Conductive silver paste formulation process optimization method, device, equipment and medium

Abstract

The application discloses a conductive silver paste formula process optimization method and device, equipment and medium, relates to the technical field of data processing, and comprises the following steps: collecting silver paste samples and encoding unified material representation, building a dedicated constraint library containing silver powder, rheology, sintering and failure forbidden zone, generating candidates through a double model, predicting performance and uncertainty, screening feasible samples through constraint projection, and selecting experimental batches through multi-target scoring; closed-loop trial production testing, adaptive updating of the model combined with neighborhood desensitization data, and final determination of the optimized conductive silver paste formula and process parameters through multi-target screening. Invalid trial production can be reduced, low resistance, high adhesion and printing stability are considered, original confidential data does not need to be shared, and small sample nodes can also efficiently complete multi-target collaborative optimization.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, equipment and medium for optimizing the formulation process of conductive silver paste. Background Technology

[0002] Currently, the mainstream conductive silver paste formulation process optimization in the industry falls into two categories. The first is the traditional manual experimentation mode, which relies on the experience of R&D personnel combined with single-factor comparative experiments to adjust the silver powder ratio and process parameters one by one to carry out trial production tests. The second is a general machine learning material optimization framework, which only uses the formulation components, base temperature, and time process parameters as input features to build a simple component-performance mapping model to predict conductivity indicators. This is combined with a common generative model to produce candidate formulations, and then the experimental batches are screened through simple scoring. Some cross-institutional collaborative R&D uses a fixed-weight federated learning framework to exchange model parameters and achieve collaborative data training.

[0003] The combination space of manual experimental modes is huge, and it is difficult to balance multi-dimensional performance indicators under limited experimental budgets. It is prone to producing a large number of failed experiments such as agglomeration, network blockage, film cracking, and failure to meet adhesion standards. The R&D cycle is long and the raw material and labor costs are high. The general machine learning material optimization framework does not establish specific constraints for the strong thixotropic non-Newtonian rheology of conductive silver paste, multi-stage sintering phase transformation, and multi-scale stacking conduction mechanism of silver powder. The generated candidate formulations often exceed the printing viscosity window, the conductive network is discontinuous after sintering, and a large number of candidates are not manufacturable. At the same time, failed samples such as broken lines and low conductivity are directly discarded, and the failure rules cannot be transformed into reverse constraints. The general prediction model can only output a single indicator of resistivity and cannot simultaneously predict the risk of rheological printing and the quality of sintered film. The candidate selection lacks a multi-objective comprehensive evaluation basis. The traditional fixed-weight federated collaborative method does not distinguish between different node silver paste systems and experimental data quality differences. Low-quality node data will interfere with the model optimization effect and cannot avoid the risk of leakage of original sensitive experimental data. The generalization ability of the model is poor when developing small-sample R&D node models.

[0004] Therefore, how to generate and screen candidate conductive silver paste formulations and processes that satisfy the constraints of silver powder morphology, rheological window, sintering film formation and failure samples in a high-dimensional candidate space that couples the variables of conductive silver paste formulation, preparation process, printing film formation and sintering curing has become an urgent problem to be solved. Summary of the Invention

[0005] The main objective of this application is to provide a method, apparatus, equipment, and medium for optimizing conductive silver paste formulations and processes, aiming to solve the technical problem of how to generate and screen candidate conductive silver paste formulations and processes that meet the constraints of silver powder morphology, rheological window, sintering film formation, and failed samples.

[0006] To achieve the above objectives, this application proposes a method for optimizing the formulation and process of conductive silver paste, comprising: Acquire local node conductive silver paste sample data and encode the sample data into a unified material representation; A silver paste constraint library is generated based on the unified material representation and preset process boundaries. The silver paste constraint library includes silver powder morphology constraints, rheological window constraints, sintering film formation constraints, and failure sample exclusion zones. The silver paste constraint library, target performance conditions, and unified material representation are input into the candidate generator and performance predictor, respectively, to obtain the candidate conductive silver paste material representation and the corresponding performance prediction value and prediction uncertainty. Feasibility projection is performed on the candidate conductive silver paste material representations according to the silver paste constraint library to obtain feasible candidate conductive silver paste material representations; The candidate acquisition score is calculated based on the performance prediction value, prediction uncertainty, constraint residual of the silver paste constraint library, preset experimental cost and distance between candidates, and the next round of experimental batches is selected from the feasible candidate conductive silver paste material representations based on the candidate acquisition score. The formula process instruction is generated and executed according to the next round of experimental batches to obtain experimental test results, failure reasons and neighborhood node summary data. Adaptive parameter updates are performed on the performance predictor and candidate generator according to the experimental test results, failure reasons and neighborhood node summary data to obtain the updated performance predictor and updated candidate generator. Based on the updated performance predictor and the updated candidate generator, candidate conductive silver paste formulations and process parameters that meet the target performance requirements are generated within the feasible domain of the silver paste constraint library, and the optimized conductive silver paste formulations and process parameters are determined through multi-objective screening.

[0007] In one embodiment, the step of acquiring local node conductive silver paste sample data and encoding the sample data into a uniform material representation includes: Acquire local node conductive silver paste sample data, wherein the sample data includes silver powder composition data, resin carrier data, solvent and additive data, process data, printing film data, structural characterization data, rheological property data and conductivity test data; The process data in the sample data are subjected to dimension unification and process window normalization to obtain standardized process data. The rheological performance data in the sample data are curve encoded to obtain a rheological vector; The conductivity test data in the sample data are sequentially encoded to obtain a conductivity vector. The silver powder composition data, resin carrier data, solvent and additive data are mapped into a formulation vector; The standardized process data is mapped to a process vector, and the structural characterization data is mapped to a structural vector; The rheological vector and electrical conductivity vector are concatenated with the formulation vector, process vector, and structural vector to obtain a unified material representation.

[0008] In one embodiment, the step of inputting the silver paste constraint library, the target performance conditions, and the unified material representation into the candidate generator and the performance predictor, respectively, to obtain the candidate conductive silver paste material representation and the corresponding performance prediction value and prediction uncertainty, includes: The unified material representation is input into the performance predictor for latent variable mapping and feature extraction to obtain the material latent variable representation. The latent variables of the material are input into the conductive network predictor, rheological printing window predictor, sintering film predictor and adhesion predictor of the performance predictor, respectively, to obtain the volume resistivity prediction value, sheet resistance prediction value, rheological window prediction value, adhesion prediction value and the prediction uncertainty corresponding to each prediction value. The target performance conditions and the silver paste constraint library are input into the candidate generator, and reverse denoising sampling is performed under the feasible region constraint of the silver paste constraint library to obtain the candidate conductive silver paste material representation. The candidate conductive silver paste material representation is input into the performance predictor to perform performance prediction, and the corresponding performance prediction value and prediction uncertainty of the candidate conductive silver paste material representation are obtained. Based on the silver paste constraint library, a feasibility projection verification is performed on the candidate conductive silver paste material representation. Candidates that violate the forbidden zones of failed samples in the silver paste constraint library are eliminated, and the verified candidate conductive silver paste material representation and the corresponding performance prediction value and prediction uncertainty are obtained.

[0009] In one embodiment, the step of performing a feasibility projection on the candidate conductive silver paste material representation based on the silver paste constraint library to obtain a feasible candidate conductive silver paste material representation includes: From the silver paste constraint library, silver powder morphology constraints, rheological window constraints, sintering film formation constraints, and failure sample exclusion zones are extracted to obtain a set of constraint parameters. Based on the set of constraint parameters, the formulation vector in the candidate conductive silver paste material representation is nonnegated, normalized, and truncated to upper and lower limits to obtain the candidate conductive silver paste material representation after formulation verification. Based on the rheological window constraint and sintering film formation constraint in the set of constraint parameters, the process vector in the candidate conductive silver paste material representation after formula verification is subjected to boundary truncation and differentiable projection processing to obtain the candidate conductive silver paste material representation after process verification. Based on the forbidden zone constraints of the failed samples in the set of constraint parameters, the latent space distance between the candidate conductive silver paste material representation after process verification and the center of the failed sample cluster is calculated to obtain the forbidden zone distance value. When the distance value of the restricted area is greater than the preset restricted area radius threshold, the candidate conductive silver paste material representation after process verification is determined as a feasible candidate conductive silver paste material representation.

[0010] In one embodiment, the step of calculating a candidate acquisition score based on the performance prediction value, prediction uncertainty, constraint residuals of the silver paste constraint library, preset experimental cost, and distance between candidates, and selecting the next batch of experiments from the feasible candidate conductive silver paste material representations based on the candidate acquisition score, includes: The low resistivity benefit term is calculated based on the volume resistivity prediction value and the sheet resistance prediction value in the performance prediction value, and the high adhesion benefit term is calculated based on the adhesion prediction value in the performance prediction value, thus obtaining a set of performance benefit terms. The uncertainty exploration term is obtained by weighted summation of the uncertainties of each performance index in the predicted uncertainty; The constraint violation penalty term is calculated based on the constraint residuals of the silver paste constraint library, and the diversity distance term is calculated based on the latent space distance between the feasible candidate conductive silver paste material representation and the selected batch. A multi-objective acquisition function is constructed based on the set of performance benefit terms, uncertainty exploration terms, constraint violation penalty terms, diversity distance terms, and preset experimental costs to obtain the candidate acquisition scores corresponding to each feasible candidate conductive silver paste material representation. Based on the candidate acquisition scores, the feasible candidate conductive silver paste materials are sorted in descending order to obtain a candidate priority sequence; Based on the candidate priority sequence and the preset batch size, candidates that satisfy the condition that the distance between candidates is greater than the preset diversity distance threshold are selected in turn to obtain the next round of experimental batches.

