Treatment process of sulfur-free copper-clad plate for automobile plate

Abstract

The invention provides a treatment process of a sulfur-free copper-clad plate for an automobile plate. The treatment process comprises a multi-parameter coupling closed-loop control method based on multi-modal sensor acquisition, data normalization and abnormity elimination, deposition dynamic fingerprint database construction, principal component analysis dimensionality reduction and support vector regression. By monitoring the concentration, temperature and flow velocity of reaction liquid and the state of a substrate in real time and fusing thickness distribution, surface roughness and binding force performance indexes, dynamic prediction of deposition performance and weighted optimization of deposition time are achieved, an optimal deposition time threshold value is determined in combination with a Pareto optimization algorithm, automatic control is achieved, and the optimal deposition time is obtained. According to the method, the thickness uniformity, the production efficiency and the parameter adaptive capacity of the deposition process can be remarkably improved, online model incremental learning is supported, and the stability and the accurate control effect of a process system are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of process control of sulfur-free electroless copper plating, and in particular to a processing technology of sulfur-free copper-clad plate for automobile plates. BACKGROUND

[0002] The current sulfur-free copper-clad plate chemical deposition process is widely used in high-reliability manufacturing fields such as automobile electronics. The mainstream technology route adopts step-by-step setting of key process parameters such as deposition temperature, reaction solution concentration and flow rate, and matches them through fixed or experience-based deposition time to obtain the target copper layer thickness and surface quality. Common process control mainly relies on manual setting of deposition time or single parameter feedback adjustment. Some advanced production lines introduce PID regulation loop to realize dynamic control of temperature and flow rate, but the deposition time parameter is still input in a preset or batch correction manner, lacking fine management of the complex coupling relationship between process parameters. The industry development trend focuses on improving deposition uniformity, batch stability and process automation level, and tries to integrate online sensing, data-driven modeling and other technologies to promote process intelligent upgrading. At present, the process control technology suitable for the field of sulfur-free electroless copper plating mainly strengthens the closed-loop regulation of single parameters such as temperature and concentration. A few researches explore the cooperation between parameters by using simple statistical models or experience tables. However, in the complex sulfur-free chemical system, the copper deposition rate and the acquisition of high uniformity layer are influenced by the cooperation of multiple parameters. Most of the existing methods fail to effectively model the nonlinear dynamic relationship between deposition time and multiple parameters of the reaction solution, and it is also difficult to deduce the whole process change of deposition behavior according to real-time data. Specifically, for high-performance sulfur-free copper-clad plates for automobiles, the thickness distribution and adhesion consistency of the copper layer are strict, and the process regulation is not accurate, which leads to high batch fluctuation and material defect risk. SUMMARY

[0003] The present application provides a processing technology of sulfur-free copper-clad plate for automobile plates to solve the above technical problems.

[0004] The technical solution of the present application is as follows: a processing technology of sulfur-free copper-clad plate for automobile plates, comprising: S1: collecting reaction solution concentration, temperature and flow rate parameter data in the chemical deposition process of sulfur-free copper-clad plate, recording corresponding deposition time reference value and substrate state information, and forming an initial process parameter set; S2: performing normalization processing and outlier rejection operation on the initial process parameter set to generate a standardized parameter matrix, eliminating the influence of different dimensions on subsequent modeling; S3: Based on the multiple sets of process parameters-performance corresponding data obtained by small batch deposition experiment, a deposition dynamic fingerprint library is constructed, and the deposition dynamic fingerprint library includes a ternary mapping relationship of parameter combination, deposition time and thickness distribution, surface roughness and bonding force; S4: Principal component analysis dimension reduction processing is performed on the standardized parameter matrix in the deposition dynamic fingerprint library, key feature factors affecting deposition performance are extracted, and a multi-parameter coupled response model is trained based on a support vector regression algorithm; S5: Real-time parameter data of reaction liquid concentration, temperature and flow rate in the current deposition process are collected, and the real-time parameter data are input into the multi-parameter coupled response model to generate a deposition performance prediction curve; S6: The optimal deposition time threshold is calculated according to the deposition performance prediction curve, and the optimal deposition time threshold satisfies the weighted optimization condition of thickness uniformity index and production efficiency index; S7: The optimal deposition time threshold is input into the deposition control system to perform dynamic adjustment of process parameters, so that the reaction liquid concentration, temperature, flow rate and deposition time form a closed loop control; S8: The deposition process data and final state quality data after adjustment are collected, and online incremental learning is performed on the deposition dynamic fingerprint library and the multi-parameter coupled response model to update the model parameter weight and the feature factor mapping relationship.

[0005] The processing process of the sulfur-free copper-clad plate for automobile plates provided by the application has the following beneficial effects: (1) The application realizes the collaborative evolution control between deposition time and other key process parameters by constructing a "deposition dynamic fingerprint library" and establishing a multi-parameter coupled response model based on machine learning. By structurally storing the multi-dimensional parameter combination in the historical deposition process and the corresponding physical performance results, a traceable and comparable fingerprint data set is formed, providing a high-quality sample basis for model training. On this basis, the core feature factors affecting the deposition behavior are extracted by principal component analysis, effectively reducing the input space redundancy and improving the modeling efficiency and generalization ability. Further, support vector regression or lightweight neural network is used to establish a "parameter-time-performance" ternary mapping relationship, which can predict the quality trend under different deposition times according to the current system state in real-time working condition, and inversely solve the optimal process window. This mechanism breaks through the traditional static setting mode, changes the passive execution variable to the active optimization variable, significantly improves the adaptive ability and control accuracy of the deposition process, and greatly improves the uniformity, bonding force stability and consistency performance of the copper-clad plate deposition layer across batches; (2) The deposition time management method proposed in the application has online learning and continuous optimization capability, and builds a closed-loop evolving intelligent process control system, effectively overcoming the problem of insufficient robustness of the prior art in the face of small fluctuations in raw materials, equipment aging or environmental disturbance. Unlike the limitation that the traditional model is fixed once deployed, the application continuously collects new production data during actual operation, automatically identifies abnormal samples and updates the fingerprint library content, and then periodically re-trains or incrementally adjusts the prediction model parameters, ensuring that it always reflects the latest production line real state. This self-evolution architecture not only enhances the adaptability to complex industrial scene changes, but also reduces the dependence on external manual intervention and frequent calibration and debugging, and is especially suitable for the demand for high-performance copper-clad plate high reliability and large-batch continuous manufacturing in the automotive electronics field. At the same time, by introducing a lightweight modeling strategy, the algorithm can run efficiently in the embedded control system, meeting the time constraint requirements of real-time decision-making on site, avoiding the impact of calculation delay on production line rhythm. The overall scheme can realize rapid deployment and migration application without complex parameter adjustment, has good scalability and engineering practicability, and provides a stable, intelligent and sustainable optimization process support platform for high-end electronic material manufacturing. BRIEF DESCRIPTION OF DRAWINGS

[0006] Figure 1 A flowchart of a processing process of a sulfur-free copper-clad plate for an automobile plate according to the present application; Figure 2 A sub-flowchart of a processing process of a sulfur-free copper-clad plate for an automobile plate according to the present application; Figure 3 Another sub-flowchart of a processing process of a sulfur-free copper-clad plate for an automobile plate according to the present application. DETAILED DESCRIPTION

[0007] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are used only to explain the present application, and cannot be understood as limiting the present application.

[0008] The disclosure below provides many different embodiments or examples for implementing different structures of the present application. In order to simplify the disclosure of the present application, the components and settings of a specific example are described below. Of course, they are only examples and the purpose is not to limit the present application. In addition, the present application can repeatedly refer to numerals and / or reference letters in different examples, and such repetition is for the purpose of simplification and clarity, which itself does not indicate the relationship between the various embodiments and / or settings discussed.

[0009] As Figure 1As shown, the present application provides a process for processing a sulfur-free copper-clad plate for automobile plates, specifically comprising: S1: Collecting reaction liquid concentration, temperature, flow rate parameter data in the chemical deposition process of the sulfur-free copper-clad plate, recording the corresponding deposition time reference value and substrate state information, and forming an initial process parameter set; S2: Performing normalization processing and outlier rejection operation on the initial process parameter set to generate a standardized parameter matrix, eliminating the influence of different dimensions on subsequent modeling; S3: Based on the multiple sets of process parameter-performance corresponding data obtained from small-batch trial deposition experiments, a deposition dynamic fingerprint library is constructed, which contains the ternary mapping relationship of parameter combination, deposition time and thickness distribution, surface roughness, and bonding force; S4: Performing principal component analysis dimension reduction processing on the standardized parameter matrix in the deposition dynamic fingerprint library, extracting key feature factors affecting deposition performance, and training a multi-parameter coupled response model based on a support vector regression algorithm; S5: Real-time collection of reaction liquid concentration, temperature, flow rate parameter data in the current deposition process, and input of the real-time parameter data into the multi-parameter coupled response model to generate a deposition performance prediction curve; S6: Calculating the optimal deposition time threshold value according to the deposition performance prediction curve, which satisfies the weighted optimization condition of thickness uniformity index and production efficiency index; S7: Inputting the optimal deposition time threshold value into the deposition control system to perform dynamic adjustment of process parameters, so that the reaction liquid concentration, temperature, flow rate and deposition time form a closed-loop control; S8: Collecting adjusted deposition process data and final state quality data, performing online incremental learning on the deposition dynamic fingerprint library and the multi-parameter coupled response model, and updating the model parameter weight and feature factor mapping relationship.

