method, conduct 1000 simulations by randomly perturbing the workpiece positions (within the range of ±2 mm), calculate the standard deviation of the electric field intensity at each point, and generate the electric field distribution uncertainty interval EFUI.

[0086] Based on the Dynamic Electric Field Distribution Prediction Model (DEFM), calculate the Initial Electric Field Strength (IEF) at each point on the workpiece surface. Combining the microscopic topography data in the workpiece feature vector, identify the Electric Field Hotspot Region (EHR) where the electric field distribution is abnormal. When the local electric field strength exceeds 1.5 times the average value or is lower than 0.6 times the average value, mark this region as the hotspot region. Apply the improved Butler-Volmer equation to establish the mapping relationship between the micro-region electric field strength and the theoretical current density. The improvement lies in considering the local electric field distortion effect, and the equation is: i = i0 × [exp(αaFη / RT) - exp(-αcFη / RT)] × f(E_local / E_avg), where i is the current density, i0 is the exchange current density, αa and αc are the anodic and cathodic transfer coefficients (both 0.5), F is the Faraday constant, η is the overpotential, R is the gas constant, T is the absolute temperature, and f(E_local / E_avg) is the local electric field correction function, expressed as: f(x) = x γ , where γ is the electric field sensitivity coefficient, determined experimentally to be 1.25. Considering the cathodic polarization effect, correct the theoretical current density to obtain a more accurate Actual Current Density Distribution in the Micro-region (ACDD): ACDD = i_theory × [1 - δ×exp(-t / τ)], where t is the electroplating time, τ is the polarization time constant (about 30 seconds), and δ is the polarization coefficient (about 0.2).

[0087] Step Four: Dynamic Monitoring and Prediction of Coating Growth.

[0088] Design an alternating current impedance measurement circuit with frequency scanning capabilities, with a frequency range of 100 Hz - 100 kHz, divided into 10 logarithmically equally spaced points, and a signal amplitude of 10 mV (far lower than the normal electroplating voltage and does not interfere with the normal electroplating process). Arrange 16 platinum reference electrodes in the electrolytic cell to form a 4×4 grid sampling array, with each electrode maintaining a 10 mm distance from the nearest workpiece. Develop an electrolytic cell partition modulation strategy, divide the electrolytic cell into 9 regions (3×3 grid), and independently control signal injection and acquisition for each region. Design a sequential differential sampling scheme with a sampling interval of 5 seconds, perform differential processing on the resistance spectrum data at adjacent time points to generate Differential Resistance Spectrum Data (DERS): DERS(f, t) = ERS(f, t) - ERS(f, t - Δt), where ERS(f, t) represents the resistance spectrum data at frequency f at time t, and Δt = 5 seconds.

[0089] Extract the frequency-domain features of the differential resistance spectroscopy data DERS. Use wavelet transform to decompose the signal and extract the energy distribution in different frequency bands as the key frequency features KFF. Based on the principle of electrochemical impedance spectroscopy (EIS), construct an equivalent circuit model ECM. For the gold plating process, adopt the Randles equivalent circuit, including Rs: solution resistance; Rct: charge transfer resistance; Cdl: double-layer capacitance; Zw: Warburg impedance (representing the diffusion process). The model expression is: Z(ω) = Rs + Rct / (1 + jωCdlRct) + σw / sqrt(jω), where Z(ω) is the total impedance, ω is the angular frequency, j is the imaginary unit, and σw is the Warburg coefficient. Apply the non-linear regression algorithm to fit the parameters of the equivalent circuit model and generate the electroplating process parameters EPP. The specific fitting method uses the Levenberg-Marquardt algorithm, and the objective function is: F = ∑|Z_meas(ωi) - Z_model(ωi)| 2 / |Z_meas(ωi)| 2 , where Z_meas is the measured value and Z_model is the model predicted value. Establish the mapping relationship between the electroplating process parameters and the coating growth state, and focus on the change trends of Rct (charge transfer resistance) and Cdl (double-layer capacitance): a decrease in Rct indicates an increase in electroplating activity and a faster coating growth rate; an increase in Cdl indicates an increase in the effective surface area and a more uniform coating growth. According to these relationships, calculate the coating growth rate index DGR and the uniformity index DGU: DGR = k1 / Rct; DGU = k2×Cdl / Cdl_0, where k1 and k2 are proportionality coefficients and Cdl_0 is the initial double-layer capacitance value.

