A method, medium and system for determining production parameters of ABS multilayer sheet

Abstract

The present invention provides a method, medium and system for determining production parameters of an ABS multilayer sheet, belonging to the technical field of ABS multilayer sheet production, comprising: obtaining production data, including production parameters and product quality, and normalizing the data; fitting the relationship between each production parameter and product quality according to the normalized data, and establishing a mathematical model; inputting the production parameters into the model, calculating the difference between the estimated quality and the actual quality, and screening out matching and deviation data; using a data augmentation method, processing the deviation data, and generating a third augmented production data set; using the augmented data set, training a bidirectional neural network model to obtain a parameter-quality relationship basic model, and optimizing and updating it into a parameter-quality relationship model; according to product quality requirements, using the parameter-quality relationship model to output multiple groups of reference production parameters, thereby solving the technical problem that the optimization of production parameters becomes more complicated during large-scale production, and empirical parameter adjustment is difficult to meet production needs.

Description

Technical Field

[0001] The present invention belongs to the technical field of ABS multilayer sheet production, and in particular, relates to a method, medium and system for determining production parameters of ABS multilayer sheet. Background Art

[0002] Composite materials are new materials made of two or more materials with different properties through certain process methods. Among many composite materials, ABS plastic composite sheets are widely used in home appliances, decoration, automobiles and other fields due to their excellent comprehensive performance. ABS plastic composite sheets are made of ABS resin, toughening agent, filler and other raw materials through extrusion process, and their final performance depends on the ratio of raw materials and the setting of process parameters.

[0003] At present, the production of ABS plastic composite sheets mainly relies on the production staff to adjust parameters based on experience. This empirical parameter adjustment is easily subject to the subjective judgment of the production staff and it is difficult to ensure the stability and consistency of product quality. Especially in large-scale production, the optimization of production parameters becomes more complicated, and empirical parameter adjustment is difficult to meet production needs. Summary of the invention

[0004] In view of this, the present invention provides a method, medium and system for determining production parameters of ABS multilayer sheets, which can solve the technical problems in the prior art that the optimization of production parameters becomes more complicated during large-scale production, and empirical parameter adjustment is difficult to meet production needs.

[0005] The present invention is achieved in that:

[0006] A first aspect of the present invention provides a method for determining production parameters of an ABS multilayer sheet material, comprising the following steps:

[0007] S10, obtaining production data of multiple groups of ABS multilayer sheets, including production parameters and product quality, and normalizing the production data to obtain normalized production data, wherein the product quality includes three categories: appearance quality, mechanical performance quality, and thermal performance quality; the production parameters include two categories: process parameters and raw material parameters; wherein each category of product quality and each category of production parameters include multiple data;

[0008] S20, fitting a separate order relationship between each production parameter and each product quality as a basic relationship according to the normalized production data;

[0009] S30, establishing a mathematical model according to the basic relationship, wherein the mathematical model includes three sub-models, namely, an appearance quality sub-model, a mechanical performance quality sub-model, and a thermal performance quality sub-model, and fitting the normalized production data to obtain a production mathematical model;

[0010] S40, for each set of normalized production data, inputting the production parameters into the production mathematical model to obtain product quality, recorded as estimated quality, and calculating the difference between the estimated quality and the original production quality;

[0011] S50, recording the normalized production data with a discrimination lower than a preset first discrimination threshold as matching production data, recording the normalized production data with a discrimination higher than or equal to the preset first discrimination threshold as deviation production data, augmenting the production parameters of the deviation production data by using a data augmentation method, and calculating the estimated quality by using the production mathematical model to form a first augmented production data set; optionally, the first discrimination threshold is 0.1;

[0012] S60, screening the production data with a distinction lower than a second distinction threshold in the augmented production data set as a second augmented production data set, merging the second augmented production data set with all matching production data to record as a third augmented production data set; optionally, the second distinction threshold is 0.2;

[0013] S70, using the third augmented production data set, training a bidirectional neural network model to obtain a parameter-quality relationship basic model, wherein the bidirectional neural network model includes an appearance quality branch, a mechanical property quality branch, a thermal property quality branch, a fusion branch, and a constraint branch;

[0014] S80, using all matching production data as a verification set, optimizing and updating the parameter-quality relationship basic model to obtain a parameter-quality relationship model;

[0015] S90, obtaining product quality requirements of the current ABS multilayer sheet production task, inputting them into the parameter quality relationship model, and calculating multiple groups of production parameters as reference production parameters to output to production personnel.

[0016] The appearance quality, mechanical performance quality and thermal performance quality are all obtained through experiments, and the specific methods of obtaining them include:

[0017] Appearance quality is tested by visual inspection, gloss meter and roughness meter;

[0018] The mechanical performance quality is tested by impact test, tensile test, three-point bending test and hardness tester;

[0019] The thermal performance quality is tested using a heat deformation temperature tester or a thermal expansion coefficient tester or a thermal conductivity tester as specialized equipment.

[0020] Wherein, the appearance quality at least or all includes: glossiness, color uniformity, surface flatness, embossing marks;

[0021] The mechanical properties include at least or all of: impact resistance, tensile strength, flexural modulus, hardness;

[0022] The thermal performance qualities at least or all include: thermal deformation temperature, thermal expansion coefficient, and thermal conductivity.

[0023] Wherein, the process parameters at least or all include: extruder head temperature, mold temperature, extrusion speed, and molding parameters.

[0024] Among them, the raw material parameters at least or all include: ABS resin model, toughening agent model and dosage, filler model and dosage, colorant model and dosage.

[0025] Among them, the fusion branch embeds a genetic algorithm module for global optimization search of the weight parameters of the neural network; the initial population of the genetic algorithm module is a set of weight parameter sets obtained from each step of training of the appearance quality branch, the mechanical performance quality branch, and the thermal performance quality branch.

[0026] Among them, the fitness function of the genetic algorithm module is the mean square error between the predicted value and the true value, and the elimination function is natural selection, and individuals with high fitness are retained in each generation.

[0027] Among them, the constraint branch is specifically a neural network structure based on rule constraints, which is used to introduce the feasible range of each parameter into the model; the genetic algorithm module is used to feed back the optimized weight parameter set to each branch as the weight input for the next step of training.

