A multi-link interactive precision injection molding control system

Abstract

The present invention belongs to the technical field of intelligent manufacturing, and particularly relates to a multi-link interactive precision injection molding control system, including: a parameter setting module, a PLC industrial control module, and a machine learning module; the machine learning module is used to obtain independent variables and target variables from the historical data of the injection molding process, combine them into a data set, construct a non-linear regression model, and based on the color space of the spectrophotometer and the non-linear model, break through the dependence on traditional experience and improve the color consistency of products. Through the process combination of the parameter setting module, the PLC industrial control module, and the machine learning module, the present invention realizes a full-process closed-loop architecture from process parameter input, precise control to parameter prediction; promotes the development of the injection molding control system towards precision and intelligence. The control system uses a reverse inference unit, which can reverse-derive the optimal process parameter combination according to the target color value, thereby solving the problems of low efficiency and high cost of the traditional trial-and-error method.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent manufacturing, and specifically, relates to a multi-link interactive precision injection molding control system. Background Art

[0002] With the continuous improvement of the requirements for the appearance quality of injection molded products in various industries, the hybrid color linkage injection molding technology has gradually become a hot topic in the industry because it can achieve complex color forms and personalized designs. However, traditional injection molding control systems face significant challenges when dealing with hybrid color injection molding processes:

[0003] Although the traditional PLC system can achieve single-shot glue table control, it is difficult to predict the color of the injection molded product in the scenario of hybrid color injection molding of two glue tables, resulting in inconsistent product colors.

[0004] In addition, in the development of new color products, it is necessary to rely on the experience of engineers to repeatedly trial and error to adjust the color combination and process parameters (such as melting temperature, injection pressure, etc.), resulting in a long development cycle and high cost for new color products. Summary of the Invention

[0005] To solve the technical problems of predicting the color of injection molded products in the hybrid color injection molding scenario; and the development of new color products relying on the experience of engineers to repeatedly trial and error and adjust, the present invention provides a multi-link interactive precision injection molding control system.

[0006] The object of the present invention can be achieved by the following technical solutions:

[0007] A multi-link interactive precision injection molding control system includes a parameter setting module, a PLC industrial control module, and a machine learning module;

[0008] The parameter setting module is used to set the process parameters of each injection glue table respectively, and the process parameters include melting temperature, injection speed, injection pressure, injection stroke, holding pressure time, and holding pressure;

[0009] The PLC industrial control module is used to judge according to the injection molding process and condition status, and then convert the process parameters into control instructions and output them to the corresponding components;

[0010] The machine learning module includes a data acquisition unit, a data preprocessing unit, a model training unit, and a color prediction unit;

[0011] The data acquisition unit is used to obtain independent variables and target variables from the historical data of the injection molding process, and combine the independent variables and target variables into a data set; the independent variables include the process parameters of each injection glue table and the resin raw material color; the target variable is the product color;

[0012] A data preprocessing unit preprocesses data, including data cleaning, data standardization, and dataset division; after the dataset division, it includes a training set and a test set; the ratio of the training set to the test set is 2:8 or 3:7;

[0013] A model training unit is used to construct a non - linear regression model and use the training set data to train the model;

[0014] A color prediction unit is used to input a new dataset into the trained non - linear regression model to predict the product color;

[0015] Preferably, both the resin raw material color and the product color are measured in the Lab color space or the HSB color space.

[0016] Preferably, the non - linear regression model is a decision tree regression model or a random forest regression model.

[0017] Preferably, in the model training unit, the specific process of constructing a decision tree regression model includes:

[0018] Dividing the training set into two subsets by a recursive partitioning method to construct the tree - like structure of the decision tree;

[0019] a. Divide the training set into two subsets by a recursive partitioning method to construct the tree - like structure of the decision tree;

[0020] b. For each leaf node of the decision tree, select a feature and a splitting point to minimize the mean squared error of the divided subsets;

[0021] c. For the divided subsets, repeat the splitting process of steps a - b until the stopping condition is met, i.e., the construction of the decision tree is completed;

[0022] d. When the stopping condition is met, take the current subset as the final leaf node, and the average value of the target variable in the current subset is the final predicted value.

[0023] Preferably, the stopping condition is that the depth of the decision tree reaches the preset maximum depth or the number of samples in the subset is less than the minimum number of samples.

