Model training method and process parameter optimization method based on industrial internet
By training multiple base models and meta-models and combining them with the Industrial Internet, the challenges of multi-objective and multi-mold data processing in injection molding process optimization were solved, resulting in more efficient process parameter optimization and improved production efficiency.
Patent Information
- Application Number
- CN202511614397.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-05
- Publication Date
- 2026-02-03
AI Technical Summary
Existing injection molding process optimization methods mainly rely on a single optimization algorithm, which makes it difficult to achieve multi-objective collaborative optimization and cannot fully consider the differences in structure, material and process characteristics of different molds, and lacks intelligent decision-making capabilities.
Multiple base models are trained using a training dataset. Meta-models are then trained based on the prediction results of the base models to obtain a process parameter optimization model that balances multi-objective optimization and adaptive processing of multi-mold data. The Industrial Internet is then used for mold performance evaluation and process parameter adjustment.
It improves the overall efficiency of injection molding production, enhances the model's prediction accuracy and intelligent decision-making ability for process parameter optimization, and adapts to the differences in multi-mold data.
Smart Images
Figure CN121456475A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of intelligent manufacturing and industrial internet, and particularly relates to a model training method and an industrial internet-based process parameter optimization method. BACKGROUND
[0002] In modern manufacturing, injection molding is a widely used production technology for producing various plastic products. With the intensification of market competition and the continuous improvement of consumer requirements for product quality, the optimization of injection molding process parameters has become a key link for enterprises to improve product competitiveness. By optimizing the injection molding process parameters, not only can the product quality be improved to ensure the dimensional accuracy, appearance quality and mechanical properties of the products, but also the production cycle can be significantly shortened and the production efficiency can be improved.
[0003] Currently, injection molding process optimization mainly relies on single optimization algorithms such as genetic algorithm and particle swarm algorithm. Although single optimization algorithms can optimize specific targets such as quality or energy consumption to a certain extent, their optimization process is often limited to single targets or local optimal solutions, lacking the ability to optimize multiple targets cooperatively.
[0004] In actual production, in addition to considering multiple targets, the optimization of injection molding process also needs to consider factors such as the structural, material and process characteristic differences of different molds, and existing optimization algorithms lack the ability to differentiate multiple mold data, making it impossible to make intelligent decisions based on mold characteristics. Therefore, there is an urgent need for a process parameter optimization method that can comprehensively consider multi-objective optimization and adaptively process multiple mold data. SUMMARY
[0005] To solve the above problems, the application provides a model training method and an industrial internet-based process parameter optimization method.
[0006] In a first aspect, the application provides a model training method, which comprises:
[0007] training a plurality of base models using a training data set, the training data set comprising process parameters and product quality data of a plurality of molds of an injection molding machine; the plurality of base models predicting different target variables of the injection molding machine;
[0008] based on the trained plurality of base models, using a validation data set to obtain a prediction result for each mold;
[0009] training a meta-model based on the prediction result of each mold to obtain a process parameter optimization model for the injection molding machine.
[0010] In a possible implementation, the plurality of base models comprises at least two types of base models, and different types of base models correspond to different variable types of target variables;
[0011] The method further comprises:
[0012] For each variable type, determining a base model of each target variable under the variable type based on a model dictionary associated with the variable type; the model dictionary comprising a plurality of models adapted to the variable type.
[0013] In a possible implementation, the determining of the base model of each target variable under the variable type based on the model dictionary associated with the variable type comprises:
[0014] For any target variable under the variable type, performing cross-validation training on the models in the model dictionary using the training data set and the target variable;
[0015] Determining an initial base model of the target variable from the model dictionary based on performance data of the plurality of models;
[0016] Grid searching for optimal hyperparameters of the initial base model;
[0017] Updating the initial base model using the optimal hyperparameters to obtain the base model of the target variable.
[0018] In a possible implementation, the plurality of base models comprises a first base model, a second base model, and a third base model.
[0019] The target variable predicted by the first base model is a production cycle, the target variable predicted by the second base model is energy consumption, and the target variable predicted by the third base model is product quality.
[0020] The production cycle and the energy consumption are continuous variables, and the first base model and the second base model are regression models.
[0021] The product quality is a discrete variable, and the third base model is a classification model.
[0022] In a possible implementation, the training of the meta-model based on the prediction results of each mold to obtain the process parameter optimization model of the injection molding machine comprises:
[0023] Based on the prediction results of each mold, obtaining model evaluation indexes of each base model when verifying each mold;
[0024] Taking the plurality of model evaluation indexes corresponding to each mold as secondary input features, and training the meta-model according to the secondary input features to obtain the process parameter optimization model; wherein the process parameter optimization model is used to evaluate performance differences of the plurality of molds of the injection molding machine.
[0025] In one possible implementation, training a meta-model based on the prediction results of each mold to obtain the process parameter optimization model of the injection molding machine includes:
[0026] The prediction results of each mold are used as secondary input features, and the meta-model is trained based on the secondary input features to obtain the process parameter optimization model; wherein, the process parameter optimization model is used to evaluate the accuracy of each target variable.
[0027] Secondly, this application provides a process parameter optimization method based on the Industrial Internet, the method comprising:
[0028] A process parameter optimization model is used to determine the prediction weights of multiple molds for an injection molding machine. The prediction weights are positively correlated with the performance of the molds.
[0029] The system schedules production tasks for the injection molding machine using molds whose predicted weight is greater than or equal to a preset weight threshold, and optimizes the process parameters of molds whose predicted weight is less than the preset weight threshold.