[0011] In one embodiment, the step of generating and executing the formulation process instruction based on the next round of experimental batches to obtain experimental test results, failure reasons, and neighborhood node summary data, and then performing adaptive parameter updates on the performance predictor and candidate generator based on the experimental test results, failure reasons, and neighborhood node summary data to obtain updated performance predictors and updated candidate generators, includes: A silver paste preparation instruction is generated based on the formula parameters in the next round of experimental batches, and the silver paste preparation instruction is sent to the preparation equipment so that the preparation equipment performs a dispersion and grinding operation to obtain the silver paste to be tested. A printing instruction is generated based on the printing parameters in the next round of experimental batches, and the printing instruction is sent to the printing equipment so that the printing equipment can print the silver paste to be tested into a film to obtain a printed film layer. A curing instruction is generated based on the curing parameters in the next round of experiments, and the curing instruction is sent to the curing equipment so that the curing equipment can sinter or cure the printed film layer to obtain a cured film layer. A test command is generated and sent to the test equipment, so that the test equipment performs conductivity test, adhesion test and rheology test on the cured film layer, and obtains the experimental test results and the reasons for failure. The model summary, pseudo-label sample summary, synthetic candidate statistics, and validation evaluation metrics are received from the neighboring nodes to obtain the neighboring node summary data. The experimental test results and failure reasons are fed back to the local node to update the local sample library and the forbidden zone of failed samples. The material system similarity, verification quality score and constraint violation rate of the neighboring node and the local node are calculated based on the neighboring node summary data. The aggregation weight is calculated based on the material system similarity, verification quality score and constraint violation rate, and the performance predictor and the candidate generator are adaptively updated based on the aggregation weight to obtain the updated performance predictor and the updated candidate generator.

[0012] In one embodiment, the step of generating candidate conductive silver paste formulations and process parameters that meet the target performance requirements within the feasible region of the silver paste constraint library based on the updated performance predictor and the updated candidate generator, and determining the optimized conductive silver paste formulation and process parameters through multi-objective screening, includes: The target performance conditions are input into the updated candidate generator, and reverse denoising sampling and feasibility projection are performed under the feasible region constraints of the silver paste constraint library to obtain a set of candidate conductive silver paste formulations and process parameters. The candidate conductive silver paste formulations and process parameters are input into the updated performance predictor to obtain the predicted volume resistivity, predicted sheet resistance, predicted adhesion, predicted rheological window value, and the prediction uncertainty corresponding to each candidate conductive silver paste formulation and process parameter. The constraint residuals of each candidate conductive silver paste formulation and process parameters are calculated based on the silver paste constraint library to obtain the constraint satisfaction state. The comprehensive performance score is calculated based on the predicted volume resistivity, predicted sheet resistance, predicted adhesion, predicted rheological window value and predicted uncertainty. Based on the constraint satisfaction status, comprehensive performance score and preset experimental cost, a multi-objective screening function is constructed to obtain the multi-objective screening values ​​of each candidate conductive silver paste formulation and process parameter. Based on the multi-objective screening values, the candidate conductive silver paste formulations and process parameter sets are sorted in descending order to obtain a candidate priority sequence. The candidate conductive silver paste formulation and process parameters with the highest multi-objective screening value are selected from the candidate priority sequence and determined as the optimized conductive silver paste formulation and process parameters.

[0013] Furthermore, to achieve the above objectives, this application also proposes a conductive silver paste formulation process optimization device, which includes: The data encoding module is used to acquire local node conductive silver paste sample data and encode the sample data into a unified material representation. The constraint generation module is used to generate a silver paste constraint library based on the unified material representation and preset process boundaries. The silver paste constraint library includes silver powder morphology constraints, rheological window constraints, sintering film formation constraints, and failure sample exclusion zones. The prediction module is used to input the silver paste constraint library, target performance conditions, and unified material representation into the candidate generator and performance predictor, respectively, to obtain the candidate conductive silver paste material representation and the corresponding performance prediction value and prediction uncertainty. The feasible projection module is used to perform feasible projection on the candidate conductive silver paste material representation according to the silver paste constraint library to obtain a feasible candidate conductive silver paste material representation. The acquisition selection module is used to calculate the candidate acquisition score based on the performance prediction value, prediction uncertainty, constraint residual of the silver paste constraint library, preset experimental cost and distance between candidates, and select the next round of experimental batches from the feasible candidate conductive silver paste material representations based on the candidate acquisition score; The feedback update module is used to generate and execute the formula process instructions according to the next round of experimental batches, obtain experimental test results, failure reasons and neighborhood node summary data, and perform adaptive parameter updates on the performance predictor and candidate generator according to the experimental test results, failure reasons and neighborhood node summary data, to obtain the updated performance predictor and updated candidate generator. The results module is used to generate candidate conductive silver paste formulations and process parameters that meet the target performance requirements within the feasible domain of the silver paste constraint library based on the updated performance predictor and the updated candidate generator, and to determine the optimized conductive silver paste formulations and process parameters through multi-objective screening.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the conductive silver paste formulation process optimization method described above.

[0015] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the conductive silver paste formulation process optimization method described above.

[0016] This application collects silver paste samples and encodes them for unified material characterization. It establishes a dedicated constraint library including silver powder, rheology, sintering, and failure-prone areas. A dual-model approach is used to generate candidate samples and predict performance and uncertainty. Feasible samples are screened through constraint projection, and experimental batches are selected through multi-objective scoring. Closed-loop pilot-scale testing is conducted, and the model is adaptively updated using desensitized neighborhood data. Finally, the optimized conductive silver paste formulation and process parameters are determined through multi-objective screening. This approach reduces invalid pilot-scale production, balances low resistance, high adhesion, and printing stability, eliminates the need to share original confidential data, and enables efficient multi-objective collaborative optimization even with small sample nodes. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the first embodiment of the conductive silver paste formulation and process optimization method of this application; Figure 2 This is a flowchart illustrating the second embodiment of the conductive silver paste formulation and process optimization method of this application; Figure 3 This is a schematic diagram of the module structure of the conductive silver paste formulation and process optimization device of this application; Figure 4 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the conductive silver paste formulation process optimization method in the embodiments of this application.

[0019] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0021] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0022] Currently, the mainstream conductive silver paste formulation process optimization in the industry falls into two categories. The first is the traditional manual experimentation mode, which relies on the experience of R&D personnel combined with single-factor comparative experiments to adjust the silver powder ratio and process parameters one by one to carry out trial production tests. The second is a general machine learning material optimization framework, which only uses the formulation components, basic temperature, time and process parameters as input features to build a simple component-performance mapping model to predict conductivity indicators, and combines it with a common generative model to produce candidate formulations. Then, the experimental batches are screened through simple scoring. Some cross-institutional collaborative R&D uses a fixed-weight federated learning framework to exchange model parameters and achieve collaborative data training.

[0023] The combination space of manual experimental modes is huge, and it is difficult to balance multi-dimensional performance indicators under limited experimental budgets. It is prone to producing a large number of failed experiments such as agglomeration, network blockage, film cracking, and failure to meet adhesion standards. The R&D cycle is long and the raw material and labor costs are high. The general machine learning material optimization framework does not establish specific constraints for the strong thixotropic non-Newtonian rheology of conductive silver paste, multi-stage sintering phase transformation, and multi-scale stacking conduction mechanism of silver powder. The generated candidate formulations often exceed the printing viscosity window, the conductive network is discontinuous after sintering, and a large number of candidates are not manufacturable. At the same time, failed samples such as broken lines and low conductivity are directly discarded, and the failure rules cannot be transformed into reverse constraints. The general prediction model can only output a single indicator of resistivity and cannot simultaneously predict the risk of rheological printing and the quality of sintered film. The candidate selection lacks a multi-objective comprehensive evaluation basis. The traditional fixed-weight federated collaborative method does not distinguish between different node silver paste systems and experimental data quality differences. Low-quality node data will interfere with the model optimization effect and cannot avoid the risk of leakage of original sensitive experimental data. The generalization ability of the model is poor when developing small-sample R&D node models.

[0024] Therefore, how to generate and screen candidate conductive silver paste formulations and processes that satisfy the constraints of silver powder morphology, rheological window, sintering film formation and failure samples in a high-dimensional candidate space that couples the variables of conductive silver paste formulation, preparation process, printing film formation and sintering curing has become an urgent problem to be solved.

[0025] Based on the above, this application also provides a method for optimizing the formulation and process of conductive silver paste, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the conductive silver paste formulation and process optimization method of this application.

[0026] In this embodiment, the conductive silver paste formulation process optimization method includes steps S10~S70: Step S10: Obtain local node conductive silver paste sample data and encode the sample data into a unified material representation.

[0027] It should be noted that step S10 includes: First, acquiring local node conductive silver paste sample data, which includes silver powder composition data, resin carrier data, solvent and additive data, process data, printing film data, structural characterization data, rheological property data, and conductivity test data. The need to collect so much data stems from the fact that the final performance of the conductive silver paste is determined by the formulation, process, structure, and test performance. The absence of any one type of data will prevent the subsequent model from establishing a complete mapping relationship. For example, recording only the silver powder content without recording the rheological curve makes it impossible to determine whether the formulation will cause screen clogging or line breakage during screen printing.

[0028] Secondly, the process data in the sample data were subjected to dimension unification and process window normalization to obtain standardized process data. The mixing time was in minutes, the dispersion speed was in revolutions per minute, and the grinding gap was in micrometers. By normalizing the maximum and minimum values, each process parameter was mapped to the [0,1] interval to eliminate the influence of dimension differences on model training and avoid large numerical parameters from masking the role of small numerical parameters.