[0028] The appearance quality branch specifically includes the following structure:

[0029] Input layer: process parameter node + raw material parameter node;

[0030] The first hidden layer: 64 fully connected nodes, using ReLU activation;

[0031] Second hidden layer: 32 fully connected nodes, parallel connection of a residual branch directly mapped from the first layer, using ReLU activation;

[0032] The third hidden layer: 16 fully connected nodes, receiving features and residual branches from the second layer, using ReLU activation;

[0033] Output layer: appearance quality index node, no activation function;

[0034] The mechanical performance quality branch specifically includes the following structure:

[0035] Input layer: process parameter node + raw material parameter node;

[0036] Encoder: bidirectional LSTM layer, 128 units;

[0037] Attention layer: uses the hidden state from the encoder to calculate the attention score;

[0038] Decoder: unidirectional LSTM layer, 128 units, receiving the attention-weighted encoder output as input;

[0039] Output layer: mechanical performance quality indicator node, using tanh activation;

[0040] The thermal performance quality branch specifically includes the following structure:

[0041] Input layer: process parameter node + raw material parameter node;

[0042] Encoder: Bidirectional LSTM layer, 256 units, connected in parallel with a FiLM adjustment layer;

[0043] Decoder: unidirectional LSTM layer, 256 units, receives the encoder output adjusted by FiLM, and connects an attention branch in parallel to focus on local features;

[0044] Output layer: thermal performance quality indicator node, no activation function;

[0045] The specific structure of the fusion branch is:

[0046] Input layer: the output nodes of the three branches are connected in cascade;

[0047] The first hidden layer: 64 fully connected nodes, connected in parallel with a diffusion branch to capture long-range dependencies, using ReLU activation;

[0048] The second hidden layer: 32 fully connected nodes, receiving features from the first layer and the diffusion branch output, using ReLU activation;

[0049] Output layer: 1 node, representing the comprehensive quality score, no activation function.

[0050] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for determining production parameters of an ABS multilayer board.

[0051] A third aspect of the present invention provides a system for determining production parameters of an ABS multilayer board, comprising the above-mentioned computer-readable storage medium.

[0052] Compared with the prior art, the method, medium and system for determining production parameters of ABS multilayer sheet provided by the present invention have the following beneficial effects:

[0053] 1. Establish a comprehensive ABS plastic composite sheet production data set, including process parameters, raw material parameters, and all-round product performance indicators such as appearance quality, mechanical performance quality, and thermal performance quality, laying a solid foundation for subsequent modeling and analysis. By normalizing the production data, the dimension and value differences between the indicators are eliminated, further improving the applicability of the data.

[0054] 2. A comprehensive production mathematical model was constructed, which consists of three parts: appearance quality sub-model, mechanical performance quality sub-model and thermal performance quality sub-model. This integrated model can comprehensively predict the product quality of ABS plastic composite sheets under given production parameters, which is closer to actual production needs than the existing single performance prediction model.

[0055] 3. The strategy of data augmentation and model optimization was adopted to effectively improve the generalization ability of the model. By targeted augmentation of deviation data, the missing data in the original data were made up, so that the model could more accurately capture the complex relationship between production parameters and product quality. At the same time, the model was further optimized using high-quality matching data, which further improved the prediction performance.

[0056] 4. It realizes the reverse deduction from the target product quality to the production parameters that meet the requirements, and provides reliable parameter formula suggestions for production personnel. This method greatly shortens the production debugging cycle and significantly improves production efficiency.

[0057] In summary, the solution of the present invention solves the technical problem that in the prior art, when large-scale production occurs, the optimization of production parameters becomes more complicated and empirical parameter adjustment is difficult to meet production needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0059] Figure 1 A flow chart of the method provided by the present invention;

[0060] Figure 2 is the relationship between the extruder barrel temperature and the normalized glossiness;

[0061] Figure 3 It is the structure diagram of the appearance quality sub-model;

[0062] Figure 4 It is the structure diagram of mechanical performance quality sub-model;

[0063] Figure 5 It is the structure diagram of the thermal performance quality sub-model;

[0064] Figure 6 Schematic diagram of the production parameter optimization process. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0066] like Figure 1 FIG. 1 is a flow chart of a method for determining production parameters of an ABS multilayer sheet provided by the present invention. The method comprises the following steps:

[0067] S10, obtaining production data of multiple groups of ABS multilayer sheets, including production parameters and product quality, and normalizing the production data to obtain normalized production data, wherein the product quality includes three categories: appearance quality, mechanical performance quality, and thermal performance quality; the production parameters include two categories: process parameters and raw material parameters; wherein each category of product quality and each category of production parameters include multiple data;

[0068] S20, fitting a separate order relationship between each production parameter and each product quality as a basic relationship based on the normalized production data;

[0069] S30, establishing a mathematical model based on the basic relationship, the mathematical model includes three sub-models, namely, an appearance quality sub-model, a mechanical performance quality sub-model, and a thermal performance quality sub-model, and fitting the normalized production data to obtain a production mathematical model;

[0070] S40, for each set of normalized production data, input the production parameters into the production mathematical model to obtain product quality, recorded as estimated quality, and calculate the difference between the estimated quality and the original production quality;

[0071] S50, recording the normalized production data whose discrimination is lower than a preset first discrimination threshold as matching production data, recording the normalized production data whose discrimination is higher than or equal to the preset first discrimination threshold as deviation production data, augmenting the production parameters of the deviation production data by using a data augmentation method, and calculating the estimated quality by using a production mathematical model to form a first augmented production data set;

[0072] S60, screening the production data with a distinction lower than a second distinction threshold in the augmented production data set as a second augmented production data set, merging the second augmented production data set with all matching production data, and recording the result as a third augmented production data set;

[0073] S70, using the third augmented production data set, training a bidirectional neural network model to obtain a parameter-quality relationship basic model, the bidirectional neural network model including an appearance quality branch, a mechanical performance quality branch, a thermal performance quality branch, a fusion branch, and a constraint branch;

[0074] S80, using all matching production data as a verification set, optimizing and updating the parameter-quality relationship basic model to obtain a parameter-quality relationship model;

[0075] S90, obtaining product quality requirements of the current ABS multilayer sheet production task, inputting them into the parameter quality relationship model, and calculating multiple groups of production parameters as reference production parameters to output to production personnel.