[0030] Thirdly, this application also provides a method for optimizing process parameters based on the Industrial Internet, the method comprising:
[0031] A process parameter optimization model was used to determine the correlation between the target variable with the highest accuracy value and each process parameter.
[0032] From multiple process parameters, at least one target process parameter with a correlation greater than or equal to a preset correlation threshold is selected, and at least one of the target process parameters is adjusted.
[0033] Fourthly, this application provides a model training apparatus, comprising: a first training module, a verification module, and a second training module, wherein:
[0034] The first training module is used to train multiple base models using a training dataset, which includes: process parameters and product quality data of multiple molds of the injection molding machine; the multiple base models predict different target variables of the injection molding machine;
[0035] The verification module is used to obtain the prediction results of each mold based on multiple trained base models and a verification dataset.
[0036] The second training module is used to train a meta-model based on the prediction results of each mold to obtain an optimized model of the process parameters of the injection molding machine.
[0037] In one possible implementation, the first training module is further configured to:
[0038] For each variable type, a base model for each target variable under that variable type is determined based on the model dictionary associated with that variable type; the model dictionary includes multiple models adapted to that variable type.
[0039] In one possible implementation, the first training module is further configured to:
[0040] For any target variable under the variable type, use the training dataset and the target variable to perform cross-validation training on the models in the model dictionary;
[0041] Based on the performance data of multiple models, the initial base model of the target variable is determined from the model dictionary;
[0042] The optimal hyperparameters of the initial base model are obtained through a grid search.
[0043] The initial base model is updated using the optimal hyperparameters to obtain the base model of the target variable.
[0044] In one possible implementation, the first training module is further configured to;
[0045] The target variable predicted by the first basic model is the production cycle, the target variable predicted by the second basic model is energy consumption, and the target variable predicted by the third basic model is product quality.
[0046] The production cycle and the energy consumption are continuous variables, and the first basic model and the second basic model are regression models;
[0047] The product quality is a discrete variable, and the third basis model is a classification model.
[0048] In one possible implementation, the second training module is further configured to:
[0049] Based on the prediction results of each mold, obtain the model evaluation index of each base model when validating each mold;
[0050] The multiple model evaluation indicators corresponding to each mold are used as secondary input features, and the meta-model is trained based on the secondary input features to obtain the process parameter optimization model; wherein, the process parameter optimization model is used to evaluate the performance differences of multiple molds of the injection molding machine.
[0051] In one possible implementation, the second training module is further configured to:
[0052] The prediction results of each mold are used as secondary input features, and the meta-model is trained based on the secondary input features to obtain the process parameter optimization model; wherein, the process parameter optimization model is used to evaluate the accuracy of each target variable.
[0053] Fifthly, this application provides a process parameter optimization device based on the Industrial Internet, comprising: a first prediction module and a first optimization module, wherein:
[0054] The first prediction module is used to determine the prediction weights of multiple molds of the injection molding machine using a process parameter optimization model, wherein the prediction weights are positively correlated with the performance of the molds;
[0055] The first optimization module is used to schedule production tasks for the injection molding machine using molds whose predicted weight is greater than or equal to a preset weight threshold, and to optimize the process parameters of molds whose predicted weight is less than the preset weight threshold.
[0056] Sixthly, this application provides a process parameter optimization device based on the Industrial Internet, comprising: a second prediction module and a second optimization module, wherein:
[0057] The second prediction module is used to use a process parameter optimization model to determine the correlation between the target variable with the highest accuracy value and each process parameter.
[0058] The second optimization module is used to select at least one target process parameter from multiple process parameters whose correlation is greater than or equal to a preset correlation threshold, and to adjust at least one of the target process parameters.
[0059] In a seventh aspect, this application also provides an electronic device, comprising: at least one processor and a memory, wherein:
[0060] The memory stores computer-executed instructions;
[0061] The at least one processor executes computer execution instructions stored in the memory, causing the at least one processor to perform the method as described in any of the preceding statements.
[0062] Eighthly, this application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any of the foregoing embodiments.
[0063] Ninthly, this application also provides a computer program product, including a computer program that, when executed by a processor, can implement the steps of the method as described in any of the preceding claims.
[0064] The model training method and industrial internet-based process parameter optimization method provided in this application train multiple base models using a training dataset. Then, the trained base models are used to determine the prediction results for each mold in the validation dataset. Based on the prediction results for each mold and the corresponding process parameters in the training dataset, a meta-model is trained to obtain the process parameter optimization model for the injection molding machine. Subsequently, based on the process parameter optimization model, molds with lower weights or target variables with lower accuracy are identified, and the process parameters for molds with lower weights are adjusted, or the process parameters with higher correlation to the target variables with lower accuracy are adjusted. Through this approach, the obtained process parameter optimization model possesses both multi-objective optimization and adaptive processing capabilities for multi-mold data, improving the model's prediction accuracy and simultaneously enhancing the overall efficiency of injection molding production. Attached Figure Description
[0065] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0066] Figure 1 A flowchart illustrating a model training method provided in this application embodiment. Figure One ;
[0067] Figure 2 A flowchart illustrating a process parameter optimization method based on the Industrial Internet provided in this application embodiment. Figure One ;
[0068] Figure 3 A flowchart illustrating a process parameter optimization method based on the Industrial Internet provided in this application embodiment. Figure Two ;
[0069] Figure 4 This is a schematic diagram of the structure of a model training device provided in an embodiment of this application;
[0070] Figure 5 A schematic diagram of the structure of a process parameter optimization device based on the Industrial Internet provided in this application embodiment. Figure One ;
[0071] Figure 6 A schematic diagram of the structure of a process parameter optimization device based on the Industrial Internet provided in this application embodiment. Figure Two ;
[0072] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.