[0029] Then, the rheological performance data in the sample data is curve-encoded to obtain a rheological vector. Specifically, 20-50 sampling points are taken at logarithmic intervals within a shear rate range of 0.1 mPa·s to 1000 mPa·s. The viscosity values ​​of each sampling point are dimensionless according to the reference viscosity of 1000 mPa·s and then concatenated to form a rheological vector. At the same time, low shear viscosity, high shear viscosity, shear thinning index, and thixotropic recovery rate are extracted as auxiliary features and incorporated into the vector. This is because a single viscosity value cannot determine whether the thixotropic recovery characteristics meet the requirements of continuous printing. Only by encoding the entire curve can the rheological behavior of the silver paste from standing to printing and then to recovery be completely preserved.

[0030] Next, the conductivity test data in the sample data are sequence-encoded to obtain a conductivity vector. Specifically, the volume resistivity, sheet resistance, film thickness, and resistance change rate after aging test are concatenated into a conductivity vector. For the four-probe test curve, the mean and standard deviation of the test point sequence are taken as statistical features and incorporated. This is done because volume resistivity and sheet resistance are affected by film thickness and test point distribution, and a single mean may mask the problem of local conductivity inhomogeneity. Sequence encoding can simultaneously reflect the overall conductivity level and film uniformity.

[0031] Next, the silver powder composition data, resin carrier data, solvent and additive data are mapped into a formulation vector. Specifically, the mass fractions of flake silver powder, spherical silver powder, nano silver, resin, solvent, dispersant, leveling agent, curing agent, and coupling agent are arranged in a fixed order, and each component, after normalization, satisfies the constraint that the sum is 1. This is because there are interdependent relationships between the proportions of each component; too high a silver powder content will lead to insufficient resin and decreased adhesion, while too low a content will prevent the formation of a permeation network. The fixed-order vector representation allows the model to learn the influence of different component combinations on performance.

[0032] The standardized process data is then mapped to process vectors, and the structural characterization data is mapped to structural vectors. Specifically, the process vectors include mixing time, dispersion speed, grinding gap, printing speed, screen mesh count, curing temperature, and holding time, while the structural vectors include film thickness, porosity, sintering neck size, linewidth retention rate, and interface roughness parameters. This separate mapping helps the model distinguish between controllable parameters of the preparation process and observable parameters of the film formation result, avoiding prediction confusion caused by conflating process parameters with structural features.

[0033] Finally, the rheological vector and conductivity vector are concatenated with the formulation vector, process vector, and structure vector to obtain a unified material representation. Specifically, the concatenation order is: formulation vector, process vector, structure vector, rheological vector, and conductivity vector. Each sub-vector is mapped to the same 128-dimensional dimension through an independent projection layer and then weighted and concatenated through a feature fusion layer. The weights are learned during the training process. This is equivalent to establishing a complete technical profile for each sample, from which subsequent models can extract cross-dimensional correlation features. For example, it can identify the optimal rheological window when a specific silver powder morphology is combined with a specific dispersion speed, thus providing a comprehensive information foundation for candidate generation and performance prediction.

[0034] It's important to understand that unified material representation refers to mapping the formulation, process, structural, rheological, and conductivity information in conductive silver paste sample data into a multi-dimensional vector representation in the same vector space. This representation can be used for subsequent model training, candidate generation, and performance prediction. Silver powder composition data refers to the proportion, particle size distribution parameters, and surface treatment agent type information of various silver powders such as flake silver powder, spherical silver powder, and nano-silver in conductive silver paste. Resin carrier data refers to the type, molecular weight, glass transition temperature, and mass percentage of binder resins such as epoxy resin and acrylic resin in the formulation. Solvent and additive data refers to the type, volatility characteristics, and addition ratio information of solvents, dispersants, leveling agents, curing agents, and coupling agents. Process data refers to controllable parameters in the dispersion preparation process, such as mixing time, dispersion speed, three-roll mill gap, and premixing temperature. Printing film formation data refers to printing process parameters and film formation status records, such as printing speed, screen mesh count, squeegee pressure, printing line shape, and film thickness. Structural characterization data refers to microstructural parameters such as film thickness, porosity, silver particle sintering neck size, linewidth retention, and interface morphology characteristics. Rheological property data refers to rheological test data such as viscosity values ​​at different shear rates, thixotropic recovery curves, yield stress, and shear thinning index. Electrical conductivity test data refers to electrical property data such as volume resistivity, sheet resistance, four-probe test curves, and aging stability test sequences.

[0035] Step S20: Generate a silver paste constraint library based on the unified material representation and preset process boundaries.

[0036] It should be noted that the silver paste constraint library includes silver powder morphology constraints, rheological window constraints, sintering film constraints, and a no-go zone for failed samples.

[0037] Specifically, firstly, the value range and data distribution of each sub-vector in the unified material representation are analyzed. Combined with the preset process boundaries, initial constraint parameters are extracted from the manual rule library, historical failure sample library, printing process specification library, and sintering and curing equipment boundary library to construct silver powder morphology constraints, rheological window constraints, and sintering film formation constraints. The silver powder morphology constraint limits the silver powder gradation range through flake diameter-to-thickness ratio, particle size distribution, specific surface area, and multi-scale filling ratio. The silver powder morphology constraint includes the silver powder multi-scale morphology vector, specifically represented as follows: in Represents the multi-scale morphology vector of silver powder; These represent the silver powder particle size corresponding to a cumulative volume fraction of 10%, a cumulative volume fraction of 50%, and a cumulative volume fraction of 90%, respectively. This indicates the flake diameter to thickness ratio, which is the ratio of the planar dimension to the thickness of the flake silver powder. This refers to the specific surface area, which is the total surface area per unit mass of silver powder. These represent the mass fractions of flake-shaped silver powder, spherical silver powder, and nano-silver powder in the total mass of silver powder, respectively. This morphology vector allows for the differentiation of the effects of different silver powder gradations on packing density, permeation threshold, and film adhesion for the same silver content. The rheological window constraint uses the Herschel-Bulkley model to describe the strongly nonlinear rheological behavior; the specific formula is as follows: in This represents shear stress, indicating the silver paste at the shear rate. The ability to resist shear deformation; The yield stress represents the minimum shear stress that the silver paste must overcome to begin flowing from a static state. It represents the consistency coefficient, reflecting the viscosity of the silver paste; The shear thinning index indicates the degree of shear thinning in the silver paste. When... If the deviation from the target range or the thixotropic recovery time exceeds the threshold, the candidate is deemed to have a risk of screen clogging, edge collapse, or printing line breakage. Sintering film formation constraints simultaneously consider solvent evaporation, resin crosslinking, and the growth process of the silver particle sintering neck. The degree of curing is described using an Arrhenius-type equation. in Indicates the degree of curing. =0 indicates that it is not cured. =1 indicates complete curing; Indicates the curing time; Represents absolute temperature, and represents the thermodynamic temperature during the curing process; It represents the pre-exponential factor, the frequency factor of the reaction rate constant, and reflects the collision frequency at which the reaction occurs; Indicates activation energy; This represents the gas constant. The Arrhenius-type equation describes the resin curing kinetics.

[0038] The degree of silver particle sintering neck growth is expressed as follows: in This indicates the degree of sintering neck growth, representing the relative development of sintering necks between silver particles. The closer it is to 1, the more fully the sintering necks have grown and the tighter the connection between particles. Indicates the sintering neck size between adjacent silver particles; The equivalent radius of the silver particles is represented by , and the spherical equivalent radius of the silver particles is represented by . For non-spherical particles, the volume equivalent sphere radius is taken. Secondly, cluster analysis is performed on agglomeration samples, sedimentation samples, network blockage samples, broken line samples, cracked samples, and samples with insufficient adhesion from historical failure samples. The cluster centers and exclusion zone radii of the failure samples are extracted to construct exclusion zone constraints for failure samples. This is because historical failure data contains the distribution patterns of infeasible formulations. Transforming failure samples into exclusion zone constraints can prevent the generator from repeatedly recommending known failed formulation combinations. For example, if a certain type of high silver content, low resin ratio formulation repeatedly cracks at a specific sintering temperature, then that area should be marked as an exclusion zone. Then, the constraints of silver powder morphology, rheological window, sintering film formation, and failure sample exclusion zone are fused to obtain a silver paste constraint library. The silver powder morphology constraint limits the silver powder gradation range through the aspect ratio, particle size distribution, specific surface area, and multi-scale filling ratio. The rheological window constraint limits printability through the deviation of the shear rate-viscosity curve. The sintering film formation constraint limits the continuity of film formation through film porosity, sintering neck size, and degree of curing. The multi-scale morphology vector of silver powder (formulation design) → determines the rheological behavior (printability) and sintering activity (low-temperature feasibility) → the rheological window constraint ensures the printing quality (film uniformity) → the film quality affects the heat and mass transfer conditions of curing / sintering → the degree of curing and the degree of sintering neck growth jointly determine the continuity and adhesion of the conductive network after film formation → ultimately manifested as target performance such as volume resistivity, sheet resistance, and adhesion. These four aspects are not independent constraints, but constitute a complete coupled chain of "formulation-process-structure-performance". Finally, the constraint library is configured to be invoked in all three stages: candidate generation, candidate screening, and acquisition function calculation. This ensures that the unified material representation outputs constraint residuals and constraint satisfaction states after being input into the constraint library. This ensures that each step is constrained by physical feasibility and historical failure experience, avoiding the generation of candidate formulations that violate basic physical laws or known failure modes.

[0039] Step S30: Input the silver paste constraint library, target performance conditions, and unified material representation into the candidate generator and performance predictor, respectively, to obtain the candidate conductive silver paste material representation and the corresponding performance prediction value and prediction uncertainty.

