Intelligent prediction method for quality of injection molding product

Abstract

The invention discloses an intelligent injection molding product quality prediction method, which comprises the steps of collecting product structure information, mold parameters and injection molding equipment operation parameters, and constructing a scientific mold testing process window based on simulation analysis; designing an orthogonal experiment scheme to generate a simulation data set, and forming a multi-source data training sample in combination with field test model data; training a quality prediction model by using a distributed gradient boosting tree model, inputting process parameters, and outputting quality indexes such as product gram weight, volume shrinkage and buckling deformation; the model interpretability is realized through a shop value analysis method, and the influence weight of the process parameters on the quality indexes is evaluated; the model is deployed in a production control system or a cloud platform, the product quality is predicted in real time, a dynamic feedback mechanism is established, and when the error between actual detection data and a predicted value exceeds a threshold value, model retraining and parameter correction are automatically triggered to form closed-loop control. The method effectively improves the accuracy, transparency and engineering adaptability of quality prediction of the injection molding product, and can be popularized.

Description

Technical Field

[0001] The present invention relates to a quality prediction and modeling method in the cross - field of intelligent manufacturing and polymer molding process, specifically belonging to the intelligent prediction technology of product quality in the injection molding process. In particular, it relates to a method that integrates mold flow simulation analysis and on - site trial - mold data, and realizes high - precision and multi - target prediction of key quality indicators of injection molded products by constructing an interpretable data - driven prediction model, which is applicable to the injection production scenario with complex process parameter control and real - time quality feedback. Background Art

[0002] Injection molding is one of the most widely used processing technologies in the production of thermoplastic polymer materials products, with significant advantages such as high molding efficiency, high automation degree, and precise product structure. However, in the actual injection production process, product quality is affected by the coupling of multiple factors such as raw material characteristics, mold structure, and process parameter settings, resulting in frequent problems such as weight deviation, warpage deformation, and volume shrinkage of products during the molding process, seriously affecting the consistency and stability of products.

[0003] Traditional injection process debugging mainly relies on the experience of operators, and adjusts process parameters by repeated trial - molding to meet quality requirements. This method has a long cycle, low efficiency, and the parameter adjustment process lacks systematicness, making it difficult to achieve quantitative prediction and active control of product quality. With the development of intelligent manufacturing and data - driven modeling technologies, researchers have begun to try to apply artificial intelligence algorithms to injection process modeling and optimization. For example, mapping models between process parameters and quality are established using methods such as BP neural networks and support vector machines to achieve molding parameter recommendation or quality trend prediction.

[0004] However, the existing methods still have the following technical bottlenecks:

[0005] (1) Most models rely on a single type of data (such as simulation data or on - site trial - mold data), lacking a heterogeneous data fusion mechanism, resulting in limited model generalization ability;

[0006] (2) The prediction models are mostly of the "black - box" structure, lacking the interpretability of the influence of process parameters on quality, and it is difficult to provide effective decision - making support for production engineering;

[0007] (3) After the model is deployed, it often lacks a dynamic correction mechanism and cannot adapt to production condition changes under different mold structures, material properties, or environmental fluctuations, restricting its stability and scalability in the actual scenarios of multi - variety and small - batch production.

[0008] Therefore, there is an urgent need to develop a quality prediction method that is oriented to complex injection molding conditions, supports multi-source data fusion, has model interpretability, and can be dynamically updated, so as to realize the intelligent transformation of the injection molding process from "experience-driven" to "data-driven", and improve the real-time performance, stability, and engineering feasibility of product quality control. Summary of the Invention

[0009] Aiming at the technical problems existing in the existing injection molding quality prediction methods, such as poor model interpretability, single source of training data, and lack of dynamic adaptation ability after model deployment, the present invention proposes an intelligent quality prediction method for injection molding products that integrates simulation and trial molding data, has process interpretability, and supports dynamic model correction, aiming to improve the accuracy, transparency, and continuous adaptability of product quality prediction in the injection molding production process.

[0010] To solve the above problems, the present invention provides an intelligent quality prediction method for injection molding products. The core lies in integrating mold flow simulation data and on-site trial molding data, constructing a quality prediction model with interpretability and dynamic correction ability, and deploying it in the injection molding production process to achieve quality prediction and feedback optimization control in continuous production. The method includes the following steps:

[0011] Step S10: Obtain product and process input information. Collect the three-dimensional structure model of the injection molding product, mold design parameters, and injection molding equipment operation configuration, including but not limited to product volume, number of mold cavities and arrangement form, cooling water channel distribution, mold material type, injection molding machine model, screw diameter, maximum injection speed and pressure, etc. This information serves as the basis for subsequent simulation modeling and process window construction.