[0076] The specific implementation methods of the above steps are described in detail below:

[0077] The specific implementation method of step S10 is:

[0078] First, we obtain the production data of multiple sets of ABS multilayer sheets, including production parameters and product quality. These production data cover two categories: process parameters and raw material parameters, as well as three categories: appearance quality, mechanical performance quality, and thermal performance quality. For example, process parameters include factors such as extruder temperature, mold temperature, extrusion speed, and molding pressure. Raw material parameters include factors such as ABS resin model, toughening agent model and dosage, filler model and dosage, and colorant model and dosage. Product quality includes indicators such as gloss, color uniformity, surface flatness, embossing marks, impact resistance, tensile strength, bending modulus, hardness, heat deformation temperature, thermal expansion coefficient, and thermal conductivity.

[0079] Secondly, the obtained production data is normalized to unify the numerical range of each indicator, laying the foundation for subsequent mathematical modeling and analysis. The normalization method can be minimum-maximum normalization, that is, subtract the minimum value of each indicator value from the indicator, and then divide it by the range of the indicator (maximum value minus minimum value), and map all indicator values ​​to the [0,1] interval. This can eliminate the influence of the dimension and value differences between the indicators, making the subsequent modeling more stable and reliable.

[0080] In general, the purpose of this step is to establish a comprehensive and standardized ABS multi-layer sheet production data set, including three major indicators: process parameters, raw material parameters and product quality, and pre-process the data to lay the foundation for subsequent modeling and analysis.

[0081] The specific implementation method of step S20 is:

[0082] Based on the normalized production data obtained in step S10, a separate order function relationship between each production parameter (process parameter and raw material parameter) and each product quality (appearance quality, mechanical property quality, thermal property quality) is fitted as a basic relationship model. Here, the polynomial regression method is used to establish a basic mathematical model between production parameters and product quality.

[0083] Specifically, for any production parameter and a product quality indicator , the following basic relationship model can be established:

[0084] ;

[0085] in, , , ... is the coefficient to be fitted, and n is the order of the polynomial, which can be determined according to the actual fitting effect. Through the least squares method or other regression analysis methods, the basic relationship model between each pair of production parameters and product quality can be fitted.

[0086] These basic relationship models reflect the influence of a single production parameter on a single product quality indicator, laying the foundation for the subsequent establishment of a more complex comprehensive mathematical model. The purpose of this step is to explore the basic functional relationship between production parameters and product quality, which is conducive to a deep understanding of the mechanism of action of each factor.

[0087] The specific implementation method of step S30 is:

[0088] Based on the basic relationship model established in step S20, three sub-models are further constructed, corresponding to appearance quality, mechanical performance quality and thermal performance quality. Specifically:

[0089] 1. Appearance quality sub-model: The input of this model is process parameters and raw material parameters, and the output is the product appearance quality indicators, such as gloss, color uniformity, surface flatness, etc. A feedforward neural network structure can be used, and the input layer contains all process parameter and raw material parameter nodes. After feature extraction and combination of multiple hidden layers, the predicted output of appearance quality indicators is finally obtained.

[0090] 2. Mechanical performance quality sub-model: This model adopts the structure of encoding-attention-decoding. The input layer is the process parameters and raw material parameters. The bidirectional LSTM encoding layer extracts features, the attention layer calculates the importance weight of the features, and finally the unidirectional LSTM decoding layer predicts the product's impact resistance, tensile strength, bending modulus, hardness and other mechanical performance indicators.

[0091] 3. Thermal performance quality sub-model: This model also adopts an encoding-decoding structure, but the encoding layer uses a bidirectional LSTM to connect the FiLM adjustment layer in parallel. In addition to the unidirectional LSTM, the decoding layer also has an attention mechanism branch to better characterize the influence of process parameters and raw material parameters on thermal properties such as heat deformation temperature, thermal expansion coefficient, and thermal conductivity.

[0092] The three sub-models can learn the complex nonlinear mapping relationship between production parameters and corresponding product quality through training. The three sub-models are integrated to form a comprehensive production mathematical model, which can predict the overall product quality of ABS multi-layer sheets under given production parameter conditions.

[0093] The specific implementation of step S40 is:

[0094] For each set of normalized production data, the production parameters (process parameters and raw material parameters) are input into the production mathematical model established in step S30 to obtain the product quality predicted by the model, which is recorded as "estimated quality". Then, the difference between this estimated quality and the actual product quality recorded in the original production data is calculated, that is, the "distinction".

[0095] Specifically, the square error can be used as a measure of discrimination:

[0096] ;

[0097] Where n is the number of product quality indicators, represents the predicted value of the i-th quality indicator, Indicates the corresponding actual value.

[0098] Through this step, the degree of agreement between the model prediction and the actual measurement for each set of production data can be quantified. Data with less differentiation indicates that the model's prediction of these production situations is more accurate and can be regarded as "matching data", while data with greater differentiation indicates that the model's prediction of these situations is biased and can be regarded as "deviant data". This provides a basis for subsequent data augmentation and model optimization.

[0099] The specific implementation method of step S50 is:

[0100] According to the discrimination calculated in step S40, the production data is divided into two categories:

[0101] 1. Matching production data: data with a discrimination lower than the preset first discrimination threshold θ1. For this part of data, it can be considered that the model has predicted the product quality well and no further processing is required.

[0102] 2. Deviation production data: data with a discrimination greater than or equal to the first discrimination threshold θ1. This part of the data indicates that the model has a large error in predicting these production situations and needs further processing.

[0103] For the deviation production data, the production parameters are expanded by using data augmentation methods to increase the diversity of training samples. Specifically, Gaussian distribution or other random perturbation methods can be used to generate new data points around the production parameters of the deviation data. Then, these newly generated data are input into the production mathematical model established in step S30, and the inference quality is calculated to form the first augmented production data set.

[0104] The purpose of this step is to make up for the missing production conditions in the original data set by effectively augmenting the deviation data, so as to provide richer and more comprehensive samples for subsequent model training.

[0105] The specific implementation of step S60 is:

[0106] The first augmented production data set generated in step S50 is further screened and adjusted to further improve the quality of the data and the generalization performance of the model.

[0107] The specific steps are as follows:

[0108] 1. Calculate the discrimination of each group of data in the first augmented data set, using the same method as step S40.

[0109] 2. Set a second discrimination threshold θ2, and select data with a discrimination lower than θ2 as the second augmented data set. This part of data is merged with the matching production data determined in step S50 to form the final third augmented production data set.

[0110] 3. The second discrimination threshold θ2 should be smaller than the first threshold θ1, that is, θ2<θ1. This ensures that the data quality in the third augmented data set is high, including both the original matching data and the screened augmented data, covering various situations that may occur in the actual production process to the greatest extent.