[0024] Preferably, in the model training unit, the specific process of constructing a random forest regression model includes:

[0025] Randomly sample from the training set by Bootstrap sampling to form multiple sample subsets;

[0026] Each sample subset is used to train a decision tree, and when splitting the nodes of each decision tree, randomly select m features;

[0027] Repeat the above process for the divided sample subsets to construct n decision trees until the stopping condition is met;

[0028] Each decision tree outputs a predicted value, and the final predicted value is obtained by calculating the average of the predicted values of all decision trees.

[0029] Preferably, it further includes: inputting test set data into the non-linear regression model to output the predicted value of the product color; evaluating the performance of the model by calculating the mean square error of the Lab color space;

[0030] Among them, the formula for calculating the mean square error of the Lab color space is:

[0031] ;

[0032] In the formula, is the predicted value of the product color; is the actual value of the product color; n is the number of samples;

[0033] According to the evaluation results, adjust the model parameters to optimize the performance.

[0034] Preferably, it further includes a reverse inference unit for reverse inferring the optimal process parameters required to achieve the product color; its specific process includes:

[0035] S1) After the decision tree regression model is trained, save the structure and parameters of the decision tree;

[0036] S2) Randomly initialize a set of independent variable data;

[0037] S3) Input the independent variable data into the decision tree regression model to obtain the predicted color;

[0038] S4) Calculate the errors of the L, a, and b components between the predicted color and the target color respectively;

[0039] S5) Establish the mapping relationship between the L, a, and b components and the independent variable data, where the L component (brightness) is related to the process parameters; the a component is related to the resin raw material color of the first injection station; the b component is related to the resin raw material color of the second injection station;

[0040] S6) Use the optimization algorithm to optimize the errors of each component to minimize the errors of each component;

[0041] S7) Select the independent variable data with the minimized error, and this independent variable data includes the resin raw material color and the optimal process parameters.

[0042] Preferably, in step S6), the optimization algorithm is the gradient descent method or the grid search method; the gradient descent method gradually reduces the error by iteratively updating the process parameters; the grid search method enumerates all possible parameter combinations within the feasible range of the process parameters and selects the combination that minimizes the error.

[0043] Preferably, the specific process of using the gradient descent method to optimize the error of each component includes:

[0044] Synchronously update the process parameters through the gradient descent method to gradually reduce the error; where the error of each component is:

[0045] ; ; ;

[0046] In the formula, is the predicted value of the product color; is the target value of the product color.

[0047] Calculate the gradients of the errors of each component with respect to the independent variable data respectively: The calculation formula is:

[0048] ; ; ;

[0049] In the formula, represents calculating the gradient by the finite difference method; , , represents the gradient values of each component;

[0050] Update the independent variable data using the gradient values; the update formula is:

[0051] ;

[0052] represents the learning rate, which is used to control the update step size; represents the process parameter;

[0053] Then repeat steps S3 - S6); until the error converges or the maximum number of iterations is reached.

[0054] Advantages of the present invention:

[0055] 1. Through the process combination of the parameter setting module, the PLC industrial control module, and the machine learning module, a full - process closed - loop architecture from process parameter input, precise control to parameter prediction is achieved; promoting the development of the injection molding control system towards precision and intelligence.

[0056] 2. Based on the color space and non - linear model of the spectrophotometer, breaking through the traditional experience dependence, improving the color consistency of products. Through the reverse inference unit, the system can reverse - derive the optimal process parameter combination according to the target color value, solving the problems of low efficiency and high cost of the traditional trial - and - error method. Significantly shortening the machine - tuning cycle for new product development and improving production flexibility. Description of the Drawings

[0057] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0058] Figure 1 It is a structural framework diagram of a multi-link interactive precision injection molding control system of the present invention.

[0059] Figure 2 It is a flowchart for constructing a decision tree regression model in a multi-link interactive precision injection molding control system of the present invention.

[0060] Figure 3 It is a flowchart of a reverse speculation unit in a multi-link interactive precision injection molding control system of the present invention. Specific embodiments

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0062] Please refer to Figures 1 - 3 As shown, a multi-link interactive precision injection molding control system includes a parameter setting module, a PLC industrial control module, and a machine learning module;

[0063] The parameter setting module is used to set the process parameters of each injection unit respectively. The process parameters include melting temperature, injection speed, injection pressure, injection stroke, holding time, and holding pressure; among them, the parameter setting module supports independent parameter configuration for multiple injection units, and combines with the PLC industrial control module to dynamically adjust the execution instructions in real time to ensure the coordinated operation of multi-link devices.