[0090] Based on the workpiece electric field disturbance matrix FEDM, a resistance spectrum interference model ESIM in a multi-workpiece environment is constructed. By applying the independent component analysis (ICA) algorithm, the independent resistance spectrum components IESC of each workpiece group are separated from the mixed resistance spectrum signals. Compared with traditional ICA, in this embodiment, spatial prior information is introduced, and the physical position relationship of the workpieces is used to assist signal separation: S = W × X, where S is the separated signal source matrix, X is the mixed observation signal matrix, and W is the separation matrix (determined by maximizing non-Gaussianity). A signal processing method based on Kalman filtering is designed to eliminate the resistance spectrum fluctuations caused by non-plating growth factors. The state equation and the observation equation are set as: x(t + 1) = Ax(t) + w(t); z(t) = Hx(t) + v(t), where x represents the state vector (including the true resistance spectrum parameters), z is the observation vector, A is the state transition matrix, H is the observation matrix, and w and v are the system noise and the observation noise respectively. Through the Kalman filtering update and prediction steps, purified resistance spectrum data PERS is obtained. Combining with the intra-group difference matrix IDM, the purified resistance spectrum data is further refined to obtain more accurate group plating growth features GDGF: GDGF(g) = ∑ w_i × PERS(i), where g represents the workpiece group, and w_i is the weight of workpiece i within the group (related to the similarity with the representative workpiece).

[0091] Integrate the plating growth status data and the group plating growth characteristics to construct an initial plating growth history database DGDB, which contains the following information: timestamp, workpiece group identifier, resistance spectrum parameters (Rs, Rct, Cdl, Zw), plating growth rate index (DGR), uniformity index (DGU), and electroplating current parameters. Apply the long short-term memory network (LSTM) to train the plating growth time series prediction model TPML based on the plating growth history database. The network structure includes: Input layer: 8 neurons (corresponding to the input feature dimension); LSTM layer 1: 64 neurons, dropout rate 0.2; LSTM layer 2: 32 neurons, dropout rate 0.2; Fully connected layer: 16 neurons, activation function ReLU; Output layer: 2 neurons (corresponding to DGR and DGU). The training parameters are: batch size: 32; learning rate: 0.001, using the Adam optimizer; sequence length: 10 time steps (50 seconds). Combine the workpiece feature vector and the actual current density distribution in the micro-region to develop a micro-region plating growth predictor RDGP that considers micro-region characteristics. Divide the workpiece surface into multiple micro-regions, and adjust the prediction parameters for each micro-region according to its characteristics: DGR_local = DGR × f_shape × f_material × f_current, where f_shape, f_material, and f_current are the shape factor, material factor, and current factor, respectively, calculated according to local characteristics. Design a dynamic confidence interval estimation method to provide a prediction uncertainty evaluation PUE for the prediction results. Adopt the Monte Carlo Dropout technique to keep the dropout activation during the inference stage, perform 50 forward propagations, calculate the mean and standard deviation of the prediction results, and obtain a 95% confidence interval.

[0092] Step 5: Generate a differentiated current distribution strategy.

[0093] Define an electroplating quality index system QIS including plating thickness uniformity, surface roughness, adhesion strength, etc. The specific indicators include: TU: plating thickness uniformity, deviation < ±5%; SR: surface roughness, Ra < 0.2μm; AH: adhesion (adhesion strength), > 15N / mm 2 ; BS: surface brightness, reflectivity > 90%; PD: plating defect density, < 2 per cm 2Based on the requirements of the electroplating process, assign index weights IW for each quality index: TU: 0.35; SR: 0.25; AH: 0.20; BS: 0.10; PD: 0.10. Combine the workpiece dataset to set differentiated target quality parameters TQP for different workpiece groups. For example, for the "U"-shaped small workpiece group in the precision connection part, set a higher thickness uniformity requirement (deviation < ±3%). Construct an association model between electroplating quality and current parameters to form a quality-parameter mapping relationship (QPM). For each quality index, establish its functional relationship with key electroplating parameters: TU = f1(CD, PF, DC, CT); SR = f2(CD, PF, DC); AH = f3(CD, CT, TE); BS = f4(CD, PF, DC, TE); PD = f5(CD, PF, TEMP), where CD represents current density, PF represents pulse frequency, DC represents duty cycle, CT represents electroplating time, TE represents electrolyte stirring intensity, and TEMP represents electroplating solution temperature. These functional relationships are established through historical data analysis and electroplating experiments and are described using a multiple regression model.