[0040] It should be noted that step S30 includes: First, inputting the unified material representation into the performance predictor for latent variable mapping and feature extraction to obtain the material latent variable representation. Specifically, inputting the unified material representation into the encoder for nonlinear mapping to obtain a low-dimensional continuous material latent variable representation. in Indicates the first Material latent variables for each conductive silver paste sample; Indicates a unified material representation; Represents the encoder network, with parameters as follows: ; This indicates the preset latent variable dimension. In this example, the encoder uses a multi-layer fully connected network structure, and the input dimension is the uniform material representation dimension. The output dimension is the preset latent variable dimension. The activation function uses a modified linear unit, which extracts the nonlinear coupling features between the formula vector, process vector, structure vector, rheological vector and conductivity vector by dimensionality reduction layer by layer. For example, the interaction effect between the silver powder flake diameter-to-thickness ratio and the dispersion speed is used. This compresses high-dimensional heterogeneous data into a semantically consistent low-dimensional space and avoids feature redundancy and dimensional conflicts caused by direct splicing.

[0041] Secondly, the latent variable representations of the materials are input into the conductive network predictor, rheological printing window predictor, sintering film predictor, and adhesion predictor of the performance predictor, respectively, to obtain the predicted values ​​of volume resistivity, sheet resistance, rheological window, and adhesion, as well as the prediction uncertainties corresponding to each predicted value. Specifically, the latent variable representations of the materials are input in parallel into four sub-predictors, each consisting of an independent multilayer fully connected network, which outputs the predicted values ​​of volume resistivity, sheet resistance, rheological window, and adhesion, respectively. The prediction uncertainties are calculated using the variance of the integrated model, with the specific formula as follows: in Indicates the first The conductive silver paste sample in the first Prediction uncertainty of performance indicators; This indicates the number of ensemble models and the number of independent sub-models that make up the ensemble predictor. Indicates the index of the number of ensemble models; Indicates the first The ensemble sub-model for the first The conductive silver paste sample in the first Predicted output values ​​for each performance metric; Indicates the first One candidate conductive silver paste sample; Indicates the first The candidate sample at the th The predicted mean of the performance indicators is derived from The arithmetic mean of the predicted values ​​of the ensemble sub-models is obtained; Indicates the first The observation noise variance of each performance indicator reflects the inherent uncertainty caused by measurement errors of the test equipment and fluctuations in the experimental environment. In this example, the conductive network predictor focuses on the silver powder gradation and sintering neck characteristics among the latent variables, while the rheological printing window predictor focuses on the shear thinning index and thixotropic recovery correlation. By separating the predictor structure, gradient interference between different performance indicators is avoided, thus improving the prediction accuracy of each specific indicator.

[0042] Then, the target performance conditions and the silver paste constraint library are input into the candidate generator. Reverse denoising sampling is performed under the feasible region constraints of the silver paste constraint library to obtain the candidate conductive silver paste material representation. Specifically, the candidate generator adopts a conditional diffusion model structure, and the forward denoising process is as follows: in Indicates that in the known number of... Step latent variables Under the conditions, the first Step latent variables The probability distribution; This represents a normal (Gaussian) distribution, with the first parameter being the mean and the second parameter being the covariance matrix. Indicates the first The noise variance of the step is a preset variance scheduling parameter; Let represent the identity matrix, and let represent Gaussian noise with independent and homoscedasticity in each dimension.

[0043] In the reverse generation stage, with the target volume resistivity Target side obstruction Target adhesion and silver paste constraint library Using the feasible region as a condition, the target region is gradually recovered from standard Gaussian noise through a denoising network: in This represents the mean prediction value from the reverse denoising process; Indicates the first Step signal retention coefficient; Indicates the preceding Step-by-step cumulative signal retention coefficient; This represents a denoising neural network, whose output is the predicted noise component; Indicates the initial latent variable To the The percentage of the original signal retained after each step of accumulation.

[0044] Next, the candidate conductive silver paste material representation is input into the performance predictor for performance prediction, obtaining the corresponding performance prediction values ​​and prediction uncertainties. Specifically, the candidate material representation output by the candidate generator is mapped into latent variables by an encoder, and then input in parallel into the aforementioned conductive network predictor, rheological printing window predictor, sintering film predictor, and adhesion predictor to obtain the corresponding volume resistivity prediction value, sheet resistance prediction value, rheological window prediction value, adhesion prediction value, and various prediction uncertainties. This step is equivalent to performing a virtual test on the generated candidates, quickly evaluating whether they meet the target performance requirements before being put into actual experiments, and filtering out candidates whose predicted performance deviates from the target value by more than a preset threshold (e.g., the volume resistivity prediction value is higher than the target value by 20%), thereby reducing invalid experimental batches.

[0045] Finally, a feasibility projection verification is performed on the candidate conductive silver paste material representations based on the silver paste constraint library. Candidates that violate the forbidden zones of failed samples in the silver paste constraint library are eliminated, resulting in the verified candidate conductive silver paste material representations and their corresponding performance prediction values ​​and prediction uncertainties. Specifically, the candidate material representations are mapped to latent space coordinates, and their Euclidean distances to the centers of each failure cluster in the forbidden zones of failed samples are calculated. The formula for the specific penalty value of the forbidden zones of failed samples is as follows: in Indicating candidate conductive silver paste samples The penalty value for the forbidden zone of failed samples is set. The larger the penalty value, the closer the candidate is to the known failure mode, and it should be eliminated or downweighted. Represents the number of failure clusters, a positive integer, representing the total number of failure pattern categories obtained after cluster analysis of historical failure samples; Indicating candidate conductive silver paste samples The coordinate vector in the latent space has the same dimension as the encoder output dimension. Indicates the first The centers of each failure cluster are obtained by taking the mean or center point after clustering historical failure samples. Indicates the first The radius of the forbidden zone of each failure cluster is a positive real number, and the value of the radius is determined based on the distribution density of the failure samples in the latent space. The L2 norm represents the straight-line distance between two points in the latent space. This represents the truncation function, ensuring that the penalty is zero when the distance between the candidate and the center of the failed cluster is greater than the radius of the restricted area; a positive penalty is only applied when the distance is less than the radius. This indicates that the candidate has fallen into the failure zone and is therefore eliminated. Simultaneously, the projection operator is used; the specific formula is: in Indicates the first The output results of the candidate latent variables after feasibility projection; This represents the feasible region defined in the silver paste constraint library. The projection operator; Indicates the first One candidate latent variable; The variable represents the independent variable that minimizes the objective function; the operation represents finding the optimal solution. This represents the feasible region defined by the silver paste constraint library. Candidate latent variables; This represents the square of the Euclidean norm (L2 norm squared). The remaining candidates are projected onto the feasible region boundary. The validated and retained candidates, along with their corresponding performance predictions and prediction uncertainties, constitute the validated candidate set, which is used for subsequent multi-objective acquisition function ranking.

[0046] Step S40: Perform feasibility projection on the candidate conductive silver paste material representations according to the silver paste constraint library to obtain feasible candidate conductive silver paste material representations.

[0047] It should be noted that step S40 includes: first, extracting silver powder morphology constraints, rheological window constraints, sintering film constraints, and failure sample exclusion zones from the silver paste constraint library to obtain a set of constraint parameters. Specifically, the morphology constraints of silver powder include the flake diameter-to-thickness ratio range (2-50), particle size distribution range (D10 0.5-2μm, D50 2-10μm, D90 10-30μm), and specific surface area threshold (0.5-5m² / g). The rheological window constraints include the upper limit of low shear viscosity (50000mPa·s), the lower limit of high shear viscosity (5000mPa·s), and the upper limit of thixotropic recovery time (60s). The sintering film formation constraints include the curing temperature window (80-200℃) and the upper limit of film porosity (15%). The failure sample exclusion zone includes the center coordinates of each failure cluster and the corresponding exclusion zone radius (0.5-2.0). The reason why it is necessary to extract these four types of constraints at the same time is that the candidate feasibility verification needs to cover four dimensions: formulation design, process execution, film formation quality, and historical failure experience. The absence of any one of these will lead to a verification blind spot. For example, only verifying the formulation ratio while ignoring the rheological window may generate candidates with excessive viscosity that may cause screen blockage.

[0048] Secondly, based on the constraint parameter set, the formulation vector in the candidate conductive silver paste material representation is subjected to nonnegation, normalization, and upper / lower limit truncation to obtain the candidate conductive silver paste material representation after formulation verification. Specifically, nonnegation sets the negative mass fraction components to zero, normalization makes the sum of the proportions of all components equal to 1, and upper / lower limit truncation restricts the silver content to the range of 60%-85%, the resin content to the range of 5%-25%, and the solvent and additive content to the range of 5%-20%.

[0049] Then, based on the rheological window constraint and sintering film formation constraint in the constraint parameter set, boundary truncation and differentiable projection processing are performed on the process vector in the candidate conductive silver paste material representation after formulation verification, resulting in the candidate conductive silver paste material representation after process verification. Specifically, boundary truncation limits the mixing time to 10-120 min, the dispersion speed to 500-3000 rpm, the grinding gap to 5-50 μm, and the curing temperature to within the allowable range of the equipment. Differentiable projection uses the Sigmoid function to approximate hard truncation, with the following formula: in This represents the output value of the process parameters after differential projection; Indicates the upper limit boundary value of the process parameter; Indicates the lower limit boundary value of the process parameter; This represents the Sigmoid activation function.

[0050] Next, based on the forbidden zone constraints of the failure sample set, the latent space distance between the candidate conductive silver paste material representation after process verification and the center of the failure sample cluster is calculated to obtain the forbidden zone distance value. Specifically, the candidate is mapped to latent space coordinates, the distance to the center of each failure cluster is calculated using the Euclidean norm, and the minimum distance is taken as the forbidden zone distance value.

