Foaming injection molding process optimization and product performance prediction method based on artificial intelligence

Abstract

The invention provides a foaming injection molding process optimization and product performance prediction method based on artificial intelligence, and aims to realize process optimization of product appearance and performance in a foaming injection molding process through an artificial intelligence algorithm. According to the method, light transmittance is used for indirectly representing the mechanical performance of a product, and the method comprises the following steps: performing mold testing analysis through an orthogonal experiment to obtain a process parameter-microstructure parameter-performance parameter data set; an LSTM model and an XGBoost model are adopted to construct correlation models of the technological parameters and the microstructures and correlation models of the microstructures and the product performance respectively; an optimization algorithm of LSTM-EGA and an optimization algorithm of XGboost-SEGA are constructed respectively, and the optimization algorithms are used for reversely deducing technological parameters according to the target parameters; the light transmittance of the product is adjusted according to the requirement for the mechanical performance of the product, and the optimal forming process is reversely solved through a model algorithm; and performing on-machine mold testing, adjusting a model structure based on a test result, and re-recommending a process. According to the prediction method, high equipment investment is avoided, and the test process is greatly simplified, so that the test cost can be remarkably reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of injection molding, and particularly relates to a foaming injection molding process optimization and performance prediction method based on artificial intelligence. BACKGROUND

[0002] In the field of modern polymer material molding, foaming injection molding process has become one of the core technologies for manufacturing automobile lightweight parts and high-performance packaging materials due to its unique microcellular structure advantage. The synergistic optimization of the apparent quality (such as surface roughness, weld mark, dimensional stability) and mechanical properties (such as tensile strength, impact toughness) of the product directly determines its market competitiveness and service reliability in industrial applications.

[0003] However, the multi-physical field coupling characteristics (such as melt flow-foaming agent diffusion-cooling solidification kinetics coupling) of the process result in a highly nonlinear relationship between process parameters and product performance. The traditional trial method faces three technical bottlenecks: first, the interaction mechanism of key process parameters (melt temperature, injection rate, holding pressure, etc.) is not clear, making it difficult to establish an accurate parameter model; second, the experience-driven optimization strategy has defects such as long trial period (usually 15-30 iterations) and low data utilization rate (less than 30% effective information collection rate); third, existing empirical formulas (such as Arrhenius type viscosity model) are difficult to represent the dynamic evolution process of the foaming system, making it difficult to analyze the quantitative relationship between process parameters and microstructure (cell diameter distribution, cell density, etc.), ultimately leading to the inability to effectively evaluate whether the product service performance (such as mechanical strength) meets the design and use requirements in the design and production stages, ultimately negatively affecting product engineering efficiency.

[0004] In recent years, data-driven methods have provided new ideas for the above challenges. Deep neural networks (DNN) can establish a nonlinear mapping from the process parameter space to the performance space due to their strong feature extraction capability; Bayesian optimization and other intelligent algorithms can achieve global optimization of high-dimensional parameter space. However, existing research still has inherent limitations: ① the cross-scale correlation mechanism of process-structure-performance is not clear; ② the synergistic verification system of experimental data and numerical simulation is not perfect; ③ there is a lack of interpretable proxy models to guide process adjustment.

[0005] For the characterization of product performance, current research has shown that product microcellular diameter, product skin thickness, and product mechanical performance and light transmission performance are closely related. Within a certain range, the larger the product cell diameter, the lower the light transmission rate and the lower the mechanical performance. Therefore, it is reasonable to build a correlation model between product light transmission rate and mechanical performance using product microstructure features as intermediate variables.

[0006] To solve the above problems, the present application proposes a foaming injection molding process optimization and performance prediction method based on multi-modal data fusion of artificial intelligence. By constructing a three-dimensional correlation model of process parameters-microstructure characterization-macroscopic performance, a quantifiable process parameter feature map and an interpretable prediction criterion are established. This method not only realizes the rapid optimization of process parameter combination (predicting more than 50% reduction in trial molding period), but also reveals the mechanism of process parameters on product performance through the coupling relationship of cell structure-transmittance-mechanical properties, providing theoretical support and engineering practice guidance for intelligent manufacturing of high-performance foam injection molded products. SUMMARY