[0094] Based on the optimized workpiece grouping OG and the group coating growth characteristics GDGF, determine the basic group current parameters GCP for each workpiece group. The initial parameter settings are as follows: long and large: current density 2.8 A / dm 2 , pulse frequency 80 Hz, duty cycle 65%; long and medium: current density 3.0 A / dm 2 , pulse frequency 85 Hz, duty cycle 70%; long and small: current density 3.2 A / dm 2 , pulse frequency 90 Hz, duty cycle 75%; "L"-shaped large: current density 2.6 A / dm 2 , pulse frequency 75 Hz, duty cycle 60%; "L"-shaped small: current density 2.9 A / dm 2 , pulse frequency 80 Hz, duty cycle 70%; "U"-shaped large: current density 2.5 A / dm 2 , pulse frequency 70 Hz, duty cycle 55%; "U"-shaped small: current density 2.7 A / dm 2 , pulse frequency 75 Hz, duty cycle 65%, Apply a multi-objective optimization algorithm to optimize the group current parameters according to the quality-parameter mapping relationship QPM and the target quality parameters TQP. Use the weighted sum method to transform the multi-objective problem into a single-objective problem: F(x) = ∑ IW_i × [TQP_i - f_i(x)] 2 / TQP_i 2, where \(x\) represents the current parameter vector, \(IW_i\) represents the weight of index \(i\), \(TQP_i\) represents the target value of index \(i\), and \(f_i\) represents the prediction function of index \(i\). The particle swarm optimization (PSO) algorithm is used to solve the optimal parameters. After 100 iterations with 50 particles, the optimized group current parameters are obtained. Considering the mutual interference intensity coefficient MIFC, interference compensation is performed on the optimized parameters. The compensation formula is: \(CD_{adj} = CD_{opt}×(1 + \sum MIFC(g,j)×\lambda_j)\), where \(CD_{adj}\) is the adjusted current density, \(CD_{opt}\) is the optimized current density, \(MIFC(g,j)\) is the interference coefficient of group \(g\) affected by group \(j\), and \(\lambda_j\) is the compensation coefficient (generally from -0.05 to 0.05). After interference compensation, the anti-interference group current parameters RIGCP are obtained. Combining the prediction uncertainty evaluation PUE, parameter tolerance intervals are designed for each workpiece group: \(CD_{range} = CD_{adj} ± k×\sigma_{CD}\); \(PF_{range} = PF_{adj} ± k×\sigma_{PF}\); \(DC_{range} = DC_{adj} ± k×\sigma_{DC}\); where \(k\) is the confidence level parameter (set to 1.96 for a 95% confidence interval), and \(\sigma\) represents the standard deviation of the corresponding parameter.

[0095] Based on the output of the actual current density distribution in the micro-region ACDD and the micro-region coating growth predictor RDGP, the micro-region regulation target MACT that requires differential processing is identified. When the deviation of the predicted local coating growth rate exceeds ±15% or the deviation of the uniformity index exceeds ±10%, the micro-region is marked as the target area to be regulated. A dedicated pulse waveform template library PWTL for different types of micro-regions is designed. The templates include: standard square wave: suitable for flat regions with a constant duty cycle; trapezoidal wave: suitable for transition regions with slow rising and falling edges; exponential wave: suitable for deep holes and grooves with exponential rising / falling characteristics; composite wave: suitable for complex-shaped regions, composed of basic waveforms.

[0096] Select the most suitable waveform template for each micro-region regulation target and optimize the waveform parameters according to the micro-region characteristics. The selection strategy is based on a decision tree model, considering factors such as the shape characteristics, curvature, depth-to-width ratio, etc. of the micro-region. For example: For high-curvature sharp corners: Select an exponential wave with a short rise time and a long fall time; For deep holes: Select an exponential wave with a long rise time and a short fall time; For grooves: Select a trapezoidal wave with moderate rise / fall times. The waveform parameter optimization is based on the micro-region current density requirement, adjusting the pulse frequency (PF), duty cycle (DC), rise time (RT), and fall time (FT). Generate a micro-region pulse parameter set MRPP. Ensure the coordination between the pulse parameters of each micro-region to avoid mutual interference. Check the parameter differences between adjacent micro-regions. When the frequency difference > 20 Hz or the duty cycle difference > 15%, introduce a smooth transition zone and establish a parameter gradient. Form a complete workpiece micro-region current modulation scheme MCMS.