[0064] The PLC industrial control module is used to judge according to the injection process and condition status, and then convert the process parameters into control instructions and output them to the corresponding components;

[0065] The machine learning module includes a data acquisition unit, a data preprocessing unit, a model training unit, and a color prediction unit;

[0066] The data acquisition unit is used to obtain independent variables and target variables from the historical data of the injection process, and combine the independent variables and target variables into a data set; the independent variables include the process parameters of each injection unit and the color of the resin raw material; the target variable is the product color;

[0067] A data preprocessing unit preprocesses data, including data cleaning, data standardization, and dataset division; after the dataset is divided, it includes a training set and a test set; the ratio of the training set to the test set is 2:8 or 3:7.

[0068] Specifically, in data preprocessing, the standardization of Lab color values needs to process the L, a, and b channels separately. The steps of Z-score standardization: calculate the mean and standard deviation for each component (L, a, b) respectively, and then perform standardization.

[0069] For each feature (such as barrel temperature, injection speed, injection pressure, etc.), calculate its mean and standard deviation: then for each feature value, use the Z-score standardization formula to calculate the standardized value.

[0070] In addition, during the process of standardizing the parameters (mean, standard deviation or minimum, maximum) of the test set, the calculation results of the training set must be used to ensure data consistency.

[0071] A model training unit is used to build a non-linear regression model and use the training set data to train the model;

[0072] A color prediction unit is used to input a new dataset into the trained non-linear regression model to predict the product color;

[0073] Specifically, the control system of the multi-link interactive precision injection molding machine proposed by the present invention performs combined injection molding by defining two injection units, mixes the resin raw materials of different single colors from the two injection units for injection molding, and forms a colored injection molded product. By controlling the process parameters of the melting, injection, and holding pressure actions of the two injection units, the color of the injection molded product shows diversity. Among them, the main process parameters include:

[0074] Melting temperature: The melting temperature of the resin raw material inside the barrel of the injection unit.

[0075] Injection speed: The speed at which the screw injects the molten plastic into the mold cavity.

[0076] Injection pressure: The pressure required to inject the molten plastic into the mold cavity.

[0077] Injection stroke: The distance that the screw moves during the injection process, which directly determines the amount of plastic injected into the mold cavity.

[0078] Holding pressure time: The duration of continuing to apply pressure to the plastic in the mold cavity after the injection is completed.

[0079] Holding pressure: The pressure applied to the mold cavity during the holding pressure stage. This pressure is usually lower than the injection pressure but needs to be maintained for a period of time to ensure that the plastic fully fills the mold cavity.

[0080] Since other injection molding process parameters do not have a relevant impact on the product color, they can be controlled according to the single-shot nozzle injection molding process parameters.

[0081] In the specific implementation process, the parameter setting module is the medium for the system to interact with the user, allowing the user to set various process parameters of the injection molding machine. It adopts the HMI (Human Machine Interface) software framework, which is written in a high-level language, separating the front-end UI design from the background PLC industrial control logic code; the HMI software reads or writes data from the PLC industrial control through communication protocols such as ADS. During the process of writing and reading various process parameters of the injection molding machine, the HMI software communicates with the PLC in a batch reading and dynamic switching manner, thus optimizing the data transmission efficiency.

[0082] The PLC industrial control module supports the IEC 61131 standard and provides multiple programming languages, such as Instruction List (IL), Ladder Diagram (LD), Function Block Diagram (FBD), Sequential Function Chart (SFC), and Structured Text (ST), to control the logic control and motion sequence control of each component of the injection molding machine. Through the process combination of the parameter setting module, the PLC industrial control module, and the machine learning module, a full-process closed-loop architecture from process parameter input, precise control to parameter prediction is achieved; promoting the development of the injection molding control system towards precision and intelligence.

[0083] Furthermore, both the resin raw material color and the product color are measured in the Lab color space or the HSB color space.

[0084] Specifically, the Lab color space is described by three parameters measured using a spectrophotometer (IPS):

[0085] L: Represents the brightness of the color. The higher the L value, the brighter the color.