[0007] To solve the above problems, the present application innovatively constructs a transmittance-mechanical property multi-scale correlation model under the premise of complex and lagging product mechanical strength measurement, and proposes a process optimization method based on deep feature decoupling. Based on the theoretical framework of light scattering theory and microstructure correlation, the method establishes a quantitative mapping relationship between transmittance and cell morphology parameters (cell diameter, cell density, skin thickness), indirectly representing the mechanical performance degradation mechanism of the product. Further, by integrating deep learning models and optimization algorithms, cross-modal feature extraction from molding process parameters to product transmittance features is achieved, ultimately forming a process optimization decision system with interpretability and generalization ability. The method includes the following steps:

[0008] Step S10: Obtain the material grade of the product and its physical, chemical and mechanical properties, as well as the mold structure parameters (such as gate location, cavity number and cooling water layout) and other key information;

[0009] Step S20: Perform injection molding experiments based on material properties and mold structure to determine the range of key process parameters, including injection speed, holding pressure, melt temperature, holding time, barrel temperature, and foaming agent content, to construct the process window of injection molding parameters;

[0010] Step S30: Set up an orthogonal experiment table within the process window to analyze the product trial molding, obtain the molded product, and perform microstructure analysis and performance analysis to construct a process parameter-microstructure parameter-service performance parameter dataset;

[0011] Step S40: Construct a three-dimensional data matrix of process parameters, microstructure parameters and performance parameters, and use machine learning models (including LSTM, XGBoost, etc.) to mine the nonlinear relationships between parameters, and establish a "process-structure-performance" mapping model;

[0012] Step S50: Construct a genetic algorithm driven reverse optimization framework, taking the target performance parameters as the constraint boundary, and selecting the optimal process that meets the target data through multi-objective coding and Pareto frontier search;

[0013] Step S60: Actual machine test is carried out by using the recommended process, microstructure data and product performance data are extracted, and the accuracy of the method is verified.

[0014] Step S70: Adjust the model structure based on the test results and re-recommend the process.

[0015] Preferably, the present study adopts single factor experimental design method, taking injection molding process parameters (foaming agent mass fraction, injection rate, mold cavity temperature, barrel setting temperature, holding pressure and holding duration) as independent variables, and systematically explores the correlation mechanism between them and product surface defects (such as sink marks, weld marks, warpage deformation, etc.). By quantitatively analyzing the influence of each process parameter on the appearance quality evaluation index of the product (such as surface roughness, defect area ratio), the product forming process window that meets the constraint of no surface defects is established.

[0016] Preferably, the determination step of the product forming process window includes the determination of injection speed, mold temperature and barrel temperature, holding pressure, and holding time, wherein the melt temperature and mold temperature are determined according to the recommended temperature of the material supplier.

[0017] Preferably, the process parameters to be optimized, such as injection speed, melt temperature, mold temperature, holding pressure, holding time, and foaming agent content, are used as factors, and the effective value range of the optimized parameters obtained by injection molding experiment is divided into 5 levels. A 6-factor 5-level orthogonal experiment table is used to determine 25 process parameter combinations for calculation, and a multi-batch systematic test is used to obtain a multi-dimensional data set. The result list contains the following data:

[0018] Process parameter set: injection speed, mold temperature, barrel temperature, holding pressure, holding time, and foaming agent content;

[0019] Microstructure characteristics: cell diameter, cell density, and surface thickness, which are quantitatively characterized by microscope combined with image analysis method;

[0020] Optical performance index: transmittance, which is determined by color spectrum haze meter after measuring light flux and obtained by formula calculation;

[0021] The calculation formula of transmittance is as follows:

[0022]

[0023] Where G represents transmittance, G t represents incident light flux, G d represents total transmitted light flux through the sample product.

[0024] Preferably, the orthogonal test structure data obtained in step S30 is divided into two data sets, one is a process parameter-product microstructure data set, which records in detail the microstructure characteristics of the product under different process parameter conditions; the other is a product microstructure data-product performance data set, which focuses on the correlation between the product microstructure and the final performance of the product.

[0025] Subsequently, further subdivision processing is carried out for the two data sets, which are divided into training set, validation set and test set. The training set is used for the preliminary training of the model, and the proportion of 70% makes the model learn the potential law in the data; the validation set is used for adjusting the hyperparameters of the model, and the proportion of 15% is used for optimizing the performance of the model; the test set accounts for 15%, which is used to comprehensively evaluate the generalization ability and prediction accuracy of the model after the model training is completed.