[0012] Step S20: Construct a simulation process window and generate a simulation data set. Based on the collected model and parameter information, establish a digital mold model in the mold flow analysis software, carry out a scientific trial molding process, determine the effective value range of key process parameters such as injection speed, holding pressure, holding time, and melt temperature item by item, and form a scientific process window. Within this window range, construct process combination schemes by the orthogonal experimental design method, batch execute mold flow simulation, and output the simulated quality indicators (weight, warpage curvature, volume shrinkage rate, etc.) of the product under each process combination to form a structured simulation data set.

[0013] Step S30: Construct an on-site experimental data set. According to the process parameter combinations recommended by the model or set by experience, carry out actual trial molding experiments on the injection molding machine, collect the actual process setting parameters and the quality inspection results of the molded products, ensure that the data covers different typical process combination conditions, and form a representative measured sample data set. This data is consistent with the simulation data in structure, both including input parameters and output quality indicators.

[0014] Step S40: Integrate data and build a prediction model. Normalize the features, handle missing values, and unify the formats of the simulation data and the measured data, and then merge them to form a model training dataset. Using the product quality indicators as the response variables (output layer) and the process parameters as the input variables (input layer), build a distributed gradient boosting tree (XGBoost) model to achieve multi-objective joint prediction of the target quality indicators (such as gram weight, shrinkage rate, and warpage amount).

[0015] Step S50: Achieve model interpretability and feature contribution analysis. Introduce the SHAP value analysis method during the model training process to evaluate the contribution degree of each input variable, and output the ranking of the relative influence degree of different process parameters on each quality indicator, so as to realize interpretable modeling of model prediction, which is convenient for engineers to identify key factors and assist in optimizing and adjusting parameters.

[0016] Step S60: Introduce the genetic algorithm for multi-objective optimization. Based on the model prediction accuracy, introduce the genetic algorithm to perform multi-objective optimization on the model structure and parameter combinations, and set the objective function to minimize the gram weight error, warpage deformation rate, and volume shrinkage rate. Consider the process window boundary constraints during the optimization process to improve the global optimal ability and generalization performance of the prediction model.

[0017] Step S70: Deploy the model and establish an online prediction and adaptive correction mechanism. Deploy the trained prediction model to the injection molding production line control system or the edge embedded module to achieve quality prediction and process monitoring of each batch of products during the continuous production process. When the system detects that the prediction error exceeds the set threshold, automatically collect the measured data of the corresponding batch, and perform retraining and parameter update of the model, so as to build a closed-loop quality prediction and adaptive optimization system.

[0018] Preferably, the machine learning model is a distributed gradient boosting tree model constructed by the XGBoost algorithm. The input features include process parameters such as injection speed, mold temperature, melt temperature, holding pressure, and holding time, and the output results include product gram weight, volume shrinkage rate, and warpage deformation amount at key positions.

[0019] More preferably, the training dataset is divided into a training set, a validation set, and a test set in the ratio of 70%, 15%, and 15% after preprocessing to improve the generalization ability and stability of the model and avoid overfitting.

[0020] More preferably, during the model training process, the grid search and cross-validation methods are used to optimize the model parameters. The parameters to be tuned include learning rate, subsample ratio, maximum depth of the tree, and number of weak classifiers, etc., to obtain the optimal prediction performance.

[0021] As a further preference, the model interpretability analysis adopts the SHAP value method to output the contribution degree distribution of each input feature to the target quality index, visually present the influence path of key process parameters, and can be used for abnormal process diagnosis and expert-assisted decision-making.

[0022] As a further preference, the optimization parameters of the genetic algorithm include population size, number of generations, crossover probability, and mutation probability. The optimization objective function is the weighted sum of squared residuals between the predicted value and the measured value, and the constraint condition is set that each process parameter shall not exceed the process window boundary.

[0023] As a further preference, the prediction model is integrated into the injection molding equipment control platform through an embedded system or an edge computing module, or provides prediction services through an industrial cloud platform in the form of an API, realizing real-time invocation and prediction feedback of the model in the actual injection molding production process.

[0024] As a further preference, in the model application stage, the system automatically collects the actual quality data of each batch of products after each production cycle and conducts an error evaluation with the prediction results; when the error exceeds a preset threshold (such as 2%), the model retraining process is automatically triggered, and a new round of corrected prediction parameters are output for the next batch of production processes, forming a dynamic learning and feedback closed-loop mechanism.

