PCB circuit etching parameter prediction method and device, computer equipment and storage medium

Abstract

The invention discloses a PCB circuit etching parameter prediction method and device, computer equipment and a storage medium. The method comprises the following steps: collecting multi-source data related to etching quality; preprocessing the multi-source data to obtain a data set; defining a model initial feature set and a training label set; setting the seed number of a random number generator; dividing the data set according to a proportion to form a training set and a verification set; respectively training a random forest model and a lightgbm model by using the training set; screening out an important feature set, and carrying out hyper-parameter tuning to obtain an intermediate prediction model; training a meta learning device to obtain a target prediction model; obtaining to-be-predicted sample data, and performing feature value processing on the to-be-predicted sample data according to the definition of the initial feature set to obtain a feature vector; and inputting the feature vector into a target prediction model to generate an etching parameter prediction value. By implementing the method provided by the invention, the accuracy and stability of etching parameter prediction can be improved.

Description

Technical Field

[0001] This invention relates to the field of PCB line etching parameter prediction technology, and more specifically to PCB line etching parameter prediction methods, apparatus, computer equipment and storage media. Background Technology

[0002] In PCB (Printed Circuit Board) manufacturing, etching is a core step that determines circuit precision and product quality. Etching quality directly affects the width accuracy, lateral etching degree, and circuit uniformity, thus determining the electrical performance and reliability of the PCB. Etching parameters mainly include the etching equipment's line speed, upward spray pressure, and downward spray pressure; precise control of these parameters is crucial for ensuring etching quality.

[0003] Currently, the industry mainly uses the following methods to control etching parameters: 1. Manual Adjustment of Settings: This method heavily relies on the experience and technical expertise of the etching machine operators. Due to differences in operator experience and skills, parameter settings are highly subjective and unstable. Significant differences in parameter settings by different operators make it difficult to guarantee consistent etching quality, thus affecting the overall quality of the PCB products.

[0004] 2. Prediction methods based on a single model: KNN (Nearest Neighbor) Model: While the KNN model can perform sensitive analysis on local data, it is more suitable for small-scale, low-noise datasets. When dealing with large-scale, complex data, the KNN model is computationally inefficient and easily affected by noisy data, resulting in insufficient prediction accuracy.

[0005] Linear regression models have limited ability to model complex nonlinear relationships, making it difficult to accurately capture the complex nonlinear relationships during the etching process. Therefore, their prediction accuracy and generalization performance are insufficient, making it difficult to adapt to the needs of etching parameter prediction under different production conditions.

[0006] In summary, the existing methods for controlling etching parameters have the following problems: over-reliance on human experience leads to poor stability and consistency in parameter settings; in addition, single prediction models have limitations in handling complex nonlinear relationships, resulting in insufficient prediction accuracy and generalization performance.

[0007] Therefore, there is a need for an etching parameter prediction method that can overcome the above-mentioned defects, reduce reliance on human experience, improve the accuracy and stability of etching parameter prediction, and thus improve the overall quality and production efficiency of PCB manufacturing. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, apparatus, computer equipment and storage medium for predicting PCB circuit etching parameters.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: Methods for predicting PCB circuit etching parameters include: Collect multi-source data related to etching quality; Preprocess the multi-source data to obtain the dataset; Define the initial feature set and training label set for the model; Set the seed number for the random number generator; The dataset is divided proportionally according to the seed number of the random number generator to form a training set and a validation set; The Random Forest model and the LightGBM model are trained using the training set. During the training process, the model takes the features in the initial feature set as input and the target parameters in the training label set as output, and learns the mapping relationship between the features and the target parameters to obtain the trained Random Forest model and LightGBM model. The trained random forest model and lightgbm model are evaluated using a validation set, and an important feature set is selected. Based on the important feature set, the hyperparameters of the trained random forest model and lightgbm model are tuned to obtain an intermediate prediction model. The intermediate prediction model is used to predict the validation set to obtain the initial prediction value. The initial prediction value is used as the input feature and the actual label of the validation set is used as the output feature to train a meta-learner based on a neural network to obtain the target prediction model. Obtain the sample data to be predicted, and perform feature value processing on the sample data to be predicted according to the definition of the initial feature set to obtain the feature vector; The feature vector is input into the target prediction model to generate predicted values ​​for etching parameters.