[0051] Finally, when the forbidden zone distance value is greater than the preset forbidden zone radius threshold, the candidate conductive silver paste material representation after process verification is determined as a feasible candidate conductive silver paste material representation. When the forbidden zone distance value is less than or equal to the preset forbidden zone radius threshold, the candidate is determined to fall into the failed sample forbidden zone, a positive penalty value is generated, and it is removed.

[0052] Step S50: Calculate the candidate acquisition score based on the performance prediction value, prediction uncertainty, constraint residual of the silver paste constraint library, preset experimental cost and distance between candidates, and select the next round of experimental batches from the feasible candidate conductive silver paste material representations based on the candidate acquisition score.

[0053] It should be noted that step S50 includes: first, calculating a low resistivity benefit term based on the predicted volume resistivity and sheet resistance values ​​in the performance prediction values, and calculating a high adhesion benefit term based on the predicted adhesion values ​​in the performance prediction values, thus obtaining a set of performance benefit terms. Specifically, the low resistivity benefit term is in logarithmic reciprocal form. The square resistance benefit term adopts a negative linear form. The adhesion benefit term is in a positive linear form. The three are combined to obtain the performance benefit item set. The reason for using a logarithmic form to handle resistivity is that volume resistivity spans multiple orders of magnitude, and linear weighting would cause the differences in the fractional part to be submerged. The logarithmic transformation gives equal weight to improvements of different orders of magnitude, while the range of variation of sheet resistance and adhesion is relatively concentrated, so a linear form is sufficient.

[0054] Secondly, the uncertainty exploration term is obtained by weighted summation of the uncertainties of each performance index in the prediction uncertainty. Specifically, the uncertainty of volume resistivity is... Uncertainty of resistance Adhesion uncertainty and rheological window uncertainty The uncertainty exploration term is obtained by weighting and summing the terms according to their respective preset weights. The specific formula is as follows: in This represents the weight of the uncertainty in volume resistivity; The weight of the sheet resistance uncertainty; This represents the weight of the adhesion uncertainty; This represents the weight of the rheological window uncertainty.

[0055] Then, a constraint violation penalty term is calculated based on the constraint residuals of the silver paste constraint library, and a diversity distance term is calculated based on the latent space distance between feasible candidate conductive silver paste materials and the selected batches. Specifically, the constraint violation penalty term is composed of a weighted average of inequality constraint residuals, equality constraint residuals, and failure exclusion penalties, while the diversity distance term is calculated based on the latent space distance between feasible candidates and the selected batches. The minimum distance between each candidate latent space is calculated using the following formula: in Indicating candidate conductive silver paste samples With the selected batch The diversity distance term between them; Indicates the selected batch The first in One candidate; Indicates the first Material latent variables for each conductive silver paste sample; Indicates the selected candidate The latent variables of the material are represented.

[0056] Next, a multi-objective acquisition function is constructed based on the performance benefit term set, uncertainty exploration term, constraint violation penalty term, diversity distance term, and preset experimental cost, yielding candidate acquisition scores corresponding to each feasible candidate conductive silver paste material representation. The multi-objective acquisition function is expressed as: in Indicates a feasible candidate Candidate collection scores; Indicates the first One feasible candidate conductive silver paste sample; This indicates the weight of the low resistivity benefit term; Indicates a feasible candidate The predicted value of volume resistivity; This indicates the weight of the low square resistance benefit term; Indicates a feasible candidate The predicted value of the sheet resistance; This indicates the weight of the high adhesion benefit item; Indicates a feasible candidate The predicted adhesion value; Indicates the weight of the uncertainty exploration term; Indicates a feasible candidate Uncertainty exploration term; Indicates a feasible candidate The prediction uncertainty vector; Indicates the weight of the penalty item for constraint violation; Indicates a feasible candidate The sum of constrained residuals; Indicates the weight of the experiment cost item; Indicates a feasible candidate The estimated experimental cost; Indicates the weight of the diversity distance term; Indicates optional candidate conductive silver paste samples With the selected batch The diversity distance term between them.

[0057] Finally, based on the candidate acquisition score, the feasible candidate conductive silver paste materials are sorted in descending order to obtain a candidate priority sequence. According to the candidate priority sequence and the preset batch size, candidates that meet the condition that the distance between candidates is greater than the preset diversity distance threshold are selected to obtain the next round of experimental batches. Specifically, starting from the head of the priority sequence, the latent space distance between the current candidate and each candidate in the selected batch is compared one by one. If the minimum distance is greater than 0.5, it is included in the batch; otherwise, it is skipped and the next candidate is checked until the batch is filled or the sequence is traversed. The preset diversity distance threshold of 0.5 is determined by the latent space dimension (128 dimensions) and the candidate distribution density statistics to ensure that the candidates within the batch have significant differences in formulation and process, avoiding resource waste caused by repeated experiments with similar formulations.

[0058] Step S60: Generate and execute the formula process instructions according to the next round of experimental batches, obtain experimental test results, failure reasons and neighborhood node summary data, and perform adaptive parameter updates on the performance predictor and candidate generator according to the experimental test results, failure reasons and neighborhood node summary data, to obtain the updated performance predictor and updated candidate generator.

[0059] It should be noted that step S60 includes: First, generating a silver paste preparation instruction based on the formulation parameters in the next experimental batch, and sending the silver paste preparation instruction to the preparation equipment so that the preparation equipment can perform dispersion and grinding operations to obtain the silver paste to be tested. Specifically, the formulation parameters include the mass fraction of flake silver powder, the mass fraction of spherical silver powder, the mass fraction of nano silver, the type and mass fraction of resin, the type and mass fraction of solvent, the type and mass fraction of dispersant, the mass fraction of leveling agent, the mass fraction of curing agent, and the mass fraction of coupling agent. The silver paste preparation instruction includes the order of adding each component, the premixing time (10-30 min), the dispersion speed (500-3000 rpm), and the three-roll mill gap (5-50 μm). After receiving the instruction, the preparation equipment adds each component to the planetary mixer in a preset order, first premixing at low speed to initially wet the silver powder and resin carrier, then dispersing at high speed to break up the silver powder agglomerates, and finally optimizing the particle size distribution by controlling the gap of the three-roll mill to obtain a silver paste to be tested with uniform viscosity and no obvious sedimentation.

[0060] Secondly, printing instructions are generated based on the printing parameters in the next round of experiments and sent to the printing equipment so that the printing equipment can print the silver paste to be tested into a film, thus obtaining a printed film layer. Specifically, the printing parameters include screen mesh count (200-400 mesh), printing speed (50-200mm / s), squeegee pressure (0.1-0.5MPa), and printing linewidth (50-500μm). The printing instructions include screen installation parameters, squeegee angle (60-75°), printing gap, and ink return speed. After receiving the instructions, the printing equipment installs the screen with the corresponding mesh count, adjusts the squeegee angle and pressure to fully fill the silver paste at the screen opening, and prints the silver paste to be tested onto the substrate surface at the set speed to obtain a printed film layer with the target linewidth and film thickness. If the printing speed is too low, the silver paste will be over-spread, resulting in increased linewidth; if it is too high, insufficient filling will cause broken lines. Through precise control of the instructions, the thickness deviation of the printed film layer is controlled within ±10%.

[0061] Then, a curing instruction is generated based on the curing parameters in the next batch of experiments, and sent to the curing equipment to sinter or cure the printed film layer to obtain a cured film layer. Specifically, the curing parameters include curing temperature (80-200℃), holding time (10-60min), heating rate (1-5℃ / min), and sintering atmosphere (nitrogen or air). The curing instruction includes temperature profile setting, atmosphere flow control (5-20L / min), and cooling method. After receiving the instruction, the curing equipment heats the printed film layer to the target temperature according to the set heating rate. Under nitrogen protection, oxidation of the silver particle surface is avoided. During the holding period, the resin gradually crosslinks and cures, and sintering necks are formed between the silver particles. After cooling to room temperature, a cured film layer with a continuous conductive network and stable adhesion is obtained. If the heating rate is too fast, the solvent will evaporate rapidly, resulting in bubbles and pores. If it is too slow, the production cycle will be prolonged. The setting of 1-5℃ / min takes into account both film quality and process efficiency.

[0062] Next, test instructions are generated and sent to the testing equipment to perform conductivity, adhesion, and rheological tests on the cured film layer, obtaining experimental test results and reasons for failure. Specifically, the test instructions include the coordinates of the four-probe test points, the specifications of the cross-cut adhesion test tape (3M 600 type), the cross-cut spacing (1mm), and the shear rate range of the rheometer test. After receiving the instructions, the testing equipment measures the sheet resistance at preset points on the surface of the cured film layer using the four-probe tester and calculates the volume resistivity. The cross-cut tester applies and removes the adhesive tape after crossing the lines to evaluate the adhesion level (0-5). The rheometer measures the viscosity-shear rate curve of the same batch of silver paste to be tested. If the volume resistivity is higher than the target threshold, it is recorded as a conductivity failure; if the adhesion level is lower than the target threshold, it is recorded as an adhesion failure; if the viscosity exceeds the printable window, it is recorded as a rheological failure. At the same time, optical microscope images are acquired to determine whether there are failure modes such as screen blockage, broken lines, cracking, or film discontinuity. The test values ​​and failure modes are encoded as experimental test results and reasons for failure.

[0063] Next, the model summary, pseudo-label sample summary, synthetic candidate statistics, and validation evaluation metrics are received from the neighboring nodes to obtain the neighboring node summary data. Specifically, the model summary includes the low-rank approximation of the encoder network parameters of the neighboring nodes (e.g., extracting the top 50 principal components through principal component analysis) and the sign distribution of the predictor network parameters (positive weight proportion, negative weight proportion, and zero weight proportion). The pseudo-label sample summary includes the histogram of the predicted label distribution of the neighboring nodes for the local untested candidates (binned by volume resistivity, sheet resistance, and adhesion). The synthetic candidate statistics include the mean vector, covariance matrix, and constraint satisfaction rate of the candidates generated by the neighboring nodes. The validation evaluation metrics include the mean square error of the neighboring node validation set, the candidate manufacturability rate, and the constraint violation rate. The above data is processed for differential privacy (adding Laplace noise with a scale parameter of 0.01) and then transmitted to the local node through the network. The local node parses the data to obtain the neighboring node summary data.