[0086] a: Represents the position of the color on the red / green axis. A positive value indicates a red bias, and a negative value indicates a green bias.

[0087] b: Represents the position of the color on the yellow / blue axis. A positive value indicates a yellow bias, and a negative value indicates a blue bias.

[0088] As an alternative implementation, the HSB color space is described by three components measured using a color sensor:

[0089] H: Represents the hue attribute of the color, with a value range of [0, 360]. 0 represents red, 120 represents green, and 240 represents blue.

[0090] S: Represents the saturation of the color, with a value range of [0, 1]. 0 represents grayscale (no color); 1 represents a fully saturated color.

[0091] B: Represents the lightness or darkness of the color, with a value range of [0, 1]. 0 represents black, and 1 represents the brightest color.

[0092] Use a spectrophotometer to accurately measure the Lab color values of the resin raw materials and products, and quantify the color characteristics through brightness (L), red-green axis (a), and yellow-blue axis (b) to solve the color difference problem of traditional RGB detection. Through these measurement methods, the color values of the resin raw materials and products can be quantified, so as to perform color matching and control.

[0093] Furthermore, the non-linear regression model is a decision tree regression model or a random forest regression model.

[0094] Specifically, the process parameters in the injection molding process include continuous variables (such as plasticizing, injection, and holding pressure) and discrete variables (such as temperature, pressure, speed, and injection stroke). The relationship between these parameters and the product color is usually non-linear. The non-linear regression model can capture complex non-linear relationships.

[0095] On the other hand, there may be interaction effects between different parameters (such as the combination of plasticizing temperature and injection speed affects the color). Non-linear regression models (such as decision trees and random forests) can automatically capture the interaction effects between variables without the need for manual design of interaction terms.

[0096] The injection molding process parameters and color data are usually high-dimensional data (multiple input features and multiple output targets). The non-linear regression model can effectively process high-dimensional data and reduce the dimension through feature selection (such as feature importance in random forests).

[0097] Decision trees can visually display the feature splitting process, and random forests can reveal key process parameters through feature importance analysis. It can effectively fit the complex relationship between the independent variable data (resin raw material color and injection molding process parameters) and the product color, and achieve high-precision color prediction.

[0098] Furthermore, in the model training unit, the specific process of constructing a decision tree regression model includes:

[0099] Divide the training set into two subsets through a recursive splitting method to construct the tree structure of the decision tree;

[0100] Each leaf node of the decision tree selects a feature and a splitting point to minimize the mean squared error (MSE) of the split subsets;

[0101] For the divided subsets, repeat the above splitting process until the stopping condition is met, i.e., the construction of the decision tree is completed;

[0102] When the stopping condition is met, the current subset is used as the final leaf node, and the average value of the target variable in the current subset is the final predicted value.

[0103] Furthermore, the stopping condition is that the depth of the decision tree reaches the preset maximum depth or the number of samples in the subset is less than the minimum number of samples.

[0104] Specifically, the decision tree regression model is a non - linear regression method based on a tree - like structure. By recursively splitting the dataset, a tree - like structure is constructed to capture the complex relationship between input features and the target variable. In the mixed two - color injection molding control system, the decision tree regression model is used to predict the product color (such as the L, a, b values in the Lab color space). Its working principle is to recursively split the dataset to construct a tree - like structure. Each leaf node selects a feature and a split point to minimize the mean squared error (MSE) of the divided subsets. Among them, the hyperparameters for initializing the decision tree include:

[0105] Maximum depth: The maximum depth of the tree.

[0106] Minimum samples for split: The minimum number of samples required for splitting a leaf node.

[0107] Minimum samples for leaf: The minimum number of samples required for a leaf node.

[0108] During the node splitting process of the decision tree, for each feature X i , all possible split points are traversed; the dataset is divided into two subsets, and the mean squared error (MSE) of the subsets is calculated; the feature and split point that minimize the mean squared error are selected as the node. When one of the following conditions is met, the splitting stops and the current subset is used as a leaf node.

[0109] The decision tree regression model can intuitively display the feature splitting process and importance. At the same time, it can capture complex non - linear relationships and improve the accuracy of color prediction. The decision tree regression model can efficiently and accurately predict the color of injection - molded products and further improve performance by optimizing hyperparameters. Its intuitiveness and interpretability provide reliable technical support for the intelligent control of the injection - molding process.