[0073] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation
[0074] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0075] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented, for example, in orders other than those illustrated or described herein.
[0076] In this application, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.
[0077] In modern manufacturing, injection molding is a key technology for producing various plastic products. Optimizing its process parameters is crucial for improving product quality, shortening production cycles, and reducing costs. However, current optimization methods mostly rely on single algorithms, such as genetic algorithms or particle swarm optimization. While these methods can optimize specific objectives to a certain extent, they struggle to achieve collaborative optimization of multiple objectives. Furthermore, when processing multi-mold data, they cannot fully consider the differences in structure, materials, and process characteristics of different molds, lacking intelligent decision-making capabilities. Therefore, there is an urgent need for a new optimization method that can comprehensively consider multi-objective optimization and adaptively process multi-mold data to meet the complex needs of actual production.
[0078] This application provides a model training method that trains multiple base models using a training dataset. Then, the trained base models are used to determine the prediction results for each mold in the validation dataset. Based on the prediction results for each mold and the corresponding process parameters in the training dataset, a meta-model is trained to obtain an optimized process parameter model for the injection molding machine. This method enables the obtained optimized process parameter model to possess both multi-objective optimization and adaptive processing capabilities for multi-mold data, improving the model's prediction accuracy and enhancing the overall efficiency of injection molding production.
[0079] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.
[0080] The technical solutions of this application and how they solve the aforementioned technical problems are described in detail below with specific embodiments. These specific embodiments may exist independently or in combination with each other. Identical or similar concepts or processes may not be repeated in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.
[0081] In modern manufacturing, injection molding is a widely used production technology for producing various plastic products. With increasing market competition and rising consumer demands for product quality, optimizing injection molding process parameters has become a crucial step for companies to enhance their product competitiveness. Optimizing injection molding process parameters can not only improve product quality, ensuring dimensional accuracy, appearance quality, and mechanical properties, but also significantly shorten production cycles and increase production efficiency. Currently, injection molding process optimization mainly relies on single optimization algorithms, such as genetic algorithms and particle swarm optimization. While single optimization algorithms can optimize specific objectives, such as quality or energy consumption, their optimization process is often limited to a single objective or a local optimum, lacking the ability to collaboratively optimize multiple objectives. In actual production, injection molding process optimization needs to consider multiple objectives simultaneously, as well as the significant differences in the structure, materials, and process characteristics of different molds. Existing optimization algorithms lack the ability to process multi-mold data differentially and cannot make intelligent decisions based on mold characteristics. Therefore, there is an urgent need for a process parameter optimization method that can comprehensively consider multi-objective optimization and adaptively process multi-mold data.
[0082] Therefore, this embodiment uses a multi-model fusion approach to train a meta-model based on the prediction results and process parameters of each mold output by multiple base models, which include multiple target variables, to obtain a process parameter optimization model that takes into account both multi-objective optimization and adaptive processing of multi-mold data.
[0083] Figure 1 A flowchart illustrating a model training method provided in this application embodiment. Figure One .like Figure 1 As shown, the method includes:
[0084] S101. Train multiple base models using the training dataset.
[0085] The training dataset includes process parameters and product quality data of multiple molds of the injection molding machine, and the target variables of the injection molding machine predicted by the multiple base models are different.
[0086] Equipment data from the injection molding machine's production process is collected, cleaned to remove outliers and missing values, and then standardized to ensure consistent dimensions for easier model training and optimization. The data is then divided into training and validation datasets according to a predefined ratio. Both datasets contain process parameters for multiple molds used in the injection molding machine, as well as product quality data.
[0087] Based on multiple predetermined target variables, the base model corresponding to each target variable is retrieved, and each base model is trained using a training dataset. There is a one-to-one correspondence between the base model and the target variable. The base model can be selected according to the actual situation; this embodiment does not impose any limitations. For example, the base model can be a random forest, XGBoost, logistic regression, or similar model.
[0088] It should be noted that, since different types of base models are suitable for different scenarios, to improve the reliability of base model selection, the aforementioned base models include at least two types of base models. For example, at least two types of base models may include regression models and classification models.
[0089] Regression models are a statistical method that establishes a mathematical relationship between independent and dependent variables to predict and analyze the relationships between variables. They are commonly used in economic forecasting, market analysis, and product quality control. Common algorithms in regression models include decision tree regression, random forest regression, and gradient boosting regression.
[0090] A classification model is a machine learning method that establishes a mathematical relationship between feature variables and category variables to classify and predict data. Its core principle is to learn a classification rule or model based on existing data samples, enabling the model to accurately classify new, unknown data. Common algorithms include support vector machines, decision tree classification, and random forest classification.
[0091] Furthermore, considering that classification and regression models involve numerous algorithms, to improve the reliability of the prediction results output by the base model, it is necessary to select the model with the highest fitness from multiple models for each target variable as the base model for that target variable. Given that the variable type of the target variable in the injection molding production process is fixed, this embodiment determines the suitable base model from the corresponding model dictionary based on the variable type of the target variable. The model dictionary includes multiple models suitable for that variable type. The specific steps are as follows:
[0092] For any target variable under the variable type, use the training dataset and the target variable to perform cross-validation training on the models in the model dictionary;
[0093] Based on the performance data of multiple models, the initial base model of the target variable is determined from the model dictionary;
[0094] The optimal hyperparameters of the initial base model are obtained through a grid search.
[0095] The initial base model is updated using the optimal hyperparameters to obtain the base model of the target variable.