[0111] The purpose of this step is to further improve the quality of the augmented data, ensure the reliability of the data through multiple rounds of screening, and lay the foundation for subsequent model training.

[0112] The specific implementation of step S70 is:

[0113] Using the third augmented production data set generated in step S60, a bidirectional neural network model is trained as a basic model of parameter-quality relationship. The model includes the following key components:

[0114] 1. Appearance quality branch: feedforward neural network structure, the input layer is process parameters and raw material parameters, features are extracted through multiple hidden layers, and appearance quality indicators are output.

[0115] 2. Mechanical performance quality branch: encoding-attention-decoding structure, the input layer is process parameters and raw material parameters, the features are extracted through bidirectional LSTM encoding, the attention layer calculates the feature weights, and finally the mechanical performance indicators are predicted through unidirectional LSTM decoding.

[0116] 3. Thermal performance quality branch: It also adopts the encoding-decoding structure. The encoding layer is a bidirectional LSTM parallel FiLM adjustment, and the decoding layer is a unidirectional LSTM parallel attention mechanism to better model thermal performance.

[0117] 4. Fusion branch: The outputs of the three branches are cascaded and passed through a multi-layer fully connected network to finally output a single score representing the overall quality.

[0118] 5. Constraint branch: Introduce a neural network structure based on rule constraints, introduce the feasible range of each parameter into the model, and ensure that the prediction results are within a reasonable range.

[0119] 6. Genetic algorithm module: embedded in the fusion branch, used to perform global optimization search on the weight parameters of the neural network to improve the generalization performance of the model.

[0120] This bidirectional neural network model can learn the complex nonlinear relationship between process parameters, raw material parameters and appearance quality, mechanical properties quality, and thermal properties quality, laying the foundation for subsequent parameter-quality relationship model training.

[0121] The specific implementation method of step S80 is:

[0122] Using all the matching production data generated in step S60 as a verification set, the parameter-quality relationship basic model trained in step S70 is optimized and updated to obtain the final parameter-quality relationship model.

[0123] The specific optimization methods are as follows:

[0124] 1. Input the matching production data into the base model and calculate the loss between the model’s predicted output and the actual quality on the validation set.

[0125] 2. Use the back-propagation algorithm to update the weight parameters of the neural network model according to the gradient of the loss function to reduce the prediction error.

[0126] 3. Repeat the above process until the performance indicators of the model on the validation set (such as mean square error) converge to a level that meets the requirements.

[0127] 4. The final parameter-quality relationship model can accurately predict the product quality under given production parameter conditions.

[0128] The purpose of this step is to use high-quality matching production data to further optimize and improve the basic model established in step S70, so that it better fits the actual production situation and improves the prediction accuracy and generalization ability of the model.

[0129] The specific implementation of step S90 is:

[0130] When a user obtains a new ABS multilayer sheet production task and needs to determine the appropriate production parameters, the following steps can be taken:

[0131] 1. Obtain the product quality requirements of the current production task, including the target indicators of appearance quality, mechanical performance quality and thermal performance quality.

[0132] 2. Input these quality target values ​​into the parameter-quality relationship model trained in step S80.

[0133] 3. Based on the quality target, the model will calculate multiple sets of production parameter formulas that meet the quality requirements through reverse deduction. These production parameters include process parameters (such as extruder temperature, mold temperature, extrusion speed, molding pressure, etc.) and raw material parameters (such as ABS resin model, toughening agent model and dosage, filler model and dosage, colorant model and dosage, etc.).

[0134] 4. Output these reference production parameters to production personnel as a basis for guiding production. Production personnel can make minor adjustments based on these reference parameters according to actual conditions to ensure that the final product quality meets the requirements.

[0135] This step realizes the reverse deduction from product quality goals to production parameters, providing production personnel with reliable parameter formula suggestions, greatly reducing the cost and time of trial and error, and improving production efficiency.

[0136] In general, the method for determining the production parameters of ABS multilayer sheets proposed in the present invention, by establishing a comprehensive production data set, constructing a mathematical model of parameter-quality relationship, and performing effective data augmentation and model optimization, finally achieves the reverse deduction of the production parameter formula that meets the requirements from the product quality target. This method makes full use of advanced technologies such as machine learning and neural networks, and has important application value in improving the production quality and efficiency of ABS multilayer sheets.

[0137] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for determining production parameters of an ABS multilayer board.

[0138] A third aspect of the present invention provides a system for determining production parameters of an ABS multilayer board, comprising the above-mentioned computer-readable storage medium.

[0139] In order to better understand and implement the present invention, the following provides an embodiment 1 of the specific steps of the computer-readable program running in a computer-readable medium or a computer-like device of the present invention: In this embodiment 1, for step S10, the purpose of this step is to establish a comprehensive and standardized ABS multi-layer sheet production data set. Specifically, this step includes the following mathematical processing:

[0140] First, obtain the production data of multiple sets of ABS multi-layer sheets, including production parameters and product quality. Group production data, each group of data includes Process parameters , Raw material parameters as well as Product quality indicators . It can be expressed in matrix form, namely:

[0141] ; ;

[0142] ;

[0143] Secondly, the obtained production data is normalized to make the numerical range of each indicator uniform. The minimum-maximum normalization can be used, that is, each indicator value is subtracted from the minimum value of the indicator, and then divided by the range of the indicator (maximum value minus minimum value), and all indicator values ​​are mapped to [0,1] Within the range.

[0144] For any indicator , and its normalization process is as follows:

[0145] ;

[0146] Through the above steps, the normalized production parameter matrix can be obtained , raw material parameter matrix and product quality matrix .

[0147] As for step S20, the purpose of this step is to explore the basic functional relationship between production parameters and product quality. Specifically, a separate order relationship model between each production parameter and each product quality index can be fitted, that is:

[0148] ;

[0149] in, Indicates Product quality indicators, Indicates Production parameters, are the coefficients to be fitted, is the order of the polynomial.

[0150] The least square method or other regression analysis methods can be used to fit the basic relationship model between each pair of production parameters and product quality. Through these basic relationship models, the influence of a single production parameter on a single product quality indicator can be explored.

[0151] For step S30, the purpose of this step is to construct a comprehensive production mathematical model. Based on the basic relationship model established in step S20, three sub-models can be further constructed, corresponding to appearance quality, mechanical performance quality and thermal performance quality.