[0010] The step S1: collecting reaction liquid concentration, temperature, flow rate parameter data in the chemical deposition process of the sulfur-free copper-clad plate, and recording the corresponding deposition time reference value and substrate state information to form an initial process parameter set. Specifically comprising: S1.1: Based on the multi-modal sensor network deployed in the chemical deposition production line of the sulfur-free copper-clad plate, real-time monitoring data of reaction liquid concentration, temperature, and flow rate are obtained to construct a multi-dimensional input feature space of the process; The input condition is the multi-modal sensor network already laid out in the chemical deposition production line of the sulfur-free copper-clad plate, and the sensor nodes include reaction liquid concentration monitoring unit, temperature acquisition unit and flow rate detection unit, respectively corresponding to electrochemical concentration sensor array, thermocouple and infrared temperature measurement combination module, and ultrasonic flowmeter sensing module; A multi-modal data fusion acquisition method (parameters: sensor sampling frequency setting value, synchronous clock signal source) is used to realize the parallel acquisition of reaction liquid concentration, temperature, and flow rate, and to maintain the consistency and comparability of the time stamp; Further, through a synchronous triggering mechanism (parameter: acquisition instruction cycle issued by the centralized controller), the unified starting point of different sensor acquisition channels is realized, the synchronization of multi-dimensional signals is ensured, and the original monitoring data set is obtained; Further, a physical layer noise suppression algorithm (parameter: low-pass filter cutoff frequency configured according to sensor type) is used for signal preprocessing of each acquisition channel data, to remove high-frequency random noise and maintain the authenticity of the physical quantity change trend; Further, through a feature vector construction method (parameters: sequence length, time window size of concentration, temperature, and flow rate), the multi-channel data after filtering are structured and combined according to the time window to form a high-dimensional input feature vector set with time stamp, which is used as the multi-dimensional input feature space basic data of the process; Through the above multi-modal sensor network data acquisition and feature construction processing method, the original sensing signal is converted into structured feature data complete in spatial and time dimensions, realizing the high-precision initial parameter input effect required for subsequent deposition time coupling model training; For example, in a continuous production line for chemical deposition of sulfur-free copper clad plate for automobile panels, the sampling frequency of the reaction liquid concentration sensor array is configured as 20 Hz, the sampling frequency of the thermocouple and infrared temperature measurement module is configured as 10 Hz, the sampling frequency of the ultrasonic flowmeter is set as 5 Hz, and the synchronous acquisition cycle of the centralized controller is 1 s. Each channel starts data acquisition at the same time under the trigger of the centralized controller. The concentration signal is processed by a low-pass filter with a cutoff frequency of 2 Hz, the temperature signal is processed by a low-pass filter with a cutoff frequency of 1 Hz, and the flow rate signal is processed by a low-pass filter with a cutoff frequency of 0.5 Hz, and the waveform distortion is significantly reduced. A time window with a length of 60 points is constructed within the acquisition cycle, and the concentration, temperature, and flow rate signals are synchronously spliced into a 180-dimensional high-dimensional input feature vector according to the time stamp, and the time index of the current deposition stage is marked, and input into S1.2 to execute the electrochemical impedance spectroscopy online calibration module. In this embodiment, the output structured feature data realizes high-precision capture of concentration fluctuation peak and temperature change inflection point, significantly improving the representativeness and stability of subsequent model input; S1.2: Perform online calibration processing based on electrochemical impedance spectroscopy on the acquired reaction liquid concentration data to eliminate the influence of sensor drift and cross interference on concentration measurement accuracy, and obtain the corrected reaction liquid concentration value; The raw signal of the reaction liquid concentration output by the multi-modal sensor network is collected by an electrochemical impedance spectrum acquisition module (parameters: frequency scanning range 1Hz~100kHz, voltage excitation amplitude 10mV) to obtain the impedance spectrum curve of the reaction liquid and generate a raw impedance data matrix; Further, an equivalent circuit fitting algorithm (parameters: Randles circuit model fitting criterion, least squares error convergence threshold 1e 6) is used to solve the nonlinear mapping function between the impedance spectrum and the reaction liquid concentration and obtain a fitted concentration value vector; Further, a zero-phase digital filtering algorithm (parameters: filter type FIR, order 512, cutoff frequency 0.45 times the Nyquist frequency) is used to suppress high-frequency interference in the fitted concentration value vector and generate a smoothed concentration data sequence; Further, a temperature compensation algorithm (parameters: compensation coefficient calibrated by a temperature-sensitive factor κ, reference temperature 25 degrees Celsius) is used to correct the drift of the concentration value caused by changes in the temperature of the reaction liquid and output a temperature-corrected concentration value sequence; Further, a multi-channel cross-correlation analysis method (parameters: number of concentration signal channels 3, cross-correlation threshold 0.85) is used to evaluate the consistency of the concentration signals output by different sensors and obtain a cross-interference weight coefficient matrix; The temperature-compensated concentration value sequence is weighted and fused by the weight coefficient matrix to convert it into a corrected reaction liquid concentration value, achieving dual suppression of sensor drift and cross-interference and improving the accuracy of concentration measurement; For example, on a certain automobile copper-clad plate chemical deposition production line, an electrochemical impedance spectrum module with a frequency scanning range of 1Hz~100kHz and an excitation voltage of 10mV is used to collect reaction liquid concentration data, and the Randles model is selected as the standard for equivalent circuit fitting. The mean square error of the fitting result is 5e 7, and after FIR zero-phase filtering (cutoff frequency 22.5kHz), the high-frequency noise amplitude of the signal is reduced from the original 0.12 units to 0.02 units. The temperature compensation coefficient κ is set to 0.0025 concentration units / degree Celsius, and within the working temperature range of 25~45 degrees Celsius, the corrected concentration drift is controlled within 0.005 units. The cross-correlation values of the three-channel concentration signals are 0.89, 0.87, and 0.92, respectively, and the concentration value after weight fusion is reduced by about 70% in the long-term running, ensuring the high reliability of the subsequent process parameter matrix input; S1.3: Use thermocouples and infrared temperature measurement devices to collect reaction liquid temperature at multiple points simultaneously, and perform weighted average filtering on the collected data to obtain spatially representative deposition environment temperature characteristic values; S1.4: Collecting the flow rate signal of the reaction liquid based on the ultrasonic flow meter, and performing time domain sliding window integration and frequency domain fast Fourier transform processing on the flow rate signal to extract the flow rate fluctuation characteristic parameters; S1.5: Extracting the deposition time reference value according to the deposition process history record, and combining the contact angle data and surface roughness measurement value output by the substrate surface cleanliness detector to generate the substrate state information feature vector; Using a process database retrieval algorithm (parameters: historical deposition batch number, deposition environment classification label), the deposition process history record calling function is realized to obtain the verified deposition time data of different batches of sulfur-free copper-clad plate samples; Further, through a time series pattern recognition method (parameters: deposition time record sequence, process environment disturbance index), the deposition time reference value extraction function is realized, and reference time data representing the current working condition is obtained; Further, using the contact angle measurement method (parameters: surface cleanliness detector optical imaging data, droplet volume setting value), the surface wettability quantitative analysis is realized, and the static contact angle value is calculated through image segmentation and boundary fitting algorithm to generate surface cleanliness index data; Further, using the white light interferometry algorithm (parameters: probe scanning path, sample surface reflection signal), the three-dimensional morphology acquisition of the substrate surface roughness is realized, and the Sa value calculation based on the ISO 25178 standard is performed to generate the surface roughness quantitative data; Further, through the parameter feature fusion algorithm (parameters: deposition time reference value, contact angle data, Sa value data), the uniform coding of different source process parameters is realized, and they are combined into a one-dimensional feature vector as the output of the substrate state information feature vector; Through the processing method based on the feature fusion algorithm, the historical deposition time and surface physical characteristic index of the previous step are converted into substrate state information feature vectors that can be used as model input, realizing the correlation description effect of deposition time and substrate state; Exemplarily, under the condition that the production line of the sulfur-free copper-clad plate is classified as "high temperature and medium speed", the historical process database is searched and the deposition records corresponding to the batch numbers 20231105 and 20231106 are called to obtain the deposition time sequence data of 20.3 min and 20.5 min, respectively. By the time sequence pattern recognition method, the average value of the approximate working condition is selected as the reference, and the result is the reference value 20.4 min. In the contact angle test of the same batch of samples, the surface cleanliness detector sprays 2 μL of droplets, and the optical system is used to collect the droplet boundary image. The contact angle is calculated to be 78.2° by boundary fitting. In the white light interferometry, a scanning range of 5 mm x 5 mm is adopted, and the step interval is 10 μm. The Sa value is calculated to be 0.58 μm. The reference time 20.4 min, the contact angle 78.2° and the Sa value 0.58 μm are input into the feature fusion algorithm, and the substrate state information feature vector is output by linear normalization and code combination. By inputting the vector into the subsequent S1.6 structured splicing step, the matching accuracy of the deposition time and the substrate physical characteristics can be significantly improved, and the initial feature consistency of the multi-parameter coupled response model can be enhanced; S1.6: The modified reaction liquid concentration value, deposition environment temperature feature value, flow rate fluctuation feature parameter and substrate state information feature vector are structured and spliced to form an initial process parameter set matrix containing process parameters and state information.