[0096] For example, the cross-validation training process can be as follows: Divide the training dataset into 5 non-overlapping subsets in a 5:1 ratio, take 4 subsets as training subsets, and take the remaining subset as validation subset. For each subset, first standardize the input features using Z-score, and construct sample-label pairs using a sliding window method. The window length is consistent with the injection molding cycle to preserve temporal characteristics. If the model is a random forest or gradient boosting tree, set the number of decision trees to ≥200, the maximum depth to ≤20, the minimum number of samples per node to ≥5, and use the mean squared error (MSE) as the splitting criterion. If the model is a neural network, use a 3-layer fully connected structure (the number of hidden neurons is 128, 64, and 32 respectively), the activation function is ReLU, the output layer is a linear unit, the loss function is set to MSE, the optimizer is Adam (lr=1e-3, β1=0.9, β2=0.999), the batch size is 64, the maximum number of training epochs is 300, and an early stopping strategy is introduced (patience=20, stop if the MSE of the validation subset does not decrease). After each training fold is completed, the predicted and actual values of the validation fold are recorded for subsequent calculation of the average MSE.
[0097] The process of determining the initial base model can be as follows: In addition to the average MSE, the performance data also includes the Normalized-RMSE and MAE as auxiliary indicators; if the MSEs are the same and the difference is <1%, the Normalized-RMSE is further compared; if they are still the same, the MAE is compared; and finally, the model with the smallest overall error is determined as the initial base model.
[0098] Furthermore, the process of searching for the optimal hyperparameters is as follows: the grid search space is predefined according to the model type, for example:
[0099] Random Forest: n_estimators∈{200,400,600}, max_depth∈{10,20,30}, min_samples_split∈{5,10,20};
[0100] XGBoost:n_estimators∈{200,400,600}, max_depth∈{3,5,7}, learning_rate∈{0.01,0.05,0.1}, subsample∈{0.7,0.9,1.0}.
[0101] Neural network: The number of neurons in hidden layers 1 / 2 / 3 are scanned to {64,128,256}, {32,64,128}, and {16,32,64} respectively, with dropout_rate ∈ {0.1,0.2,0.3} and learning rate ∈ {1e-4,5e-4,1e-3}; Grid search uses 5-fold cross-validation, and the training process of step
[0092] is repeated for each set of hyperparameters, and the average MSE is recorded; if the validation error of three consecutive sets of parameters decreases by less than 0.1%, the search is terminated early to save computing power.
[0102] After determining the optimal hyperparameters, retrain the model once using the complete training dataset (without further validation layers) to obtain the final base model. After training, calculate MSE, R², and MAE on a separately reserved test set (accounting for 10% of the original data, which was not involved in the training and hyperparameter tuning process) and write them into the model report to ensure reproducibility and traceability.
[0103] Specifically, for any target variable under each variable type, 5-fold cross-validation is performed on each model in the associated model dictionary using the training dataset and the target variable to obtain the performance data of each model, i.e., the mean squared error, which will be referred to as MSE hereafter. The MSE with the smallest value is selected from multiple MSEs, and the initial base model for the target variable is determined based on the model corresponding to the smallest MSE. Then, hyperparameter tuning is performed on the selected initial base model, i.e., grid search is used to determine the optimal hyperparameters of the initial base model. Specifically, a pre-defined hyperparameter grid is traversed, with each grid point representing a specific set of hyperparameter configurations. For each set of hyperparameters, 5-fold cross-validation is performed, that is, the dataset is divided into 5 equal parts, 4 parts are used to train the initial base model, and the remaining part is used for validation, for a total of 5 rounds of training and validation. The MSE of each set of hyperparameters in each round of validation is recorded, and its average value is calculated to evaluate the model performance under that hyperparameter combination. Finally, the hyperparameter combination with the smallest average MSE is selected as the optimal hyperparameters of the initial base model. After determining the optimal hyperparameters of the initial base model, the initial base model is updated based on the optimal hyperparameters to obtain the final base model of the target variable.
[0104] In one possible implementation, the injection molding process is analyzed, and the overall efficiency of injection molding production is typically evaluated through the product's production cycle, energy consumption during production, and product quality. Therefore, multiple target variables can be production cycle, energy consumption, and product quality. Since there is a one-to-one correspondence between the target variables and the base models, the corresponding base models include: a first base model, a second base model, and a third base model. The first base model predicts the production cycle as the target variable, the second base model predicts energy consumption as the target variable, and the third base model predicts product quality as the target variable. The production cycle and energy consumption are continuous variables, and the first and second base models are regression models. Product quality is a discrete variable, and the third base model is a classification model.
[0105] Furthermore, since the target variable includes two variable types—continuous and discrete—the corresponding model dictionaries also include a first model dictionary and a second model dictionary. The first model dictionary corresponds to the continuous variable type; that is, it is the model dictionary for regression models. The second model dictionary corresponds to the discrete variable type; that is, it is the model dictionary for classification models. For example, the first model dictionary includes: XGBoost, CatBoost, LightGBM, Linear Regression, Ridge Regression, Lasso Regression, SVR, Random Forest, Decision Tree, Extra Trees, AdaBoost, Gradient Boosting, and K-Nearest Neighbors Regression. The second model dictionary includes: CatBoostClassifier, DecisionTreeClassifier, and LGBMClassifier.
[0106] S102. Based on multiple trained base models, use the validation dataset to obtain the prediction results for each model.
[0107] In this step, after training each base model, prediction results for each mold are obtained using the validation dataset based on the trained base models. The prediction results include the target variable output by each base model based on the process parameters of each mold.
[0108] For example, the multiple base models include: a first base model, a second base model, and a third base model. The target variable predicted by the first base model is the production cycle, the target variable predicted by the second base model is energy consumption, and the target variable predicted by the third base model is product quality. The prediction results for each mold obtained through the multiple base models are: predicted production cycle, predicted energy consumption, and predicted product quality.