[0110] Furthermore, in the model training unit, the specific process of constructing a random forest regression model includes:

[0111] Randomly sample from the training set by Bootstrap sampling to form multiple sample subsets;

[0112] Each sample subset is used to train a decision tree. When splitting nodes in each decision tree, m features are randomly selected;

[0113] Repeat the above process for the divided sample subsets to build n decision trees until the stopping condition is met;

[0114] Each decision tree outputs a predicted value. By calculating the average of the predicted values of all decision trees, the final predicted value is obtained.

[0115] Specifically, the random forest regression model improves the accuracy and robustness of the model by constructing multiple decision trees and integrating their prediction results. Among them, the hyperparameters of the initial random forest regression model include:

[0116] Number of trees: The number of decision trees in the forest.

[0117] Maximum depth: The maximum depth of each tree.

[0118] Minimum number of samples for splitting: The minimum number of samples required for node splitting.

[0119] Number of features randomly selected for each tree: Usually the square root of the total number of features.

[0120] In the process of random sampling, the size of each sample subset is usually the same as that of the training set. However, due to sampling with replacement, some samples may be repeatedly selected. Each tree uses different Bootstrap samples and randomly selected features to improve the diversity of samples; thus improving the generalization ability of the model.

[0121] Furthermore, it also includes: Inputting the test set data into the non - linear regression model to output the predicted value of the product color; Evaluating the performance of the model by calculating the mean square error in the Lab color space;

[0122] Among them, the formula for calculating the mean square error in the Lab color space is:

[0123] ;

[0124] In the formula, is the predicted value of the product color; is the actual value of the product color; n is the number of samples;

[0125] According to the evaluation results, adjust the model parameters to optimize the performance.

[0126] Specifically, the mean squared error in the Lab color space is the difference between the predicted values and the actual values of the L, a, and b components in the Lab color space, and the mean squared error is combined to statistically measure the overall prediction accuracy of the model. The smaller the MSE, the higher the prediction accuracy of the model. If the MSE is large, it indicates that there may be underfitting or overfitting problems in the model. The process of optimizing the model parameters is as follows:

[0127] According to the evaluation results, adjust the hyperparameters of the decision tree regression model to optimize the performance.

[0128] Retrain the model using the adjusted hyperparameters and evaluate the performance.

[0129] Use cross-validation to further optimize the hyperparameters.

[0130] As an alternative implementation, the differences between the predicted values and the actual values of the H, S, and B components in the HSB color space can also be used, and the mean squared error or the mean absolute error is combined to statistically measure the overall prediction accuracy of the model.

[0131] Furthermore, it also includes a reverse inference unit for reverse inferring the optimal process parameters required to achieve the product color; the specific process includes:

[0132] S1) After the decision tree regression model is trained, save the structure and parameters of the decision tree;

[0133] S2) Randomly initialize a set of independent variable data;

[0134] S3) Input the independent variable data into the decision tree regression model to obtain the predicted color;

[0135] S4) Calculate the errors of the L, a, and b components between the predicted color and the target color respectively;

[0136] S5) Establish the mapping relationship between the L, a, and b components and the independent variable data. Among them, the L component (brightness) is related to the process parameters; the a component is related to the resin raw material color of the first injection station; the b component is related to the resin raw material color of the second injection station;

[0137] S6) Use an optimization algorithm to optimize the errors of each component to minimize the errors of each component;

[0138] S7) Select the independent variable data with the minimized error, and this independent variable data includes the resin raw material color and the optimal process parameters.

[0139] In the specific implementation process, the goal of the reverse inference unit is to minimize the error between the predicted color and the target color through an optimization algorithm, so as to find the optimal process parameters. By giving the target product color Y = (L, a, b) (taking the Lab color space as an example), it finds a set of optimal independent variable data X = (X1, X2, …, X n ), such that the color predicted by the model is as close as possible to the target color.

[0140] An independent mapping relationship is established between the three components of the Lab color value and the independent variable data (process parameters and raw material color), including:

[0141] The L component (luminance): is related to the process parameters (such as melt temperature, injection speed, etc.); minimize the luminance error E L .

[0142] The a component (red / green axis): is related to the resin raw material color (Lab value) of the first injection stage; minimize the red / green axis error E a .