[0011] The step S2: performing normalization processing and outlier rejection operation on the initial process parameter set to generate a standardized parameter matrix, eliminating the influence of different dimensions on subsequent modeling. Specifically, it includes: S2.1: Perform minimum-maximum normalization processing on the reaction liquid concentration, temperature and flow rate parameter data in the initial process parameter set to eliminate the interference of different parameter dimension differences on the modeling process and obtain a normalized parameter vector sequence; S2.2: Perform outlier detection on the normalized parameter vector sequence based on the Z-score method to calculate the standard deviation and mean value of each parameter dimension, identify and eliminate abnormal data points exceeding the 3σ threshold to improve the representativeness and modeling robustness of the data set; S2.3: Perform missing value interpolation processing on the parameter vector sequence after removing outliers, adopt an interpolation strategy based on K-Nearest Neighbors algorithm, and fill in the missing items based on the weighted average of similar samples to obtain a complete parameter matrix; For the parameter vector sequence after removing outliers, an interpolation strategy based on K-Nearest Neighbors (KNN) algorithm is adopted (parameters: neighbor number k, distance measurement method is Euclidean distance), to realize the identification of missing values and the selection of optimal neighborhood sample set; Further, by distance weighting calculation method (parameter: weight function is the inverse form of distance), the contribution of neighbor samples to missing items is selected and the weighting coefficient matrix for interpolation calculation is obtained. Further, the weighted average interpolation method (formula as follows) is used to realize accurate calculation of missing values:

[0012] wherein, ′ is the interpolated missing value, is the weighting coefficient of the i-th neighbor sample, is the corresponding parameter value of the i-th neighbor sample; Further, the missing items in each vector are replaced by the calculated interpolation values through matrix replacement operation, realizing the completion of the parameter vector sequence and generating a complete parameter matrix; Through the above K-nearest neighbor weighted average interpolation algorithm, the non-complete data after removing outliers in the previous step is converted into a standardized parameter matrix with complete structure, which can be directly used for subsequent covariance analysis, realizing the guarantee of data integrity and subsequent modeling accuracy. For example, for the reaction liquid parameter data matrix after removing outliers, set the neighbor number k to 5, the Euclidean distance as the similarity measurement method between samples, and the neighborhood sample distance of the missing value position as 0.3, 0.5, 0.4, 0.6, and 0.2, respectively. After inverse weighting processing, the weighting coefficients are 3.33, 2.00, 2.50, 1.67, and 5.00, respectively. According to the above formula, assuming that the corresponding neighbor reaction liquid concentration values are 1.25, 1.30, 1.28, 1.26, and 1.31 mol·L -1 , the interpolation calculation process is:

[0013] The calculation result obtains an interpolated concentration value of about 1.29 mol·L -1 , which replaces the original missing position to form a complete matrix with high stability and consistency index in the subsequent covariance calculation stage, significantly improving the modeling accuracy of the deposition dynamic fingerprint library; S2.4: Perform covariance matrix calculation on the complete parameter matrix, analyze the correlation between each parameter dimension, and identify highly correlated parameters for subsequent principal component analysis to provide dimension reduction basis. S2.5: Based on the eigenvalue decomposition result of the covariance matrix, construct the eigenvector basis of the standardized parameter matrix to form an orthogonal transformation matrix of the parameter space to support the dimension reduction operation of the subsequent principal component analysis.

[0014] For example, Figure 2As shown, step S3 involves constructing a dynamic deposition fingerprint database based on multiple sets of process parameter-performance correspondence data obtained from small-batch trial deposition experiments. This database includes a ternary mapping relationship between parameter combinations, deposition time and thickness distribution, surface roughness, and adhesion. Specifically, it includes: S3.1: Based on the process parameter combination data obtained from small-batch pilot deposition experiments, including reaction solution concentration, temperature, flow rate and deposition time, perform data structuring and organization operations to generate a parameter vector set to form the input dimension basis of the deposition dynamic fingerprint library; The input data includes raw records of reaction solution concentration, temperature, flow rate and deposition time collected from small-batch trial deposition experiments, which serve as the basis for the input dimensions of the deposition dynamic fingerprint database; A structured data modeling method (parameters: data field definitions include reaction solution concentration, temperature, flow rate, and deposition time) is adopted to achieve field standardization and type constraint mapping of the original experimental parameters, converting non-numerical records into a computable numerical format; Furthermore, by using a time series resampling algorithm (parameter: sampling interval is uniformly set to 1s), time alignment processing of data collected by multiple sensors is achieved, and a synchronous recording sequence of reaction liquid concentration, temperature, flow rate, and deposition time on a unified time reference is obtained; Furthermore, a multidimensional feature encoding method is adopted (parameter: normalization range is set to [0,1]) to realize the vectorized encoding of parameters with different physical dimensions and generate a parameter vector data structure containing four dimensions; Furthermore, a missing value interpolation algorithm (based on the K-nearest neighbor weighted average strategy, with a neighborhood size of k=5) is adopted to fill in the missing parameters caused by short-term communication interruption of the sensor and generate a parameter matrix that has passed integrity verification. By using parameter dictionary mapping and matrix transpose processing, the multidimensional parameter matrix from the previous step is transformed into a set of data vectors that adapt to the input dimension of the dynamic fingerprint database, thus realizing the structured input construction of parameter-performance mapping; For example, in a set of pilot deposition experiments, the reaction solution concentration was recorded as 0.25 mol / L to 0.35 mol / L by an online electrochemical impedance spectroscopy calibration sensor, the temperature was recorded as 293 K to 298 K by an infrared thermometer, the flow rate was measured as 0.55 m / s to 0.65 m / s by an ultrasonic flow meter, and the deposition time was set as 180 s to 240 s by the process control system. When standardizing the above data, "mol / L" was uniformly converted to "mol·L". -1 "Floating-point numbers and K values ​​are uniformly converted to Kelvin temperature scale floating-point numbers, and the unit of flow rate is uniformly set to m·s." -1When resampling time series records for different parameters, a sampling interval of 1 second is set. Linear interpolation is used to fill in missing time point data, so that each parameter forms a quadruple record on the same time base. A maximum-minimum normalization formula is used.

[0015] in These are the original sample values. The minimum value of this dimension. To achieve the maximum value in this dimension, the dimensions of multidimensional parameters are unified and normalized. The missing value imputation stage employs a K-nearest neighbor weighted average strategy, calculating the mean of the five samples with the smallest Euclidean distance to the missing record parameter as the imputation value, and performing data consistency verification. The final output parameter vector set contains multiple four-dimensional vectors, each vector element corresponding to the corrected reaction solution concentration, temperature, flow rate, and deposition time, providing complete and well-organized input data features for the subsequent mapping performance output of the deposition dynamic fingerprint database. S3.2: Thickness distribution detection was performed on the sulfur-free copper clad laminate samples obtained after each group of trial deposition experiments. The spatial distribution characteristics of the thickness of the deposited layer were measured by a step meter or X-ray fluorescence method, and a thickness distribution feature vector was generated as one of the output dimensions of the deposition dynamic fingerprint database. For sulfur-free copper-clad laminate samples obtained after small-batch trial deposition experiments, the thickness distribution detection subsystem's step meter or X-ray fluorescence measurement module is called to collect the thickness values ​​of the deposited layer at different locations, forming a spatial distribution raw data matrix. A two-dimensional spatial interpolation algorithm (parameters: bicubic interpolation kernel function, grid resolution 0.1 mm) is used to achieve continuous reconstruction of the original thickness sampling point data across the entire substrate surface, and obtain a high-resolution thickness distribution data map. Furthermore, by using a thickness distribution normalization algorithm (parameters: minimum thickness is set with reference to the minimum value of the entire sample, maximum thickness is set with reference to the maximum value of the entire sample), the uniformity of the thickness data scale is achieved, and the dimensional differences of thickness levels between different samples are eliminated. Furthermore, the formula for calculating the thickness uniformity index is adopted:

[0016] Combined with the thickness standard deviation, the uniformity definition formula is used:

[0017] in, The average thickness For thickness standard deviation, It serves as a thickness uniformity index, enabling the quantification of the degree of uniformity in the spatial distribution of thickness. Furthermore, based on a spatial distribution statistical vectorization algorithm (parameter: number of partitions N is 1 to 32, preferably 16 partitions), the continuous thickness distribution data is divided into preset regions, the mean thickness and fluctuation coefficient of each region are calculated, and the results are concatenated into a structured thickness distribution feature vector. This feature vector includes the global mean, overall fluctuation, and thickness changes in local regions, serving as one of the output dimensions of the deposition dynamic fingerprint database; Through the above spatial interpolation, normalization, uniformity index calculation and feature vectorization processing, the original thickness distribution measurement results are transformed into high-dimensional structured thickness feature data that can be used for model training, thereby significantly improving the accuracy of sedimentation performance characterization and data utilization. For example, in a pilot deposition experiment, a profilometer was used to measure the deposition layer thickness at 256 coordinate points on the substrate surface. The measured values ​​ranged from 18.2 μm to 22.5 μm, with an average value of... μm, standard deviation μm. A bicubic interpolation algorithm was used to reconstruct a two-dimensional thickness distribution map with 0.1 mm spacing from the data points. After min-max normalization, the uniformity index was calculated as follows: ≈0.958. The global mean, global fluctuation coefficient, and thickness mean of 16 local regions were arranged according to the region sequence to obtain a thickness distribution feature vector of length 18, which was input into the deposition dynamic fingerprint database. This data significantly improved the accuracy of thickness prediction and the effect of deposition uniformity control in the subsequent training of multi-parameter coupled response models. S3.3: Surface roughness is detected for each group of deposition samples. Surface morphology data is collected using a white light interferometer or profilometer, and the Sa value is calculated based on the ISO 25178 standard to generate surface roughness characteristic parameters, which serve as one of the output dimensions of the deposition dynamic fingerprint database. The input conditions are the unique identifier information of each group of sediment samples, the corresponding sedimentation process parameters, and the thickness distribution feature vector obtained through thickness detection; The white light interferometer measurement method (parameters: measurement wavelength range 400~700nm, vertical resolution 1nm) is used to achieve full-domain scanning of the micro-morphology of the deposited sample surface and generate raw interferometric data containing the height field matrix. Furthermore, by using a profilometer measurement method (parameters: probe tip diameter 2μm, measurement step 0.5μm), linear profile curves of the same surface are acquired, and a sequence of height values ​​of spatial sampling points is obtained to supplement the local curvature information of the interference data; Furthermore, through a topography data fusion algorithm (parameter: weighting coefficients are determined by the overlap rate of the measurement area), spatial interpolation fusion of white light interferometric data and contour curve data is achieved, and a three-dimensional surface topography dataset in a unified coordinate system is generated. Furthermore, the surface roughness calculation method based on ISO 25178 (parameters: sampling area 1 mm², removal of morphology trend term) was used to calculate the Sa value of the three-dimensional surface morphology dataset.

[0018] S3.4: Perform adhesion tests on each group of deposited samples, evaluate the adhesion strength between the deposited layer and the substrate using the cross-cut method or the tensile method, and quantify it into adhesion strength level or tensile strength value to generate adhesion strength characteristic parameters, which serve as one of the output dimensions of the deposition dynamic fingerprint library. For the bonding strength evaluation process of each group of deposited samples, the cross-cut test or tensile test is used as the detection method to ensure the comparability of the bonding strength data between the deposited layer and the substrate under different process conditions. When using the cross-cut test, several parallel cuts with uniform spacing are made on the surface of the deposit layer to form a regular grid area. The adhesion ratio is calculated by using the coating peeling at the intersection of the cuts, and the adhesion strength level data is generated. Furthermore, during the tensile test, a standard tensile fixture is installed at the end of the sample, and the tensile rate is controlled to apply an axial load until the interface between the deposited layer and the substrate fails. The failure load value is recorded and converted into tensile strength. A digital signal acquisition module is used to synchronously acquire the original test signals generated by the cross-cut test and the tensile test, and low-pass filtering is performed to eliminate high-frequency noise interference, so as to obtain a smoothed bonding force test curve. Furthermore, the load value at the bonding failure point is extracted using a peak detection algorithm, and the bonding force characteristic parameters are calculated in conjunction with the sample cross-sectional area. The calculation formula is as follows:

[0019] in, For the characteristic parameters of bonding force, This is the failure load value. This represents the cross-sectional area of ​​the sample. Furthermore, by using data normalization processing methods, the dimensionality of the bonding force characteristic parameters obtained from different testing methods is unified, and the cross-cut method grade value and the tensile method strength value are mapped to the same numerical range to form a set of bonding force parameters that can be fused. A feature fusion algorithm is used to perform weighted averaging of the two types of binding force feature parameters. The weight coefficients are set according to the detection accuracy and stability of each method to generate comprehensive binding force feature parameters. The above algorithm is used to convert the bonding force test signal from the previous step into bonding force feature data of the dynamic fingerprint database, thereby achieving standardization and structuring of the bonding force output dimension. For example, in the testing of a batch of deposited samples, the cross-cut method used a 6×6 grid with a cut spacing of 1 mm. Four grids were detected as detached, and the bonding strength grade, after mapping normalization, was 0.92. In the tensile test, the sample cross-sectional area was... mm², the failure load was extracted as by the peak detection algorithm. N, the characteristic parameter of the bonding force is calculated as follows: MPa. A weighted average fusion algorithm was used, with the cross-grid method weight coefficient set to 0.4 and the tensile method weight coefficient set to 0.6. The comprehensive bonding force characteristic parameter results are as follows: (Results after unifying numerical ranges). After structuring, the bonding force characteristic parameters of this batch of samples, along with the corresponding thickness distribution parameters and surface roughness parameters, were entered into the deposition dynamic fingerprint database. Verification showed that the relative deviation of the bonding force characteristic data between different testing methods was significantly reduced, achieving stability and repeatability of bonding force data fusion. S3.5: Combine the parameter vector set with the corresponding thickness distribution feature vector, surface roughness feature parameter, and bonding force feature parameter through a ternary mapping to construct a deposition dynamic fingerprint database table structure, so as to form a structured mapping relationship between deposition time, reaction conditions, and deposition performance. The structured parameter vector set, thickness distribution feature vector, surface roughness feature parameters and bonding force feature parameters are used as input data objects. A multi-dimensional feature indexing method (parameter: quadruplet index key, key value includes reaction solution concentration, temperature, flow rate and deposition time) is used to achieve matching indexing for different process conditions and performance indicators. Furthermore, by using a relational database schema design method (parameters: ternary mapping main table, foreign key association performance index sub-table), the logical association between process parameter combinations and deposition performance data is realized, and a searchable table structure model is obtained; Furthermore, a data schema normalization algorithm (parameter: third normal form constraint rule) is adopted to achieve redundancy elimination and consistency maintenance of process parameter fields and performance index fields in the database table, and to generate table structure definitions that conform to process data management standards; Furthermore, based on the numerical domain characteristics of performance indicators, a field type optimization algorithm (parameters: floating-point precision setting for thickness distribution field, double-precision setting for roughness field, and integer level setting for bonding force field) is used to achieve a balance between storage efficiency and data precision, and to generate an optimized storage field configuration. Furthermore, a multidimensional indexing algorithm (parameters: B+ tree index combined with hash key) is adopted to achieve rapid combination retrieval of deposition time and reaction liquid parameters, thereby improving the data retrieval efficiency for subsequent model training; By using mapping combination and index optimization processing, the parameters and performance data from the previous step are transformed into a structured deposition dynamic fingerprint database table, realizing a stable mapping relationship between deposition time, reaction conditions and deposition performance. For example, in the sulfur-free copper clad laminate chemical deposition process, an experimental group with a reaction solution concentration of 1.5 mol / L, a temperature of 60℃, a flow rate of 1.2 m / s, and a deposition time of 35 min was selected. Its thickness distribution feature vector was configured as [12.3, 12.5, 12.4] μm, surface roughness feature parameter as 0.85 μm, and adhesion feature parameter as 145 MPa, forming a parameter quadruple index key. This index key is recorded in the main table of a relational database, while a foreign key is linked to a performance index sub-table. When using third normal form normalization, the process parameters and performance indicators are separated into different tables to reduce redundant storage. The field type optimization algorithm sets the thickness distribution field to float(5,2), the roughness field to double, and the adhesion field to int. The multidimensional index construction algorithm builds a B+ tree index for the index key in the main table and a hash index for the deposition time field, enabling queries for deposition times in the 30-40 min range to return matching results in milliseconds. In this embodiment, the model training task can quickly retrieve 20 samples that meet the interval conditions from the database, significantly shortening the model training time and significantly improving the process optimization iteration cycle.