[0109] S103. Based on the prediction results of each mold, train a meta-model to obtain the process parameter optimization model of the injection molding machine.
[0110] In this step, during actual production, the model used to adjust injection molding process parameters needs to take into account multi-objective optimization, as well as the differences in structure, material, and process characteristics of different molds. Therefore, this embodiment determines two types of input information based on prediction results and process parameters, and trains the meta-model using the two types of input information to obtain process parameter optimization models that focus on different directions.
[0111] One type of input information helps the meta-model learn the differences between different molds, determining which molds have more accurate predictions and which have larger prediction errors. This helps the meta-model assign higher weights to more accurate molds during prediction, improving the overall prediction accuracy of the process parameter optimization model trained on the meta-model. Therefore, the process parameter optimization model trained based on this input information focuses on identifying the molds that require process parameter optimization from among the multiple molds in the injection molding machine.
[0112] For example, data preparation: Secondary input feature vector construction: For each mold m, collect the evaluation metrics generated by each base model in the 5-fold cross-validation, and concatenate them in the order of "base model-metric" to obtain a fixed-length vector Xm. If a base model is missing a metric, it is filled with the global mean of the training set. Label construction: Use the actual product quality pass / fail (0 / 1) in the validation compromise as the sample-level label y; for mold-level tasks, take the F1-score of all validation folds of the mold as the mold-level label Ym;
[0113] Model selection: A 2-layer TabNet ("attention-based" interpretable network) was adopted, with the following parameters: feature_dim=128, output_dim=64, n_steps=3, n_independent=2, n_shared=2, gamma=1.3;
[0114] Training configuration: Loss function: BCEWithLogitsLoss (mold-level task) or MSELoss (F1-score regression task), switch according to task;
[0115] Optimizer: AdamW, lr=2e-3, weight_decay=1e-4; batch size: 256; maximum epochs: 500, early stop patience=30, monitor validation set AUC or RMSE;
[0116] Post-training processing: Save the optimal ckpt, and calculate AUC≥0.85 or RMSE≤0.05 on the test set (10% independent data) to solidify the model; otherwise, automatically roll back and re-search the grid for lr and weight_decay.
[0117] Another type of input information focuses on the target variable itself, enabling the process parameter optimization model trained based on this input information to determine the accuracy of prediction for each target variable and to perform targeted optimization on process parameters that are highly correlated with target variables with low accuracy.
[0118] Specific training methods for the meta-model of "target variable accuracy":
[0119] Secondary input feature vector construction: For each target variable t of each mold m, calculate the MAPE by taking the base model prediction value and the true value to obtain the vector X_{m,t}; after concatenating all target variables, form a |m|×|target variable| sample;
[0120] Label construction: The negative value of MAPE is used as the label y_{m,t} of "accuracy score". The higher the score, the more accurate the accuracy.
[0121] Model selection: LightGBM, 800 trees, depth ≤ 8, learning rate 0.05, feature_fraction=0.8, bagging_fraction=0.8, bagging_freq=5;
[0122] Training configuration: Loss function: regression_l2;
[0123] Verification method: 5-fold group-K-fold (grouped by mold ID to prevent data leakage);
[0124] Early stopping: Stop if the RMSE of the validation set does not decrease within 50 rounds;
[0125] Post-training processing: Save the model with the best iteration rounds; calculate the predicted score and the true score on the independent test group, and R² ≥ 0.75 before it can be solidified; otherwise, automatically increase the number of trees to 1200 and retrain.
[0126] For the process parameter optimization model focusing on different molds, the generation process is as follows: based on the prediction results of each mold, obtain the model evaluation index of each base model when verifying each mold; take the multiple model evaluation indexes corresponding to each mold as secondary input features, and train the meta-model according to the secondary input features to obtain the process parameter optimization model; wherein, the process parameter optimization model is used to evaluate the performance differences of multiple molds of the injection molding machine.
[0127] The model evaluation metric for each base model during the validation of each mold is determined based on the predicted value of that base model in the prediction results and the product quality data of that mold in the validation dataset. This embodiment does not limit the specific selection of the model evaluation metric for each base model. For example, the model evaluation metric for the base model corresponding to the production cycle can be the mean squared error of the cycle or the mean absolute percentage error of the cycle. The model evaluation metric for the base model corresponding to energy consumption can be the mean squared error of energy consumption or the mean absolute percentage error of energy consumption. The model evaluation metric for the base model corresponding to product quality can be the accuracy of product quality. After obtaining the model evaluation metric for each base model during the validation of each mold, multiple model evaluation metrics corresponding to each mold are used as secondary input features. A meta-model is then trained based on these secondary input features to obtain a process parameter optimization model. The labels of the meta-dataset are the same as the labels of the validation dataset.
[0128] The process for obtaining the process parameter optimization model that focuses on the target variable itself is as follows: the prediction result of each mold is used as a secondary input feature, and the meta-model is trained based on the secondary input feature to obtain the process parameter optimization model; wherein, the process parameter optimization model is used to evaluate the accuracy of each target variable.
[0129] Specifically, a suitable meta-model is selected from multiple models. The input information of the meta-model is the prediction result of each mold, and the output information is the correlation between the target variable with the highest accuracy and each process parameter. The meta-model is trained using the prediction results of multiple molds to obtain the final process parameter optimization model used to evaluate the accuracy of each target variable. Then, the process parameters with high accuracy and high correlation with the target variable to be optimized are optimized in a targeted manner.
