Method for optimizing end face moisture resistance of impregnated paper based on machine learning

Abstract

The present application relates to a machine learning-based impregnated paper end face moisture-proof performance optimization method, comprising the following steps: S1: obtaining the initial process sample of the to-be-processed plate and uniformly pretreating to obtain the end face multi-modal feature vector; S2: obtaining the end face moisture-proof performance label and constructing the supervised learning training sample; S3: constructing the end face moisture-proof performance prediction model and training based on the supervised learning training sample to obtain the trained end face moisture-proof performance prediction model; S4: taking the trained end face moisture-proof performance prediction model as the basic model, performing sensitivity analysis on the adjustable process parameters, and obtaining the key process parameter set; S5: taking the key process parameter set as the optimization variable, constructing the multi-objective process parameter optimization model, and obtaining the optimized process parameter scheme; S6: according to the optimized process parameter scheme, implementing the differentiated impregnated paper and end face sealing and protection process control on the to-be-processed plate. The present application effectively improves the impregnated paper end face moisture-proof performance optimization efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing, and in particular to a method for optimizing the moisture-proof performance of the end face of impregnated paper based on machine learning. Background Technology

[0002] Impregnated paper-faced engineered wood panels are widely used in furniture, cabinetry, flooring, and interior decoration due to their good decorative properties, wear resistance, ease of processing, and moderate cost. However, when used in humid environments, moisture can easily penetrate the substrate through cut edges, openings, or weak edges, leading to problems such as end-face swelling, warping of the veneer layer, detachment of the edge sealant, localized cracking, and even mold growth. Compared to the flat surface of the panel, the end faces expose more wood fibers and pores, resulting in generally weaker moisture resistance. This makes them a key weak point affecting the durability and lifespan of impregnated paper-faced panels.

[0003] Existing end-face moisture-proofing processes typically involve adjusting the resin content and pick-up amount of the impregnated paper, controlling the volatile matter in the impregnated paper, optimizing hot-pressing temperature and pressure, applying an end-face primer, applying edge-sealing adhesive, and adjusting edge-sealing pressure. However, existing process parameters often rely on operator experience or single-factor experiments, making it difficult to fully consider the coupled effects of factors such as substrate density, moisture content, pore structure, impregnation solution formulation, hot-pressing conditions, edge-sealing material properties, and the temperature and humidity of the manufacturing environment. Therefore, when there are batch variations in substrate, fluctuations in environmental conditions, or significant differences in product structure, traditional processes often fail to achieve stable end-face moisture-proofing effects. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to provide a machine learning-based method for optimizing the moisture-proof performance of the impregnated paper end face, thereby effectively improving the optimization efficiency of the moisture-proof performance of the impregnated paper end face.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A machine learning-based method for optimizing the moisture-proof performance of impregnated paper end faces includes the following steps: S1: Obtain the initial process sample of the board to be processed and perform unified preprocessing to obtain the end face multimodal feature vector; S2: Obtain the end-face moisture-proof performance label, and associate the end-face multimodal feature vector with the corresponding end-face moisture-proof performance label to construct supervised learning training samples; S3: Construct a prediction model for the moisture-proof performance of the end face, and train it based on supervised learning training samples to obtain the trained prediction model for the moisture-proof performance of the end face. S4: Using the trained end-face moisture-proof performance prediction model as the base model, perform sensitivity analysis on adjustable process parameters to obtain a set of key process parameters; S5: Set the key process parameters as optimization variables, construct a multi-objective process parameter optimization model, and obtain the optimized process parameter scheme; S6: Based on the optimized process parameter scheme, implement differentiated impregnation paper and end-face sealing process control for the boards to be processed.

[0006] The machine learning-based system for optimizing the moisture-proof performance of impregnated paper end faces includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the machine learning-based method for optimizing the moisture-proof performance of impregnated paper end faces as described above.

[0007] A computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the steps of the method described above.

[0008] The present invention has the following beneficial effects: 1. This invention collects parameters of the substrate, impregnated paper, impregnating liquid formulation, hot-pressing bonding, end-face sealing, and manufacturing environment to construct a multimodal feature vector. Combined with the end-face moisture-proof performance prediction model, it enables joint prediction of end-face water absorption rate, thickness expansion rate, edge sealing peel strength, and failure risk. This allows for quantitative prediction of end-face moisture-proof performance before process execution. 2. This invention conducts sensitivity analysis on adjustable process parameters such as impregnation liquid formulation, impregnation paper pick-up amount, volatile matter, hot pressing parameters, end-face base coating amount, edge sealing adhesive coating amount, and edge sealing pressure. It then screens out a set of key process parameters that significantly affect the moisture-proof performance of the end face and constructs a multi-objective process parameter optimization model with the goals of minimizing end-face water absorption rate, minimizing thickness expansion rate, minimizing failure risk, maximizing edge sealing peel strength, and minimizing manufacturing cost. This model seeks the Pareto optimal balance among multiple mutually restrictive performance and cost objectives, avoiding the decline of other performance or increase in manufacturing cost due to the one-sided pursuit of a certain indicator. This ensures that the process parameter scheme has both moisture-proof performance guarantee and economy. Attached Figure Description

