A method for evaluating and predicting the quality of thick-walled transparent polymer products using infrared temperature measurement

Abstract

The present invention provides a method for evaluating and predicting the quality of thick-walled transparent polymer products by using infrared temperature measurement. The characteristic parameters (first-order color moment, second-order color moment, one-dimensional information entropy, two-dimensional information entropy) of infrared thermal images are introduced into the product quality evaluation, which improves the comprehensiveness and accuracy of the evaluation. A relationship model between the temperature information (i.e., characteristic parameters) provided by infrared thermal images and product quality is established by using support vector regression (SVR), and the parameters of this relationship model are optimized by using cross-validation to obtain an evaluation model with optimized parameters. The evaluation model has high evaluation accuracy for product quality and good robustness. Using the product quality evaluation method of the present invention to evaluate products can accurately guide the process parameters in the product processing to obtain high-quality products.

Description

Technical Field

[0001] The present invention belongs to the technical field of quality assessment, and particularly relates to a method for evaluating and predicting the quality of thick-walled transparent polymer products by using infrared temperature measurement. Background Art

[0002] Injection molding, also known as injection or injection molding, is an important method for processing and forming polymer materials. In recent years, more and more plastic products have been applied in industries such as electronics, medical, and automotive. This benefits from the many advantages of injection molding technology. The four most critical points are as follows: (1) Injection molding can relatively easily design and produce relatively precise and complex parts with high efficiency; (2) There are various material selection options for injection molding; (3) The processing and material costs of injection molding are relatively low, and the degree of automation is relatively high; (4) The waste of injection molding is less and more environmentally friendly. There are many factors affecting the quality of injection molding products, such as the temperature of the polymer melt, filling rate, pressure, and mold design, etc. Among them, the cooling time of injection molding parts accounts for the largest proportion of the production cycle in the entire injection molding process.

[0003] Generally speaking, whether the temperature control of the mold cavity is appropriate will largely affect the final forming effect and production efficiency of injection molding parts. This is because the cooling rate, crystallization, orientation, shrinkage, and uniformity of injection molding parts are all related to the cavity and melt temperature. Starting from the perspective of the temperature field to optimize the processing technology of injection molding, the first and most important thing is to achieve accurate perception and characterization of the temperature field. Although there are many traditional temperature assessment methods, such as the commonly used thermocouple measurement and ultrasonic measurement, these methods have some problems more or less. The former has a simple structure and low price, but has a low spatial resolution and can only perform fixed-point temperature measurement and cannot perform large-range and high-precision temperature monitoring; the latter has the advantages of strong penetration and high sensitivity, but it is not easy to measure parts with complex structures.

[0004] As a special injection molding part, thick-walled transparent polymers have higher requirements for the temperature field in the injection molding technology. The production process requires a more uniform temperature field, and traditional temperature measurement methods cannot meet these requirements. Summary of the Invention

[0005] To solve the problems existing in the prior art, the present invention provides a different idea to detect the temperature of the melt surface during the injection molding process and evaluate the temperature field, and relate its two-dimensional temperature distribution map to the quality of the product, providing a means for quality assessment and prediction to improve the product quality of injection molding parts by locally controlling temperature parameters and improving process parameters. This method can use the infrared thermal image of the product when the single layer is opened to evaluate and predict the final quality of the product and give an evaluation score, so as to guide product production and process optimization.

[0006] A method for evaluating and predicting the quality of thick-walled transparent polymer products using infrared temperature measurement comprises the following steps:

[0007] (1) Collect infrared thermal images of each layer of multiple thick-walled transparent polymer products and extract ROI images;

[0008] (2) Convert each ROI image into a grayscale matrix and calculate the eigenvalue of each ROI image;

[0009] For each product, the average value of the feature value of the ROI image corresponding to each layer is used as the final feature of the infrared thermal image of the product;

[0010] (3) Evaluate the health of each product and give each product an evaluation score;

[0011] (4) The final features and evaluation scores of the infrared thermal images of all products are used as the training set. The training set is trained using support vector regression, and cross-validation is used to optimize the training model parameters to obtain the evaluation model;

[0012] (5) Collect infrared thermal images of each layer of the product to be evaluated, process them according to steps (1) to (2) to obtain the final features of the product to be evaluated, and use the final features as input to evaluate and predict the quality of the product to be evaluated using the evaluation model.