[0020] like Figure 3 As shown, step S4 involves performing principal component analysis to reduce the dimensionality of the standardized parameter matrix in the deposition dynamic fingerprint database, extracting key feature factors affecting deposition performance, and training a multi-parameter coupled response model based on the support vector regression algorithm. Specifically, this includes: S4.1: Perform principal component analysis (PCA) dimensionality reduction on the standardized parameter matrix in the deposition dynamic fingerprint database to extract the principal component loading matrix of deposition performance influencing factors, so as to reduce parameter redundancy and retain key process features; S4.2: Calculate the contribution weight vector of each parameter in the deposition performance based on the principal component loading matrix to quantify the influence of different process parameters on thickness distribution, surface roughness and adhesion. S4.3: Use the contribution weight vector to perform a weighted linear combination transformation on the original standardized parameter matrix to generate a set of deposition process feature vectors in a low-dimensional feature space, so as to improve the model training efficiency and generalization ability. The input consists of the contribution weight vectors of each parameter to the deposition performance obtained after principal component analysis and the original standardized parameter matrix. The weighted mapping product algorithm (parameters: contribution weight vector w, original standardized parameter matrix X) is used to weight and correct each parameter dimension according to its proportion of influence on deposition performance, resulting in a weighted matrix. ; Furthermore, through a linear combination transformation method (parameter: weighting matrix) The selected principal component eigenvector set P is used to project the weighted parameter matrix onto a low-dimensional feature space, thus obtaining the transformation matrix. ; Furthermore, a feature normalization processing algorithm (parameter: transformation matrix T) is adopted to normalize the low-dimensional feature vectors, so that each feature component has a uniform scale range, thereby eliminating the differences in feature dimensions of different principal components. The method reconstructs the feature vector set (parameter: normalized low-dimensional feature matrix). This generates a set of feature vectors for the deposition process in a low-dimensional feature space, and ensures that each sample maintains the representativeness of deposition performance information in the new feature space. By using weighted linear combination and low-dimensional mapping, the contribution weights and the original standardized matrix from the previous step are transformed into a low-dimensional feature vector set of the deposition process, thereby simultaneously improving the model training efficiency and generalization ability. For example, in one implementation, the contribution weight vector w takes the value [0.42, 0.35, 0.23], and the original standardized parameter matrix X is a 100×3 dimension matrix, where the first column is the concentration standardized value, the second column is the temperature standardized value, and the third column is the flow rate standardized value. The weighted matrix is ​​calculated using a weighted mapping product algorithm. :

[0021] Matrix multiplication, with column-wise weights, yields a weighted version of a 100×3 matrix. The eigenvector set P (a 3×2 matrix) of the first two principal components obtained from principal component analysis is then selected, and a linear combination transformation is performed. This yields a low-dimensional feature matrix of 100×2. Z-score normalization is then performed on each column of T to generate the normalized feature matrix. Matrix. As the feature vector set of the deposition process in the low-dimensional feature space is input into the support vector regression model, the prediction stability of the validation set after training is significantly improved, and the model can maintain high consistency between deposition time and performance indicators under different deposition conditions. S4.4: Based on the feature vector set of the deposition process and the corresponding deposition time and performance index data, construct a training sample set to serve as the input-output pair of the support vector regression (SVR) model; Based on the low-dimensional sedimentation process feature vector set obtained in the previous step and the corresponding sedimentation time and performance index data, the input and output mapping relationship required for model training is generated by using a data pairing organization method. The feature vector index matching algorithm (parameters: feature vector ID, timestamp index) is used to achieve precise association between the feature vectors of the deposition process and the corresponding deposition time records, so that each feature vector is uniquely mapped to a deposition time sample unit; Furthermore, by using a performance index aggregation algorithm (parameters: thickness distribution characteristic value, surface roughness parameter value, and bonding force measurement value), the deposition time sample unit is bound to the multi-dimensional performance parameter output, and a three-dimensional data record structure is obtained. Furthermore, a unified time-scale resampling method (parameters: target time step Δt, interpolation method: cubic spline interpolation) is adopted to align the performance parameters of sample units with different deposition times on a unified time dimension and generate a smoothed performance sequence. Furthermore, based on the input specification of the support vector regression model, the smoothed sedimentation process feature vector set is used as the input matrix X, and the corresponding sedimentation time and performance index are jointly encoded into the output matrix Y, thereby realizing the structured construction of input-output pairs; The sample consistency verification algorithm (parameter: Mahalanobis distance threshold) is used. This transforms the results of the previous step into a training sample set that meets the requirements of SVR modeling, achieving the technical effect of data integrity and consistency during the model training phase. For example, in a copper-clad laminate deposition process for automotive electronics, a low-dimensional feature vector set with a dimension of 6 has been obtained. The deposition time ranges from 0 to 1800 seconds, and the performance indicators include four items: mean thickness, thickness variance, surface roughness Sa, and adhesion grade. A feature vector index matching algorithm is used to establish a one-to-one correspondence between the feature vector set and the deposition time records, for example, feature vector numbering. The corresponding deposition time is 900 seconds. Through a performance metric aggregation algorithm, [the following data is collected / sorted]. The corresponding mean thickness of 5.2 μm, variance of 0.03 μm², Sa value of 27 nm, and bonding strength level of 3 were bound together into a complete output record. Based on a resampling step size of Δt = 60 seconds, a cubic spline function was used to interpolate the mean thickness and Sa value over time, ensuring a smooth and continuous performance curve at a uniform time scale. The 6-dimensional input matrix X and the 4-dimensional output matrix Y were reassembled into a training sample set according to SVR requirements, and the Mahalanobis distance of each sample was calculated, removing outliers with a distance greater than 3.5. The training dataset obtained after the above processing maintained the accurate correspondence between high-dimensional features and multi-objective performance indicators. After being used for SVR model training, it significantly improved the stability and accuracy of deposition time prediction under different working conditions. S4.5: The radial basis function (RBF) kernel function is used to perform support vector regression modeling on the training sample set to establish a multi-parameter coupled response model between the concentration of the deposition reaction solution, temperature, flow rate and deposition time; S4.6: Perform cross-validation and hyperparameter tuning on the multi-parameter coupled response model to obtain the optimal combination of model parameters, and evaluate the prediction accuracy and stability of the model under different working conditions; S4.7: Deploy the optimized multi-parameter coupled response model into the deposition control system to achieve dynamic matching and closed-loop control of deposition time and reaction liquid parameters.