[0009] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0010] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: refer to Figure 1 In this embodiment, a method for optimizing the moisture-proof performance of impregnated paper end faces based on machine learning is provided, including the following steps: S1: Obtain the substrate parameters, impregnated paper parameters, impregnating liquid formula parameters, hot pressing and bonding parameters, end face sealing parameters, and manufacturing environment parameters of the board to be processed, form an initial process sample, and perform unified preprocessing to obtain the end face multimodal feature vector; S2: Obtain the end-face moisture-proof performance label, and associate the end-face multimodal feature vector with the corresponding end-face moisture-proof performance label to construct supervised learning training samples; S3: Construct a prediction model for the moisture-proof performance of the end face, and train it based on supervised learning training samples to obtain the trained prediction model for the moisture-proof performance of the end face. S4: Using the trained end-face moisture-proof performance prediction model as the basic model, sensitivity analysis is performed on adjustable process parameters such as impregnation liquid formulation, impregnation paper pick-up amount, volatile matter, hot pressing parameters, end-face base coating amount, edge sealing adhesive coating amount and edge sealing pressure to obtain a set of key process parameters. S5: Set the key process parameters as optimization variables, and construct a multi-objective process parameter optimization model with the optimization objectives of minimizing end face water absorption rate, minimizing end face thickness expansion rate, minimizing end face failure risk, maximizing edge sealing peel strength, and minimizing manufacturing cost, and obtain the optimized process parameter scheme. S6: Based on the optimized process parameter scheme, implement differentiated impregnation paper and end-face sealing process control for the boards to be processed.

[0011] In this embodiment, the initial process sample of the board to be processed is obtained and uniformly preprocessed to obtain the end face multimodal feature vector. Specifically, the following steps are taken: The substrate parameters, impregnated paper parameters, impregnation solution formulation parameters, hot-pressing parameters, end face sealing parameters, and manufacturing environment parameters of the board to be processed are obtained. The substrate parameters include the board type, thickness, density, moisture content, water absorption rate, thickness expansion rate, internal bond strength, static bending strength, surface density, core porosity, end face roughness, end face pore characteristics, and substrate batch information. The impregnated paper parameters include paper basis weight, thickness, air permeability, fiber composition, paper moisture content, resin pick-up amount, resin distribution uniformity, and impregnated paper volatiles. The impregnation solution formulation parameters include resin type, solid content, viscosity, pH value, type and amount of curing agent, type and amount of penetrant, type and amount of toughening agent, type and amount of waterproofing agent, and type and amount of releasing agent. The hot-pressing parameters include hot-pressing temperature, hot-pressing pressure, hot-pressing time, pressure curve, pressure release method, press speed, and curing time after lamination; the end-face sealing parameters include end-face cutting method, end-face sanding grit, type of end-face primer, primer amount, primer viscosity, primer drying time, type of edge sealing adhesive, edge sealing adhesive application amount, edge sealing tape material, edge sealing tape thickness, edge sealing temperature, edge sealing pressure, edge sealing speed, and curing time; the manufacturing environment parameters include production area temperature, relative humidity, dust concentration, storage time, and storage environment temperature and humidity; the collected parameters are categorized by board batch, production time, process section number, and... Product numbers are associated to form initial process samples corresponding to individual sheets to be processed, which are then subjected to unified preprocessing. For continuous parameters, outlier identification, missing value completion, dimensional unification, and standardization are employed. For categorical parameters, tag coding, unique thermal coding, or embedded coding are used. For end-face images, microscopic images, or infrared detection images, cropping, noise reduction, brightness correction, scale unification, and feature enhancement are performed. For time-series process data such as temperature, pressure, speed, and humidity, time alignment is performed according to a unified sampling period, and statistical features such as mean, extreme values, fluctuation amplitude, rate of change, and stabilization time are extracted. Data with batch differences, measurement errors, or missing information can be corrected or removed based on samples from the same batch, adjacent process records, and preset quality rules to improve the reliability and consistency of sample data. Based on unified preprocessing, features are extracted and fused from parameters from different sources to obtain end-face multimodal feature vectors. The end-face multimodal feature vector includes at least the substrate structure features, impregnation features of impregnated paper, impregnation liquid formulation features, hot pressing features, end-face sealing features, manufacturing environment features, and end-face texture features, pore features, and edge sealing adhesive layer continuity features obtained from image data.