[0152] The appearance quality sub-model can adopt a feedforward neural network structure, with the input layer being the process parameter and raw material parameter nodes. , after multiple hidden layers extract features, the output layer is the appearance quality index The loss function of this sub-model can be defined as: ;

[0153] The mechanical performance quality sub-model adopts the structure of encoding-attention-decoding. The input layer is also the process parameters and raw material parameters. The features are extracted through the bidirectional LSTM encoding layer. h =[ h ⃗ ; h ← ] , the attention layer calculates feature weights ,Finally, the mechanical performance index is predicted through the unidirectional LSTM decoding layer The loss function can be defined as: ;

[0154] The thermal performance quality sub-model also adopts the encoding-decoding structure. The encoding layer uses a bidirectional LSTM parallel FiLM adjustment layer. ,The decoding layer is a unidirectional LSTM parallel attention mechanism, outputting thermal performance indicators , the loss function is: ;

[0155] The outputs of the three sub-models are cascaded and passed through a multi-layer fully connected network with fusion branches to finally obtain the comprehensive quality score. , the loss function of the fusion branch is defined as: ;

[0156] in, is the actual comprehensive quality score.

[0157] During the training process, a joint optimization method can be used to minimize the weighted sum of the above four loss functions: ;

[0158] in, is the weight coefficient of each sub-model loss.

[0159] Through the above methods, a comprehensive production mathematical model can be constructed, which can predict the overall product quality under given production parameters.

[0160] As for step S40, the purpose of this step is to quantify the degree of agreement between the model prediction and the actual measurement for each set of production data. Specifically, the square error can be used as a measure of the degree of distinction: ;

[0161] in, Indicates The first The predicted value of a quality indicator, Indicates the corresponding actual value. Data with smaller discrimination indicates that the model's prediction of these production situations is more accurate and can be regarded as "matching data", while data with larger discrimination indicates that the model's prediction of these situations is biased and can be regarded as "deviation data".

[0162] As for step S50, the purpose of this step is to make up for the missing production conditions in the original data set by effectively augmenting the deviation data. Specifically, Gaussian distribution can be used for random perturbation to generate new data points around the production parameters of the deviation data:

[0163] ;

[0164] ;

[0165] in, and is the variance of Gaussian distribution, which can be adjusted according to the actual situation. These newly generated data are input into the production mathematical model constructed by S30, and their inference quality is calculated to form the first augmented production data set.

[0166] As for step S60, the purpose of this step is to further improve the quality of the augmented data. First, calculate the discrimination of each group of data in the first augmented data set , the same method as S40. Then set the second discrimination threshold , filter out the ones with a discrimination lower than The data is used as the second augmented data set. This part of the data is combined with the matching production data determined in S50 to form the final third augmented production data set. , ensuring that the data quality of the third augmented data set is high.

[0167] As for step S70, the purpose of this step is to train a bidirectional neural network model as a basic model of parameter-quality relationship. The model includes appearance quality branch, mechanical performance quality branch, thermal performance quality branch, fusion branch and constraint branch.

[0168] The appearance quality branch adopts a feedforward neural network structure, with the input layer being the process parameter and raw material parameter nodes, extracting features through multiple hidden layers, and the output layer being the appearance quality index. The loss function of this branch can be defined as: ;

[0169] The mechanical performance quality branch adopts an encoding-attention-decoding structure. The input layer is the process parameters and raw material parameters. The features are extracted through the bidirectional LSTM encoding layer. h =[ h ⃗ ; h ← ] , the attention layer calculates feature weights ,Finally, the mechanical performance indicators are predicted through the unidirectional LSTM decoding layer. The loss function is: ;

[0170] The thermal performance quality branch also adopts the encoding-decoding structure. The encoding layer uses a bidirectional LSTM parallel FiLM adjustment layer. ,The decoding layer is a unidirectional LSTM parallel attention mechanism, which outputs thermal performance indicators. The loss function is: ;

[0171] The outputs of the three branches are cascaded and passed through the multi-layer fully connected network of the fusion branch to finally obtain the comprehensive quality score , the loss function of the fusion branch is defined as: ;

[0172] in, is the actual comprehensive quality score.

[0173] During the training process, a joint optimization method can be used to minimize the weighted sum of the above four loss functions: ;

[0174] A genetic algorithm module is also embedded in the fusion branch to perform global optimization search on the weight parameters of the neural network and improve the generalization performance of the model. The constraint branch introduces a neural network structure based on rule constraints, introduces the feasible range of each parameter into the model, and ensures that the prediction results are within a reasonable range.

[0175] As for step S80, the purpose of this step is to use high-quality matching production data to further optimize and improve the basic model established in step S70. Specifically, the back propagation algorithm can be used to update the weight parameters of the neural network according to the prediction error of the model on the validation set: ;

[0176] in, are the parameters of the model, is the learning rate. Repeat the above process until the performance indicators (such as mean square error) of the model on the validation set converge to a level that meets the requirements, and the final parameter-quality relationship model is obtained.

[0177] As for step S90, the purpose of this step is to achieve the reverse derivation from product quality target to production parameters. q * =[ q 1 * , q 2 * ,..., q L * ] The parameter-quality relationship model obtained through training in step S80 is inputted.

[0178] According to the parameter-quality relationship model obtained by training in step S80, the idea of ​​back propagation can be used to obtain the target product quality. Starting from this, reverse calculation is performed on the production parameter formula that meets the quality requirements. Specifically, the following optimization problem can be defined: ;

[0179] in, is the product quality predicted by the model, and They are process parameters and raw material parameters respectively.

[0180] The optimization problem can be solved by gradient descent method, iteratively updating process parameters and raw material parameters until the predicted product quality is With the goal Achieve a sufficiently small error. The final result is and That is, the reference production parameters that meet the given product quality requirements.

[0181] Specifically, the principle of the present invention is: using the idea of ​​machine learning, a comprehensive production mathematical model covering all aspects of product performance is established, and through in-depth mining and effective use of existing production data, automatic optimization from target product quality to production parameters that meet requirements is achieved. The key to this method is:

[0182] 1. Make full use of the value of existing production data. Most manufacturing companies have accumulated a large amount of production data, but it is difficult for existing parameter optimization methods to fully tap the potential of these data. The present invention establishes a standardized data set covering process parameters, raw material parameters and comprehensive product quality indicators by comprehensively collecting and normalizing production data, laying the foundation for subsequent modeling and analysis.