[0013] In the above step (1), the infrared thermal image is the thermal image taken by the infrared thermal imager when the mold is opened after each layer of injection and cooling is completed during the product injection molding process. Since the principle of the infrared thermal imager is to perform photoelectric conversion on the infrared radiation energy of the target to be measured detected by the lens, convert it into an electronic signal, and then simulate a pseudo-color image based on the intensity calculation of the electronic signal, it is different from the true color image with three color components of R, G, and B. The actual parameter of the pseudo-color is a single index value, that is, after the infrared thermal image is grayed, the gray value of each pixel corresponds to a unique temperature value. Subsequent processing of the infrared image is performed after the image is grayed, which has no effect on the actual temperature calculation.

[0014] Support Vector Regression (SVR) is often used to analyze data for classification or regression problems. It is robust to outliers and the decision model is only related to the support vectors. Therefore, most training samples do not need to be retained after training. It has high prediction accuracy and excellent generalization ability. At the same time, this method has high training accuracy for small samples.

[0015] The product quality evaluation method of the present invention extracts the characteristic information (characteristic values) contained in the thermal image of the product when each layer of mold is opened, and at the same time evaluates the health of the product. Using the evaluation score and the above characteristic information as the training set to train the evaluation model, it links the temperature information with the product quality and establishes a product quality evaluation model; subsequently, this evaluation model is used to evaluate the quality of the product to be evaluated, which has guiding significance for the production of the product and the process optimization in the injection molding process of thick-walled transparent polymers. Moreover, this method obtains the product temperature field information based on infrared temperature measurement technology, and has many advantages such as large area, non-contact temperature measurement, high temperature measurement efficiency, high resolution, and fast response speed, which are not possessed by traditional temperature measurement means.

[0016] When predicting the product quality using the method of the present invention, prediction can be carried out when the injection of the first few layers is just completed. According to the obtained results, it can be evaluated whether it is necessary to continue the injection molding, thereby significantly reducing the production time, improving the production efficiency, and reducing the cost.

[0017] Preferably, in step (2), the characteristic values include the first-order color moment, the second-order color moment, the one-dimensional information entropy, and the two-dimensional information entropy.

[0018] The color moment of an image is an internal feature of the image, which is difficult to distinguish by the naked eye. For a random variable R, its probability distribution can be uniquely represented and described by its moments of each order. If the gray values of each pixel point of an image are regarded as a probability distribution, then the moments of each order of the image can also be used to describe the image.

[0019] The first-order color moment refers to the operation of averaging the gray values of each pixel point after graying the thermal image. This parameter can reflect the brightness of an image as a whole. The larger the first-order color moment, the brighter the image. The calculation formula of the first-order color moment is:

[0020]

[0021] Among them, E represents the first-order color moment; I and J are the horizontal and vertical pixel numbers of the infrared thermal image respectively; G ij is the gray value of the pixel point with horizontal and vertical coordinates i and j in the infrared thermal image.

[0022] The second-order color moment of the infrared thermal image is calculated by taking the square root of the second-order central moment of the pixel. This parameter can reflect the distribution breadth of the image color. The smaller the second-order color moment, the more uniform the color distribution of the image. The formula for calculating the second-order color moment is as follows:

[0023]

[0024] Where σ represents the second-order color moment; E represents the first-order color moment; I and J are the horizontal and vertical pixel numbers of the infrared thermal image respectively; G ij is the grayscale value of the pixel with horizontal and vertical coordinates i and j in the infrared thermal image.

[0025] The concept of information entropy was first proposed by Shannon, drawing on concepts from thermodynamics. He defined the average amount of information in a message after redundancy has been eliminated as "information entropy." The underlying idea is that "events with low probability of occurrence contain more information," and this concept provides a method for quantitatively measuring information. Simply put, if an event is certain to occur, its information entropy is 0; conversely, if the event is uncertain, its information entropy is higher. In other words, lower information entropy indicates less uncertainty about the event and more information contained in the signal.

[0026] Shannon quantified the amount of information of an event. The amount of information of an event I(x) is defined as follows:

[0027] I(x)=-log2P(x)

[0028] Among them, P(x) is the probability of the event occurring.

[0029] After quantifying the information, we can calculate the information entropy, which is actually the expected value of the amount of information. The formula for discrete quantities is as follows:

[0030]

[0031] Among them, H1(x) represents one-dimensional information entropy; L represents the number of various possibilities of the event; P l (x) represents the probability of a certain possibility of the event occurring.