[0010] The present invention also provides a PCB circuit etching parameter prediction device, comprising: Collection unit, used to collect multi-source data related to etching quality; The preprocessing unit is used to preprocess multi-source data to obtain a dataset; Define the unit, used to define the initial feature set and training label set of the model; A setting unit is used to set the seed number for the random number generator; The partitioning unit is used to divide the dataset proportionally according to the number of seeds of the random number generator to form training and validation sets; The training unit is used to train the random forest model and the lightgbm model respectively using the training set. During the training process, the model takes the features in the initial feature set as input and the target parameters in the training label set as output, and learns the mapping relationship between the features and the target parameters to obtain the trained random forest model and lightgbm model. The selection and tuning unit is used to evaluate the trained random forest model and lightgbm model using the validation set, and to select important feature sets. Based on the important feature sets, the hyperparameters of the trained random forest model and lightgbm model are tuned to obtain intermediate prediction models. The prediction training unit is used to predict the validation set using the intermediate prediction model to obtain the initial prediction value. The initial prediction value is used as the input feature and the actual label of the validation set is used as the output feature to train a neural network-based meta-learner to obtain the target prediction model. The acquisition and processing unit is used to acquire the sample data to be predicted and perform feature value processing on the sample data to be predicted according to the definition of the initial feature set to obtain the feature vector; The input generation unit is used to input the feature vector into the target prediction model to generate predicted values ​​for the etching parameters.

[0011] The present invention also provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the above-described method.

[0012] The present invention also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0013] The advantages of this invention compared to existing technologies are as follows: by collecting multi-source data related to etching quality and using machine learning models for parameter prediction, the invention can automatically learn the patterns in the data, thereby reducing reliance on human experience and improving the stability and reliability of parameter settings. In addition, by combining the ensemble learning method of random forest model and lightGBM model, and further optimizing it using a meta-learner based on neural network, the invention can better capture the complex nonlinear relationships in the etching process and significantly improve prediction accuracy.

[0014] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram illustrating an application scenario of the PCB line etching parameter prediction method provided in this embodiment of the invention. Figure 2 A flowchart illustrating the PCB line etching parameter prediction method provided in this embodiment of the invention; Figure 3 A schematic block diagram of a PCB line etching parameter prediction device provided in an embodiment of the present invention; Figure 4 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0019] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0020] Please see Figure 1 and Figure 2 , Figure 1 This is a schematic diagram illustrating an application scenario of the PCB line etching parameter prediction method provided in this embodiment of the invention. Figure 2This is a schematic flowchart illustrating the PCB line etching parameter prediction method provided in this embodiment of the invention. The method is applied to a server that interacts with a terminal. Through innovative ensemble learning methods and optimization strategies, it effectively solves the problems of reliance on human experience, insufficient prediction accuracy, and weak generalization ability in existing technologies. It provides the PCB manufacturing industry with an efficient, accurate, and stable etching parameter prediction method, demonstrating significant progress and important practical application value.

[0021] Figure 2 This is a flowchart illustrating the PCB circuit etching parameter prediction method provided in an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S200.

[0022] S110. Collect multi-source data related to etching quality; Specifically, the following multi-source data related to etching quality were collected: Product features: Surface copper requirement value, actual surface copper value, copper thickness deviation value (actual value - required value), minimum product line width, minimum film line spacing, residual copper rate on top surface, residual copper rate on bottom surface, required line width on top surface, required line width on bottom surface.

[0023] Material characteristics: Type of resist (wet film type, dry film type).

[0024] Equipment parameters: line speed, upward spray pressure, downward spray pressure.

[0025] Chemical parameters: chemical temperature, copper ion concentration, hydrochloric acid (HCl) concentration, oxidant concentration, and specific gravity.

[0026] Quality parameters: actual value of top surface line width, actual value of bottom surface line width.

[0027] These data are acquired through automated production management systems (MES), sensor data collection, or manual recording to ensure the comprehensiveness and accuracy of the data.