[0130] Correlation calculation and targeted optimization:
[0131] The SHAP value is used to interpret the LightGBM metamodel, and the global importance |SHAP| of each process parameter to the "accuracy score" feature is extracted.
[0132] Sort by |SHAP| in descending order and take the top-3 process parameters as "targeted optimization parameters";
[0133] The DOE table is automatically generated in the MES system, with a parameter variation range of ±5% and a step size of 1%, and sent to the injection molding machine controller to perform closed-loop verification.
[0134] If the target variable MAPE decreases by ≥10% after verification, then the new parameter is fixed; otherwise, it is rolled back and logged.
[0135] This application provides a model training method. Multiple base models are trained using a training dataset. Then, the trained base models are used to determine the prediction results for each mold in the validation dataset. Based on the prediction results for each mold and the corresponding process parameters in the training dataset, a meta-model is trained to obtain an injection molding machine process parameter optimization model. This method enables the obtained process parameter optimization model to possess both multi-objective optimization and adaptive processing capabilities for multi-mold data, improving the model's prediction accuracy and enhancing the overall efficiency of injection molding production. Furthermore, two types of secondary input features are defined based on the prediction results for each mold. A meta-model is trained based on these secondary input features and the corresponding process parameters, resulting in two types of process parameter optimization models. This ensures that when optimizing the injection molding process using the process parameter optimization model, multiple objectives are considered, and the differences in structure, material, and process characteristics of different molds are taken into account.
[0136] Figure 2 A flowchart illustrating a process parameter optimization method based on the Industrial Internet provided in this application embodiment. Figure One This embodiment provides a detailed explanation of the steps involved in optimizing the process parameters of multiple molds based on a process parameter optimization model. For example... Figure 2 As shown, the method includes:
[0137] S201. Using a process parameter optimization model, determine the prediction weights of multiple molds for the injection molding machine. The prediction weights are positively correlated with the performance of the molds.
[0138] The prediction weights are positively correlated with the performance of the mold.
[0139] In this step, the process parameter optimization model is trained based on multiple model evaluation indicators obtained during the verification of each mold using multiple base models. The specific generation process has been described in detail in the above embodiments and will not be elaborated further here.
[0140] Specifically, the actual process parameters of each mold in the injection molding machine, as well as the corresponding actual product quality data, such as actual production cycle, actual energy consumption, and actual product quality, are obtained. The actual production cycle of each mold is output as multiple base models, yielding the model evaluation index for each base model during the mold validation process. The training process of multiple base models and the determination process of model evaluation indices have been described in detail in the aforementioned embodiments and will not be elaborated upon here. The objective loss function within the process parameter optimization model is called to obtain the index weight corresponding to each model evaluation index. Based on the multiple model evaluation indices and index weights of the mold, the prediction weight of the mold is determined. The higher the prediction weight, the better the performance of the corresponding mold, that is, the more suitable the process parameters of the mold are.
[0141] For example, by calling the objective function of the process parameter optimization model, the weights of the production cycle mean square error (a) are determined to be 0.00865325, the production cycle mean absolute percentage error (b) is -0.0163502, the energy consumption average error (c) is 0.00133376, the energy consumption average absolute percentage error (d) is -0.03770715, and the product quality accuracy (e) is 0.0627139. Based on multiple base models, the multiple model evaluation indicators for mold a are determined as follows: production cycle mean square error is 43.1371, production cycle mean absolute percentage error is 0.9218, energy consumption average error is 0.0041, energy consumption average absolute percentage error is 0.8892, and product quality accuracy is 0.88.
[0142] The prediction weight of mold a is calculated as follows: 0.00865325×43.1371-0.0163502×0.9218+0.00133376×0.0041-0.03770715×0.8892+0.0627139×0.88=0.3804.
[0143] S202, Schedule production tasks for the injection molding machine using molds whose predicted weight is greater than or equal to a preset weight threshold, and optimize the process parameters of molds whose predicted weight is less than the preset threshold.
[0144] In this step, given the predicted weight of each mold in the injection molding machine, when the machine needs to operate, it prioritizes scheduling the molds with predicted weights greater than or equal to a preset weight threshold, and with higher values. Molds with predicted weights less than the preset weight threshold indicate poor performance, requiring optimization and adjustment of their process parameters.
[0145] This application provides a process parameter optimization method based on the Industrial Internet. It determines the actual process parameters and actual product quality data of each mold in an injection molding machine, and identifies multiple model evaluation indicators for that mold using multiple base models. Based on the process parameter optimization model, it determines the weight of each model evaluation indicator. Then, based on the multiple model evaluation indicators and their weights, it determines the predicted weight of the corresponding mold. When the injection molding machine calls up a mold, it recommends molds with predicted weights greater than or equal to a preset weight threshold, and adjusts the process parameters of molds with predicted weights less than the preset weight threshold. Through this method, closed-loop control enables dynamic adjustment of injection molding process parameters, adapting to dynamic production environments while fully utilizing the characteristics of different molds, thus improving the generalization ability and prediction accuracy of the process parameter optimization model.
[0146] Figure 3A flowchart illustrating a process parameter optimization method based on the Industrial Internet provided in this application embodiment. Figure Two This embodiment provides a detailed explanation of the steps involved in targeted optimization of the process parameters of the injection molding machine's target variable based on a process parameter optimization model. For example... Figure 3 As shown, the method includes:
[0147] S301. Using a process parameter optimization model, determine the correlation between the target variable with the highest accuracy value and each process parameter.