[0026] Preferably, the sample feature matrix in the process parameter-product microstructure data set is as follows:

[0027]

[0028] x i1 , x i2 , x i3 , x i4 , x i5 , x i6 respectively represent the material temperature, mold temperature, injection speed, holding pressure, holding time and content of foaming agent in the orthogonal experimental result data.

[0029] The target variable matrix is as follows:

[0030]

[0031] Among them, y j1 , y j2 , y j3 respectively represent the cell diameter, cell density and skin thickness in the orthogonal experimental result data table.

[0032] Among them, the sample feature matrix in the product microstructure data-product performance data set is as follows:

[0033]

[0034] x i1 , x i2 , x i3 , respectively represent the cell diameter, cell density and skin thickness in the orthogonal experimental result data.

[0035] The target variable matrix is as follows:

[0036]

[0037] wherein y j1 represents the light transmittance in the orthogonal experiment result data table.

[0038] Preferably, a double deep learning architecture is constructed to realize cross-scale correlation modeling of process parameters-microstructure-optical performance, specifically including:

[0039] LSTM-feature extraction module: taking the time-series process parameter set (injection rate, mold temperature, blowing agent concentration, etc.) as input, capturing the dynamic evolution characteristics of the process parameters through the gated recurrent unit, and outputting a high-dimensional microstructure feature vector Z = {cell diameter, cell density, skin thickness};

[0040] XGBoost-performance prediction module: taking the microstructure feature Z as input, using the gradient boosting decision tree ensemble learning framework to establish a nonlinear mapping relationship, and realizing quantitative prediction of light transmittance;

[0041] Cross-modal fusion mechanism: through attention weight distribution strategy (Attention Mechanism) to optimize feature transmission path, and strengthen the conduction effect analysis of process parameter disturbance on optical performance.

[0042] Preferably, the steps of constructing the LSTM model include: setting network hyperparameters (learning rate, number of hidden layer units, time step, dropout rate, and batch size), initializing the weight matrix (such as input gate, forget gate, and output gate) and bias term of the gated recurrent unit (GRU), constructing a multi-layer LSTM network architecture, and optimizing the time-series feature extraction capability through the backpropagation algorithm.

[0043] Preferably, the steps of constructing the XGboost model include: setting model parameters, including learning rate, maximum depth of numbers, subsampling ratio, column sampling ratio, etc., initializing the model and constructing decision trees.

[0044] Preferably, NS-GA algorithm is used for parameter optimization, and the steps of constructing NS-GA algorithm include: setting population size, evolution times, crossover probability, and mutation probability; wherein the expressions of the objective functions are respectively:

[0045]

[0046] wherein y is the ideal product light transmittance set by the user, is the product light transmittance predicted by the model according to the product microstructure parameters within the product forming process window range;

[0047]

[0048] wherein y iis the product microstructure parameter recommended based on the ideal light transmittance, is the product microstructure parameter predicted according to the process parameters within the product forming process window.

[0049] Preferably, the mechanical properties of the product are characterized by the light transmittance characteristics of the product, specifically, based on the physical properties of the product and optical principles, there is an inherent correlation between the changes in the light transmittance characteristics of the product and the mechanical properties of the product. Through experimental research and theoretical analysis, the light transmittance characteristic change rule corresponding to the product in different mechanical property states is determined. The mechanical properties of the DOE data are tested, and the light transmittance data of the product under different mechanical property indexes (such as tensile strength, bending strength, impact strength, etc.) are recorded, and a correlation database between the mechanical properties of the product and the light transmittance characteristics is established. Then, according to the mechanical property requirements of the product, the required light transmittance requirements of the product are determined, and then the process parameters are optimized through an artificial intelligence model.