[0028] In other words, by comprehensively collecting multi-source data related to etching quality, a rich information foundation is provided for subsequent model training, which helps to improve the accuracy and generalization ability of the prediction model.

[0029] S120. Preprocess the multi-source data to obtain the dataset; Specifically, preprocessing includes data cleaning and theoretical equipment line rate calculation.

[0030] Data cleaning includes removing records with missing values ​​to ensure data integrity. Simultaneously, outlier stratification criteria are established based on rules, and identified abnormal data records are removed. For example, thresholds for actual copper thickness and equipment line speed are set, and outlier data records exceeding these thresholds are removed.

[0031] Theoretical equipment line speed calculation refers to calculating the theoretical equipment line speed using a specific formula, based on the actual line width, the required line width, and the current equipment line speed, to guide the adjustment of equipment parameters. Specifically, the method involves calculating the theoretical equipment line speed based on the relationship between the line width deviation (the difference between the actual and required values) and the equipment line speed, thus guiding the adjustment of equipment operating parameters.

[0032] More specifically, the theoretical equipment line speed is calculated based on the actual line width, the required line width, and the equipment line speed.

[0033] The specific formula is as follows: Where avg (top face line width requirement value - top face line width actual value, bot face line width requirement value - bot face line width actual value) is the average line width deviation.

[0034] The key-value pairs are specifically: Actual copper thickness ≤ 55μm: 0.1m / min line speed variation, 3μm line width variation.

[0035] Actual copper thickness ≤ 75μm: 0.1m / min line speed variation, 4μm line width variation.

[0036] Actual copper surface value ≤ 90μm: 0.1m / min line speed variation value, 5μm line width variation value.

[0037] Actual copper surface value > 90μm: 0.1m / min line speed variation value, 6μm line width variation value.

[0038] In other words, the data preprocessing step effectively removed noisy data and outliers, ensuring data quality and consistency. Simultaneously, by calculating the theoretical line speed of the equipment, more instructive features were provided for model training, improving the model's predictive performance.

[0039] S130. Define the initial feature set and training label set for the model; Specifically, the following initial feature set and training label set for the model are defined: Initial feature set: Product features: required copper surface value, actual copper surface value, copper thickness deviation value, minimum product line width, minimum film line spacing, residual copper rate on top surface, residual copper rate on bottom surface, required line width on top surface, required line width on bottom surface.

[0040] Material characteristics: Type of resist (wet film type, dry film type).

[0041] Equipment parameters: line speed, upward spray pressure, downward spray pressure.

[0042] Chemical parameters: chemical temperature, copper ion concentration, hydrochloric acid concentration, oxidant concentration, and specific gravity.

[0043] Training label set: Target parameters: theoretical equipment linear speed, upward spray pressure, downward spray pressure.

[0044] In other words, a clear feature set and label set provide clear input and output targets for model training, ensuring that the model can learn the accurate mapping relationship between features and target parameters, thereby improving the accuracy of prediction.

[0045] S140, Set the seed number for the random number generator; Specifically, a seed number is set for the random number generator (e.g., a seed number of 42) to ensure the repeatability and consistency of data partitioning. Setting the seed number provides a fixed basis for randomness in subsequent data partitioning, ensuring consistent results each time.

[0046] In other words, by setting the seed number of the random number generator, the stability and repeatability of data partitioning are ensured, making the model training and validation process more reliable.

[0047] S150. Divide the dataset proportionally according to the number of seeds of the random number generator to form a training set and a validation set; Specifically, based on the set seed number of the random number generator, the dataset is divided into a training set and a validation set in a certain ratio (e.g., 8:2). The training set is used for model training, and the validation set is used for model performance evaluation. Data partitioning tools (such as train_test_split) are used to perform the partitioning, ensuring the randomness and representativeness of the data.

[0048] In other words, reasonable data partitioning ensures that the model can learn the patterns in the data during training, while the validation set is used to evaluate the model's generalization ability and avoid overfitting or underfitting.