[0012] In this embodiment, obtaining the end-face moisture-proof performance label and constructing a training sample set, and associating the end-face multimodal feature vector with the corresponding end-face moisture-proof performance label to construct supervised learning training samples, is specifically as follows: The end-face moisture-proof performance of the board material after impregnation with paper and end-face sealing is tested to obtain the end-face moisture-proof performance label; the performance label includes end-face water absorption rate, end-face thickness expansion rate, end-face edge sealing peel strength, end-face adhesive layer integrity, end-face crack or blistering level, end-face mildew level, and end-face failure risk level; wherein, the end-face water absorption rate can be determined based on the mass change of the sample before and after water absorption, the end-face thickness expansion rate can be determined based on the thickness change of the adjacent area of ​​the end face before and after immersion, and the edge sealing peel strength can be obtained through tensile or peel tests; accelerated moisture resistance tests are conducted under specified temperature, humidity, and immersion time conditions, and end-face cracking, delamination, warping, blistering, and mildew are recorded through visual inspection, image recognition, or manual inspection. Failure status; For test results with different dimensions and evaluation methods, the moisture-proof performance labels of the end face are uniformly processed; for continuous indicators such as end face water absorption rate, thickness expansion rate, and peel strength, outlier removal, dimension normalization, or interval mapping can be performed; for discrete indicators such as end face appearance defect level, adhesive layer integrity level, and failure risk level, they can be classified into categories such as qualified, low-risk, medium-risk, and high-risk labels according to preset quality standards; comprehensive evaluation rules are set according to end face water absorption rate, end face thickness expansion rate, edge sealing peel strength, and failure defect degree to form a comprehensive scoring label or risk classification label to characterize the overall moisture-proof performance of the end face; finally, the end face multimodal feature vector corresponding to each board is matched and associated with its end face moisture-proof performance label obtained through the same test standard and testing process, according to product number, board batch, production time, process batch, and sample location, to form a supervised learning training sample.

[0013] In this embodiment, the end-face moisture-proof performance prediction model includes a multimodal feature encoding layer, an attention fusion layer, a shared feature extraction layer, and a multi-task prediction layer, as detailed below: The multimodal feature encoding layer divides the end-face multimodal feature vector into six input classes: x=[x (b) ,x (p) ,x (l) ,x (h) ,x (e) ,x (v) ]; Where, x (b) For substrate parameter characteristics, x (p) For the parameter characteristics of impregnated paper, x (l) As a characteristic of the impregnation solution formulation, x (h) For the characteristics of hot-pressed bonding parameters, x (e) For end-face sealing parameter characteristics, x (v)For manufacturing environment and end face image features; For the m-th feature x (m) Feature mapping is performed using fully connected coding units: ; in, Let m be the weight matrix corresponding to the m-th class of features. Here, φ(·) is the bias term, and φ(·) is the nonlinear activation function; each mode is mapped to the same dimension d after encoding. h h (m) This represents the latent feature representation of the m-th modality obtained after mapping; The attention fusion layer, to reflect the differences in the impact of different process modes on the moisture-proof performance of the end face, applies attention weighting to the encoding results of each mode; the attention score for the m-th feature is: ; Among them, W a Let b be the attention mapping matrix. a Here, u is the bias vector, and u is the attention weight vector; The attention weights for each modality are calculated using the Softmax function: ; Where M is the number of modes, α m The contribution weight for the m-th class feature; Attention score for the k-th class feature; The modality-coded features are fused according to their corresponding weights to obtain a fused feature vector: ; To preserve complete information from different modalities, the weighted features and the original concatenated features are jointly input: ; The fused feature vector z f Inputting the shared feature extraction network yields a deep characterization related to the end-face moisture-proof performance: ; in, For the shared layer weight matrix, p is the shared layer bias vector. d The proportion of random inactivation; the shared feature r is used to characterize the coupling relationship between the substrate pore structure, resin penetration and curing state, hot-press densification degree, edge sealing adhesive layer bonding state and environmental humid and hot exposure conditions. The multi-task prediction layer, based on shared feature r, sets up three regression prediction branches and one classification prediction branch, wherein: The branch for predicting end-face water absorption rate is: ; The prediction branch for end face thickness expansion rate is as follows: ; The prediction branch for edge sealing peel strength is as follows: ; The end-face failure risk prediction branch is: ; in, To predict the water absorption rate at the end face, W w1 W w2 For the branch weight matrix of end-face water absorption rate prediction, b w1 b w2 The branch bias vector for predicting end-face water absorption rate; To predict the end face thickness expansion rate, W t1 W t2 b is the branch weight matrix for predicting the end face thickness expansion rate. t1 b t2 The branch bias vector for predicting the end face thickness expansion rate; To predict the edge-sealing peel strength, W p1 W p2 For the branch weight matrix of edge sealing peel strength prediction, b p1 b p2 Predict the branch bias vector for edge banding peel strength; W is the predicted probability vector for the end-face failure risk category. r1 W r2 For the branch weight matrix of end-face failure risk prediction, b r1 b r2 This is the branch bias vector for predicting end-face failure risk. The risk level of end face failure is determined based on the highest probability category: ; Where C represents the number of risk categories; Indicates the predicted category of end-face failure risk; This represents the probability value that the model predicts the end-face failure risk belongs to the c-th risk category.