[0050] Compared with the prior art, the present application has the following technical advantages: the present application provides a foaming injection molding process optimization and performance prediction method based on artificial intelligence, which adopts single factor experiment to construct a forming process window to meet the appearance quality requirements of the product, and then performs mold analysis through DOE experiment, which can accurately combine deep learning network and optimization algorithm to optimize the process parameters, and effectively improve the product performance index. The training data range of the prediction model is determined by the process window, which solves the problem of requiring a large amount of training data for the model, reduces the testing of invalid parameters in the injection molding experiment, improves the effectiveness of the training data, improves the reliability of the prediction results, and improves the interpretability of the results. Finally, it is worth emphasizing that the method of characterizing the mechanical properties of the product according to the light transmittance used in the present method is an innovative performance characterization means. In the traditional performance test process, a large number of complex and expensive equipment and processes are often required to obtain relevant data about the mechanical properties of the product, which undoubtedly leads to high testing costs. The present method uses the inherent correlation between the light transmittance and the mechanical properties of the product, and by detecting and analyzing the light transmittance characteristics of the product, the mechanical properties of the product can be indirectly and accurately inferred. The present method not only avoids many cumbersome steps and high equipment investment in traditional testing methods, but also greatly simplifies the testing process, thereby significantly reducing the testing cost, saving a large amount of resources and funds for enterprises in product quality control and research and development, and has important application value and practical significance. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly express the implementation mode of the technical solutions described in the present application, the drawings are described in detail.

[0052] Figure 1 is a flow chart of a foaming injection molding process optimization and performance prediction method based on artificial intelligence provided by an embodiment of the present application;

[0053] Figure 2 is a viscosity curve diagram of an embodiment of the present application;

[0054] Figure 3 is a schematic diagram of a mixed model provided by an embodiment of the present application;

[0055] Figure 4 is a curve diagram of the relationship between light transmittance and product mechanical strength provided by an embodiment of the present application. DETAILED DESCRIPTION

[0056] The present application will be described below in conjunction with specific embodiments. The purpose of the specific embodiments is to illustrate the present application, not to limit the present application. The terms used herein have the meanings commonly used in the art. Therefore, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present application belongs. The raw materials, instruments and equipment used in the present application can be purchased from the market or can be prepared by existing methods.

[0057] As Figure 1 , the present application proposes a foaming injection molding process optimization and performance prediction method based on artificial intelligence. By using product light transmittance to represent product mechanical performance, a light transmittance-mechanical performance multi-scale correlation model is constructed, and a process optimization method based on deep feature decoupling is proposed. By integrating deep learning models and optimization algorithms, cross-modal feature extraction from process parameter space to light transmittance features is achieved, forming a process optimization decision system with both interpretability and generalization ability. The specific embodiments are as follows:

[0058] Step S10: Obtain the material grade of the product and its physical, chemical and mechanical properties, as well as the mold structure parameters (such as gate location, cavity number and cooling water layout) and other key information;

[0059] Step S20: Perform injection molding experiments according to the material properties and mold structure to determine the key process parameter range, including injection speed, holding pressure, melt temperature, holding time, barrel temperature, foaming agent content, etc., and construct the process window of injection molding parameters;

[0060] Step S30: Set up an orthogonal experiment table within the process window to analyze the product trial molding, obtain the molded product and perform microstructure analysis and performance analysis to construct the process parameter-microstructure parameter-service performance parameter dataset;

[0061] Step S40: Construct a three-dimensional data matrix of process parameters, microstructure parameters and performance parameters, use machine learning models (including LSTM, XGBoost, etc.) to mine the nonlinear relationship between parameters, and establish a "process-structure-performance" mapping model;

[0062] Step S50: Construct a genetic algorithm driven reverse optimization framework to target performance parameters as constraint boundaries, and screen the optimal process that meets the target data through multi-objective coding and Pareto frontier search, etc.

[0063] Step S60: Perform actual on-machine test molding test using the recommended process, extract microstructure data and product performance data, and verify the accuracy of the method;

[0064] Step S70: Adjust the model structure based on the test results and re-recommend the process.

[0065] In this example, PC material (920108-M210004_1) injection molded products are taken as the object, and the product structure is as shown in Figure 2

[0066] I. Construction of molding process window

[0067] The experimental machine adopts Changfei ZE1200 full-electric injection molding machine, and the mold temperature, barrel temperature, injection speed, holding time, holding pressure, and foaming agent content are determined. Among them, the mold temperature and barrel temperature are determined according to the recommended temperature of the material manufacturer. In this example, the material temperature range is set to [270, 285], and the mold temperature range is [65, 85].