[0049] S160. Use the training set to train the random forest model and the lightgbm model respectively. During the training process, the model takes the features in the initial feature set as input and the target parameters in the training label set as output, and learns the mapping relationship between the features and the target parameters to obtain the trained random forest model and lightgbm model. Specifically, the random forest model and the lightgbm model are trained using the training set respectively: Random Forest Model: Initialize the parameters of the random forest model, such as the number of trees (e.g., 100), maximum depth (e.g., 10), minimum number of samples required for splits (e.g., 2), and minimum number of samples for leaf nodes (e.g., 1).

[0050] The random forest model is trained by using the initial feature set in the training set as input and the target parameters in the training label set as output.

[0051] Random forest models learn the complex nonlinear relationship between features and target parameters by constructing multiple decision trees.

[0052] LightGBM model: Initialize the parameters of the LightGBM model, such as the learning rate (e.g., 0.1), the number of iterations of the boosting tree (e.g., 100), the maximum depth (e.g., 7), the maximum number of leaves in the tree (e.g., 31), the minimum number of samples in the leaf nodes (e.g., 20), the maximum number of bins for feature splits (e.g., 255), the L1 regularization parameter (e.g., 0.01), and the L2 regularization parameter (e.g., 0.01).

[0053] The LightGBM model is trained using the initial feature set in the training set as input and the target parameters in the training label set as output.

[0054] The LightGBM model uses a gradient boosting algorithm to progressively optimize model performance, handle large-scale data, and improve training efficiency.

[0055] In other words, by training the Random Forest Model and the Light Gradient Boosting Machine Model, and leveraging the respective advantages of the two models (the overfitting ability of Random Forest and the efficiency of Light Gradient Boosting Machine Model), a foundation was laid for subsequent model optimization, thereby improving the accuracy and stability of predictions.

[0056] S170. Use the validation set to evaluate the trained random forest model and lightgbm model, and select important feature sets. Based on the important feature sets, perform hyperparameter tuning on the trained random forest model and lightgbm model to obtain intermediate prediction models. Specifically, the trained random forest model and lightgbm model are evaluated using a validation set: Model Evaluation: The trained Random Forest and LightGBM models are evaluated using a validation set, and the error between the predicted values ​​and the actual labels is calculated. Appropriate evaluation metrics (such as mean squared error and mean absolute error) are selected to measure model performance.

[0057] Feature Importance Analysis: This section analyzes the feature importance of the Random Forest and LightGBM models to identify the most important feature sets that contribute significantly to predictions. For example, the Random Forest model calculates feature importance using Gini impurity, while the LightGBM model uses its built-in feature importance evaluation tool.

[0058] Hyperparameter tuning: Based on the important feature set, the hyperparameters of the Random Forest model and the LightGBM model are tuned to find the optimal combination of hyperparameters. Methods such as Bayesian optimization or grid search are used for tuning to ensure that the model performs better than the initial model on the validation set.

[0059] More specifically, using the initial feature set from the training set as input and the target parameters from the training label set as output, a random forest model is trained to obtain the initial random forest prediction model rf1. Simultaneously, the permutation importance algorithm is called to evaluate the importance of features in the prediction model rf1 using validation set data. `feature_rf` ​​is defined as the set of features with importance values ​​greater than 0 output by the permutation importance algorithm. Furthermore, the hyperparameters of the random forest model are tuned using the BayesSearchCV algorithm. Specific hyperparameters include: the number of trees, maximum depth, minimum number of samples required for splits, and minimum number of samples required for leaf nodes. Finally, the optimal parameter prediction model rf2 is output after tuning.

[0060] Using the initial feature set from the training set as input and the target parameters from the training label set as output, the LightGBM model is trained to obtain the initial LightGBM prediction model IGB1. Simultaneously, the importance ranking algorithm is invoked to evaluate the importance of features in the prediction model IGB1 using validation set data. `feature_lgb` is defined as the set of features with importance values ​​greater than 0 output by the importance ranking algorithm. Furthermore, the hyperparameters of the LightGBM model are tuned using the BayesSearchCV algorithm. Specific hyperparameters include: learning rate, number of iterations for boosting the tree, maximum depth, maximum number of leaves in the tree, minimum number of samples per leaf node, maximum number of bins for feature splitting, L1 regularization parameter (alpha), and L2 regularization parameter (lambda). Finally, the optimally tuned prediction model IGB2 is output.