[0032] The present invention uses information entropy as one of the characteristic parameters of infrared thermal images to measure the uncertainty of the color occurrence of each pixel in the image, thereby judging whether its temperature field is uniform. First, after converting the infrared image into a grayscale image, the number of times the grayscale value identical to the grayscale value of each pixel appears in the entire image is counted. After calculating the probability, the probability of occurrence of the grayscale value of each pixel is normalized to calculate the information entropy of the entire image, that is, the one-dimensional information entropy. The larger the one-dimensional information entropy, the more uneven the color distribution in the image, that is, the more uneven the temperature field distribution of the target to be measured shown in the infrared thermal image.

[0033] One-dimensional information entropy; the calculation formula is as follows:

[0034]

[0035] Among them, H1(x) represents the one-dimensional information entropy; L represents the type of grayscale value in the grayscale matrix; Pl (x) represents the probability of the occurrence of the l-th gray value.

[0036] The information entropy reflects the statistical result of the entire image and represents the average amount of information in a group of images. For a specific image, the information entropy is unique. The one-dimensional information entropy can represent the overall distribution uniformity of the data in the image, but it does not consider the change trend and continuity of its distribution. Therefore, the present invention introduces the concept of two-dimensional information entropy to characterize this.

[0037] The extraction process of the two-dimensional information entropy of the infrared thermal image is as Figure 3 shown:

[0038] Extract the temperature field information on the surface of the target to be measured through an infrared thermal imager. In the present invention, the target to be measured is a thick-walled transparent polymer product. Using this pseudo-color infrared thermal image, after converting it into a grayscale image, obtain the grayscale value of each pixel point of the entire image and construct a grayscale matrix, and the number of grayscale levels is set to N. Calculate the two-dimensional information entropy feature of the image according to the obtained grayscale matrix for subsequent training of the model. As Figure 3 shown, assume that the grayscale value at the position of a certain element (x, y) in the grayscale matrix is G1(i, j), and the grayscale value at the position of another element (i + a, j + b) is G2(i + a, j + b). If the grayscale value of each element in the grayscale matrix and the grayscale value of each element in all other elements in the (a, b) direction (i.e., the slope is ) form a combination, the number of types of this combination is N 2 possibilities. Count the number of occurrences of each type of combination, and normalize the number of occurrences of each type of combination to the probability P pq of the occurrence of each pair of combinations (i.e., the grayscale value combination is (p, q)), and a new temperature probability distribution matrix [P pq can be constructed. N×N Based on this matrix, the two-dimensional information entropy H2 of the infrared thermal image can be extracted, and its calculation formula is:

[0039]

[0040] where H2 represents the two-dimensional information entropy; N represents the number of grayscale levels in the grayscale matrix; P pq represents the probability of the occurrence of the grayscale value combination of (p, q), where p and q are the grayscale values in the grayscale matrix respectively. The two-dimensional information entropy calculated in this formula is actually the two-dimensional information entropy in the set slope direction.

[0041] As a further preference, extract the two-dimensional information entropy in the directions of slope values of 0, ∞, +1, and -1 respectively, and use the average value of the two-dimensional information entropy in all the extracted slope directions as the final two-dimensional information entropy of the infrared thermal image.

[0042] In fact, it is not comprehensive to construct a combination of the gray value of an element in the gray matrix and the gray values of any other element in a single direction, and it is impossible to measure the overall two-dimensional information entropy of the entire image. Therefore, in the present invention, for the entire gray value matrix, two-dimensional information entropy is extracted in the directions with slopes of 0, ∞, +1, and -1, as Figure 4 shown. The average value of the information entropy in the four slope directions obtained is used as the final characteristic parameter for measuring the two-dimensional information entropy of the entire infrared thermal image.

[0043] Preferably, in step (3), the shrinkage warpage, light transmittance, and residual stress are used as indicators to evaluate the health of the product, and the final quality of the product is scored in the form of weighted scoring to obtain the evaluation score of the product.

[0044] There are many quality evaluation indicators and standards for thick-walled transparent polymer products. After screening and classifying numerous evaluation indicators, the present invention summarizes and concludes the three most important indicators as follows: 1) high profile accuracy and less shrinkage warpage (shrinkage warpage); 2) low residual stress (residual stress); 3) high light transmittance (light transmittance). Among them, indicator 1) mainly evaluates the closeness between the shape of the product and the ideal shape. The smaller the shape deformation, the better the product quality. In addition, the larger the surface error and the higher the deformation degree, the higher the optical distortion of the product. Indicator 2) can measure the mechanical properties of the product. The lower the residual stress, the better the tensile, bending, impact resistance and other properties of the product, and the product life will also be improved. In addition, for transparent products, the residual stress is closely related to the optical properties. The larger the residual stress, the worse the optical properties. Indicator 3) measures the light transmittance of the product to visible light.