[0068] 1.1 Injection speed determination. In injection molding, PC material melt is a non-Newtonian fluid, and its viscosity changes with shear rate. Injection speed affects shear rate, and thus affects viscosity. Therefore, by determining the viscosity steady state interval to select the appropriate injection speed, i.e. single factor test of injection speed and record the peak pressure of screw and filling time, draw the relationship curve of shear rate and effective viscosity, and obtain the viscosity steady state interval as the injection speed process interval.

[0069] In this example, the injection speed results are shown in Table 1. According to the relationship curve drawn, the interval where the slope change rate of the adjacent two points of the curve is less than 0.1 is set as the injection speed recommended process interval, so the injection speed interval is determined as [51.6, 76.1];

[0070] Table 1 Single factor test data of injection speed

[0071]

[0072] 1.2 Holding time determination: The product weight curve reflects the change of product weight during injection molding. When the product weight curve tends to be flat and the weight basically no longer increases, it indicates that the holding pressure is sufficient to compensate for shrinkage. At this time, the corresponding holding time is the appropriate holding time. By analyzing the product weight curve, the holding time is determined. The holding time process window obtained in this experiment is [3, 4]. ​

[0073] 1.3 Other parameter determination: single factor experiment was carried out, only one process parameter value was changed each time, and the value was gradually increased from a lower value, a series of injection molding experiments were carried out. After each experiment, the appearance quality of the product was evaluated according to the evaluation standard of product appearance quality, and the appearance indexes such as color uniformity, surface roughness, presence or absence of flash or crack were comprehensively evaluated. Based on these evaluation results, the value range of the corresponding parameters that can stably obtain better product appearance quality was found out, and the recommended parameter interval of each related process parameter was determined. The holding pressure interval obtained in this experiment is [45, 60], and the foaming agent content is [2%, 4%].

[0074] II. Data set acquisition

[0075] An orthogonal experiment table was set up in the process window, and trial molding analysis was carried out; the orthogonal experiment was an experiment table 2 of 6 factors and 5 levels, in this example, the division of the orthogonal experiment table was carried out by using python library, 25 groups of experimental data were obtained, and the taguchi type orthogonal table is shown in the following table:

[0076] Table 2 Orthogonal experiment of 6 factors and 5 levels

[0077]

[0078] The 25 groups of DOE experimental data were used for injection molding experiment, 5 trial moldings were carried out for each experiment, and the average value was taken as the experimental result. The results required to be extracted in this example include microstructure parameters (cell diameter, cell density, surface layer thickness) and product performance parameters (transmittance).

[0079] 2.1 Microstructure data acquisition

[0080] (1) Cell diameter and density characterization

[0081] Field emission scanning electron microscope (FE-SEM) combined with image binarization analysis technology was used, and the specific process was as follows:

[0082] Sample preparation: after brittle fracture of the sample in liquid nitrogen, gold spraying treatment (accelerating voltage 5kV, current 10mA, film thickness 10nm) was carried out to enhance the electrical conductivity;

[0083] Microimaging: SEM images (resolution 512x512 pixels) of the cross section of the sample were obtained under the conditions of accelerating voltage 15kV and working distance 5mm;

[0084] Image processing: based on Image Pro Plus software, the cell area was extracted by gray threshold segmentation (Otsu algorithm), and the cell diameter distribution and cell density (n=N / A, N is the total number of cells, A is the statistical area) were calculated;

[0085] Error control: 5 fields of each group of samples were randomly selected for repeated measurement, and the average value was taken as the final result (confidence interval 95%).

[0086] (2) Skin thickness measurement

[0087] Image Pro Plus software was used to analyze and calculate the micro-test photos of the samples, and the skin thickness data (i.e. non-cellular skin layer) was obtained.

[0088] 2.2 Product performance parameter test (transmittance)

[0089] According to the standard of ASTM D1003-13, the optical performance characterization was completed by using WGT-S transmittance and haze tester, and the specific steps were as follows:

[0090] Sample preparation: cut the sample into 50mm x 50mm x 3mm size, and wipe the surface with alcohol to remove dirt;

[0091] Instrument calibration: use standard transmittance plate (transmittance ≥ 99%) to correct the zero point of spectrophotometer (wavelength range 380-780nm);

[0092] Transmittance measurement: place the sample at the entrance of the integrating sphere, record the full waveband transmittance light intensity I t And the incident light intensity I0, calculate the transmittance according to the formula: (λ = 550nm center wavelength)

[0093] Data correction: deduct the loss of sample surface reflection (reflectivity R = 4%, corrected according to Fresnel formula), take the average value of 5 measurements (standard deviation ≤ 0.5%).