[0061] The combination of the optimal parameter prediction model rf2 and the optimal parameter prediction model lgb2 constitutes the intermediate prediction model.

[0062] In other words, through model evaluation and hyperparameter tuning, the model performance was further optimized, the model's accuracy and generalization ability were improved, and the model was able to show good prediction results on different datasets.

[0063] S180. Use the intermediate prediction model to predict the validation set to obtain the initial prediction value. Use the initial prediction value as the input feature and the actual label of the validation set as the output feature to train a meta-learner based on a neural network to obtain the target prediction model. Specifically, initial predictions are generated by using a tuned random forest model and a LightGBM model (intermediate prediction model) to make predictions on the validation set. These initial predictions reflect the model's preliminary predictions of the validation set data.

[0064] Meta-learner training (e.g., Meta-Learner): A neural network-based meta-learner is trained using initial predicted values ​​as input features and the actual labels from the validation set as output features. The meta-learner learns the relationship between the output of the intermediate prediction model and the actual labels, further optimizing the prediction results to obtain the target prediction model. The network structure of the meta-learner can include an input layer, hidden layers (e.g., two layers, each with 128 nodes, using ReLU activation function), and an output layer, using mean squared error as the loss function and Adam as the optimization algorithm.

[0065] More specifically, the optimized random forest model rf2 and lightGBM model lgb2 are used to predict the results on the validation set, outputting predicted values ​​rf_pred and lgb_pred respectively. Using rf_pred and lgb_pred as input features and the actual labels of the validation set as output features, a neural network-based meta-learner is trained.

[0066] The meta-learner learns the error mapping relationship between the predicted values ​​rf_pred from the random forest model, the predicted values ​​lgb_pred from the lightGBM model, and the actual labels on the validation set. It outputs the optimal weights w1 and w2 that satisfy w1 + w2 = 1, ultimately forming a weighted fusion target prediction model. best_pred=w1×rf_pred+w2×lgb_pred.

[0067] In other words, by relearning the output of the intermediate prediction model through the meta-learner, the prediction accuracy and generalization ability of the model are further improved, and high-precision prediction of etching parameters is achieved.

[0068] S190. Obtain the sample data to be predicted, and perform feature value processing on the sample data to be predicted according to the definition of the initial feature set to obtain the feature vector; Specifically, the data to be predicted is acquired. This data can be from new production batches or PCB boards awaiting processing. Data sources include the MES system, sensor data collection, or manual recording. Simultaneously, feature value processing is performed on the data to be predicted according to the definition of the initial feature set, including data cleaning and normalization, to ensure that the feature processing method for the samples to be predicted is consistent with that of the training set, thus obtaining feature vectors.

[0069] In other words, by standardizing the data to be predicted, we can ensure that the feature processing method is consistent with that of the training set, thereby improving the model's ability to adapt to new data.

[0070] S200. Input the feature vector into the target prediction model to generate predicted values ​​for etching parameters.

[0071] Specifically, the processed feature vectors are input into the target prediction model to generate predicted etching parameters. These predicted values ​​can guide the setting of etching parameters in the actual production process, improving production efficiency and product quality. Furthermore, the predicted etching parameters are output to the MES system or related production equipment for operators to reference and use. The predicted etching parameters can be directly used to adjust equipment parameters, such as equipment line speed, upper spray pressure, and lower spray pressure, ensuring high quality and efficiency in the etching process.

[0072] More specifically, the trained model-related files are stored and integrated with the MES system. Prediction sample data, including `feature_rf` ​​and `feature_lgb`, is acquired. Following the feature vector data processing method used in the training set, the feature values ​​of the prediction samples are processed to obtain feature vectors. These processed feature vectors are then substituted into the target prediction model to output predicted etching parameters. The output predicted etching parameters are then sent to a database, where employees can query them using relevant equipment and apply them to production.

[0073] In other words, the final target prediction model can generate high-precision etching parameter prediction values ​​based on the input feature vector, providing PCB manufacturers with scientific and intelligent parameter setting solutions, and significantly improving production efficiency and product quality.