[0064] Then, the experimental test results and failure reasons are fed back to the local node to update the local sample library and the forbidden zone of failed samples. Based on the summary data of the neighboring nodes, the similarity of the material system between the neighboring nodes and the local node, the verification quality score and the constraint violation rate are calculated. Specifically, the volume resistivity, sheet resistance, adhesion, viscosity curves, and failure modes from the experimental test results are encoded into a new unified material representation and added to the local sample library. At the same time, the latent space coordinates corresponding to the failure reasons are added to the failure sample cluster. If the number of samples in the cluster exceeds a preset threshold (10), the cluster center and the radius of the restricted area are recalculated, and the restricted area of ​​the failure samples is updated. The material system similarity is obtained by weighted summation of the cosine similarity of the silver powder system (cosine of the angle between the vectors of the proportion of flake silver powder, the proportion of spherical silver powder, and the proportion of nano silver powder, with a weight of 0.4), the resin carrier similarity (resin type identification matching, with a weight of 0.2), and the overlap of the printing process window (the intersection ratio of the curing temperature range, with a weight of 0.4), with a value range of [-1,1]. The verification quality score is obtained by multiplying the exponential decay function of the mean square error of the verification set by the candidate manufacturability, with a value range of [0,1]. The constraint violation rate is directly taken from the proportion of candidates that violate the silver paste constraint library generated by the neighboring nodes.

[0065] Finally, aggregation weights are calculated based on material system similarity, verification quality score, and constraint violation rate. Adaptive parameter updates are then performed on the performance predictor and candidate generator based on these aggregation weights, resulting in updated performance predictors and candidate generators. Specifically, aggregation weights are calculated based on the three indicators mentioned above. A weighted average update is then performed on the local performance predictor network parameters and candidate generator network parameters based on these aggregation weights. This update method allows local nodes to retain their own data characteristics while absorbing knowledge from neighboring regions with similar material systems, high verification quality, and low constraint violation rates, resulting in updated performance predictors and candidate generators.

[0066] Step S70: Based on the updated performance predictor and the updated candidate generator, candidate conductive silver paste formulations and process parameters that meet the target performance requirements are generated within the feasible domain of the silver paste constraint library, and the optimized conductive silver paste formulations and process parameters are determined through multi-objective screening.

[0067] Specifically, the target performance conditions are first input into the updated candidate generator. The generator combines the feasible region of the silver paste constraint library to complete sampling and feature projection, and outputs multiple sets of candidate conductive silver paste formulations and process parameters. Then, all candidates are sent to the updated performance predictor to obtain the predicted values ​​of each performance for each candidate. Subsequently, the constraint residuals of each candidate are calculated in combination with the silver paste constraint library. A scoring function is built based on the performance prediction values, constraint residuals, and preparation costs. All candidates are sorted according to the scoring results. Finally, the top-ranked candidate is selected as the optimized conductive silver paste formulation and process parameters.

[0068] This embodiment collects silver paste samples and encodes them for unified material characterization. A dedicated constraint library is built, including silver powder, rheology, sintering, and failure-prone areas. Candidate samples are generated using a dual-model approach, and performance and uncertainty are predicted. Feasible samples are screened through constraint projection, and experimental batches are selected through multi-objective scoring. Closed-loop trial production testing is conducted, and the model is adaptively updated using desensitized neighborhood data. Finally, the optimized conductive silver paste formulation and process parameters are determined through multi-objective screening. This approach reduces invalid trial production, balances low resistance, high adhesion, and printing stability, eliminates the need to share original confidential data, and allows for efficient multi-objective collaborative optimization even with small sample nodes.

[0069] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 The conductive silver paste formulation process optimization method step S70 further includes steps S201~S206: Step S201: Input the target performance conditions into the updated candidate generator, and perform reverse denoising sampling and feasibility projection under the feasible region constraints of the silver paste constraint library to obtain a set of candidate conductive silver paste formulations and process parameters.

[0070] Specifically, the target performance conditions include the upper limit of volume resistivity, the upper limit of sheet resistance, the lower limit of adhesion, and the rheological window requirements. The updated candidate generator uses these target values ​​as condition vectors and gradually converges to the region that meets the target performance conditions in the latent space, starting from standard Gaussian noise. After each denoising step, the latent variables are mapped to the feasible region of the silver paste constraint library by the projection operator. Candidates that violate the silver content boundary, fall into the failure sample exclusion zone, or exceed the rheological window are corrected or eliminated. Finally, the data is decoded into a formula vector and a process vector to obtain the candidate set. In this example, the size of the candidate set is controlled between 50 and 200 to balance the diversity of generation with the efficiency of subsequent screening calculations. The updated candidate generator is used instead of the initial model because after multiple rounds of experimental feedback and neighborhood knowledge aggregation, the generator fits the distribution of the local material system more accurately and can generate candidates that are closer to the real feasible region, reducing invalid sampling.

[0071] Step S202: Input the candidate conductive silver paste formulation and process parameter set into the updated performance predictor to obtain the predicted volume resistivity, predicted sheet resistance, predicted adhesion, predicted rheological window value, and the prediction uncertainty corresponding to each candidate conductive silver paste formulation and process parameter.

[0072] Specifically, the updated performance predictor performs forward computation in parallel for each candidate in the candidate set. The conductive network predictor extracts the resistivity prediction value from the silver powder gradation and sintering neck feature dimensions. The rheological printing window predictor extracts the rheological window prediction value from the particle morphology and resin system interaction feature dimensions. The adhesion predictor extracts the adhesion prediction value from the resin crosslinking and interface bonding feature dimensions. The prediction uncertainty is calculated by the variance of the ensemble model. The number of ensemble models is 5-10. Each sub-model is trained using different random initializations or data subsampling. The greater the discrepancy in the prediction values, the more likely the candidate is to be located at the cognitive boundary of the model, and the higher the uncertainty. This step is equivalent to performing a virtual test on the generated candidates, quickly screening them before putting them into actual experiments, and avoiding sending candidates that obviously do not meet the target requirements into the preparation stage.

[0073] Step S203: Calculate the constraint residuals of each candidate conductive silver paste formulation and process parameters based on the silver paste constraint library to obtain the constraint satisfaction state, and calculate the comprehensive performance score based on the predicted volume resistivity, predicted sheet resistance, predicted adhesion, predicted rheological window value and predicted uncertainty.

[0074] Specifically, the constraint residuals consist of inequality constraint residuals, equality constraint residuals, rheological window residuals, film continuity residuals, and a weighted average of failure exclusion penalty, as shown in the following formula: in This represents the sum of constraint residuals for the candidate conductive silver paste samples; Indicates a unified material representation; Indicates the first Weighting coefficients of inequality constraints; Indicates the index of inequality constraints; Indicates the first The term inequality constraint function outputs the constraint deviation. This represents a truncation function that only penalizes candidates that violate the constraints, and penalizes zero when the constraints are satisfied. Indicates the index of equality constraints; Indicates the first Equality constraint function; Indicates the first The term equality constraint function outputs the absolute value of the constraint deviation. The weighting coefficients represent discrete, non-manufacturable indicators; This is an indicator function representing discrete unmanufacturable conditions. It outputs 1 when the candidate involves an unavailable combination of raw materials or is outside the equipment range, and 0 otherwise. The weighting coefficients represent the residuals of the rheological window; This represents the rheological window residual function, used to calculate the integral deviation between the candidate viscosity curve and the printable viscosity window. The weighting coefficient representing the residual error in film formation continuity; The residual function representing film continuity is used to calculate the deviation between the candidate film porosity, sintering neck size, and continuity constraint targets. The normalized residuals are summed into a single scalar. Constraint satisfaction is represented by Boolean values ​​or hierarchical forms: zero or below a preset threshold indicates complete satisfaction; above the threshold indicates partial violation; and falling into the failure zone indicates severe violation. The comprehensive performance score uses a weighted approach: the lower the predicted volume resistivity and sheet resistance values ​​are compared to the target upper limit, the higher the predicted adhesion value is compared to the target lower limit, and the closer the predicted rheological window value is to the center of the target interval, the higher the score. Prediction uncertainty within a reasonable range is given bonus points to preserve exploratory value, while exceeding a reasonable range results in deductions to suppress unreliable predictions. This scoring mechanism transforms multi-objective optimization into a comparable single scalar, avoiding the complex trade-offs of the Pareto front.

[0075] Step S204: Construct a multi-objective screening function based on constraint satisfaction status, comprehensive performance score and preset experimental cost to obtain multi-objective screening values ​​for each candidate conductive silver paste formulation and process parameters.

[0076] Specifically, the multi-objective screening function adopts a linear weighted form, with the constraint satisfaction state having the highest weight, followed by the comprehensive performance score, and the experimental cost having a lower weight. Candidates that seriously violate the constraints are directly assigned the lowest screening value and eliminated. The cost item is estimated by the precious metal content in the formula and the process complexity. Formulas with high silver content have higher costs, and multi-step grinding processes have higher complexity. This weight allocation reflects an optimization strategy that prioritizes physical feasibility, followed by performance compliance, and takes cost control into account, so that the screening results not only meet basic manufacturing requirements but also have the potential for performance improvement.

[0077] Step S205: Sort the candidate conductive silver paste formulations and process parameter sets in descending order based on the multi-objective screening values ​​to obtain the candidate priority sequence.