[0014] In this embodiment, the trained end-face moisture-proof performance prediction model is used as the base model to perform sensitivity analysis on adjustable process parameters and obtain a set of key process parameters, as detailed below: The trained end-face moisture-proof performance prediction model is denoted as f. θ(·), taking the multimodal feature vector of the end face of the board to be analyzed as input, outputs the end face water absorption rate, end face thickness expansion rate, edge sealing peel strength, and end face failure risk; the input features are divided into a set of adjustable process parameters and a set of non-adjustable condition parameters [u,z]; where u=[u1,u2,…,u K [ ] represents a set of adjustable process parameters, including impregnation solution formulation parameters, impregnated paper pick-up amount, volatile matter, hot-pressing temperature, hot-pressing pressure, hot-pressing time, end-face primer application amount, edge sealing adhesive application amount, and edge sealing pressure, etc.; z represents non-adjustable or batch-fixed parameters such as substrate properties, basic properties of impregnated paper, production environment, and product specifications; the model output is: ; in, To predict the water absorption rate at the end face, To predict the thickness expansion rate of the end face, To predict the edge banding peel strength, To predict the probability of end face failure; For each adjustable process parameter u k While keeping other process parameters and unadjustable condition parameters constant, the local sensitivity of this parameter to various performance indicators is calculated using the local perturbation method. ; For continuous forecasting indicators, the formula for calculating local sensitivity is: ; in, Indicates the adjustable process parameter u in the i-th sample. k Local sensitivity to the j-th performance metric, j∈{w,t,p}; u i,-k Indicates excluding u i,k Other adjustable process parameters besides Δu k The perturbation step size is the parameter. For the probability of end-face failure risk, its local sensitivity is: ; To eliminate the influence of different parameter dimensions and value ranges on the sensitivity results, the local sensitivity is transformed into a dimensionless standardized sensitivity: ; in, This represents the allowable adjustment range for the k-th process parameter. This represents the preset maximum allowable value for the k-th process parameter; This represents the preset minimum allowable value of the k-th process parameter; For the range of variation of the corresponding performance indicators, This represents the maximum value of the corresponding performance indicator; This represents the minimum value of the corresponding performance index; ε is a minimal constant to prevent the denominator from being zero; the global sensitivity is obtained by averaging the standardized sensitivity of all samples. ; Where N is the number of samples used for sensitivity analysis; Construction parameter u k Overall sensitivity index: ; Where, ω w ω t ω p and ω r The weights are respectively: end face water absorption rate, end face thickness expansion rate, edge sealing peel strength, and end face failure risk; This represents the normalized sensitivity of the k-th process parameter to the end-face water absorption rate; This represents the normalized sensitivity of the k-th process parameter to the end face thickness expansion rate; This represents the normalized sensitivity of the k-th process parameter to the edge-sealing peel strength; This represents the normalized sensitivity of the k-th process parameter to the risk of end-face failure; For any two adjustable process parameters u k and u l Perform second-order difference calculations to obtain the interaction sensitivity. The interaction sensitivity of each performance index is averaged and weighted to obtain the parameter pairs (u). k ,u l Interaction impact index: ; in, For standardized interaction sensitivity; if If the value exceeds the preset interaction threshold, then u will be... k and u l As a pair of coupled process parameters, it is retained to avoid neglecting the synergistic effects between parameters such as hot-pressing temperature and pressure, primer amount and edge-sealing adhesive amount, and impregnated paper pick-up amount and volatile matter in subsequent optimizations; ω j This represents the weighting coefficient corresponding to the moisture-proof performance index of the j-th end face; A set of key process parameters is constructed based on the comprehensive sensitivity index and the interaction influence index. Let the comprehensive sensitivity threshold be... The interaction sensitivity threshold is The set of key process parameters is as follows: ;in, Take all G kThe top q% quantile values. The final set of key process parameters includes the impregnation liquid formulation components, impregnated paper pick-up amount, volatile matter, hot pressing temperature, hot pressing pressure, hot pressing time, end face primer amount, edge sealing adhesive application amount, and edge sealing pressure, all of which have a significant impact on the moisture-proof performance of the end face.