[0025] Compared with the prior art, the intelligent prediction method for the quality of injection molding products proposed by the present invention has significant advantages in terms of method structure, modeling ability, and engineering adaptability, which are mainly reflected in the following aspects:

[0026] 1. The data fusion modeling mechanism improves the model accuracy and generalization ability. The present invention first unifies the structure and fuses the modeling of mold flow simulation data and on-site trial mold measurement data. Compared with the traditional prediction model relying on a single data source, the fitting ability of the model to the actual working conditions and the generalization ability to different process parameter combinations are significantly improved, solving the problem of model inaccuracy caused by data deviation.

[0027] 2. Introducing an interpretable modeling mechanism enhances engineering transparency. By constructing a prediction model based on the XGBoost algorithm and introducing the SHAP value analysis method, a quantifiable and visual causal relationship model between process parameters and quality indicators is established, breaking the "black box" limitation of traditional neural networks, providing a traceable and intervenable quality control basis for the engineering site, and facilitating process optimization and knowledge transfer.

[0028] 3. Constructing a dynamic correction and closed-loop learning mechanism enhances the model stability. The present invention designs an error-driven dynamic correction strategy to realize real-time evaluation and automatic update of the model performance during continuous production, enabling the model to continuously adapt to variable factors such as working condition fluctuations, material differences, or mold aging, and maintaining the long-term stability of prediction accuracy and control effect.

[0029] 4. Multi-objective optimization and boundary constraints enhance prediction reliability. By introducing a genetic algorithm to perform multi-objective optimization on the model, while minimizing the gram weight error, shrinkage rate error, and warpage error, the process window boundary is introduced as a parameter search limit condition to ensure that the optimization results are stable and physically feasible within the implementable range.

[0030] 5. Flexible deployment and strong engineering promotion. The prediction model constructed by the present invention has good deployment compatibility. It can be integrated into the on-site injection molding equipment control system for edge deployment, or deployed on the industrial cloud platform to provide remote quality prediction services through API interfaces, supporting rapid application and large-scale promotion in complex production scenarios with multiple modes and varieties.

[0031] In summary, the method of the present invention not only significantly improves the accuracy and interpretability of injection molding quality prediction, but also takes into account the dynamic evolution ability of the model and the industrial deployment requirements, with remarkable technical advancement and broad engineering application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] To further describe the technical solutions of the present invention more clearly, the embodiments of the present invention are further described in conjunction with the drawings, where:

[0033] Figure 1 is the overall flowchart of the intelligent prediction method for the quality of injection molding products according to the present invention;

[0034] Figure 2 is the three-dimensional structure schematic diagram of a typical injection molded product adopted in the embodiment of the present invention;

[0035] Figure 3 is the schematic diagram of the shap value analysis result in the embodiment, showing the contribution degree ranking of process parameters to quality indicators;

[0036] Figure 4 is the system structure schematic diagram of the model deployment and feedback correction mechanism according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0037] For the convenience of further understanding the technical solutions of the present invention, the present invention is described in detail below in conjunction with the drawings and embodiments. This embodiment is only used to clarify the principle and implementation path of the present invention, and is not a limitation on the protection scope of the present invention.

[0038] As Figure 1 shown, the present invention proposes an intelligent prediction method for the quality of injection molding products, which integrates mold flow simulation data and on-site trial mold data, constructs an interpretable machine learning prediction model, and establishes a dynamic feedback correction mechanism to realize real-time prediction and continuous optimization of the quality indicators of injection molded products.

[0039] Step S10: Obtain product and process input information

[0040] In this embodiment, the injection molded product of ABS material LGABSHI121_1 is taken as the research object, and the product structure is as Figure 2 shown, the product volume is 31.2 cm 3 , the number of mold cavities is 4, the arrangement form is rectangular arrangement, the cooling water channel distribution is spiral water channel, the mold material type is 45 steel, the injection molding machine model is Hai Tian - MA900, the screw diameter is 36 mm, the maximum injection speed is 120 cm 3 / s, and the maximum injection pressure is 196 Mpa.

[0041] Step S20: Construct a simulation process window and generate a simulation data set

[0042] Based on the product structure, mold design and equipment parameters, 5 key process parameters closely related to the product quality are selected: material temperature, mold temperature, injection speed, holding pressure, and holding time.