[0097] Integrate the anti-interference group current parameters RIGCP and the workpiece micro-region current modulation scheme MCMS to form a preliminary multi-level current distribution strategy MLCDS. This strategy includes two levels: Group level: Set the basic current parameters for each workpiece group; Micro-region level: Based on the group parameters, perform differential modulation on specific micro-regions. Apply electroplating process simulation technology to verify the effectiveness of the multi-level current distribution strategy. Use the finite element method to simulate the electroplating process and predict the coating thickness distribution and quality indicators. According to the simulation results, adjust the parameters to optimize the electroplating effect. Design a parameter adjustment strategy for different stages of the electroplating process to form a complete timing control strategy TSCS. Divide the electroplating process into three stages: Initial stage (0 - 30 seconds): Use a lower current density (80% of the base value) to establish the initial coating; Main growth stage (30 seconds - 90% of the total time): Apply the complete multi-level control strategy; Trimming stage (the last 10% of the time): Reduce the current density (70% of the base value) and increase the pulse frequency (120% of the base value) to improve the surface finish. Develop alternative solutions for abnormal situations to enhance the adaptability of the control strategy. Design response strategies for common abnormal situations: Abnormal resistance spectrum fluctuation: Temporarily switch to the safe parameter mode and reduce the current density by 20%; Temperature anomaly: Adjust the pulse duty cycle to reduce heat generation; Workpiece position offset: Recalculate the electric field distribution and update the current parameters. Finally, generate a complete current distribution control strategy CCDCS, including: The basic current parameters of 7 workpiece groups; The micro-region differential modulation scheme for workpieces within each group; The timing control strategy for the entire electroplating process; The response strategy for abnormal situations.

[0098] To verify the effectiveness of this embodiment, three typical workpieces were selected for comparative tests: a complex "U"-shaped small connector (high-precision requirement), a long-strip large connector (conventional requirement), and an "L"-shaped large connector (mixed requirement). The electroplating was carried out using the method of this embodiment, the traditional constant current method, and the simple grouping method respectively. The electroplating time was 15 minutes, and the gold thickness target was 1.5 μm. The results showed that the thickness deviation of this embodiment was ±3.2%, the traditional method was ±12.7%, and the simple grouping method was ±7.5%. The surface roughness Ra value of this embodiment was 0.15 μm, the traditional method was 0.28 μm, and the simple grouping method was 0.21 μm. The bonding strength of this embodiment was 17.8 N / mm 2 , the traditional method was 14.2 N / mm 2 , and the simple grouping method was 15.6 N / mm 2 . The surface brightness (reflectivity) of this embodiment was 93%, the traditional method was 85%, and the simple grouping method was 89%. The coating defect density of this embodiment was 0.8 per cm 2 , the traditional method was 3.5 per cm 2 , and the simple grouping method was 1.9 per cm 2 . Considering all indicators, this embodiment is superior to the traditional method and the simple grouping method in all aspects, especially in the thickness uniformity and defect control of workpieces with complex shapes, with the most significant effects.

[0099] This embodiment effectively solves the signal interference problem during the co-bath electroplating of multiple workpieces through temporal difference and spatial partition sampling, and realizes the high-precision real-time monitoring of the coating growth process. Based on the multi-level similarity measurement and the improved spectral clustering algorithm, the precise grouping of workpieces with similar features and the quantitative characterization of the intra-group differences are realized. By designing different pulse waveforms for the complex micro-regions on the workpiece surface, the problem of inconsistent coating quality in the micro-regions in traditional electroplating technology is solved. Integrating group-level parameter control and micro-level pulse modulation, the all-round differential current control from macro to micro is realized.