[0015] In this embodiment, the multi-objective process parameter optimization model is constructed with the following optimization objectives: minimum end-face water absorption rate, minimum end-face thickness expansion rate, minimum end-face failure risk, minimum manufacturing cost, and maximum edge sealing peel strength. Specifically: ; in: ; The optimization variables are constrained by process boundary constraints, quality threshold constraints, equipment capacity constraints, and formula ratio constraints, specifically as follows: ; in, and These are the lower and upper limits of the allowable parameters for the k-th critical process parameter, respectively. and These are the maximum allowable values ​​for end face water absorption rate, end face thickness expansion rate, end face failure risk, and edge sealing peel strength, respectively. The non-dominated solutions obtained from the multi-objective optimization algorithm are filtered to form a set of Pareto optimal process parameter schemes. : ; Among them, symbols Pareto dominance means that for two candidate process parameter schemes, if one scheme is not inferior to the other scheme on all optimization objectives and is superior to the other scheme on at least one optimization objective, then the scheme dominates the other scheme. For u key Another key process parameter vector for comparison.

[0016] In this embodiment, the candidate schemes in the Pareto optimal process parameter scheme set are normalized, and the final optimized process parameter scheme is determined by a comprehensive optimization function. The normalized minimization objective function is: ; in, and Let f be the minimum and maximum values ​​of the j-th objective function in the Pareto optimal solution set, respectively, and ε be a minimal constant to prevent the denominator from being zero; j This represents the j-th original objective function value corresponding to the current key process parameter scheme; The comprehensive optimization function is expressed as: ; Where, ω j The weights for the j-th optimization objective; D(u key ,u ref ) represents candidate solution u key Compared with the current stable production process scheme ref The process adjustment distance between them is used to suppress excessive parameter jumps, and it can be expressed as:

[0017] In the formula, α k The adjustment penalty weight for the k-th critical process parameter is λ, where λ is the process stability penalty coefficient; u k,ref For the reference process parameter vector and u k The corresponding reference parameter values; Ultimately, the Pareto optimal solution, which minimizes the overall optimization function, is chosen as the optimized process parameter scheme. ; The optimized process parameter scheme u opt This includes recommended settings for each key process parameter, allowable adjustment range, corresponding predicted end-face moisture resistance performance index, predicted edge banding peel strength, predicted failure risk, and unit board manufacturing cost.

[0018] In this embodiment, based on the optimized process parameter scheme, differentiated impregnation paper and end-face sealing process control are implemented for the boards to be processed, specifically as follows: For each batch of boards to be processed, the corresponding substrate parameters, impregnation paper parameters, manufacturing environment parameters, and optimized process parameter scheme are obtained. optBased on the type of board material, initial moisture content, density, surface roughness, end porosity, basis weight of impregnated paper, impregnation solution formulation, ambient temperature and humidity, and equipment operating status, the boards to be processed are classified into different processes, and corresponding differentiated process control instructions are generated. These differentiated process control instructions include the set values ​​and allowable fluctuation ranges for impregnated paper pickup amount, volatile matter, impregnation solution component ratio, hot-pressing temperature, hot-pressing pressure, hot-pressing time, end-face base coating amount, edge sealing adhesive coating amount, edge sealing pressure, and edge sealing curing time. For different board conditions, relevant process parameters are adjusted according to the optimized process parameter scheme to avoid problems such as high end-face water absorption, increased thickness expansion rate, insufficient edge sealing peel strength, or increased manufacturing costs caused by using uniform fixed process parameters. In the impregnated paper preparation and hot-pressing lamination processes, based on... The process parameters are optimized to control the impregnation solution formula, impregnation paper pick-up amount, and volatile matter content, ensuring that the resin content, penetration state, and curing reaction conditions in the impregnated paper match the porosity characteristics of the substrate end face. The hot-pressing temperature, pressure, and time are also controlled to ensure a stable bond between the impregnated paper and the board surface, while reducing micropores in the end face area caused by excessive water loss, insufficient resin migration, or incomplete local curing. In the end face sealing process, the amount of primer, primer penetration time, edge sealing adhesive application amount, edge sealing pressure, and curing time are controlled in stages based on the end face porosity, surface roughness, and predicted failure risk. For boards with high end face porosity, high substrate moisture content, or high predicted failure risk, the penetration coverage of the end face primer and the effective application amount of the edge sealing adhesive are increased, while the edge sealing pressure and curing time are adjusted within a preset safety range. During the process execution, the actual parameters of the impregnation, drying, hot pressing, end-face primer coating, and edge sealing processes are collected online to form an actual process parameter vector u. act and with optimized process parameter scheme u opt By comparing the results, we obtain the parameter deviation vector: ; When any process parameter deviation exceeds its corresponding allowable fluctuation range, the control system outputs a correction command to perform closed-loop adjustment on the corresponding execution equipment; the closed-loop adjustment includes adjusting the impregnation liquid supply, impregnation paper running speed, drying temperature, hot pressing temperature, hot pressing pressure, hot pressing time, primer supply, edge sealing adhesive supply, edge sealing pressing pressure, or curing time.