[0043] On the basis of scientific trial mold analysis, combined with equipment capabilities and material characteristics, the value ranges of each parameter are determined as follows:

[0044] Material temperature: 230–250 °C

[0045] Mold temperature: 40–80 °C

[0046] Injection speed percentage: 45%–85%

[0047] Holding pressure percentage: 16%–40%

[0048] Holding time: 3–5 s

[0049] Within the above parameter ranges, the orthogonal experimental design method is used to set up an orthogonal table with five factors and five levels. The experimental scheme is shown in the following table.

[0050]

[0051] Using Moldex3D as the simulation analysis tool, 25 groups of process parameter combination simulations in the table are executed in sequence, and the product quality indicators corresponding to each group of parameters are extracted, including product weight, volume shrinkage rate, and warpage deformation amount, to form a structured simulation data set, with a total of 25 groups of samples.

[0052] Step S30: Construct a field experiment data set

[0053] According to the results of the simulation data analysis, typical process parameter combinations were selected, and on-site mold trial experiments were carried out on Haitian injection molding machines. During the mold trial process, the actually set process parameters and the quality indicators actually detected for the products were collected. An electronic balance (accuracy 0.001 g) was used to measure the product weight, and a coordinate measuring machine (accuracy 0.01 mm) was used to obtain the key dimensions and calculate the volume shrinkage rate and warpage deformation amount. A total of 20 groups of measured data were collected, and the specific data are shown in the following table.

[0054]

[0055] Step S40: Integrate the data and build a prediction model

[0056] Merge the simulation data set and the mold trial data set to form a complete training data set.

[0057] The data preprocessing process includes: 1) Missing value handling: Use the average value of neighboring samples to fill in the data to avoid data structure damage caused by missing values, making the data set tend to be complete in scale and structure, facilitating subsequent data analysis and modeling calculations; 2) Outlier removal: Use the 3σ principle to remove abnormal samples; this method can improve the model accuracy and prediction ability, and avoid model deviation and overfitting caused by the addition of abnormal samples. 3) Feature normalization: Perform 0–1 interval normalization processing on the 5 input process parameters. Since the magnitudes of the 5 input process parameters vary greatly, normalization processing can avoid some features having an excessive impact on the model due to their large value ranges, thereby improving the model stability and generalization ability. By preprocessing the data samples, the modeling problems caused by missing values, outliers, and dimensional differences in the data set in industrial production can be effectively solved, and it has good application effects for scenarios in industrial production where the data quality is controllable, the parameter magnitudes vary significantly, and both efficiency and robustness need to be considered.

[0058] After the preprocessing is completed, for the industrial scenario with a moderate amount of current data and high process stability requirements, the data set is divided into a 70% training set, a 15% validation set, and a 15% test set. Among them, 70% of the data supports the model to fully capture the relationship between process parameters and product quality; 15% of the validation set guides the adjustment of model parameters to avoid overfitting production noise; 15% of the test set independently verifies the generalization ability of the model to ensure its practicality in the real production line.

[0059] Input feature matrix X (5 items): Material temperature (°C); Mold temperature (°C); Injection speed (mm / s); Holding pressure (MPa); Holding time (s).

[0060] Output response matrix Y (3 items): Product weight (g); Volume shrinkage rate (%); Warpage deformation amount (mm).

[0061] Step S50: Achieve model interpretability and feature contribution analysis

[0062] Build an injection molding quality prediction model using the distributed gradient boosting tree (XGBoost). The model training parameters are set as follows: the learning rate is set to 0.05 to avoid overfitting the model due to local noise; the maximum tree depth is set to 5, which can balance the prediction accuracy and training time, and retain sufficient non-linear expression ability to capture the non-linear relationship between key parameters; the number of weak classifiers is set to 400, which can make full use of data features to obtain a stable optimal solution while avoiding overfitting; the subsample ratio is set to 0.7, that is, 70% of the samples are randomly selected during the training of each tree, which can increase the model diversity and reduce the dependence on a specific sample distribution; the minimum weight sum is set to 0.5 to prevent the model from being overly sensitive to low-frequency outliers (such as occasional defective product data); the minimum split loss function value is set to 0.01, thus generating a simpler tree structure, which is suitable for the implicit demand for model interpretability in industrial scenarios. Finally, five-fold cross-validation and grid search method are used to optimize the hyperparameters. During the training process, the mean square error is used as the objective function to build a high-precision and low-overfitting prediction model under the constraints of noise, dimension difference and limited samples of industrial data.