[0100] The present invention solves the problem of differential control for groups of workpieces with similar features. A multi-level similarity metric function for the electroplating process is constructed to calculate the similarity of workpieces in terms of macroscopic shape, microscopic structure, and material properties. A weighted mechanism for electroplating disturbance sensitivity is introduced, enabling enhanced weights to be obtained for regions with high curvature and high current density fluctuations, thereby enabling the identification of minor differences within a group of workpieces of the same specification. Based on the comprehensive similarity, a spectral clustering algorithm and an adaptive grouping adjustment algorithm are applied to accurately group the workpieces, and representative workpieces are selected through centrality analysis. For the workpieces within each optimized workpiece group, the feature deviation from the representative workpiece is calculated, and a within-group difference distribution model is constructed to provide a basis for subsequent differential current parameter adjustment, thereby achieving the accurate identification and control of minor differences within a group of workpieces with similar features. The problem of mutual interference of micro-region current parameters for workpieces with complex shapes is solved. An adaptive waveform library for the electric field distortion region is constructed, including special waveforms such as variable amplitude exponential waves, feedback modulation trapezoidal waves, and double-peak composite waves, respectively for the electroplating requirements of special micro-regions such as high curvature sharp corners, deep holes, and slits. Through the electric field gradient response algorithm to dynamically optimize the waveform parameters, the system can select the most suitable waveform template for each micro-region regulation target and accurately adjust the parameters. More importantly, a waveform interference pre-compensation technology is introduced, which can predict and eliminate the waveform crosstalk effect between adjacent micro-regions, ensuring the coordination between the pulse parameters of each micro-region, effectively solving the problem of mutual interference of micro-region current parameters, achieving precise current regulation for complex micro-regions on the workpiece surface, and improving the uniformity and quality consistency of the coating. The problem of mutual interference of electric fields in the electroplating environment of multiple workpieces is solved. By using a modified Laplace equation considering the non-linear effect of ion migration, the difference between the electric field disturbance of a single workpiece and the ideal reference electric field is accurately simulated and calculated, and a workpiece electric field interference matrix is constructed. By introducing a spatial gradient weighting factor, the system can accurately quantify the mutual interference intensity between workpieces, especially effectively compensating for the strong interference in high current density regions. Based on the mutual interference intensity coefficient, an interference propagation network model is constructed to clearly identify the main propagation paths of electric field interference, and when generating a differential current distribution strategy, the optimized current parameters are adjusted through an interference compensation mechanism, effectively eliminating the adverse effects of mutual interference of electric fields between workpieces, and enabling each workpiece to obtain a more uniform and consistent coating quality during the batch electroplating process.

[0101] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A differentiated and precise current distribution control method in a multi-workpiece batch electroplating process, characterized in that: include: Acquire feature data of multiple workpieces and perform preprocessing to obtain a workpiece data set; Based on the artifact data set, analyze the artifact difference characteristics and group them, obtain the optimized artifact grouping and calculate the intra-group difference matrix; Based on the optimized workpiece grouping and workpiece data set, the electric field distribution of the electrolytic cell is modeled and analyzed to obtain the actual current density distribution in the micro area and the workpiece electric field interference matrix; Based on the workpiece electric field interference matrix and the intra-group difference matrix, the coating growth dynamic monitoring and prediction are carried out to obtain the group coating growth characteristics; The group coating growth characteristics are read, and the differentiated current distribution control strategy is generated by combining the actual current density distribution in the micro-area and optimizing the workpiece grouping.

2. The method according to claim 1, characterized in that Get the optimized artifact grouping and calculate the intra-group difference matrix, including: Read the artifact dataset, perform dimensionality reduction and extract key features to obtain the dimensionality reduction feature representation and feature importance weights, and then construct a multi-level artifact similarity measure to obtain the comprehensive similarity; Based on the comprehensive similarity, the artifacts are grouped and representative artifacts are selected to obtain optimized artifact grouping and representative artifacts, and the intra-group differences are quantified and analyzed based on them to obtain the intra-group difference matrix.

3. The method according to claim 2, characterized in that Calculate the comprehensive similarity, including: Data is extracted from the dimension reduction feature representation and feature importance weights, and a multi-level similarity metric function for the electroplating process is used to calculate the macroscopic shape similarity, microscopic structure similarity, and material property similarity; based on this, similarity scores between workpieces are calculated to obtain a workpiece similarity matrix; According to the distribution of each level feature in the workpiece similarity matrix, combined with the electroplating disturbance sensitivity weighting mechanism, the level weight coefficient is dynamically adjusted; The comprehensive similarity between artifacts is calculated by combining the artifact similarity matrix, hierarchical weight coefficients and feature importance weights.

4. The method according to claim 2, characterized in that: Get optimized artifact grouping and representative artifacts, including: Based on the comprehensive similarity, the artifacts are preliminarily grouped using the spectral clustering algorithm to obtain the initial artifact grouping; The rationality of the initial workpiece grouping is evaluated by using the grouping evaluation index based on the sensitivity of the electroplating process, and the grouping evaluation results are obtained; Adopting the adaptive grouping adjustment algorithm, the grouping boundaries are optimized according to the grouping evaluation results to obtain the optimized workpiece grouping; For each optimized artifact group, a centrality analysis is performed and the artifact with the highest centrality is taken as the representative artifact.