[0019] The machine learning-based system for optimizing the moisture-proof performance of impregnated paper end faces includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the machine learning-based method for optimizing the moisture-proof performance of impregnated paper end faces as described above.

[0020] A computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the method steps described above.

[0021] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0022] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0023] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0024] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0025] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for optimizing the moisture-proof performance of impregnated paper end faces based on machine learning, characterized in that, Includes the following steps: S1: Obtain the initial process sample of the board to be processed and perform unified preprocessing to obtain the end face multimodal feature vector; S2: Obtain the end-face moisture-proof performance label, and associate the end-face multimodal feature vector with the corresponding end-face moisture-proof performance label to construct supervised learning training samples; S3: Construct a prediction model for the moisture-proof performance of the end face, and train it based on supervised learning training samples to obtain the trained prediction model for the moisture-proof performance of the end face. S4: Using the trained end-face moisture-proof performance prediction model as the base model, perform sensitivity analysis on adjustable process parameters to obtain a set of key process parameters; S5: Set the key process parameters as optimization variables, construct a multi-objective process parameter optimization model, and obtain the optimized process parameter scheme; S6: Based on the optimized process parameter scheme, implement differentiated impregnation paper and end-face sealing process control for the boards to be processed.

2. The method for optimizing the moisture-proof performance of impregnated paper end faces based on machine learning according to claim 1, characterized in that, The process involves obtaining initial process samples of the board to be processed and performing unified preprocessing to obtain end-face multimodal feature vectors. Specifically, this includes: obtaining the substrate parameters, impregnated paper parameters, impregnating liquid formula parameters, hot-pressing bonding parameters, end-face sealing parameters, and manufacturing environment parameters of the board to be processed; associating the collected parameters according to board batch, production time, process segment number, and product number to form initial process samples corresponding to a single board to be processed, and performing unified preprocessing; based on the unified preprocessing, feature extraction and fusion are performed on parameters from different sources to obtain end-face multimodal feature vectors.

3. The method for optimizing the moisture-proof performance of impregnated paper end faces based on machine learning according to claim 1, characterized in that, The process involves obtaining end-face moisture-proof performance labels and constructing a training sample set. The end-face multimodal feature vectors are then associated with the corresponding end-face moisture-proof performance labels to construct supervised learning training samples. Specifically, the end-face moisture-proof performance of the boards treated with impregnated paper facing and end-face sealing is tested to obtain end-face moisture-proof performance labels. For test results with different dimensions and evaluation methods, the end-face moisture-proof performance labels are uniformly processed. Finally, the end-face multimodal feature vector corresponding to each board is matched and associated one-to-one with its end-face moisture-proof performance label obtained through the same test standard and testing process, according to product number, board batch, production time, process batch, and sample location, to form supervised learning training samples.