[0063] After the training is completed, the SHAP value analysis method is used to calculate the contribution degree of each input feature to the prediction result, and the influence ranking of process parameters on product weight, volume shrinkage rate and warpage deformation amount is output. The SHAP value analysis results are shown as Figure 3 shown, to assist on-site engineering personnel in optimizing process parameters and tracing anomalies. Step S60: Introduce genetic algorithm for multi-objective optimization

[0064] On the basis of the completion of the model structure and parameter training, a genetic algorithm is introduced to perform multi-objective optimization on the prediction accuracy. The objective function is the weighted minimization of weight error, warpage deformation rate and volume shrinkage rate. During the optimization process, the process window boundary is added as a constraint condition to improve the prediction robustness and deployability.

[0065] Step S70: Deploy the model and establish an online prediction and adaptive correction mechanism

[0066] The prediction model supports the following two deployment methods: 1) Edge deployment: Integrate the model into the injection molding equipment control system or edge server; 2) Cloud deployment: Package it as a RESTful API interface and connect it to the MES or SCADA system.

[0067] The system collects process parameters in the production process in real time, calls the model to predict product quality indicators, and compares them with the on-site measured data. When it is detected that the average prediction error exceeds the set threshold (such as ±2%), the model retraining is automatically triggered, the training data set is incrementally updated, the model parameters are optimized, and the model is redeployed to form a dynamic learning and closed-loop feedback mechanism, such as Figure 4 shown.

[0068] In summary, in the production of typical ABS products, when the method of the present invention is applied, the average prediction error of the model test set is 1.4%, while for traditional experience-based statistical prediction or linear regression prediction, the error rate is about 2% - 10%. The method of the present invention is superior to the minimum error required by the current market. At the same time, in the continuous production verification, the prediction accuracy of this model can be stably above 95%, which is superior to the industry average prediction error of 80% - 90%. Therefore, the present invention has good engineering adaptability and industrial promotion value.

Claims

1. An intelligent prediction method for the quality of injection molding products, characterized in that, Including the following steps: Step S10: Obtain the structural information, mold parameters, and injection molding equipment operation parameters of the injection molded product. The information includes, but is not limited to, product volume, cavity structure, mold cooling system, material grade, machine model, and operation configuration, etc.; Step S20: Use mold flow simulation software to conduct simulation mold trials within a preset process range, construct an effective process window for key process parameters such as injection speed, injection pressure, holding pressure, and holding time, and form a simulation data set; Step S30: Based on the process window, construct an orthogonal experiment plan, execute multiple groups of process combination simulation simulations, and extract high-quality sample data including input parameters and product quality results; Step S40: During the on-site mold trial process, collect actual molding process parameters and corresponding product quality data to construct an experimental data set for enhancing the generalization ability and prediction accuracy of the model; Step S50: Integrate the simulation data set and the experimental data set to construct a machine learning model for predicting key product quality indicators. The model is constructed based on the distributed gradient boosting algorithm, and the importance of input features is ranked; Step S60: Introduce a genetic algorithm to perform multi-objective optimization on the machine learning model. The objective function includes minimizing the product weight deviation rate, warpage rate, and volume shrinkage rate; Step S70: Apply the trained prediction model to the continuous production process of injection molded products, conduct quality prediction and conformity determination for multiple batches of products, and dynamically correct the model parameters based on the prediction error feedback; 2. The intelligent prediction method for the quality of an injection molding product according to claim 1, characterized in that The machine learning model is a distributed gradient boosting tree model, trained and constructed using the XGBoost algorithm. The input features include process parameters, and the output results include product weight, volume shrinkage rate, and warpage rate at key positions; 3. The intelligent prediction method for the quality of an injection-molded product according to claim 2, wherein The model outputs the relative contribution degree of each input feature to the prediction result through the shap value analysis method to realize the interpretable evaluation of the influence of process parameters on product quality; 4. An intelligent prediction method for the quality of an injection molding product according to claim 1, characterized in that, The training data of the model is jointly composed of orthogonal experiment data generated by mold flow simulation and measured data obtained from on-site mold trials. The input variables include process parameters, and the output variables are the corresponding product quality indicators; 5. An intelligent prediction method for the quality of an injection molding product according to claim 1, characterized in that, During the continuous production process, the model is evaluated for accuracy based on actual detection data. When the prediction error exceeds the set threshold, the model is automatically updated to correct the parameters and improve the prediction accuracy; 6. The intelligent prediction method for the quality of an injection molding product according to claim 1, wherein The model is integrated into the injection molding equipment control platform through an embedded system, or provides an API interface in the form of a cloud service for real-time prediction and feedback control of product quality during the injection molding process.

Images