[0094] III. Model construction and optimization

[0095] Based on the obtained process-structure-performance data set, LSTM model and XGBoost model were used to construct the relationship between process-structure and structure-performance data set, as shown in Figure 3

[0096] 3.1 Process structure relationship modeling

[0097] For the time sequence characteristics of process parameters and microstructure mapping, a four-layer stacked LSTM network was used to dynamically filter key process parameter time sequence characteristics through the forgetting gate and input gate. The network introduced a Dropout layer (probability 0.2) to prevent overfitting, and a fully connected layer was used in the output layer to map the hidden state to the structure representation space. During training, Adam optimizer (learning rate 0.001) was used, and the early stopping condition was set as the loss of the validation set not decreasing for 3 consecutive rounds.

[0098] 3.2 Structure performance relationship modeling​

[0099] (1) Based on mutual information screening (threshold 0.03), structural features (such as grain size distribution and dislocation density) that are strongly correlated with performance indicators (such as strength and wear resistance) are retained;

[0100] (2) Introducing polynomial feature interaction (degree = 2) to capture the nonlinear coupling effect between features;

[0101] (3) The input data is normalized. The model parameters are set as follows: maximum tree depth 6, subsample ratio 0.8, and regularization parameter λ = 1.5 to balance model complexity and prediction accuracy.

[0102] 3.3 Model Optimization

[0103] We built an NS-GA algorithm to optimize the LSTM and XGBoost models. We recoded the LSTM network parameters, including the number of hidden units, learning rate, and batch size, and fine-tuned the XGBoost model's hyperparameters, including tree depth, learning rate, and subsampling ratio.

[0104] IV. Implementation Effect Verification

[0105] This method was applied to PC foam injection molding, characterizing the mechanical strength of the product through transmittance. Injection molding experiments were conducted to obtain a data set of product process parameters. The experimental steps were: first, a process-microdata-transmittance correlation model was constructed based on the test data; then, a fitting formula for product transmittance and mechanical strength was constructed. Finally, a reverse deduction was performed based on the product's mechanical strength to obtain a recommended process. Strength was predicted based on the process, resulting in a predicted mechanical strength value and an error analysis. The resulting data is shown in Table 3 below.

[0106] Table 3 Error analysis of mechanical strength prediction values

[0107]

[0108]

[0109] Among them, the average error between the mechanical strength prediction value and the test value is 1.744, indicating that the adopted model has high accuracy and can better predict product performance. In this example, the required product mechanical performance requirement is set to 57MPa, and then according to Figure 4The formula fitting between the light transmittance and the mechanical properties of the product is performed, the light transmittance requirement of the product is 66%, reverse solving is performed, the recommended process parameter combination is that the mold temperature is 75 DEG C, the plastic temperature is 281.25 DEG C, the injection speed is 65 mm / s, the holding pressure is 75 MPa, the holding time is 3.7 s, the foaming agent content is 3.25 wt%, a plurality of trial mold tests are performed by using the recommended process, the average value of the mechanical strength of the product is 55 MPa, the test shows that the error rate of the mechanical strength of the product obtained by using the recommended process and the standard mechanical strength is less than 5%, and the application effect is good. Compared with the current market research progress, the error rate of the mechanical strength control of most researches is 10%-20%. Compared with this, the patent has obvious advantages.

Claims

1. A foam injection molding process optimization and product performance prediction method based on artificial intelligence, characterized in that The following steps are involved: step S10: Obtain the material grade of the product and its physical, chemical and mechanical properties, as well as the gate position, number of cavities and cooling water channel layout parameters of the mold structure; Step S20: performing injection molding experiments based on material properties and mold structure parameters to determine the range of key process parameters, including injection speed, holding pressure, melt temperature, holding time, barrel temperature, and foaming agent content, to establish a process window for injection molding parameters; Step S30: setting an orthogonal experimental table within the process window to perform product trial mold analysis, obtain a molded product, and perform microstructure analysis and performance analysis to construct a process parameter-microstructure parameter-service performance parameter data set; Step S40: constructing a three-dimensional data matrix of process parameters, microstructure parameters, and performance parameters, using a machine learning model to mine the nonlinear relationship between the parameters, and establishing a mapping model between process-structure-performance; Step S50: constructing a genetic algorithm-driven reverse optimization framework, taking the target performance parameters as constraint boundaries, and screening the optimal process that meets the target data through multi-objective coding and Pareto frontier search; Step S60: Performing actual on-machine mold testing using the recommended process to extract microstructure data and product performance data to verify the accuracy of the method; Step S70: Adjust the model structure and re-recommend the process based on the test results.