[0045] The final quality of the product is evaluated in the form of weighted scoring, which is used as the final single indicator for measuring the product quality, that is, the dependent variable of the regression prediction model.

[0046] The planar shrinkage warpage amount is measured by a dial indicator. Fix the dial indicator so that it is placed vertically. Place the plastic part (product) on a moving platform with two degrees of freedom in the X and Y directions, with the measured plane facing up. Rotate the two handles of the moving platform to move the position, so that the dial indicator measures the height value of a specific point on the plastic part. The relevant measuring instrument device is as Figure 5 shown. After obtaining the values of multiple points, the planar shrinkage warpage amount can be calculated.

[0047] The light transmittance is measured by an optical transmittance measuring instrument. The LH-221 optical transmittance measuring instrument is selected, which can be used to measure the transmittance of visible light and infrared light through transparent polymer products. This measuring instrument complies with the standard of GB5137.2-2002 "Test Method Standard for Optical Properties of Safety Glass for Motor Vehicles". The measuring part device consists of a light source, a lens, and a receiver, as Figure 6 shown.

[0048] The measurement of residual stress is characterized by the birefringence value. The ST150D polariscope developed and produced by Xiamen Eastech Precision Instruments Co., Ltd. is used to quantitatively measure the optical path difference and characterize the birefringence. As Figure 7 shown, 3 measuring points with an interval of 20 mm are selected in the product flow direction, and the measurement of the optical path difference near the gate and at the end is excluded during actual measurement. The instrument uses the Senarmont compensation method and is tested in accordance with standards such as GB / T7962.5-2010.

[0049] As a further preference, during the weighted scoring process, the weight ratios of shrinkage and warping (profile accuracy), light transmittance (light transmission ability, light transmittance), and residual stress (birefringence value) are 2:1:1.

[0050] As a preference, the specific operation of optimizing the training model parameters by cross-validation is as follows:

[0051] After dividing the training set into k parts, each part of the data is sequentially used as the validation set, and the remaining part is used as the training set for training to obtain k training models; two hyperparameters c and g in the k training models are traversed with a set step size, and the accuracy under different combinations of hyperparameters c and g is calculated using the validation set (i.e., one of the k parts); the accuracy of the k models under the same hyperparameter combination is averaged, the average accuracy under different hyperparameter combinations is evaluated, and the hyperparameter combination with the highest average accuracy is selected as the optimal parameter of the evaluation model.

[0052] The method for evaluating the quality of thick-walled transparent polymer products of the present invention establishes a relationship model, i.e., an evaluation model, between the temperature information (i.e., characteristic parameters) provided by the infrared thermal image and the product quality by using support vector regression (SVR). During the model training process, in the first step, the infrared thermal images of each layer of each collected sample are corrected for image tilt. Subsequently, ROI extraction is performed to obtain the thermal image of the entire product area, and the extracted ROI image is converted into a grayscale matrix, and then four major characteristic values are calculated; the four major characteristic values include the first-order color moment, the second-order color moment, the one-dimensional information entropy, and the two-dimensional information entropy. In the second step, the health of the collected samples themselves is evaluated. The three major index parameters for evaluating the samples include shrinkage warping, residual stress, and light transmittance. The three major indexes are weighted and scored to give the evaluation score results corresponding to each product. The data after processing (including the four major characteristic values and the evaluation scores) are used as the training set data for training to obtain a model for evaluating and predicting the product quality, and the cross-validation method is used to optimize the parameters of the trained model to obtain an evaluation model with optimized parameters. Finally, the infrared thermal images of each layer of the sample to be evaluated are collected, the four major characteristic values are extracted and used as inputs, and the quality of the sample to be evaluated is evaluated by using the obtained evaluation model.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0054] The method for evaluating and predicting the quality of thick-walled transparent polymer products by using infrared temperature measurement of the present invention introduces the characteristic parameters (the first-order color moment, the second-order color moment, the one-dimensional information entropy, and the two-dimensional information entropy) of the infrared thermal image into the product quality evaluation, improving the comprehensiveness and accuracy of the evaluation; a relationship model between the temperature information (i.e., characteristic parameters) provided by the infrared thermal image and the product quality is established by using support vector regression (SVR), and the cross-validation is used to optimize the parameters of the relationship model to obtain an evaluation model with optimized parameters; the evaluation model has high evaluation accuracy for the product quality and has good robustness. Using the product quality evaluation method of the present invention to evaluate the product can accurately guide the process parameters during the product processing to obtain high-quality products. Description of the Drawings