[0183] 2. An integrated production mathematical model was constructed. Existing methods often only focus on a single product performance indicator, while in actual production, ABS plastic composite sheets need to meet the requirements of multiple indicators such as appearance, mechanical, and thermal. The present invention proposes an integrated model including an appearance quality sub-model, a mechanical performance quality sub-model, and a thermal performance quality sub-model, which can comprehensively predict product quality and is closer to actual production.

[0184] 3. Effective data augmentation and model optimization strategies are adopted. Production data often has the problem of uneven distribution, and it is difficult to achieve good generalization performance by directly training models with raw data. The present invention significantly improves the prediction accuracy and stability of the model by targeted augmentation of deviation data and optimization of the model using high-quality matching data.

[0185] 4. The reverse deduction from the target product quality to the production parameters is realized. Most existing methods focus on predicting product quality from production parameters, while in production practice, more attention is paid to how to determine the production parameters that meet the quality requirements. Through the idea of ​​reverse optimization, the present invention can efficiently calculate the reference production parameter formula that meets the given quality target, providing a reliable decision-making basis for production personnel.

[0186] In general, the core innovation of the present invention is to make full use of existing production data resources, construct an integrated comprehensive performance prediction model, and adopt effective data augmentation and model optimization strategies to achieve automatic optimization from target product quality to production parameters, thereby greatly improving the production quality and efficiency of ABS plastic composite panels.

[0187] In order to better understand the present invention, a specific embodiment 2 of the present invention is provided below, which is cited in a specific scenario: a certain ABS plastic composite sheet production enterprise, whose production line adopts an extrusion process to manufacture ABS composite sheets. In order to ensure the stability of product quality, the enterprise decides to adopt the ABS plastic composite sheet production parameter determination method proposed in the present invention to optimize the production parameters.

[0188] 1. Data collection and preprocessing

[0189] First, the company collected 50 sets of ABS plastic composite sheet production data in the past year, including process parameters and product quality indicators. Specifically, the process parameters include 8 parameters such as extruder barrel temperature (℃), extruder screw speed (rpm), mold temperature (℃), and discharge speed (m / min); the raw material parameters include ABS resin model, toughening agent model and dosage (phr), filler model and dosage (phr), colorant model and dosage (phr), etc., a total of 6 parameters. Product quality indicators include appearance quality (gloss, color uniformity, surface flatness), mechanical properties (impact resistance, tensile strength, bending modulus) and thermal properties (heat deformation temperature, thermal expansion coefficient, thermal conductivity), a total of 9 indicators.

[0190] The 50 sets of production data collected are shown in Table 1:

[0191] Table 1 Basic production data

[0192]

[0193] The collected production data is normalized to eliminate the dimensional differences between the indicators. Taking the extruder barrel temperature as an example, the normalization formula is: ;

[0194] in, is the extruder barrel temperature, normalized to All process parameters and raw material parameters are normalized in the same way to obtain the normalized production data set. and . Product quality indicators Also normalized, we get .

[0195] 2. Establishment of basic relationship model

[0196] Based on the normalized production data set, the individual order function relationship between each production parameter and each product quality indicator is first fitted as the basic relationship model.

[0197] To fit the extruder barrel temperature ( ) and glossiness( ) as an example, a third-order polynomial fitting is used: ;like Figure 2The figure shows the effect of extruder barrel temperature on the gloss of ABS plastic composite sheet. The horizontal axis represents the extruder barrel temperature (℃), and the vertical axis represents the normalized gloss. The curve shows a third-order polynomial fitting relationship, which reflects the nonlinear relationship between temperature and gloss. It can be seen from the figure that with the increase of temperature, the gloss first increases and then decreases, reaching a maximum value at a certain temperature point. This chart helps production personnel understand and control temperature parameters to obtain ideal product gloss.

[0198] Similarly, the basic relationship models between the other 7 process parameters and 6 raw material parameters and 9 product quality indicators were fitted, totaling 72 models. These basic relationship models can reflect the influence of a single production parameter on the quality of a single product, providing a basis for the establishment of subsequent comprehensive models.

[0199] 3. Construction of comprehensive production mathematical model

[0200] Based on the above basic relationship model, the present invention constructs a comprehensive production mathematical model including three sub-models to predict appearance quality, mechanical performance quality and thermal performance quality respectively.

[0201] (1) Appearance quality sub-model

[0202] The appearance quality sub-model adopts a feedforward neural network structure, with an input layer of 8 normalized process parameter nodes and 6 normalized raw material parameter nodes. After extracting features through two hidden layers, it finally outputs 3 appearance quality indicators (gloss, color uniformity, and surface flatness). Figure 3 As shown in the figure, 8 normalized process parameter nodes and 6 normalized raw material parameter nodes are used as parameters 1-14 of the input layer, the first hidden layer is set with 10 nodes H1_1~H1_10, the second hidden layer is set with 6 hidden nodes H2_1~H2_6, and the output layer is set with 3 nodes, representing glossiness, color uniformity and surface flatness respectively. The connecting lines between the nodes represent the data flow and weight relationship. This diagram clearly shows the complexity of the model and the connection between the layers, which helps to understand the process of appearance quality prediction.

[0203] The loss function of this sub-model is defined as: ;

[0204] in, The model predicts Appearance quality indicators, is the actual value. By minimizing the loss function, the appearance quality sub-model can be trained.

[0205] (2) Mechanical performance quality sub-model

[0206] The mechanical performance quality sub-model adopts the structure of encoding-attention-decoding. The input layer is also the normalized value of process parameters and raw material parameters, and the features are extracted through the bidirectional LSTM encoding layer. h =[ h ⃗ ; h ← ] , the attention layer calculates feature weights ,Finally, three mechanical performance indicators (impact resistance, tensile strength, and bending modulus) are predicted through the unidirectional LSTM decoding layer, such as Figure 4 As shown, Figure 4 The model structure used to predict the mechanical properties of ABS plastic composite sheets is described. It contains an input layer (14 parameters), a bidirectional LSTM encoding layer, an attention layer, an LSTM decoding layer, and an output layer (impact resistance, tensile strength, flexural modulus). This structure can effectively capture the temporal relationship and mutual influence between parameters, thereby more accurately predicting mechanical properties. The connecting lines in the figure show the data flow path, which helps to understand how the model works.