[0074] The aforementioned PCB circuit etching parameter prediction method collects multi-source data related to etching quality and uses machine learning models for parameter prediction. It can automatically learn the patterns in the data, thereby reducing reliance on human experience and improving the stability and reliability of parameter settings. In addition, by combining the ensemble learning method of random forest model and lightgbm model, and further optimizing it with a meta-learner based on neural network, it can better capture the complex nonlinear relationships in the etching process and significantly improve the prediction accuracy.

[0075] Figure 3This is a schematic block diagram of a PCB circuit etching parameter prediction device 300 provided in an embodiment of the present invention. Figure 3 As shown, corresponding to the above-described PCB line etching parameter prediction method, the present invention also provides a PCB line etching parameter prediction device 300. This PCB line etching parameter prediction device 300 includes a unit for performing the above-described PCB line etching parameter prediction method, and the device can be configured in a server. Specifically, please refer to... Figure 3 The PCB line etching parameter prediction device 300 includes: Collection unit 301 is used to collect multi-source data related to etching quality; The preprocessing unit 302 is used to preprocess multi-source data to obtain a dataset; Definition unit 303 is used to define the initial feature set and training label set of the model; Setting unit 304 is used to set the seed number of the random number generator; The partitioning unit 305 is used to divide the dataset proportionally according to the number of seeds of the random number generator to form a training set and a validation set; Training unit 306 is used to train the random forest model and the lightgbm model respectively using the training set. During the training process, the model takes the features in the initial feature set as input and the target parameters in the training label set as output, learns the mapping relationship between the features and the target parameters, and obtains the trained random forest model and lightgbm model. The screening and tuning unit 307 is used to evaluate the trained random forest model and lightgbm model using the validation set, and to screen out important feature sets. Based on the important feature sets, the hyperparameters of the trained random forest model and lightgbm model are tuned to obtain intermediate prediction models. The prediction training unit 308 is used to predict the validation set using the intermediate prediction model to obtain the initial prediction value, and to train a neural network-based meta-learner with the initial prediction value as the input feature and the actual label of the validation set as the output feature to obtain the target prediction model. The acquisition and processing unit 309 is used to acquire the sample data to be predicted and perform feature value processing on the sample data to be predicted according to the definition of the initial feature set to obtain the feature vector. The input generation unit 310 is used to input the feature vector into the target prediction model to generate predicted values ​​of etching parameters.

[0076] In one embodiment, the multi-source data includes: Product characteristic data, including surface copper requirement value, actual surface copper value, copper thickness deviation value, minimum product line width, minimum film line spacing, top surface residual copper rate, bottom surface residual copper rate, top surface line width requirement value, and bottom surface line width requirement value; Material property data, including the type of corrosion resist; Equipment parameter data, including equipment line speed, upward spray pressure and downward spray pressure; The parameters of the solution include the solution temperature, copper ion concentration, hydrochloric acid concentration, oxidant concentration, and specific gravity. Quality parameter data, including the actual values ​​of the top surface line width and the bottom surface line width.

[0077] In one embodiment, the preprocessing unit 302 includes: Data cleaning and theoretical equipment line speed calculations are performed on multi-source data to obtain a dataset.

[0078] In one embodiment, the training unit 306 includes: The extraction module is used to extract an initial feature set from the training set as input and extract a training label set as the output target parameter. The first initialization training module is used to initialize the parameters of the random forest model and train it using the features and labels of the training set to learn the relationship between the features and the target parameters in order to obtain the trained random forest model. The second initialization training module is used to initialize the parameters of the LightGBM model and train it using the features and labels of the training set to learn the relationship between the features and the target parameters in order to obtain the trained LightGBM model.

[0079] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned PCB line etching parameter prediction device 300 and each unit can be referred to the corresponding description in the foregoing method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0080] The aforementioned PCB line etching parameter prediction device 300 can be implemented as a computer program, which can, for example, Figure 4 It runs on the computer device shown.

[0081] Please see Figure 4 , Figure 4 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a server, wherein the server can be a standalone server or a server cluster composed of multiple servers.

[0082] See Figure 4The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0083] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a PCB line etching parameter prediction method.