[0143] The b component (yellow / blue axis): is related to the resin raw material color (Lab value) of the second injection stage; minimize the yellow / blue axis error Eb.

[0144] Its working mechanism is that the a and b components of the product after mixing are controlled by the a and b components of the two resin raw material colors, and the L component (luminance) of the product color after mixing is controlled by the process parameters. Based on the color space and non-linear model of the spectrophotometer, it breaks through the dependence on traditional experience and improves the color consistency of the product. Through the reverse inference unit, the system can reverse-derive the optimal process parameter combination according to the target color value, solving the problems of low efficiency and high cost of the traditional trial-and-error method. It significantly shortens the machine adjustment cycle for new product development and improves production flexibility.

[0145] Furthermore, in process S6), the optimization algorithm is the gradient descent method or the grid search method; the gradient descent method iteratively updates the process parameters to gradually reduce the error; the grid search method enumerates all possible parameter combinations within the feasible range of the process parameters and selects the combination that minimizes the error.

[0146] Specifically, the process parameters are synchronously updated by the gradient descent method to gradually reduce the error E; among them, the error of each component is:

[0147] ; ; ;

[0148] In the formula, is the predicted value of the product color; is the target value of the product color.

[0149] Calculate the gradients of each component error with respect to the independent variable data respectively: The calculation formula is as follows:

[0150] ; ; ;

[0151] In the formula, represents the calculation of the gradient by the finite difference method; , , represent the gradient values of each component;

[0152] Update the independent variable data using the gradient values; The update formula is as follows:

[0153] ;

[0154] represents the learning rate, which is used to control the update step size; represents the process parameter.

[0155] Then, repeat the above process S3 - S6); until the error converges or the maximum number of iterations is reached.

[0156] In the specific implementation process, it is also necessary to consider the constraint conditions of the process parameters, such as:

[0157] The range of the barrel temperature: Tmin ≤ T ≤ Tmax; the range of the injection speed: Vmin ≤ V ≤ Vmax; the range of the injection pressure: Pmin ≤ P ≤ Pmax, etc. Ensure that the process parameter update process does not exceed the reasonable threshold range.

[0158] Specifically, the reverse inference unit realizes the intelligent derivation from the target color to the optimal process parameters by minimizing the error between the actual value and the predicted value using an optimization algorithm based on the prediction result of the combined machine learning model.

[0159] As another alternative implementation, the grid search method is a global optimization algorithm that searches for the solution that minimizes the objective function by enumerating all possible parameter combinations in the parameter space. The specific application process is as follows:

[0160] Parameter range definition: Define the search range of each parameter according to the physical limitations of the process parameters (such as the barrel temperature range, injection pressure range).

[0161] Grid generation: Discretize each process parameter into grid points.

[0162] Traversal and evaluation: Traverse the set of different grid points X = (X1, X2, …, X n ).

[0163] Input the set X into the model to predict the product color value.

[0164] Calculate the total error E(X) between the predicted color and the target color;

[0165] ;

[0166] In the formula, is the predicted value of the product color; is the target value of the product color.

[0167] Output the optimal solution: Select the error set X as the optimal process parameter combination.

[0168] Specifically, the gradient descent method efficiently handles high-dimensional continuous parameter optimization problems and is suitable for online optimization and dynamic adjustment. The grid search method can find the global optimal solution and avoid local optima. It is suitable for offline optimization and precise parameter tuning.

[0169] The present invention proposes a multi-linkage interactive precision injection molding control system. Through the full-process closed-loop architecture of parameter setting, PLC industrial control, machine learning prediction, and reverse inference, the color diversity of the injection molding process is achieved. Based on the color space and non-linear model of the spectrophotometer, it breaks through the traditional empirical dependence and improves the color consistency of the product. Through the reverse inference unit, the system can reverse-derive the optimal process parameter combination according to the target color value, solving the problems of low efficiency and high cost of the traditional trial-and-error method. In the actual production process, it significantly shortens the machine adjustment cycle for new product development, reduces the scrap rate, and improves production flexibility.

[0170] The above content is only an example and explanation of the structure of the present invention. Those skilled in the art of the present technology make various modifications or supplements or use similar methods to replace the specific embodiments described, as long as they do not deviate from the structure of the invention or exceed the scope defined by this claim book, they should all fall within the protection scope of the present invention.