[0207] The loss function is defined as: ;

[0208] in, The model predicts Mechanical performance indicators, is the actual value.

[0209] (3) Thermal performance quality sub-model

[0210] The thermal performance quality sub-model also adopts the encoding-decoding structure. The encoding layer uses a bidirectional LSTM parallel FiLM adjustment layer. ,The decoding layer is a unidirectional LSTM parallel attention mechanism, which outputs three thermal performance indicators (thermal deformation temperature, thermal expansion coefficient, and thermal conductivity). Figure 5 As shown, Figure 5 The model structure for predicting the thermal properties of ABS plastic composite sheets is shown. It contains an input layer, a bidirectional LSTM encoding layer, a FiLM adjustment layer, an LSTM decoding layer, a parallel attention mechanism, and an output layer (heat deformation temperature, thermal expansion coefficient, thermal conductivity). This complex structure can handle nonlinear relationships and parameter interactions in thermal performance prediction. The dotted line in the figure indicates the role of the attention mechanism, which helps the model focus on the most relevant features.

[0211] The loss function is: ;

[0212] in, The model predicts Thermal performance index, is the actual value.

[0213] The outputs of the three sub-models are cascaded and passed through a multi-layer fully connected network with fusion branches to finally obtain a single score representing the overall quality. , the loss function of the fusion branch is defined as: ;

[0214] in, is the actual comprehensive quality score.

[0215] During the model training process, a joint optimization method is used to minimize the weighted sum of the above four loss functions: ;

[0216] The comprehensive production mathematical model constructed in this way can accurately predict the overall product quality of ABS plastic composite sheets under given production parameter conditions.

[0217] 4. Data augmentation and model optimization

[0218] In order to further improve the generalization ability of the model, the present invention adopts a two-step data augmentation and model optimization strategy:

[0219] (1) Data augmentation

[0220] First, calculate the discrimination of each set of production data , used to measure the difference between the model prediction results and the actual values: ;

[0221] The degree of distinction is lower than the first preset threshold The data is classified as "matching production data", and the rest is "deviation production data". For the deviation production data, Gaussian distribution is used for random perturbation to generate new data samples:

[0222] ;

[0223] ;

[0224] These newly generated data are input into the aforementioned comprehensive production mathematical model, and their inference quality is calculated to form the first augmented production data set.

[0225] (2) Second step optimization

[0226] All matching production data are used as the validation set to optimize and update the aforementioned comprehensive production mathematical model. Specifically, the back propagation algorithm is used to iteratively update the weight parameters of the neural network based on the prediction error of the model on the validation set: ;

[0227] Repeat the above process until the performance indicators of the model on the validation set (such as mean square error) converge to a level that meets the requirements, and the final parameter-quality relationship model is obtained.

[0228] 5. Production parameter optimization

[0229] In actual production, when a company receives a new order for ABS plastic composite sheets, it needs to determine the best production parameters to meet product quality requirements. The following steps can be taken:

[0230] (1) Obtain product quality objectives for new orders, including specific requirements for appearance quality, mechanical performance quality, and thermal performance quality.

[0231] (2) These quality target values ​​are input into the optimized parameter-quality relationship model mentioned above, and multiple sets of production parameter formulas that meet the quality requirements are calculated through reverse optimization, including process parameters and raw material parameters.

[0232] The specific optimization process is as follows:

[0233] Assume the target product quality is q * =[ q 1 * , q 2 * ,..., q 9 * ] , the optimization problem can be defined as:

[0234] ;

[0235] in, The product quality predicted by the model. The process parameters are updated iteratively through the gradient descent method. and raw material parameters , until the prediction quality With the goal The error is small enough. Figure 6 The figure shows the optimization process of the production parameters of ABS plastic composite sheets. The horizontal axis represents the parameter space and the vertical axis represents the quality index. The blue curve represents the optimization path, the red dot represents the initial parameters, and the green dot represents the optimal parameters. The fluctuations in the figure reflect the exploration and convergence during the optimization process.

[0236] (3) Output the optimized multiple sets of production parameter formulas that meet the quality requirements to the production personnel as a basis for guiding production. The production personnel can make appropriate adjustments based on these reference parameters according to the actual situation to ensure that the final product quality meets the standards.

[0237] For example, for a new order, the product quality target requirement is: gloss , Color uniformity , surface flatness , impact resistance , tensile strength , flexural modulus , Heat Deformation Temperature , thermal expansion coefficient , thermal conductivity .

[0238] These target values ​​are input into the optimized parameter-quality relationship model. After reverse optimization calculation, the following production parameter formula that meets the quality requirements is obtained: extruder barrel temperature: 225°C; extruder screw speed: 95rpm; mold temperature: 190°C; discharge speed: 6m / min; ABS resin model: ABS-3400; toughening agent model: CPE-250, dosage 6phr; filler model: TiO2-150, dosage 23phr; colorant model: MB-150, dosage 0.7phr; production personnel can make fine adjustments based on these reference parameters according to actual conditions to ensure that the final ABS plastic composite sheet produced meets the customer's comprehensive quality requirements.

[0239] By adopting the method for determining the production parameters of ABS plastic composite sheets proposed in the present invention, the enterprise solved the problem of difficulty in optimizing the parameters in the original production and greatly improved the product quality and production efficiency, which is specifically reflected as follows:

[0240] 1. A comprehensive production data set was established, covering process parameters, raw material parameters, and product quality indicators in terms of appearance, mechanics, thermals, etc., laying the foundation for subsequent modeling and analysis. By normalizing the data, the dimensional differences between the indicators were eliminated, improving the applicability of the data.

[0241] 2. An integrated comprehensive production mathematical model was constructed, including appearance quality sub-model, mechanical performance quality sub-model and thermal performance quality sub-model. This model can accurately predict the overall product quality of ABS plastic composite sheets under given production parameters, which is closer to actual production needs than the existing single performance prediction model.

[0242] 3. The strategy of data augmentation and model optimization was adopted to effectively improve the generalization performance of the model. The insufficiency of the original data was compensated by randomly perturbing and augmenting the deviation data; the model was optimized and updated using high-quality matching data, further improving the prediction accuracy.

[0243] 4. It realizes the reverse optimization from the target product quality to the production parameters that meet the requirements, and provides a reliable reference parameter formula for production personnel. Production personnel can make appropriate adjustments based on the actual situation, which greatly shortens the production commissioning cycle and significantly improves production efficiency.