4. The method for optimizing the moisture-proof performance of impregnated paper end faces based on machine learning according to claim 1, characterized in that, The end-face moisture-proof performance prediction model includes a multimodal feature encoding layer, an attention fusion layer, a shared feature extraction layer, and a multi-task prediction layer, as detailed below: The multimodal feature encoding layer divides the end-face multimodal feature vector into six input classes: x=[x (b) ,x (p) ,x (l) ,x (h) ,x (e) ,x (v) ]; Where, x (b) For substrate parameter characteristics, x (p) For the parameter characteristics of impregnated paper, x (l) As a characteristic of the impregnation solution formulation, x (h) For the characteristics of hot-pressed bonding parameters, x (e) For end-face sealing parameter characteristics, x (v) For manufacturing environment and end face image features; For the m-th feature x (m) Feature mapping is performed using fully connected coding units: ； in, Let m be the weight matrix corresponding to the m-th class of features. The bias term is φ(·), which is a nonlinear activation function; each mode is mapped to the same dimension d after encoding. h h (m) This represents the latent feature representation of the m-th modality obtained after mapping; The attention fusion layer performs attention weighting on the encoding results of each modality; the attention score for the m-th feature is: ； Among them, W a Let b be the attention mapping matrix. a Here, u is the bias vector, and u is the attention weight vector; The attention weights for each modality are calculated using the Softmax function: ； Where M is the number of modes, α m The contribution weight for the m-th class feature; Attention score for the k-th class feature; The modality-coded features are fused according to their corresponding weights to obtain a fused feature vector: ; To preserve complete information from different modalities, the weighted features and the original concatenated features are jointly input: ; The fused feature vector z f Inputting the shared feature extraction network yields a deep characterization related to the end-face moisture-proof performance: ； in, For the shared layer weight matrix, p is the shared layer bias vector. d The random inactivation rate; the multi-task prediction layer, based on the shared feature r, sets up three regression prediction branches and one classification prediction branch, wherein: The branch for predicting end-face water absorption rate is: ； The prediction branch for end face thickness expansion rate is as follows: ； The prediction branch for edge sealing peel strength is as follows: ； The end-face failure risk prediction branch is: ； in, To predict the water absorption rate at the end face, W w1 W w2 For the branch weight matrix of end-face water absorption rate prediction, b w1 b w2 The branch bias vector for predicting end-face water absorption rate; To predict the end face thickness expansion rate, W t1 W t2 b is the branch weight matrix for predicting the end face thickness expansion rate. t1 b t2 The branch bias vector for predicting the end face thickness expansion rate; To predict the edge-sealing peel strength, W p1 W p2 For the branch weight matrix of edge sealing peel strength prediction, b p1 b p2 Predict the branch bias vector for edge banding peel strength; W is the predicted probability vector for the end-face failure risk category. r1 W r2 For the branch weight matrix of end-face failure risk prediction, b r1 b r2 This is the branch bias vector for predicting end-face failure risk. The risk level of end face failure is determined based on the highest probability category: ； Where C represents the number of risk categories; Indicates the predicted category of end-face failure risk; This represents the probability value that the model predicts the end-face failure risk belongs to the c-th risk category.

5. The method for optimizing the moisture-proof performance of impregnated paper end faces based on machine learning according to claim 1, characterized in that, The trained end-face moisture-proof performance prediction model is used as the base model to perform sensitivity analysis on adjustable process parameters and obtain a set of key process parameters, as detailed below: The trained end-face moisture-proof performance prediction model is denoted as f. θ (·), taking the multimodal feature vector of the end face of the board to be analyzed as input, outputs the end face water absorption rate, end face thickness expansion rate, edge sealing peel strength, and end face failure risk; the input features are divided into a set of adjustable process parameters and a set of non-adjustable condition parameters [u,z]; where u=[u1,u2,…,u K [] represents the set of adjustable process parameters; z represents the fixed parameters; the model output is: ； in, To predict the water absorption rate at the end face, To predict the thickness expansion rate of the end face, To predict the edge banding peel strength, To predict the probability of end face failure; For each adjustable process parameter u k While keeping other process parameters and unadjustable condition parameters constant, the local sensitivity of this parameter to various performance indicators is calculated using the local perturbation method. ; Transform local sensitivity into dimensionless standardized sensitivity: ； in, The allowable adjustment range for the k-th process parameter. This represents the preset maximum allowable value for the k-th process parameter; This represents the preset minimum allowable value of the k-th process parameter; For the range of variation of the corresponding performance indicators, This indicates the maximum value of the corresponding performance indicator; This represents the minimum value of the corresponding performance index; ε is a minimal constant to prevent the denominator from being zero; the global sensitivity is obtained by averaging the standardized sensitivity of all samples. ； Where N is the number of samples used for sensitivity analysis; Construction parameter u k Overall sensitivity index: ； Where, ω w ω t ω p and ω r The weights are respectively: end face water absorption rate, end face thickness expansion rate, edge sealing peel strength, and end face failure risk; This represents the normalized sensitivity of the k-th process parameter to the end-face water absorption rate; This represents the normalized sensitivity of the k-th process parameter to the end face thickness expansion rate; This represents the normalized sensitivity of the k-th process parameter to the edge-sealing peel strength; This represents the normalized sensitivity of the k-th process parameter to the risk of end-face failure; For any two adjustable process parameters u k and u l Perform second-order difference calculations to obtain the interaction sensitivity. The interaction sensitivity of each performance index is averaged and weighted to obtain the parameter pairs (u). k ,u l Interaction impact index: ； in, For standardized interaction sensitivity; if If the value exceeds the preset interaction threshold, then u will be... k and u l Retained as a coupling process parameter; ω j This represents the weighting coefficient corresponding to the moisture-proof performance index of the j-th end face; A set of key process parameters is constructed based on the comprehensive sensitivity index and the interaction influence index; the comprehensive sensitivity threshold is set as follows: The interaction sensitivity threshold is The set of key process parameters is as follows: 。 6. The method for optimizing the moisture-proof performance of impregnated paper end faces based on machine learning according to claim 5, characterized in that, The multi-objective process parameter optimization model is constructed with the optimization objectives of minimizing end-face water absorption rate, minimizing end-face thickness expansion rate, minimizing end-face failure risk, minimizing manufacturing cost, and maximizing edge sealing peel strength, as detailed below: ； in: ； The optimization variables are constrained by process boundary constraints, quality threshold constraints, equipment capacity constraints, and formula ratio constraints, specifically as follows: ； in, and These are the lower and upper limits of the allowable parameters for the k-th critical process parameter, respectively. and These are the maximum allowable values ​​for end face water absorption rate, end face thickness expansion rate, end face failure risk, and edge sealing peel strength, respectively. The non-dominated solutions obtained from the multi-objective optimization algorithm are filtered to form a set of Pareto optimal process parameter schemes. : ； Among them, symbols Pareto dominance means that for two candidate process parameter schemes, if one scheme is not inferior to the other scheme on all optimization objectives and is superior to the other scheme on at least one optimization objective, then the scheme dominates the other scheme. For u key Another key process parameter vector for comparison.