[0084] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0085] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a PCB line etching parameter prediction method.

[0086] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0087] The processor 502 is used to run a computer program 5032 stored in the memory to perform the following steps: Collect multi-source data related to etching quality; preprocess the multi-source data to obtain the dataset; define the initial feature set and training label set for the model; set the seed number for the random number generator; divide the dataset proportionally according to the seed number of the random number generator to form the training set and validation set; train the random forest model and the LightGBM model respectively using the training set. During training, the model uses features from the initial feature set as input and the target parameters from the training label set as output, learning the mapping relationship between features and target parameters to obtain the trained random forest model and the LightGBM model; use the validation set to test the trained random forest model. The model is evaluated against the LightGBM model, and an important feature set is selected. Hyperparameters of the trained Random Forest model and LightGBM model are tuned based on the important feature set to obtain an intermediate prediction model. The intermediate prediction model is used to predict the validation set to obtain initial prediction values. The initial prediction values ​​are used as input features, and the actual labels of the validation set are used as output features to train a neural network-based meta-learner to obtain the target prediction model. The sample data to be predicted is obtained, and the feature values ​​of the sample data to be predicted are processed according to the definition of the initial feature set to obtain feature vectors. The feature vectors are input into the target prediction model to generate etch parameter prediction values.

[0088] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0089] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0090] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein when executed by a processor, the computer program causes the processor to perform the following steps: Collect multi-source data related to etching quality; preprocess the multi-source data to obtain the dataset; define the initial feature set and training label set for the model; set the seed number for the random number generator; divide the dataset proportionally according to the seed number of the random number generator to form the training set and validation set; train the random forest model and the LightGBM model respectively using the training set. During training, the model uses features from the initial feature set as input and the target parameters from the training label set as output, learning the mapping relationship between features and target parameters to obtain the trained random forest model and the LightGBM model; use the validation set to test the trained random forest model. The model is evaluated against the LightGBM model, and an important feature set is selected. Hyperparameters of the trained Random Forest model and LightGBM model are tuned based on the important feature set to obtain an intermediate prediction model. The intermediate prediction model is used to predict the validation set to obtain initial prediction values. The initial prediction values ​​are used as input features, and the actual labels of the validation set are used as output features to train a neural network-based meta-learner to obtain the target prediction model. The sample data to be predicted is obtained, and the feature values ​​of the sample data to be predicted are processed according to the definition of the initial feature set to obtain feature vectors. The feature vectors are input into the target prediction model to generate etch parameter prediction values.

[0091] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0092] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0093] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0094] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0095] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0096] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting PCB circuit etching parameters, characterized in that, include: Collect multi-source data related to etching quality; Preprocess the multi-source data to obtain the dataset; Define the initial feature set and training label set for the model; Set the seed number for the random number generator; The dataset is divided proportionally according to the seed number of the random number generator to form a training set and a validation set; The Random Forest model and the LightGBM model are trained using the training set. During the training process, the model takes the features in the initial feature set as input and the target parameters in the training label set as output, and learns the mapping relationship between the features and the target parameters to obtain the trained Random Forest model and LightGBM model. The trained random forest model and lightgbm model are evaluated using a validation set, and an important feature set is selected. Based on the important feature set, the hyperparameters of the trained random forest model and lightgbm model are tuned to obtain an intermediate prediction model. The intermediate prediction model is used to predict the validation set to obtain the initial prediction value. The initial prediction value is used as the input feature and the actual label of the validation set is used as the output feature to train a meta-learner based on a neural network to obtain the target prediction model. Obtain the sample data to be predicted, and perform feature value processing on the sample data to be predicted according to the definition of the initial feature set to obtain the feature vector; The feature vector is input into the target prediction model to generate predicted values ​​for etching parameters.

2. The PCB circuit etching parameter prediction method according to claim 1, characterized in that, The multi-source data includes: Product characteristic data, including surface copper requirement value, actual surface copper value, copper thickness deviation value, minimum product line width, minimum film line spacing, top surface residual copper rate, bottom surface residual copper rate, top surface line width requirement value, and bottom surface line width requirement value; Material property data, including the type of corrosion resist; Equipment parameter data, including equipment line speed, upward spray pressure and downward spray pressure; The parameters of the solution include the solution temperature, copper ion concentration, hydrochloric acid concentration, oxidant concentration, and specific gravity. Quality parameter data, including the actual values ​​of the top surface line width and the bottom surface line width.