[0022] Step S5: Real-time acquisition of reaction solution concentration, temperature, and flow rate parameters during the current deposition process, and input of the real-time parameter data into a multi-parameter coupled response model to generate a deposition performance prediction curve. Specifically, this includes: S5.1: Online sensor acquisition of reaction solution concentration, temperature, and flow rate parameters during the chemical deposition of sulfur-free copper clad laminates to obtain multi-dimensional process input data for the current deposition stage; In the chemical deposition process of sulfur-free copper clad laminate, for the reaction liquid parameters at the current deposition stage, a multi-modal online sensor group deployed on the chemical deposition production line is selected as the data acquisition component, and a combination of reaction liquid concentration sensor, temperature measurement module and flow rate monitoring device constitutes a multi-dimensional process input channel. An online concentration sensor based on electrochemical impedance spectroscopy (parameter settings: frequency scan range 0.1Hz~1MHz, amplitude 10mV) is used to achieve continuous sampling of the concentration of the reaction solution. Furthermore, by performing an amplitude-frequency characteristic fitting algorithm on the concentration sampling signal (parameter: automatic optimization of second-order fitting coefficients), transient concentration disturbances are suppressed, and a stable original concentration data sequence is obtained. A multi-point thermocouple array and infrared temperature measurement combination device (parameters: thermocouple distribution spacing 5cm, infrared temperature measurement accuracy ±0.1℃) are used to realize the function of spatial synchronous acquisition of reaction liquid temperature; Furthermore, the original sequence of temperature spatial distribution is generated by using a weighted coefficient calculation method based on multi-point temperature data (the weights are determined by the contribution of the location to the main flow path of the reaction liquid). A dual-channel ultrasonic flow meter (parameters: sampling frequency 50Hz, accuracy ±0.5%) is used to achieve instantaneous acquisition of the reaction liquid flow rate signal; Furthermore, by performing short-time autocorrelation analysis on the flow velocity signal (parameter: window length 200ms), the stationarity benchmark data of the flow velocity in the time domain is extracted, and the original flow velocity sampling sequence is generated; Through the above-mentioned multi-sensor acquisition link of concentration, temperature, and flow rate, the signal from the previous step is processed by unified timestamp alignment (parameter: synchronization deviation tolerance 5ms) and transformed into complete multi-dimensional process input data of the current deposition stage, realizing the structured input of real-time data stream of the deposition process; For example, in a certain batch of automotive electronic copper-clad laminate deposition, the frequency scan of the online concentration sensor using electrochemical impedance spectroscopy ranged from 0.1 Hz to 1 MHz, with an amplitude of 10 mV, resulting in an original concentration sampling sequence of 5000 points. A multi-point thermocouple array was arranged in the reaction tank at a spacing of 5 cm, collecting temperature data at 12 locations. Combined with an infrared temperature measurement module, data with an accuracy of ±0.1℃ was collected in key surface areas, resulting in an original temperature matrix of 12 × 1000. A dual-channel ultrasonic flow meter collected data at a frequency of 50 Hz with an accuracy of ±0.5%, calculating the autocorrelation coefficient within each 200 ms window to assess flow rate stability. The three types of data were aligned using a unified timestamp, with deviations controlled within 5 ms, forming a three-dimensional synchronous input matrix of concentration, temperature, and flow rate. This provides a complete real-time process input basis for subsequent normalization and outlier removal in S5.2. S5.2: Perform normalization and outlier removal operations on the collected raw data of reaction solution concentration, temperature, and flow rate to obtain a standardized real-time process parameter vector and eliminate the influence of different physical dimensions on the model input; S5.3: Based on the key feature factors extracted after dimensionality reduction by principal component analysis, perform feature mapping transformation on the standardized real-time process parameter vector to generate a low-dimensional feature vector suitable for the input of a multi-parameter coupled response model; S5.4: Input the low-dimensional feature vector into the multi-parameter coupled response model trained based on the support vector regression algorithm to predict the changing trends of deposition performance indicators such as thickness distribution, surface roughness, and bonding force at the current deposition time; S5.5: Based on the predicted values ​​of deposition performance indicators output by the multi-parameter coupled response model, construct a prediction curve of the evolution of deposition performance over time, and form a dynamic response relationship map between thickness uniformity and deposition rate. The input conditions are the prediction results of a multi-parameter coupled response model trained based on the support vector regression algorithm. The prediction results include the deposition performance index values ​​of thickness distribution, surface roughness, and bonding force at the current deposition time. A time series interpolation algorithm (parameters: sampling interval 0.5s, interpolation kernel function type cubic spline) is used to realize the continuous processing of discrete prediction points on the deposition time axis and obtain a continuous time series of deposition performance indicators. Furthermore, by using a multidimensional curve fitting method (parameter: the fitting function type is a three-dimensional Gaussian mixture function), the thickness distribution, roughness, and bonding force are fitted in a unified time coordinate system, and a multidimensional time evolution function set of deposition performance indicators is generated. Furthermore, a normalization transformation method (parameter: normalization range of maximum and minimum values ​​[0,1]) is adopted to unify the dimensions of each performance index, and the thickness distribution normalization curve T(t), roughness normalization curve R(t) and bonding force normalization curve B(t) are obtained. Furthermore, a joint visualization mapping algorithm (parameters: two-dimensional mapping mode, color-coded thickness uniformity, curve thickness-coded deposition rate) is used to construct a dynamic response relationship map between thickness uniformity and deposition rate, and generate a deposition performance time evolution map for subsequent optimization. By curve fitting and normalization, the predicted output of the multi-parameter coupled response model is transformed into an analytical curve of the deposition performance changing over time, thereby achieving an intuitive quantitative expression of the relationship between thickness uniformity and deposition rate, and supporting the optimized calculation of the deposition time threshold. For example, in a deposition process test of sulfur-free copper clad laminate, the model predicted the thickness distribution, roughness, and adhesion values ​​every 1 second from 0s to 600s. The thickness ranged from 24.8μm to 26.3μm, the roughness ranged from 0.12μm to 0.26μm, and the adhesion ranged from 155MPa to 182MPa. A time-series interpolation algorithm with a sampling interval of 0.5s was used to obtain a smooth time-series curve. The thickness curve fitting function was a three-dimensional Gaussian mixture function with a goodness of fit R² of 0.98. A normalization transformation was used to map the thickness T(t) to the [0,1] interval, where T(t) reached a uniformity peak at t=420s, the roughness R(t) remained low between t=410s and 430s, and the adhesion B(t) showed a slow upward trend throughout the process. During the joint visualization process, color coding shows that the peak of thickness uniformity is highly consistent with the flat area of ​​the rate curve, enabling high-precision positioning of the deposition time control window and ultimately providing accurate input curve data for the multi-objective optimization in step S6.

[0023] Step S6: Calculate the optimal deposition time threshold based on the deposition performance prediction curve. This optimal deposition time threshold satisfies the weighted optimization conditions of thickness uniformity and production efficiency. Specifically, it includes: S6.1: Based on the deposition performance prediction curve, normalized weighting is performed on the predicted thickness distribution, surface roughness and bonding force indices to obtain a comprehensive deposition quality evaluation index. S6.2: Perform sliding window correlation analysis on the comprehensive sedimentation quality assessment index and sedimentation time series to identify the inflection point interval of sedimentation quality, so as to determine the key control time domain in the sedimentation process; S6.3: Based on the preset thickness uniformity target threshold and production efficiency constraints, a multi-objective optimization function for deposition time is constructed, with deposition time as the decision variable and quality and efficiency as the objective functions for joint optimization; Based on the input data of the thickness uniformity target threshold and production efficiency constraints, a multi-objective mathematical programming method is adopted (parameters: deposition time t, thickness uniformity index). Production efficiency index This enables the establishment of a joint optimization model for deposition time. Furthermore, the method is constructed through the objective function (parameters: Target ≥ Preset Threshold , Target ≥ Preset Threshold This allows for the transformation of thickness uniformity and production efficiency into quantifiable optimization objectives, and yields initial expression data for the objective function. Furthermore, a normalized weighted processing method is adopted (parameter: weight coefficient). ,satisfy This allows for the balancing of quality and efficiency objectives with different dimensions, and the generation of dimensionless comprehensive target indicators. The calculation formula is as follows:

[0024] Furthermore, a method for constructing constraints (parameter: thickness uniformity coefficient) is employed. ≤ Deposition period T≤ This allows for the determination of the feasible solution space and the generation of the allowable interval for the deposition time variable t. By constructing a multi-objective optimization function, the quality and efficiency objective indicators of the previous step, which are normalized and weighted, are transformed into an iteratively solvable optimization model, thereby realizing a multi-objective optimization structure with deposition time as a single decision variable and thickness uniformity and production efficiency as joint objectives. For example, the deposition performance prediction curve shows the thickness uniformity index at t=42 min. =0.93, Productivity Index =128cm² / h, preset target threshold =0.90, =120cm² / h, weighting coefficient set to =0.6、 =0.4. Substitute into the formula:

[0025] The calculation result is This significantly improves the overall target value and meets the requirements. =0.05mm and The constraint is 50 min. This output result shows that a significant improvement in thickness uniformity and production efficiency can be achieved simultaneously at t=42 min, providing accurate initial solution conditions for subsequent Pareto front optimization. S6.4: The Pareto front optimization algorithm is used to solve the multi-objective optimization function of sedimentation time, and the Pareto optimal solution set of sedimentation time is generated to characterize the optimal trade-off between quality and efficiency. The input conditions include a multi-objective optimization function for deposition time constructed by step S6.3. The function takes deposition time as the decision variable and includes a comprehensive evaluation index of deposition quality and a production efficiency index as joint optimization objectives. The Pareto front optimization algorithm (parameters: multi-objective optimization function f(t), time search interval Δt, iteration step size ε) is used to traverse and solve the objective space, and record the non-dominated solution sets of quality and efficiency indices under different deposition times; Furthermore, by using a non-dominated ranking method (parameters: quality index Q, efficiency index E), the optimization results are classified into levels, and a set of candidate deposition time values ​​is obtained within each level, ensuring that each candidate value is better than other values ​​in at least one objective. Furthermore, the distribution uniformity of sedimentation time samples within the same non-dominant level is assessed using the crowding distance calculation method (formula below), and a highly dispersed solution set is generated:

[0026] in, Let be the crowding distance of the i-th solution. For quality indicators, For efficiency indicators, The maximum quality index within the solution set interval. This represents the minimum quality index within the solution set interval. This is the maximum efficiency index within the solution set interval. The minimum efficiency index within the solution set interval; Furthermore, by employing a joint screening strategy (parameters: non-dominant level, crowding distance threshold δ), the sedimentation time solution set is simplified, eliminating solutions with excessively low crowding, in order to retain representative solutions that cover the target spatial boundary. By combining the Pareto front optimization algorithm with non-dominated sorting and crowding screening, the solution result of the multi-objective optimization function in the previous step is transformed into the Pareto optimal solution set of deposition time, so as to achieve a quantitative characterization of the optimal trade-off between deposition quality and production efficiency. For example, in the optimization scenario of sulfur-free copper clad laminate chemical deposition process, the deposition time search interval is set to 40 min to 80 min, the iteration step size is 1 min, the quality index is the thickness uniformity comprehensive index (normalized 0.85~0.94), and the efficiency index is the output per unit time (normalized 0.60~0.88). The comprehensive weight of the multi-objective optimization function is determined by step S6.3, and the Pareto front optimization algorithm is input for 100 iterations. Non-dominated sorting yields 25 first-order front solutions. In the crowding distance calculation, δ is set to 0.15. After screening, 14 solutions are retained. These solutions all exhibit boundary distribution characteristics in the quality and efficiency spaces. In this solution set, the deposition time distribution is concentrated in the range of 56 min, 59 min, 62 min, and 65 min, with corresponding thickness uniformity values ​​above 0.92 and output per unit time above 0.80, significantly improving the stability of the deposition process and the production line utilization rate. S6.5: Perform weighted sorting on the Pareto optimal solution set of deposition time based on user-defined priority weights, and select the optimal deposition time threshold as the output to achieve closed-loop control input for the deposition process; When performing a priority-based weighted sort on the Pareto optimal solution set for deposition time, the input conditions include the candidate solution set for deposition time obtained by step S6.4 and the set of quality and efficiency weight parameters set by the user. The weighted normalization method is adopted (parameter: user-input thickness uniformity weight). Production efficiency weight This process standardizes the proportions of each weight within the interval [0,1], resulting in normalized weight coefficients used for sorting operations. Furthermore, a weighted comprehensive evaluation algorithm (parameter: sedimentary quality comprehensive index in the Pareto solution set) is used. With production efficiency value This process calculates the comprehensive score for each solution and obtains the corresponding weighted score sequence. The weighted score calculation formula is as follows:

[0027] in, The comprehensive score for the deposition time solution. These are the normalized weighting coefficients for the quality indicators. The normalized weighting coefficients for efficiency indicators. This is the comprehensive index of thickness uniformity corresponding to this solution. This is the production efficiency value corresponding to the solution; Furthermore, through a sorting algorithm (parameter: rating sequence) This allows for prioritization from highest to lowest overall score, and generates a candidate list of deposition times. Furthermore, by selecting an algorithm (parameter: score sequence sorting result), the deposition time with the highest score in the sorted list is selected as the optimal deposition time threshold, and the result is output to the input of the closed-loop control module of the process control system. By using the weighted sorting process described above, the Pareto optimal solution set result from the previous step is transformed into a single-valued deposition time threshold, thereby achieving the expected technical effect of balancing quality and efficiency. For example, in a sulfur-free copper clad laminate deposition process, the Pareto optimal solution set includes candidate deposition time values ​​of [480, 500, 520] seconds, with corresponding thickness uniformity indices of [0.92, 0.96, 0.94] and production efficiency values ​​of [0.88, 0.85, 0.90]. The user sets the thickness uniformity priority weight. Production efficiency priority weight After weight normalization, the weight coefficients remain unchanged. , Substituting into the weighted scoring formula, the score for a deposition time of 480 seconds is calculated. The score for a deposition time of 500 seconds. A score for a deposition time of 520 seconds. The scores were ranked as follows: 520 seconds > 500 seconds > 480 seconds. The 520-second score, with the highest score, was selected as the optimal deposition time threshold and output to the control system. Under this configuration, the thickness uniformity index at the final state of the deposition process remained above 0.94, and the production efficiency index was above 0.90, achieving a significant improvement in deposition performance and production efficiency.

[0028] Step S7: Input the optimal deposition time threshold into the deposition control system to dynamically adjust the process parameters, so that the reaction solution concentration, temperature, flow rate, and deposition time form a closed-loop control. Specifically, this includes: S7.1: Based on the deposition performance prediction curve and the optimal deposition time threshold, a deposition control parameter adjustment mapping table is constructed. The deposition control parameters include reaction solution concentration, temperature, flow rate and deposition time. A fuzzy control algorithm is used to dynamically prioritize the changing trends of multiple parameters. The input conditions are the optimal deposition time threshold obtained based on step S6, and the deposition performance prediction curve data output by step S5, including the dynamic trend matrix of thickness distribution, surface roughness and bonding force. A multi-dimensional process parameter analysis method (input parameters: deposition performance prediction curve, multi-objective optimization results) is adopted to realize the quantitative correlation mapping of control objects such as reaction solution concentration, temperature, flow rate and deposition time, and generate an initial parameter change vector set. Furthermore, by using the gradient sensitivity analysis method (parameter: slope matrix of the prediction curve), the immediate impact of each control parameter on the changes in deposition performance is measured, and a multi-parameter dynamic sensitivity coefficient table is obtained as a preliminary basis for priority ranking. Furthermore, a sedimentation control parameter adjustment mapping table is constructed, and a fuzzy control algorithm (membership function type: triangle and trapezoid combination, rule base size is 10 to 50, preferably not less than 25) is used to realize the fuzzy inference mapping of multi-parameter change trends and generate a control parameter priority sorting index. Furthermore, through fuzzy priority fusion processing (weight combination method: weighted average method, weight source: fusion of user settings and historical data), the dynamic priority sequence is output and converted into a set of parameter adjustment instructions that the control system can resolve; By using fuzzy control algorithm mapping and priority sorting, the curve analysis results of the previous step are transformed into a multi-parameter adjustment sequence, so as to achieve the expected control effect of the deposition control system under the optimal time threshold. For example, in a chemical deposition scenario for a batch of sulfur-free copper-clad laminates, the acquired performance prediction curves show that the thickness distribution tends to saturate after 180 s of deposition time, the roughness index shows a significant decreasing trend in the 150 s to 200 s range, and the adhesion reaches its peak at 170 s. The optimal deposition time threshold is calculated to be 175 s. Gradient sensitivity analysis yields a concentration sensitivity coefficient of 0.42, a temperature sensitivity coefficient of 0.35, and a flow rate sensitivity coefficient of 0.23. Fuzzy control defines membership functions for the three input variables: concentration, temperature, and flow rate, with the concentration membership degree reaching a high range of 0.8. 1.0, the high membership degree range for temperature is 0.75. 1.0, with a membership degree of 0.4 in the velocity range. 0.7. The priority order of the rule base inference output is concentration > temperature > flow rate. A weighted average method is used to combine user weights (concentration 0.5, temperature 0.3, flow rate 0.2) with historical process weights to generate the final priority sequence [concentration, temperature, flow rate]. This sequence is successfully input into the control system, causing the reaction solution to prioritize concentration adjustment within a specified time, then synchronously correct the temperature, and finally fine-tune the flow rate. This significantly improves the uniformity of deposition thickness and maintains the final binding force at its peak level. S7.2: Based on the priority order in the deposition control parameter adjustment mapping table, initialize the parameters of the PID control module in the deposition control system, input the sensor sampling values ​​of the current reaction solution concentration, temperature and flow rate, and generate the control output increment of each parameter; S7.3: Input the control output increment to the actuator of the reaction liquid supply system to perform closed-loop feedback control on the reaction liquid concentration control valve, temperature heating module and flow rate regulating pump, so that the reaction liquid parameters dynamically approach the set target value and form a multi-parameter coordinated adjustment response; S7.4: Before the deposition time reaches the optimal deposition time threshold, the deposition time control window is dynamically adjusted based on real-time feedback of reaction solution concentration, temperature and flow rate data. A sliding time window mechanism is used to control the deposition process in segments to improve deposition uniformity. S7.5: The deposition time control window is coupled with the reaction liquid parameter control output to form a closed-loop control loop. Based on the deposition process status feedback data, the deposition time and reaction liquid parameters are collaboratively corrected to maintain the stability and consistency of deposition performance.

[0029] Step S8: Collect adjusted deposition process data and final quality data, perform online incremental learning on the deposition dynamic fingerprint database and multi-parameter coupled response model, and update the model parameter weights and feature factor mapping relationship. Specifically, this includes: S8.1: Collect the adjusted deposition process data and final quality data. The deposition process data includes real-time data on reaction solution concentration, temperature, and flow rate. The final quality data includes the detection results of deposition layer thickness distribution, surface roughness, and bonding force, in order to form an incremental sample set. S8.2: Perform data cleaning and normalization on the incremental sample set, and unify the scale of each dimension parameter based on the Z-score standardization method to eliminate the influence of the difference in scale on model training and obtain a standardized incremental sample set. S8.3: Based on the standardized incremental sample set, perform feature mapping consistency verification, and use the principal component analysis algorithm to compare the feature space distribution of the current sample with the existing feature space of the deposition dynamic fingerprint database to determine whether it conforms to the existing process knowledge structure. S8.4: If the feature space consistency check passes, the standardized incremental sample set is incorporated into the deposition dynamic fingerprint database, and the parameter combination-deposition time-performance index ternary mapping relationship of the deposition dynamic fingerprint database is updated to expand the model training data base. S8.5: Based on the expanded deposition dynamic fingerprint database, an incremental learning strategy is adopted to update the parameter weights of the multi-parameter coupled response model online. The kernel function parameters and support vector set are adjusted by the incremental learning mechanism of the support vector regression model to improve the model's adaptability to new process conditions. S8.6: Perform performance verification on the updated multi-parameter coupled response model, input the historical validation set and the new sample set for cross-validation, calculate the model prediction accuracy index and stability index, and determine whether the model meets the prediction error threshold required for process control. S8.7: If the model validation results meet the error threshold requirements, the updated multi-parameter coupled response model will be deployed to the deposition control system to replace the original model for the next round of deposition time prediction and control, so as to achieve closed-loop iterative optimization of the process model.