[0078] Specifically, a stable sorting algorithm is used for sorting. For candidates with the same multi-objective screening value, they are sorted in descending order of prediction uncertainty. Candidates with high uncertainty are retained first to increase the value of information exploration. The head of the candidate priority sequence is the best candidate and the tail is the worst candidate. The sequence length is consistent with the size of the candidate set.

[0079] Step S206: Select the candidate conductive silver paste formulation and process parameters with the highest multi-objective screening value from the candidate priority sequence, and determine them as the optimized conductive silver paste formulation and process parameters.

[0080] Specifically, during selection, the potential space distance between the candidate and historically tested samples and other candidates in this round is checked to ensure that the distance is greater than the preset diversity threshold (0.5). This avoids recommending formulations that are too similar to historical successful samples, which would cause experimental redundancy. If the top candidate does not meet the diversity requirements, the process is repeated until a candidate that meets the conditions is found. The optimized conductive silver paste formulation and process parameters include the silver powder composition ratio, resin carrier type and content, solvent and auxiliary agent ratio, mixing time, dispersion speed, grinding gap, screen mesh number, printing speed, curing temperature and holding time. These parameters can be directly used to guide actual preparation experiments or production trials.

[0081] This embodiment generates a set of candidate conductive silver paste formulations and process parameters by inputting the target performance conditions into an updated candidate generator, performing reverse denoising sampling and feasibility projection under the constraints of the feasible region of the silver paste constraint library. The updated performance predictor outputs the predicted volume resistivity, sheet resistance, adhesion, rheological window value, and prediction uncertainty for each candidate. The constraint satisfaction state is calculated by combining the constraint residuals, and the performance prediction value and uncertainty are combined to calculate the score. The candidate priority sequence is obtained by sorting through a multi-objective screening function, and the optimal candidate is selected as the optimized conductive silver paste formulation and process parameters. The updated model generation and screening improves the accuracy of candidate formulations; the separation of constraint residuals and performance scores avoids misjudgments of feasibility and superiority; and the multi-objective screening and diversity check balance performance achievement, information exploration, and cost control, reducing the risk of experimental failure and R&D costs.

[0082] Based on the first embodiment of this application, this application also provides a device for optimizing the formulation process of conductive silver paste. Please refer to... Figure 3 The device includes: The data encoding module 10 is used to acquire local node conductive silver paste sample data and encode the sample data into a unified material representation.

[0083] The constraint generation module 20 is used to generate a silver paste constraint library based on a unified material representation and preset process boundaries. The silver paste constraint library includes silver powder morphology constraints, rheological window constraints, sintering film formation constraints, and failure sample exclusion zones.

[0084] The prediction module 30 is used to input the silver paste constraint library, target performance conditions and unified material representation into the candidate generator and performance predictor, respectively, to obtain the candidate conductive silver paste material representation and the corresponding performance prediction value and prediction uncertainty.

[0085] The feasible projection module 40 is used to perform feasible projection on the candidate conductive silver paste material representation based on the silver paste constraint library to obtain a feasible candidate conductive silver paste material representation.

[0086] The acquisition selection module 50 is used to calculate the candidate acquisition score based on the performance prediction value, prediction uncertainty, constraint residual of the silver paste constraint library, preset experimental cost and distance between candidates, and select the next round of experimental batches from the feasible candidate conductive silver paste material representations based on the candidate acquisition score.

[0087] The feedback update module 60 is used to generate and execute the formula process instructions according to the next round of experimental batches, obtain experimental test results, failure reasons and neighborhood node summary data, and perform adaptive parameter updates on the performance predictor and candidate generator according to the experimental test results, failure reasons and neighborhood node summary data, so as to obtain the updated performance predictor and updated candidate generator.

[0088] The results module 70 is used to generate candidate conductive silver paste formulations and process parameters that meet the target performance requirements within the feasible domain of the silver paste constraint library based on the updated performance predictor and the updated candidate generator, and to determine the optimized conductive silver paste formulations and process parameters through multi-objective screening.

[0089] The conductive silver paste formulation and process optimization apparatus provided in this application, employing the conductive silver paste formulation and process optimization method described in the above embodiments, can solve the technical problem of how to generate and screen candidate conductive silver paste formulations and processes that meet the constraints of silver powder morphology, rheological window, sintering film formation, and failed samples. Compared with the prior art, the beneficial effects of the conductive silver paste formulation and process optimization apparatus provided in this application are the same as those of the conductive silver paste formulation and process optimization method provided in the above embodiments, and other technical features in the conductive silver paste formulation and process optimization apparatus are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0090] This application provides a conductive silver paste formulation process optimization device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the conductive silver paste formulation process optimization method in the above embodiment 1.

[0091] The following is for reference. Figure 4 The diagram illustrates a structural schematic of a conductive silver paste formulation process optimization device suitable for implementing embodiments of this application. The conductive silver paste formulation process optimization device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 4 The conductive silver paste formulation process optimization equipment shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0092] like Figure 4As shown, the conductive silver paste formulation process optimization device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the conductive silver paste formulation process optimization device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the conductive silver paste formulation process optimization equipment to communicate wirelessly or wiredly with other devices to exchange data. Although various conductive silver paste formulation process optimization equipment are shown in the figures, it should be understood that implementation or possession of all of them is not required. More or fewer may be implemented alternatively.

[0093] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0094] The conductive silver paste formulation and process optimization equipment provided in this application, employing the conductive silver paste formulation and process optimization method described in the above embodiments, can solve the technical problem of how to generate and screen candidate conductive silver paste formulations and processes that meet the constraints of silver powder morphology, rheological window, sintering film formation, and failed samples. Compared with the prior art, the beneficial effects of the conductive silver paste formulation and process optimization equipment provided in this application are the same as those of the conductive silver paste formulation and process optimization method provided in the above embodiments, and other technical features in this conductive silver paste formulation and process optimization equipment are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0095] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0096] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0097] This application provides a computer-readable medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the conductive silver paste formulation process optimization method in the above embodiments.

[0098] The computer-readable medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor devices, or any combination thereof. More specific examples of computer-readable media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable medium may be any tangible medium containing or storing a program that can be executed by instructions, used by a device, or used in conjunction with it. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0099] The aforementioned computer-readable medium may be included in the conductive silver paste formulation process optimization equipment; or it may exist independently and not assembled into the conductive silver paste formulation process optimization equipment.

[0100] The aforementioned computer-readable medium carries one or more programs that, when executed by the conductive silver paste formulation process optimization device, enable the device to write computer program code for performing the operations of this application in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0101] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this application. In this regard, all blocks in the flowcharts or block diagrams may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that all blocks in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using dedicated hardware-based implementations that perform the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0102] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0103] The readable medium provided in this application is a computer-readable medium, which stores computer-readable program instructions (i.e., a computer program) for executing the above-described conductive silver paste formulation and process optimization method. This solves the technical problem of how to generate and screen candidate conductive silver paste formulations and processes that meet the constraints of silver powder morphology, rheological window, sintering film formation, and failed samples. Compared with the prior art, the beneficial effects of the computer-readable medium provided in this application are the same as those of the conductive silver paste formulation and process optimization method provided in the above embodiments, and will not be repeated here.

[0104] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the conductive silver paste formulation process optimization method described above.

[0105] The computer program product provided in this application can solve the technical problem of how to generate and screen candidate conductive silver paste formulations and processes that meet the constraints of silver powder morphology, rheological window, sintering film formation, and failed samples. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the conductive silver paste formulation and process optimization method provided in the above embodiments, and will not be repeated here.

[0106] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for optimizing the formulation and process of conductive silver paste, characterized in that, The method includes: Acquire local node conductive silver paste sample data and encode the sample data into a unified material representation; A silver paste constraint library is generated based on the unified material representation and preset process boundaries. The silver paste constraint library includes silver powder morphology constraints, rheological window constraints, sintering film formation constraints, and failure sample exclusion zones. The silver paste constraint library, target performance conditions, and unified material representation are input into the candidate generator and performance predictor, respectively, to obtain the candidate conductive silver paste material representation and the corresponding performance prediction value and prediction uncertainty. Feasibility projection is performed on the candidate conductive silver paste material representations according to the silver paste constraint library to obtain feasible candidate conductive silver paste material representations; The candidate acquisition score is calculated based on the performance prediction value, prediction uncertainty, constraint residual of the silver paste constraint library, preset experimental cost and distance between candidates, and the next round of experimental batches is selected from the feasible candidate conductive silver paste material representations based on the candidate acquisition score. The formula process instruction is generated and executed according to the next round of experimental batches to obtain experimental test results, failure reasons and neighborhood node summary data. Adaptive parameter updates are performed on the performance predictor and candidate generator according to the experimental test results, failure reasons and neighborhood node summary data to obtain the updated performance predictor and updated candidate generator. Based on the updated performance predictor and the updated candidate generator, candidate conductive silver paste formulations and process parameters that meet the target performance requirements are generated within the feasible domain of the silver paste constraint library, and the optimized conductive silver paste formulations and process parameters are determined through multi-objective screening. The step of inputting the silver paste constraint library, target performance conditions, and unified material representation into the candidate generator and performance predictor, respectively, to obtain the candidate conductive silver paste material representation and the corresponding performance prediction value and prediction uncertainty, includes: The unified material representation is input into the performance predictor for latent variable mapping and feature extraction to obtain the material latent variable representation. The latent variables of the material are input into the conductive network predictor, rheological printing window predictor, sintering film predictor and adhesion predictor of the performance predictor, respectively, to obtain the volume resistivity prediction value, sheet resistance prediction value, rheological window prediction value, adhesion prediction value and the prediction uncertainty corresponding to each prediction value. The target performance conditions and the silver paste constraint library are input into the candidate generator, and reverse denoising sampling is performed under the feasible region constraint of the silver paste constraint library to obtain the candidate conductive silver paste material representation. The candidate conductive silver paste material representation is input into the performance predictor to perform performance prediction, and the corresponding performance prediction value and prediction uncertainty of the candidate conductive silver paste material representation are obtained. Based on the silver paste constraint library, a feasibility projection verification is performed on the candidate conductive silver paste material representation. Candidates that violate the forbidden zones of failed samples in the silver paste constraint library are eliminated, and the verified candidate conductive silver paste material representation and the corresponding performance prediction value and prediction uncertainty are obtained.