[0148] In this step, this embodiment uses a process parameter optimization model to determine the accuracy of each target variable of the injection molding machine, and then optimizes the process parameters with higher accuracy and higher correlation to the target variables. Therefore, the input information of the process parameter optimization model used in this embodiment is the prediction result of each mold of the injection molding machine, that is, multiple prediction target variables of the mold obtained through multiple base models, such as: predicted production cycle, predicted energy consumption, and predicted product quality. The training process of this process parameter optimization model has been described in detail in the foregoing embodiments, and will not be described in detail here.
[0149] Then, multiple target variables are input into the process parameter optimization model to obtain the correlation between the target variable with the highest accuracy value output by the process parameter optimization model and each process parameter.
[0150] S302. From multiple process parameters, select at least one target process parameter whose correlation is greater than or equal to a preset correlation threshold, and adjust at least one of the target process parameters.
[0151] In this step, after obtaining the correlation between the target variable with the highest accuracy and each process parameter, the relationship between each correlation and a preset correlation threshold is determined. Based on the determination results, at least one target process parameter with a correlation greater than or equal to the preset correlation threshold is selected. Then, at least one target process parameter is optimized, thereby achieving targeted optimization of the target variable to improve the injection molding process.
[0152] The preset relevance threshold is set using a dual criterion of "statistical significance + business perceptibility": First, based on historical closed-loop validation data, the 99th percentile of the absolute value of the SHAP of each process parameter is calculated as the statistical upper limit. Then, combined with engineering experience, a minimum perceptible benefit of ≥0.05 (5% MAPE improvement equivalent) is taken, and the larger of the two is set as 0.10. When the amount of real-time data is insufficient, 0.10 is used as the default value. As the online validation sample accumulates, the quantile is refitted and automatically updated every 100 sets of mold data, thus ensuring that the quality improvement brought about by parameter adjustment can be clearly perceived on-site while ensuring statistical significance.
[0153] This application provides a process parameter optimization method based on the Industrial Internet. The actual process parameters of multiple molds in an injection molding machine are input into multiple base models to obtain prediction results for each mold. These prediction results are then input into a process parameter optimization model to obtain the correlation between the target variable with the highest accuracy and each process parameter. Subsequently, the target process parameters corresponding to correlations exceeding a preset correlation threshold are adjusted. Through this method, closed-loop control enables dynamic adjustment of injection molding process parameters to adapt to dynamic production environments. Simultaneously, multi-objective optimization can be achieved, improving the overall efficiency of injection molding production.
[0154] Figure 4 This is a schematic diagram of a model training device provided in an embodiment of this application. Figure 4 As shown, the model training device 40 includes: a first training module 401, a verification module 402, and a second training module 403, wherein:
[0155] The first training module 401 is used to train multiple base models using a training dataset, the training dataset including: process parameters and product quality data of multiple molds of the injection molding machine; the multiple base models predict different target variables of the injection molding machine;
[0156] The verification module 402 is used to obtain the prediction results of each mold based on multiple trained base models using a verification dataset;
[0157] The second training module 403 is used to train a meta-model based on the prediction results of each mold to obtain an optimized model of the process parameters of the injection molding machine.
[0158] In one possible implementation, the first training module 401 is further configured to:
[0159] For each variable type, a base model for each target variable under that variable type is determined based on the model dictionary associated with that variable type; the model dictionary includes multiple models adapted to that variable type.
[0160] In one possible implementation, the first training module 401 is further configured to:
[0161] For any target variable under the variable type, use the training dataset and the target variable to perform cross-validation training on the models in the model dictionary;
[0162] Based on the performance data of multiple models, the initial base model of the target variable is determined from the model dictionary;
[0163] The optimal hyperparameters of the initial base model are obtained through a grid search.
[0164] The initial base model is updated using the optimal hyperparameters to obtain the base model of the target variable.
[0165] In one possible implementation, the first training module 401 is further configured to;
[0166] The target variable predicted by the first basic model is the production cycle, the target variable predicted by the second basic model is energy consumption, and the target variable predicted by the third basic model is product quality.
[0167] The production cycle and the energy consumption are continuous variables, and the first basic model and the second basic model are regression models;
[0168] The product quality is a discrete variable, and the third basis model is a classification model.
[0169] In one possible implementation, the second training module 403 is further configured to:
[0170] Based on the prediction results of each mold, obtain the model evaluation index of each base model when validating each mold;
[0171] The multiple model evaluation indicators corresponding to each mold are used as secondary input features, and the meta-model is trained based on the secondary input features to obtain the process parameter optimization model; wherein, the process parameter optimization model is used to evaluate the performance differences of multiple molds of the injection molding machine.
[0172] In one possible implementation, the second training module 403 is further configured to:
[0173] The prediction results of each mold are used as secondary input features, and the meta-model is trained based on the secondary input features to obtain the process parameter optimization model; wherein, the process parameter optimization model is used to evaluate the accuracy of each target variable.
[0174] The model training device provided in this embodiment can execute the model training method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.
[0175] Figure 5 A schematic diagram of the structure of a process parameter optimization device based on the Industrial Internet provided in this application embodiment. Figure One .like Figure 5 As shown, the process parameter optimization device 50 includes: a first prediction module 501 and a first optimization module 502, wherein:
[0176] The first prediction module 501 is used to determine the prediction weights of multiple molds of the injection molding machine using a process parameter optimization model, wherein the prediction weights are positively correlated with the performance of the molds;
[0177] The first optimization module 502 is used to schedule production tasks for the injection molding machine using molds whose predicted weight is greater than or equal to a preset weight threshold, and to optimize the process parameters of molds whose predicted weight is less than the preset weight threshold.
[0178] This embodiment provides a process parameter optimization device based on the Industrial Internet, which can execute the process parameter optimization method based on the Industrial Internet provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.