[0030] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0031] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. A process for treating a non-sulfur copper clad board for automotive panels, characterized in that, The method comprises the following steps: S1: Collecting reaction liquid concentration, temperature, and flow rate parameter data during the chemical deposition process of sulfur-free copper-clad plate, recording corresponding deposition time reference values and substrate state information, and forming an initial process parameter set; S2: Performing normalization processing and outlier rejection operation on the initial process parameter set to generate a standardized parameter matrix; S3: Based on the multiple sets of process parameter-performance corresponding data obtained from small-batch trial deposition experiments, a deposition dynamic fingerprint library is constructed, which contains the ternary mapping relationship of parameter combination, deposition time and thickness distribution, surface roughness, and bonding force; S4: Performing principal component analysis dimension reduction processing on the standardized parameter matrix in the deposition dynamic fingerprint library, extracting key feature factors affecting deposition performance, and training a multi-parameter coupled response model based on a support vector regression algorithm; S5: Real-time collection of reaction liquid concentration, temperature, and flow rate parameter data during the current deposition process, and inputting the real-time parameter data into the multi-parameter coupled response model to generate a deposition performance prediction curve; S6: According to the deposition performance prediction curve, the optimal deposition time threshold is calculated; S7: Inputting the optimal deposition time threshold into the deposition control system to perform dynamic adjustment of process parameters; S8: Collecting adjusted deposition process data and final state quality data, performing online incremental learning on the deposition dynamic fingerprint library and the multi-parameter coupled response model, and updating model parameter weights and feature factor mapping relationships.

2. A process for treating a sulphur-free copper clad board for use in automotive vehicle panels according to claim 1, characterised in that, The step S1 specifically comprises: Based on the multi-modal sensor network deployed in the chemical deposition production line of sulfur-free copper-clad plate, real-time monitoring data of reaction liquid concentration, temperature, and flow rate are obtained, and a multi-dimensional input feature space of the process is constructed; Performing online calibration processing based on electrochemical impedance spectroscopy on the reaction liquid concentration data to obtain corrected reaction liquid concentration values; Using thermocouples and infrared temperature measuring devices to collect reaction liquid temperature at multiple points synchronously, and performing weighted average filtering processing on the collected data to obtain spatially representative deposition environment temperature feature values; Based on the ultrasonic flowmeter, the flow rate signal of the reaction liquid is collected, and time domain sliding window integration and frequency domain fast Fourier transform processing are performed on the flow rate signal to extract flow rate fluctuation characteristic parameters; According to the deposition process history record, the deposition time reference value is extracted, and combined with the contact angle data and surface roughness measurement value output by the substrate surface cleanliness detector, a substrate state information feature vector is generated; The corrected reaction liquid concentration value, the deposition environment temperature feature value, the flow rate fluctuation characteristic parameter, and the substrate state information feature vector are structured and spliced to form an initial process parameter set matrix containing process parameters and state information.

3. A process for treating a sulphur-free copper clad board for use in automotive vehicle panels according to claim 2, characterised in that, The corrected reaction liquid concentration value is obtained by adopting online automatic calibration of electrochemical impedance spectroscopy, the frequency scanning range is 1Hz to 100kHz, the voltage excitation amplitude is 10mV, the output is fitted by equivalent circuit and multi-channel cross-correlation analysis, and finally the concentration value sequence compensated by weight is fused to obtain the corrected reaction liquid concentration value.

4. A process for treating a sulphur-free copper clad board for use in automotive vehicle panels according to claim 1, characterised in that, The step S2 specifically comprises: Performing minimum-maximum normalization processing on the reaction liquid concentration, temperature, and flow rate parameter data in the initial process parameter set, a normalized parameter vector sequence is obtained; Based on the Z-score method, the normalized parameter vector sequence is subjected to outlier detection, the standard deviation and mean of each parameter dimension are calculated, and abnormal data points exceeding the 3σ threshold are identified and removed; Performing missing value imputation processing on the parameter vector sequence after removing outliers, using the interpolation strategy based on K-nearest neighbor algorithm, filling in the missing items based on the weighted average of similar samples, and obtaining the complete parameter matrix; Performing covariance matrix calculation on the complete parameter matrix, analyzing the correlation between each parameter dimension, and identifying highly correlated parameter pairs; Based on the eigenvalue decomposition result of the covariance matrix, the eigenvector basis of the standardized parameter matrix is constructed, and the orthogonal transformation matrix of the parameter space is formed.

5. A process for treating a sulphur-free copper clad board for use in automotive vehicle panels according to claim 1, wherein The step S3 specifically includes: Based on the process parameter combination data obtained from small batch deposition experiments, including reaction liquid concentration, temperature, flow rate and deposition time, data structure organization operation is performed, and parameter vector set is generated; Performing thickness distribution detection on the sulfur-free copper-clad plate samples obtained after each group of deposition experiments, measuring the thickness spatial distribution characteristics of the deposition layer, and generating a thickness distribution feature vector; Performing surface roughness detection on each group of deposition samples, collecting surface morphology data, and calculating Sa value based on ISO 25178 standard to generate surface roughness feature parameters; Performing adhesion test on each group of deposition samples to evaluate the bonding strength between the deposition layer and the substrate, and generating adhesion feature parameters; Triply mapping the parameter vector set, the corresponding thickness distribution feature vector, the surface roughness feature parameters and the adhesion feature parameters to construct a deposition dynamic fingerprint database table structure.

6. A process for treating a sulphur-free copper clad board for use in automotive vehicle panels according to claim 5, characterised in that, The step S3 further includes forming a feature vector by partitioning the thickness distribution obtained from the deposition experiment through a spatial distribution statistical vectorization algorithm, the number of regions is 1 to 32, and the normalized surface roughness and adhesion parameters are combined to form a deposition dynamic fingerprint library.

7. A process for treating a sulphur-free copper clad board for use in automotive vehicle panels according to claim 1, characterised in that, The step S4 specifically includes: Performing principal component analysis dimension reduction processing on the standardized parameter matrix in the deposition dynamic fingerprint library to extract the principal component loading matrix of the deposition performance influencing factor; Based on the principal component loading matrix, the contribution weight vector of each parameter in the deposition performance is calculated; Using the contribution weight vector to perform weighted linear combination transformation on the original standardized parameter matrix to generate a set of deposition process feature vectors in a low-dimensional feature space; Based on the deposition process feature vector set and the corresponding deposition time and performance index data, a training sample set is constructed; Using the radial basis function kernel function to perform support vector regression modeling on the training sample set, a multi-parameter coupling response model between the deposition reaction liquid concentration, temperature, flow rate and deposition time is established; Performing cross-validation and hyperparameter optimization on the multi-parameter coupling response model to obtain the optimal model parameter combination, and evaluating the prediction accuracy and stability of the model under different working conditions; The optimized multi-parameter coupling response model is deployed in the deposition control system.

8. A process for treating a sulphur-free copper clad board for use in automotive vehicle panels according to claim 1, characterised in that, The step S5 specifically includes: The concentration, temperature and flow rate parameters in the chemical deposition process of the sulfur-free copper-clad plate are collected by an online sensor to obtain multi-dimensional process input data of a current deposition stage; Normalization processing and outlier rejection operation are performed on the multi-dimensional process input data of the current deposition stage to obtain a standardized real-time process parameter vector; Based on the key feature factors extracted after principal component analysis dimension reduction, feature mapping conversion is performed on the standardized real-time process parameter vector to generate a low-dimensional feature vector suitable for input of a multi-parameter coupling response model; The low-dimensional feature vector is input into the multi-parameter coupling response model trained based on a support vector regression algorithm to predict the change trend of deposition performance indexes such as thickness distribution, surface roughness and bonding force at the current deposition time; According to the deposition performance index prediction value output by the multi-parameter coupling response model, a prediction curve of deposition performance evolution over time is constructed to form a dynamic response relationship map between thickness uniformity and deposition rate.

9. A process for treating a sulphur-free copper clad board for use in automotive vehicle panels according to claim 1, characterised in that, The optimal deposition time threshold satisfies the weighted optimization condition of the thickness uniformity index and the production efficiency index.

10. A process for treating a sulphur-free copper clad board for use in automotive vehicle panels according to claim 1, characterised in that, In S7, the deposition control parameter adjustment is performed through a fuzzy control algorithm, the input membership function type includes a triangular and trapezoidal function, the number of rule bases is 10 to 50, and the output parameters are used to dynamically adjust the initial proportion, integral and differential coefficients of the PID module.

Images