7. The method for optimizing the moisture-proof performance of impregnated paper end faces based on machine learning according to claim 6, characterized in that, The candidate schemes in the Pareto optimal process parameter scheme set are normalized, and the final optimized process parameter scheme is determined by a comprehensive selection function. The normalized minimization objective function is: ； in, and Let f be the minimum and maximum values ​​of the j-th objective function in the Pareto optimal solution set, respectively, and let ε be a minimal constant to prevent the denominator from being zero; j This represents the j-th original objective function value corresponding to the current key process parameter scheme; The comprehensive optimization function is expressed as: ； Where, ω j The weights for the j-th optimization objective; D(u key ,u ref ) represents candidate solution u key Compared with the current stable production process scheme ref The process adjustment distance between them is expressed as: ； In the formula, α k The adjustment penalty weight for the k-th critical process parameter is λ, where λ is the process stability penalty coefficient; u k,ref For the reference process parameter vector and u k The corresponding reference parameter values; Ultimately, the Pareto optimal solution, which minimizes the overall optimization function, is chosen as the optimized process parameter scheme. 。 8. The method for optimizing the moisture-proof performance of impregnated paper end faces based on machine learning according to claim 1, characterized in that, The process involves implementing differentiated impregnation paper and end-face sealing process control for the boards to be processed based on the optimized process parameter scheme. Specifically, for each batch of boards to be processed, the corresponding substrate parameters, impregnation paper parameters, manufacturing environment parameters, and optimized process parameter scheme are obtained. opt Based on the type of board, initial moisture content, density, surface roughness, end porosity, basis weight of impregnated paper, impregnation solution formulation, ambient temperature and humidity, and equipment operating status, the boards to be processed are classified into different processes, and corresponding differentiated process control instructions are generated. For different board conditions, relevant process parameters are adjusted in a targeted manner according to the optimized process parameter scheme. In the impregnation paper preparation and hot-pressing processes, the impregnation solution formulation, impregnation paper pick-up amount, and volatile matter are controlled according to the optimized process parameter scheme to match the resin content, penetration state, and curing reaction conditions in the impregnation paper with the end porosity characteristics of the substrate. The heat is also controlled. The temperature, pressure, and time of hot pressing are controlled to ensure a stable bond between the impregnated paper and the board surface, while reducing micropores in the end-face area caused by excessive water loss, insufficient resin migration, or incomplete local curing. In the end-face sealing process, the amount of primer, primer penetration time, edge sealing adhesive application amount, edge sealing pressure, and curing time are controlled in stages based on the end-face porosity, end-face roughness, and predicted failure risk. For boards with high end-face porosity, high substrate moisture content, or high predicted failure risk, the penetration coverage of the end-face primer and the effective application amount of edge sealing adhesive are increased, and the edge sealing pressure and curing time are adjusted within a preset safety range. During the process execution, the actual parameters of the impregnation, drying, hot pressing, end-face primer coating, and edge sealing processes are collected online to form an actual process parameter vector u. act and with optimized process parameter scheme u opt By comparing the results, we obtain the parameter deviation vector: ； When any process parameter deviation exceeds its corresponding allowable fluctuation range, the control system outputs a correction command to perform closed-loop adjustment on the corresponding actuator.

9. A machine learning-based system for optimizing the moisture-proof performance of impregnated paper end faces, characterized in that, It includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the machine learning-based method for optimizing the moisture-proof performance of the end face of impregnated paper as described in any one of claims 1-8.

10. A computer storage medium, characterized in that, The computer storage medium stores a plurality of instructions adapted for loading by a processor and executing the steps of the method as described in any one of claims 1-8.

Images