5. The method according to claim 1, characterized in that: Modeling and analysis of the electric field distribution of the electrolytic cell are carried out to obtain the actual current density distribution in the micro-area and the electric field interference matrix of the workpiece, including: The geometric structure and electrode arrangement of the electrolytic cell are obtained, and the reference electric field of the multi-workpiece electrolytic cell is modeled by combining the optimized workpiece grouping and workpiece data set to obtain the complete model of the multi-workpiece electrolytic cell and the ideal reference electric field distribution; based on this, the electric field interference between the workpieces is mapped and quantified to obtain the workpiece electric field interference matrix and the mutual interference intensity coefficient; Based on the workpiece electric field interference matrix and mutual interference intensity coefficient, a dynamic electric field distribution prediction model is constructed. Combined with the workpiece feature vector in the workpiece data set, micro-area electric field characteristics analysis and current density mapping are carried out to obtain the actual current density distribution in the micro-area.

6. The method according to claim 5, characterized in that Map and quantify the electric field interference between workpieces to obtain the workpiece electric field interference matrix and mutual interference intensity coefficient, including: Based on the complete model of the multi-workpiece electrolytic cell, the electric field distribution when a single workpiece exists is simulated to generate the electric field disturbance diagram of a single workpiece; Calculate the difference between the electric field disturbance map of a single workpiece and the ideal reference electric field distribution, and construct the workpiece electric field interference matrix; Combine the preset electric field interference intensity quantitative index and the workpiece electric field interference matrix to calculate the mutual interference intensity coefficient between workpieces; Based on the mutual interference intensity coefficient, the main propagation paths of electric field interference are identified and the interference propagation network model is constructed.

7. The method according to claim 1, characterized in that Dynamic monitoring and prediction of coating growth are performed to obtain group coating growth characteristics, including: Acquire differential resistance spectrum data through a multi-frequency differential resistance spectrum acquisition system; Based on the differential resistance spectrum data, the coating growth state data is obtained by mapping the resistance spectrum characteristics with the coating growth state; Combining the workpiece electric field interference matrix, differential resistance spectrum data and intra-group difference matrix, the workpiece group resistance spectrum is decoupled and interference eliminated to obtain the group coating growth characteristics; Based on the coating growth status data and group coating growth characteristics, the coating growth dynamic prediction model was used to perform simulations and obtain the prediction uncertainty assessment.

8. The method according to claim 7, characterized in that Multi-frequency differential resistance spectrum acquisition system, including: AC impedance measurement circuit with frequency scanning capability to superimpose weak AC signals without interfering with the normal electroplating process; A reference electrode array is arranged in the electrolytic cell to form a multi-point sampling network; The partition modulation strategy module realizes the acquisition of partition resistance spectra of different areas of the electrolytic cell by controlling the AC impedance measurement circuit and the reference electrode array; The time series differential sampling module performs differential processing on the resistance spectrum data at adjacent time points to generate differential resistance spectrum data.

9. The method according to claim 7, characterized in that: The steps of decoupling and eliminating interference of the resistance spectrum of the workpiece group and obtaining the growth characteristics of the group coating include: Based on the workpiece electric field interference matrix, a resistance spectrum interference model in a multi-workpiece environment is constructed; Combined with the resistance spectrum interference model, ICA is used to separate the independent resistance spectrum components of each workpiece group from the differential resistance spectrum data; The Kalman filter-based method is used to process the independent resistance spectrum components, eliminate the resistance spectrum fluctuations caused by non-plating growth factors, and obtain purified resistance spectrum data; Combined with the intra-group difference matrix, the resistance spectrum data is refined and purified to obtain the group coating growth characteristics.

10. The method according to claim 1, characterized in that The steps of generating a differentiated current sharing control strategy include: Construct multi-objective electroplating quality evaluation indicators, including quality-parameter mapping relationships; Based on the optimization of workpiece grouping and group coating growth characteristics, combined with the quality-parameter mapping relationship, a group-level current parameter optimization strategy is constructed to obtain the anti-interference group current parameters; Based on the actual current density distribution in the micro-area, a micro-area differentiated current modulation scheme is formulated to obtain the workpiece micro-area current modulation scheme; The anti-interference group current parameters and the workpiece micro-area current modulation scheme are integrated, and the multi-level current distribution control strategy is integrated to obtain a complete current distribution control strategy.

Images