Claims

1. A multi-link interactive precision injection molding control system, characterized in that: Including parameter setting module, PLC industrial control module and machine learning module; The parameter setting module is used to set the process parameters of each injection station separately, including melt temperature, injection speed, injection pressure, injection stroke, holding time and holding pressure; PLC industrial control module, which is used to judge the injection molding process and condition status, and then convert the process parameters into control instructions and output them to the corresponding components; The machine learning module includes a data acquisition unit, a data preprocessing unit, a model training unit, and a color prediction unit; A data acquisition unit is used to acquire independent variables and target variables from historical data of the injection molding process, and combine the independent variables and target variables into a data set; the independent variables include the process parameters of each injection molding station and the color of the resin raw material; the target variable is the product color; the resin raw material color and the product color are both measured in Lab color space or HSB color space; A data preprocessing unit performs preprocessing on the data, including data cleaning, data standardization and data set division; the data set after division includes a training set and a test set; A model training unit is used to build a nonlinear regression model and use the training set data to train the model; A color prediction unit, used for inputting a new data set into a trained nonlinear regression model to predict product color; the nonlinear regression model is a decision tree regression model; It also includes a reverse inference unit for reversely inferring the optimal process parameters required to achieve product color; its specific process includes: S1) After the decision tree regression model training is completed, the structure and parameters of the decision tree are saved; S2) Randomly initialize a set of independent variable data; S3) Input the independent variable data into the decision tree regression model to obtain the predicted color; S4) respectively calculating the errors of L, a, and b components between the predicted color and the target color; S5) establishing a mapping relationship between the L, a, and b components and the independent variable data, wherein the L component is related to the process parameters; the a component is related to the color of the resin raw material of the first injection molding station; and the b component is related to the color of the resin raw material of the second injection molding station; S6) optimizing the error of each component using an optimization algorithm so as to minimize the error of each component; S7) selecting independent variable data that minimizes the error, the independent variable data including the color of the resin raw material and the optimal process parameters.

2. A multi-link interactive precision injection molding control system according to claim 1, characterized in that: In the model training unit, the specific process of building a decision tree regression model includes: a. Divide the training set into two subsets through recursive segmentation method and construct the tree structure of decision tree; b. Each leaf node of the decision tree selects a feature and a split point so that the mean square error of the split subset is minimized; c. Repeat the segmentation process of process ab for the divided subsets until the stopping condition is met, that is, the construction of the decision tree is completed; d. When the stopping condition is met, the current subset is used as the final leaf node, and the average value of the target variable in the current subset is the final predicted value.

3. A multi-link interactive precision injection molding control system according to claim 2, characterized in that: The stopping condition is that the depth of the decision tree reaches a preset maximum depth or the number of samples in the subset is less than the minimum number of samples.

4. A multi-link interactive precision injection molding control system according to claim 2, characterized in that , also includes the following processes: Use the test set data to input into the decision tree regression model and output the predicted value of product color; The performance of the model is evaluated by calculating the mean square error of the Lab color space; the mean square error calculation formula of the Lab color space is: ； In the formula, is the predicted value of product color; is the actual value of the product color; n is the number of samples; Based on the evaluation results, the model parameters are adjusted to optimize the performance.

5. The multi-link interactive precision injection molding control system according to claim 1, characterized in that: In process S6), the optimization algorithm is a gradient descent method or a grid search method; the gradient descent method iteratively updates the process parameters to gradually reduce the error; the grid search method enumerates all possible parameter combinations within the feasible range of the process parameters and selects the combination that minimizes the error.

6. The multi-link interactive precision injection molding control system according to claim 5, characterized in that: The specific process of optimizing the error of each component by the gradient descent method includes: The process parameters are updated synchronously through the gradient descent method to gradually reduce the error; the error of each component is: ； ； ； In the formula, is the predicted value of product color; is the target value of product color; Calculate the gradient of each component error to the independent variable data separately: the calculation formula is: ； ； ； In the formula, Indicates that the gradient is calculated by the finite difference method; , , Represents the gradient value of each component; Use the gradient value to update the independent variable data; the update formula is: ； Represents the learning rate, which is used to control the update step size; Indicates process parameters; Repeat process S3-S6); until the error converges or the maximum number of iterations is reached.

Images