[0179] Figure 6 A schematic diagram of the structure of a process parameter optimization device based on the Industrial Internet provided in this application embodiment. Figure Two .like Figure 6 As shown, the process parameter optimization device 60 includes: a second prediction module 601 and a second optimization module 602, wherein:
[0180] The second prediction module 601 is used to use a process parameter optimization model to determine the correlation between the target variable with the highest accuracy value and each process parameter;
[0181] The second optimization module 602 is used to select at least one target process parameter from multiple process parameters whose correlation is greater than or equal to a preset correlation threshold, and to adjust at least one of the target process parameters.
[0182] This embodiment provides a process parameter optimization device based on the Industrial Internet, which can execute the process parameter optimization method based on the Industrial Internet provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.
[0183] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 7 As shown, the electronic device 70 includes at least one processor 701 and a memory 702. Optionally, the electronic device 70 also includes a communication component 703. The processor 701, memory 702, and communication component 703 are connected via a bus 704.
[0184] The memory 702 stores computer-executed instructions;
[0185] The at least one processor 701 executes computer execution instructions stored in the memory 702, causing the at least one processor 701 to perform the method as described in any of the preceding descriptions.
[0186] At least one processor 701 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.
[0187] Optionally, in specific implementations, the processor 701 and memory 702 are implemented independently. In this case, the processor 701 and memory 702 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc., but this does not imply that there is only one bus or one type of bus.
[0188] Optionally, in a specific implementation, if the processor 701 and the memory 702 are integrated on a single chip, the processor 701 and the memory 702 can communicate through an internal interface.
[0189] This application also provides a computer storage medium storing computer execution instructions, which, when executed by a processor, implement the technical solution of any of the aforementioned methods.
[0190] The aforementioned computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The computer-readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.
[0191] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Alternatively, the readable storage medium can be an integral part of the processor. Both the processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the control device of a garment handling apparatus.
[0192] The division of units described herein is merely a logical functional division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.
[0193] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0194] In addition, the functional units in the various embodiments of the present 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.
[0195] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion 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, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0196] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.
[0197] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.
Claims
1. A model training method, characterized in that, The method includes: Multiple base models are trained using a training dataset, which includes process parameters and product quality data for multiple molds of the injection molding machine; the multiple base models predict different target variables for the injection molding machine. Based on multiple pre-trained base models, the prediction results for each mold are obtained using the validation dataset; Based on the prediction results of each mold, a meta-model is trained to obtain the process parameter optimization model of the injection molding machine.
2. The method according to claim 1, characterized in that, The multiple base models include at least two types of base models, and the target variable has a different variable type corresponding to the different types of base models; The method further includes: For each variable type, a base model for each target variable under that variable type is determined based on the model dictionary associated with that variable type; the model dictionary includes multiple models adapted to that variable type.
3. The method according to claim 2, characterized in that, The determination of the base model for each target variable under the variable type based on the model dictionary associated with the variable type includes: For any target variable under the variable type, use the training dataset and the target variable to perform cross-validation training on the models in the model dictionary; Based on the performance data of multiple models, the initial base model of the target variable is determined from the model dictionary; The optimal hyperparameters of the initial base model are obtained through a grid search. The initial base model is updated using the optimal hyperparameters to obtain the base model of the target variable.
4. The method according to claim 2, characterized in that, The plurality of basis models include: a first basis model, a second basis model, and a third basis model; The target variable predicted by the first basic model is the production cycle, the target variable predicted by the second basic model is energy consumption, and the target variable predicted by the third basic model is product quality. The production cycle and the energy consumption are continuous variables, and the first basic model and the second basic model are regression models; The product quality is a discrete variable, and the third basis model is a classification model.
5. The method according to any one of claims 1-4, characterized in that, The process parameter optimization model for the injection molding machine is obtained by training a meta-model based on the prediction results of each mold, including: Based on the prediction results of each mold, obtain the model evaluation index of each base model when validating each mold; The multiple model evaluation indicators corresponding to each mold are used as secondary input features, and the meta-model is trained based on the secondary input features to obtain the process parameter optimization model; wherein, the process parameter optimization model is used to evaluate the performance differences of multiple molds of the injection molding machine.
6. The method according to any one of claims 1-4, characterized in that, The process parameter optimization model for the injection molding machine is obtained by training a meta-model based on the prediction results of each mold, including: The prediction results of each mold are used as secondary input features, and the meta-model is trained based on the secondary input features to obtain the process parameter optimization model; wherein, the process parameter optimization model is used to evaluate the accuracy of each target variable.
7. A method for optimizing process parameters based on the Industrial Internet, characterized in that, The method includes: Using the process parameter optimization model trained by the method described in any one of claims 1-5, the prediction weights of multiple molds of the injection molding machine are determined, and the prediction weights are positively correlated with the performance of the molds; The system schedules production tasks for the injection molding machine using molds whose predicted weight is greater than or equal to a preset weight threshold, and optimizes the process parameters of molds whose predicted weight is less than the preset weight threshold.
8. The method according to claim 7, characterized in that, The method further includes: Using the process parameter optimization model trained by the method described in claim 6, the correlation between the target variable with the highest accuracy value and each process parameter is determined. From multiple process parameters, at least one target process parameter with a correlation greater than or equal to a preset correlation threshold is selected, and at least one of the target process parameters is adjusted.
9. An electronic device, characterized in that, include: At least one processor and memory, wherein: The memory stores computer-executed instructions; The at least one processor executes computer execution instructions stored in the memory, causing the at least one processor to perform the method as described in any one of claims 1-8.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-8.