3. The PCB circuit etching parameter prediction method according to claim 1, characterized in that, The preprocessing of multi-source data to obtain a dataset includes: Data cleaning and theoretical equipment line speed calculations are performed on multi-source data to obtain a dataset.

4. The PCB circuit etching parameter prediction method according to claim 1, characterized in that, The method involves training a random forest model and a lightGBM model using a training set. During training, the model takes features from the initial feature set as input and target parameters from the training label set as output, learning the mapping relationship between features and target parameters to obtain the trained random forest model and lightGBM model, including: Extract the initial feature set from the training set as input, and extract the training label set as the output target parameter; Initialize the random forest model parameters and train it using the features and labels of the training set to learn the relationship between the features and the target parameters, so as to obtain the trained random forest model; Initialize the LightGBM model parameters and train it using the features and labels from the training set. Learn the relationship between the features and the target parameters to obtain the trained LightGBM model.

5. A PCB circuit etching parameter prediction device, characterized in that, include: Collection unit, used to collect multi-source data related to etching quality; The preprocessing unit is used to preprocess multi-source data to obtain a dataset; Define the unit, used to define the initial feature set and training label set of the model; A setting unit is used to set the seed number for the random number generator; The partitioning unit is used to divide the dataset proportionally according to the number of seeds of the random number generator to form training and validation sets; The training unit is used to train the random forest model and the lightGBM model respectively using the training set. During the training process, the model takes the features in the initial feature set as input and the target parameters in the training label set as output, and learns the mapping relationship between the features and the target parameters to obtain the trained random forest model and lightGBM model. The screening and tuning unit is used to evaluate the trained random forest model and lightgbm model using the validation set, and to screen out the important feature set. Based on the important feature set, the hyperparameters of the trained random forest model and lightgbm model are tuned to obtain the intermediate prediction model. The prediction training unit is used to predict the validation set using the intermediate prediction model to obtain the initial prediction value. The initial prediction value is used as the input feature and the actual label of the validation set is used as the output feature to train a meta-learner based on a neural network to obtain the target prediction model. The acquisition and processing unit is used to acquire the sample data to be predicted and perform feature value processing on the sample data to be predicted according to the definition of the initial feature set to obtain the feature vector; The input generation unit is used to input the feature vector into the target prediction model to generate predicted values ​​for the etching parameters.

6. The PCB circuit etching parameter prediction device according to claim 5, characterized in that, The multi-source data includes: Product characteristic data, including surface copper requirement value, actual surface copper value, copper thickness deviation value, minimum product line width, minimum film line spacing, top surface residual copper rate, bottom surface residual copper rate, top surface line width requirement value, and bottom surface line width requirement value; Material property data, including the type of corrosion resist; Equipment parameter data, including equipment line speed, upward spray pressure and downward spray pressure; The parameters of the solution include the solution temperature, copper ion concentration, hydrochloric acid concentration, oxidant concentration, and specific gravity. Quality parameter data, including the actual values ​​of the top surface line width and the bottom surface line width.

7. The PCB circuit etching parameter prediction device according to claim 5, characterized in that, The preprocessing unit includes: Data cleaning and theoretical equipment line speed calculations are performed on multi-source data to obtain a dataset.

8. The PCB circuit etching parameter prediction device according to claim 5, characterized in that, The training unit includes: The extraction module is used to extract an initial feature set from the training set as input and extract a training label set as the output target parameter. The first initialization training module is used to initialize the parameters of the random forest model and train it using the features and labels of the training set to learn the relationship between the features and the target parameters in order to obtain the trained random forest model. The second initialization training module is used to initialize the parameters of the LightGBM model and train it using the features and labels of the training set to learn the relationship between the features and the target parameters in order to obtain the trained LightGBM model.

9. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1 to 4.

10. A storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 4.

Images