2. The method for optimizing foam injection molding process and predicting product performance based on artificial intelligence according to claim 1, characterized in that: A single-factor experimental design method was adopted, with the mass fraction of foaming agent, injection rate, mold cavity temperature, barrel set temperature, holding stage pressure and holding duration in the injection molding process parameters as independent variables, to systematically explore the correlation mechanism between them and the surface defects of the product, such as sink marks, weld marks and warpage deformation.

3. The method for optimizing foam injection molding process and predicting product performance based on artificial intelligence according to claim 1, characterized in that: The steps for determining the product molding process window include determining the injection speed, mold temperature, barrel temperature, holding pressure, and holding time. The melt temperature and mold temperature are determined based on the recommended temperatures provided by the material supplier. The effective value range of the parameters to be optimized obtained from the injection molding experiment is divided into five levels using the process parameters to be optimized (injection speed, melt temperature, mold temperature, holding pressure, holding time, and foaming agent content). A six-factor, five-level orthogonal experimental table is used to determine 25 sets of process parameter combinations for calculation. Multi-batch systematic mold trials are used to obtain a multidimensional data set. The result list contains the following data: Process parameter set: injection speed, mold temperature, barrel temperature, holding pressure, holding time, and foaming agent content; Microstructural characteristics: cell diameter, cell density, and surface thickness, quantitatively characterized by microscopy combined with image analysis; Optical performance index: light transmittance, obtained by calculation after measuring the luminous flux with a color spectrum haze meter.

4. The method for foam injection molding process optimization and product performance prediction based on artificial intelligence according to claim 1, characterized in that: The orthogonal experimental structure data obtained in step S30 are divided into two data sets, one of which is a process parameter-product microstructure data set, which records in detail the microstructure characteristics of the product under different process parameter conditions; The other group is the product microstructure data-product performance dataset, which focuses on the relationship between product microstructure and product final performance. These two datasets are further segmented and divided into training set, validation set and test set. The training set is used for the initial training of the model, accounting for 70%, so that the model can learn the potential patterns in the data; the validation set is used to adjust the model's hyperparameters, accounting for 15%, and is used to optimize model performance; the test set accounts for 15% and is used to comprehensively evaluate the model's generalization ability and prediction accuracy after the model training is completed.

5. The method for optimizing foam injection molding process and predicting product performance based on artificial intelligence according to claim 1, characterized in that: A dual deep learning architecture is constructed to achieve cross-scale correlation modeling of process parameters, microstructure, and optical properties, including: LSTM-feature extraction module: Taking the injection rate, mold temperature, and foaming agent concentration in the time-series process parameter set as input, it captures the dynamic evolution characteristics of the process parameters through a gated recurrent unit and outputs a high-dimensional microstructure representation vector Z = {cell diameter, cell density, surface thickness}; XGBoost-performance prediction module: takes the microstructure feature Z as input, adopts the gradient boosting decision tree ensemble learning framework, establishes a nonlinear mapping relationship, and realizes the quantitative prediction of transmittance; Cross-modal fusion mechanism: Optimize the feature transfer path through the attention weight allocation strategy and strengthen the analysis of the transmission effect of process parameter disturbance on optical performance.

6. The artificial intelligence-based foam injection molding process optimization and product performance prediction method according to claim 5, characterized in that: The steps of building an LSTM model include: setting the learning rate, number of hidden layer units, time step, dropout rate and batch size in the network hyperparameters, initializing the weight matrix and bias term of the gated recurrent unit GRU, building a multi-layer LSTM network architecture and optimizing the time series feature extraction capability through the backpropagation algorithm; the steps of building an XGboost model include: setting the model parameters, including the learning rate, maximum depth of the number, subsampling ratio, column sampling ratio, etc., initializing the model and building a decision tree; and using the NS-GA algorithm for parameter optimization.

Images