[0055] Figure 1 Schematic diagram of the characteristic value extraction process of the infrared thermal image of each layer of a product in the training set;

[0056] Figure 2 Schematic diagram of the evaluation model establishment process;

[0057] Figure 3 Schematic diagram of the extraction process of the two-dimensional information entropy of the infrared thermal image;

[0058] Figure 4 Schematic diagram of the two-dimensional information entropy corresponding to different slope directions;

[0059] Figure 5 This is a diagram of the instrument for measuring plane shrinkage warpage;

[0060] Figure 6 Schematic diagram of an optical transmittance measuring instrument;

[0061] Figure 7 This is a schematic diagram of the product birefringence value (residual stress) measurement;

[0062] Figure 8 Schematic diagram of cross-validation operation. DETAILED DESCRIPTION

[0063] A method for evaluating and predicting the quality of thick-walled transparent polymer products using infrared temperature measurement comprises the following steps:

[0064] like Figure 2 As shown, first establish the evaluation model:

[0065] (1) Collect infrared thermal images of each layer of multiple thick-walled transparent polymer products and extract ROI images;

[0066] After each layer of the product is injected and cooled, it is photographed by an infrared thermal imager when the mold is opened to obtain an infrared thermal image of each layer.

[0067] (2) Convert each ROI image into a grayscale matrix and calculate the eigenvalue of each ROI image;

[0068] For each product, the average value of the feature values of the ROI image corresponding to each layer is used as the final feature of the infrared thermal image of the product.

[0069] The eigenvalues include first-order color moment, second-order color moment, one-dimensional information entropy and two-dimensional information entropy.

[0070] The color moments of an image are internal features that are difficult to discern with the naked eye. For a random variable R, its probability distribution can be uniquely represented and described using its moments of various orders. If the grayscale values of each pixel in an image are also considered a probability distribution, then the image can also be described using its moments of various orders.

[0071] The first-order color moment refers to the operation of calculating the average grayscale value of each pixel after graying the thermal image. This parameter can reflect the overall brightness of an image. The larger the first-order color moment, the brighter the image. The calculation formula of the first-order color moment is:

[0072]

[0073] Where E represents the first-order color moment; I and J are the horizontal and vertical pixel numbers of the infrared thermal image respectively; G ijis the gray value of the pixel at the horizontal and vertical coordinates i and j in the infrared thermal image.

[0074] The second-order color moment of the infrared thermal image is calculated by the square root of the second-order central moment of the pixels. This parameter can reflect the distribution breadth of the image color. The smaller the second-order color moment, the more uniform the color distribution of the image. The formula for calculating the second-order color moment is as follows:

[0075]

[0076] where σ represents the second-order color moment; E represents the first-order color moment; I and J are the numbers of horizontal and vertical pixels in the infrared thermal image respectively; G ij is the gray value of the pixel at the horizontal and vertical coordinates i and j in the infrared thermal image.

[0077] The concept of information entropy was first proposed by Shannon by borrowing the concept of thermodynamics. It defines the average amount of information excluding redundancy in information as "information entropy". Its basic idea is that "an event with a low occurrence probability contains more information". The proposal of this concept provides a method for the quantitative measurement of information. Simply put, it can be explained that if an event is certain to occur, its information entropy is 0; conversely, if it is uncertain whether this event will occur, its information entropy will be higher. In other words, the smaller the information entropy, the smaller the uncertainty of the event and the more information contained in the signal.

[0078] Shannon quantified the amount of information of an event. The amount of information I(x) of an event is defined as the following formula:

[0079] I(x) = -log2P(x)

[0080] where P(x) is the probability of the occurrence of this event.

[0081] After quantifying the information, the information entropy can be calculated. In fact, the information entropy is the expected value of the amount of information. Its formula for discrete quantities is as follows:

[0082]

[0083] where H1(x) represents the one-dimensional information entropy; L represents the number of various possibilities of this event; P l (x) represents the probability of the occurrence of a certain possibility of this event.