2. The method as described in claim 1, characterized in that, The step of acquiring local node conductive silver paste sample data and encoding the sample data into a unified material representation includes: Acquire local node conductive silver paste sample data, wherein the sample data includes silver powder composition data, resin carrier data, solvent and additive data, process data, printing film data, structural characterization data, rheological property data and conductivity test data; The process data in the sample data are subjected to dimension unification and process window normalization to obtain standardized process data. The rheological performance data in the sample data are curve encoded to obtain a rheological vector; The conductivity test data in the sample data are sequentially encoded to obtain a conductivity vector. The silver powder composition data, resin carrier data, solvent and additive data are mapped into a formulation vector; The standardized process data is mapped to a process vector, and the structural characterization data is mapped to a structural vector; The rheological vector and electrical conductivity vector are concatenated with the formulation vector, process vector, and structural vector to obtain a unified material representation.

3. The method as described in claim 1, characterized in that, The step of performing feasibility projection on the candidate conductive silver paste material representation based on the silver paste constraint library to obtain feasible candidate conductive silver paste material representations includes: From the silver paste constraint library, silver powder morphology constraints, rheological window constraints, sintering film formation constraints, and failure sample exclusion zones are extracted to obtain a set of constraint parameters. Based on the set of constraint parameters, the formulation vector in the candidate conductive silver paste material representation is nonnegated, normalized, and truncated to upper and lower limits to obtain the candidate conductive silver paste material representation after formulation verification. Based on the rheological window constraint and sintering film formation constraint in the set of constraint parameters, the process vector in the candidate conductive silver paste material representation after formula verification is subjected to boundary truncation and differentiable projection processing to obtain the candidate conductive silver paste material representation after process verification. Based on the forbidden zone constraints of the failed samples in the set of constraint parameters, the latent space distance between the candidate conductive silver paste material representation after process verification and the center of the failed sample cluster is calculated to obtain the forbidden zone distance value. When the distance value of the restricted area is greater than the preset restricted area radius threshold, the candidate conductive silver paste material representation after process verification is determined as a feasible candidate conductive silver paste material representation.

4. The method as described in claim 1, characterized in that, The step of calculating candidate acquisition scores based on the predicted performance value, prediction uncertainty, constraint residuals of the silver paste constraint library, preset experimental cost, and distance between candidates, and selecting the next batch of experiments from the feasible candidate conductive silver paste materials representations based on the candidate acquisition scores, includes: The low resistivity benefit term is calculated based on the volume resistivity prediction value and the sheet resistance prediction value in the performance prediction value, and the high adhesion benefit term is calculated based on the adhesion prediction value in the performance prediction value, thus obtaining a set of performance benefit terms. The uncertainty exploration term is obtained by weighted summation of the uncertainties of each performance index in the predicted uncertainty; The constraint violation penalty term is calculated based on the constraint residuals of the silver paste constraint library, and the diversity distance term is calculated based on the latent space distance between the feasible candidate conductive silver paste material representation and the selected batch. A multi-objective acquisition function is constructed based on the set of performance benefit terms, uncertainty exploration terms, constraint violation penalty terms, diversity distance terms, and preset experimental costs to obtain the candidate acquisition scores corresponding to each feasible candidate conductive silver paste material representation. Based on the candidate acquisition scores, the feasible candidate conductive silver paste materials are sorted in descending order to obtain a candidate priority sequence; Based on the candidate priority sequence and the preset batch size, candidates that satisfy the condition that the distance between candidates is greater than the preset diversity distance threshold are selected in turn to obtain the next round of experimental batches.

5. The method as described in claim 1, characterized in that, The steps of generating and executing the formulation process instructions based on the next round of experimental batches to obtain experimental test results, failure reasons, and neighborhood node summary data, and then performing adaptive parameter updates on the performance predictor and candidate generator based on the experimental test results, failure reasons, and neighborhood node summary data to obtain updated performance predictors and updated candidate generators, include: A silver paste preparation instruction is generated based on the formula parameters in the next round of experimental batches, and the silver paste preparation instruction is sent to the preparation equipment so that the preparation equipment performs a dispersion and grinding operation to obtain the silver paste to be tested. A printing instruction is generated based on the printing parameters in the next round of experimental batches, and the printing instruction is sent to the printing equipment so that the printing equipment can print the silver paste to be tested into a film to obtain a printed film layer. A curing instruction is generated based on the curing parameters in the next round of experiments, and the curing instruction is sent to the curing equipment so that the curing equipment can sinter or cure the printed film layer to obtain a cured film layer. A test command is generated and sent to the test equipment, so that the test equipment performs conductivity test, adhesion test and rheology test on the cured film layer, and obtains the experimental test results and the reasons for failure. The model summary, pseudo-label sample summary, synthetic candidate statistics, and validation evaluation metrics are received from the neighboring nodes to obtain the neighboring node summary data. The experimental test results and failure reasons are fed back to the local node to update the local sample library and the forbidden zone of failed samples. The material system similarity, verification quality score and constraint violation rate of the neighboring node and the local node are calculated based on the neighboring node summary data. The aggregation weight is calculated based on the material system similarity, verification quality score and constraint violation rate, and the performance predictor and the candidate generator are adaptively updated based on the aggregation weight to obtain the updated performance predictor and the updated candidate generator.

6. The method as described in claim 1, characterized in that, The step of generating candidate conductive silver paste formulations and process parameters that meet the target performance requirements within the feasible region of the silver paste constraint library based on the updated performance predictor and the updated candidate generator, and determining the optimized conductive silver paste formulation and process parameters through multi-objective screening, includes: The target performance conditions are input into the updated candidate generator, and reverse denoising sampling and feasibility projection are performed under the feasible region constraints of the silver paste constraint library to obtain a set of candidate conductive silver paste formulations and process parameters. The candidate conductive silver paste formulations and process parameters are input into the updated performance predictor to obtain the predicted volume resistivity, predicted sheet resistance, predicted adhesion, predicted rheological window value, and the prediction uncertainty corresponding to each candidate conductive silver paste formulation and process parameter. The constraint residuals of each candidate conductive silver paste formulation and process parameters are calculated based on the silver paste constraint library to obtain the constraint satisfaction state. The comprehensive performance score is calculated based on the predicted volume resistivity, predicted sheet resistance, predicted adhesion, predicted rheological window value and predicted uncertainty. Based on the constraint satisfaction status, comprehensive performance score and preset experimental cost, a multi-objective screening function is constructed to obtain the multi-objective screening values ​​of each candidate conductive silver paste formulation and process parameter. Based on the multi-objective screening values, the candidate conductive silver paste formulations and process parameter sets are sorted in descending order to obtain a candidate priority sequence. The candidate conductive silver paste formulation and process parameters with the highest multi-objective screening value are selected from the candidate priority sequence and determined as the optimized conductive silver paste formulation and process parameters.

7. A device for optimizing the formulation process of conductive silver paste, characterized in that, The apparatus is applied to the conductive silver paste formulation process optimization method as described in any one of claims 1 to 6, and the apparatus comprises: The data encoding module is used to acquire local node conductive silver paste sample data and encode the sample data into a unified material representation. The constraint generation module is used to generate a silver paste constraint library based on the unified material representation and preset process boundaries. The silver paste constraint library includes silver powder morphology constraints, rheological window constraints, sintering film formation constraints, and failure sample exclusion zones. The prediction module is used to input the silver paste constraint library, target performance conditions, and unified material representation into the candidate generator and performance predictor, respectively, to obtain the candidate conductive silver paste material representation and the corresponding performance prediction value and prediction uncertainty. The feasible projection module is used to perform feasible projection on the candidate conductive silver paste material representation according to the silver paste constraint library to obtain a feasible candidate conductive silver paste material representation. The acquisition selection module is used to calculate the candidate acquisition score based on the performance prediction value, prediction uncertainty, constraint residual of the silver paste constraint library, preset experimental cost and distance between candidates, and select the next round of experimental batches from the feasible candidate conductive silver paste material representations based on the candidate acquisition score; The feedback update module is used to generate and execute the formula process instructions according to the next round of experimental batches, obtain experimental test results, failure reasons and neighborhood node summary data, and perform adaptive parameter updates on the performance predictor and candidate generator according to the experimental test results, failure reasons and neighborhood node summary data, to obtain the updated performance predictor and updated candidate generator. The results module is used to generate candidate conductive silver paste formulations and process parameters that meet the target performance requirements within the feasible domain of the silver paste constraint library based on the updated performance predictor and the updated candidate generator, and to determine the optimized conductive silver paste formulations and process parameters through multi-objective screening.

8. A conductive silver paste formulation process optimization device, characterized in that, The device includes: a memory, a processor, and a conductive silver paste formulation process optimization program stored in the memory and running on the processor, the conductive silver paste formulation process optimization program being configured to implement the steps of the conductive silver paste formulation process optimization method as described in any one of claims 1-6.

9. A storage medium, characterized in that, The storage medium stores a conductive silver paste formulation process optimization program, which, when executed by a processor, implements the steps of the conductive silver paste formulation process optimization method as described in any one of claims 1-6.

Images