[0244] Taking a new order as an example, the product quality goals required by the customer include 9 indicators such as appearance, mechanics and thermals. These target values ​​are input into the optimized parameter-quality relationship model, and through reverse optimization calculation, the production parameter formula that meets the quality requirements is obtained, including extruder temperature, screw speed, mold temperature, discharge speed, and raw material parameters such as ABS resin, toughening agent, filler, and colorant. The production staff made minor adjustments to these reference parameters according to the actual situation, and the ABS plastic composite sheet finally produced successfully met the customer's comprehensive quality requirements.

[0245] For this enterprise, the main advantages of adopting the method of the present invention are:

[0246] 1. Product quality stability and consistency are guaranteed. In the past, due to empirical parameter adjustment, product quality was easily affected by the subjective judgment of production personnel, and it was difficult to ensure stability. However, the method of the present invention establishes a scientific parameter-quality mapping relationship model, which can accurately derive production parameters that meet the requirements according to quality goals, greatly improving the stability and consistency of product quality.

[0247] 2. Significantly improve production efficiency. Traditional empirical parameter adjustment requires repeated trial and error, which is time-consuming and labor-intensive. The method of the present invention can quickly calculate the production parameter formula that meets the quality target, shorten the production debugging cycle, greatly improve production efficiency, and bring obvious economic benefits to the enterprise.

[0248] 3. Improved intelligent production process. The present invention realizes automatic optimization from product quality objectives to production parameters, reduces the workload of production personnel, and improves the intelligent production process. With the further development of artificial intelligence technology in the future, this method can be further optimized to achieve a higher degree of automated production.

[0249] In summary, the method for determining the production parameters of ABS plastic composite sheets proposed in the present invention fully utilizes the advantages of mathematical modeling and machine learning, and has made significant progress in improving product quality stability, production efficiency and intelligent manufacturing level, which is of great significance for improving the production and manufacturing level of ABS plastic composite sheets and even other composite materials.

[0250] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for determining production parameters of ABS multilayer sheets, characterized in that: The following steps are involved: S10, obtaining production data of multiple groups of ABS multilayer sheets, including production parameters and product quality, and normalizing the production data to obtain normalized production data, wherein the product quality includes three categories: appearance quality, mechanical performance quality, and thermal performance quality; the production parameters include two categories: process parameters and raw material parameters; wherein each category of product quality and each category of production parameters include multiple data; S20, fitting a separate order relationship between each production parameter and each product quality as a basic relationship according to the normalized production data; S30, establishing a mathematical model according to the basic relationship, wherein the mathematical model includes three sub-models, namely, an appearance quality sub-model, a mechanical performance quality sub-model, and a thermal performance quality sub-model, and fitting the normalized production data to obtain a production mathematical model; S40, for each set of normalized production data, inputting the production parameters into the production mathematical model to obtain product quality, recorded as estimated quality, and calculating the difference between the estimated quality and the original production quality; S50, recording the normalized production data whose discrimination is lower than a preset first discrimination threshold as matching production data, recording the normalized production data whose discrimination is higher than or equal to the preset first discrimination threshold as deviation production data, augmenting the production parameters of the deviation production data by using a data augmentation method, and calculating the estimated quality by using the production mathematical model to form a first augmented production data set; S60, selecting production data with a distinction lower than a second distinction threshold in the augmented production data set as a second augmented production data set, and merging the second augmented production data set with all matching production data to record the result as a third augmented production data set; S70, using the third augmented production data set, training a bidirectional neural network model to obtain a parameter-quality relationship basic model, wherein the bidirectional neural network model includes an appearance quality branch, a mechanical property quality branch, a thermal property quality branch, a fusion branch, and a constraint branch; S80, using all matching production data as a verification set, optimizing and updating the parameter-quality relationship basic model to obtain a parameter-quality relationship model; S90, obtaining product quality requirements of the current ABS multilayer sheet production task, inputting them into the parameter quality relationship model, and calculating multiple groups of production parameters as reference production parameters to output to production personnel.

2. The method for determining production parameters of ABS multilayer sheet according to claim 1, characterized in that: The appearance quality, mechanical performance quality and thermal performance quality are all obtained through experiments, and the specific methods of obtaining them include: Appearance quality is tested by visual inspection, gloss meter and roughness meter; The mechanical performance quality is tested by impact test, tensile test, three-point bending test and hardness tester; The thermal performance quality is tested using a heat deformation temperature tester or a thermal expansion coefficient tester or a thermal conductivity tester as specialized equipment.

3. The method for determining production parameters of ABS multilayer sheet according to claim 1, characterized in that: The appearance quality at least or all includes: glossiness, color uniformity, surface flatness, embossing marks; The mechanical properties include at least or all of: impact resistance, tensile strength, flexural modulus, hardness; The thermal performance qualities at least or all include: thermal deformation temperature, thermal expansion coefficient, and thermal conductivity.

4. The method for determining production parameters of ABS multilayer sheet according to claim 1, characterized in that: The process parameters at least or all include: extruder head temperature, mold temperature, extrusion speed, and molding parameters.

5. The method for determining production parameters of ABS multilayer sheet according to claim 1, characterized in that: The raw material parameters at least or all include: ABS resin model, toughening agent model and dosage, filler model and dosage, colorant model and dosage.

6. The method for determining production parameters of ABS multilayer sheet according to claim 1, characterized in that: The fusion branch is embedded with a genetic algorithm module for performing global optimization search on the weight parameters of the neural network; the initial population of the genetic algorithm module is a set of weight parameter sets obtained from each step of training of the appearance quality branch, the mechanical performance quality branch, and the thermal performance quality branch.

7. The method for determining production parameters of ABS multilayer sheet according to claim 6, characterized in that: The fitness function of the genetic algorithm module is the mean square error between the predicted value and the true value, and the elimination function is natural selection, and individuals with high fitness are retained in each generation.

8. The method for determining production parameters of ABS multilayer sheet according to claim 7, characterized in that: The constraint branch is specifically a neural network structure based on rule constraints, which is used to introduce the feasible range of each parameter into the model; the genetic algorithm module is used to feed back the optimized weight parameter set to each branch as the weight input for the next step of training.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for determining production parameters of an ABS multilayer sheet material according to any one of claims 1 to 8.

10. A system for determining production parameters of ABS multi-layer board materials, comprising the computer-readable storage medium according to claim 9.

Images