[0084] This embodiment uses information entropy as one of the characteristic parameters of an infrared thermal image to measure the uncertainty of the color of each pixel in the image, thereby determining whether the temperature field is uniform. First, after converting the infrared image to a grayscale image, the number of times the grayscale value identical to the grayscale value of each pixel appears in the entire image is counted. After calculating the probability, the probability of occurrence of each pixel's grayscale value is normalized to calculate the information entropy of the entire image, i.e., the one-dimensional information entropy. The greater the one-dimensional information entropy, the more uneven the color distribution in the image, that is, the more uneven the temperature field distribution of the target under test as displayed by the infrared thermal image.

[0085] One-dimensional information entropy; the calculation formula is as follows:

[0086]

[0087] Among them, H1(x) represents the one-dimensional information entropy; L represents the type of grayscale value in the grayscale matrix; P l (x) represents the probability of the occurrence of the lth grayscale value.

[0088] Information entropy reflects the statistical results of the entire image, reflecting the average amount of information in a group of images. For a specific image, information entropy is unique. One-dimensional information entropy can indicate the overall uniformity of the data distribution in the image, but it does not consider the changing trend and continuity of this distribution. Therefore, this embodiment introduces the concept of two-dimensional information entropy to represent this.

[0089] The extraction process of two-dimensional information entropy of infrared thermal image is as follows: Figure 3 As shown:

[0090] The temperature field information of the target surface is extracted by an infrared thermal imager. In this embodiment, the target is a thick-walled transparent polymer product. The pseudo-color infrared thermal image is converted into a grayscale image, and the grayscale value of each pixel in the entire image is obtained and constructed into a grayscale matrix with the number of grayscale levels set to N. The two-dimensional information entropy features of the image are calculated based on the obtained grayscale matrix for subsequent model training. Figure 3 As shown, assuming that the gray value (temperature value) of a certain element (x, y) in the gray matrix is G1(i, j), and the gray value (temperature value) of another element (i+a, j+b) is G2(i+a, j+b), if the gray value of each element in the gray matrix and the element in the (a, b) direction (that is, the slope is ) forms a combination of gray values of each element in all other elements of the matrix. There are N types of such combinations. 2 Count the number of times each type of combination appears, and normalize the number of times each type of combination appears to be the probability P of each pair of combinations (i.e., the gray value combination is (p, q)). pq, a new temperature probability distribution matrix [P pq can be constructed. N×N , based on this matrix, the two-dimensional information entropy H2 of the infrared thermal image can be extracted, and its calculation formula is:

[0091]

[0092] where H2 represents the two-dimensional information entropy; N represents the number of gray levels in the gray matrix; P pq represents the probability of the gray value combination of (p, q) appearing, where p and q are the gray values in the gray matrix respectively. The two-dimensional information entropy calculated by this formula is actually the two-dimensional information entropy in the set slope direction.

[0093] Actually, it is not comprehensive to construct combinations of the gray value of an element in the gray matrix and the gray values of any other elements in a single direction, and it cannot measure the overall two-dimensional information entropy of the entire image. Therefore, in this embodiment, the two-dimensional information entropy is extracted for the entire gray value matrix in the directions of slopes of 0, ∞, +1, -1, as shown in Figure 4 . The average value of the information entropies in the four slope directions obtained is used as the final characteristic parameter for measuring the two-dimensional information entropy of the entire infrared thermal image.

[0094] (3) Evaluate the health of each product and give the evaluation score of each product;

[0095] Among them, the health of the product is evaluated with shrinkage warping (profile accuracy), light transmittance (transparency, light transmission ability) and residual stress (birefringence value) as indicators, and the final quality of the product is scored in the form of weighted scoring to obtain the evaluation score of the product.

[0096] The final quality of the product is evaluated in the form of weighted scoring, which is used as the final single index for measuring the product quality, that is, the dependent variable of the regression prediction model, as shown in Table 1:

[0097] Table 1 Dependent variable of the SVR model

[0098]

[0099] There are many indicators and standards for evaluating the quality of thick-walled transparent polymer products. After screening and classifying numerous evaluation indicators, the present invention summarizes the three most important indicators as follows: 1) high profile accuracy, less shrinkage and warping (shrinkage and warping); 2) low residual stress (residual stress); 3) high light transmittance (light transmittance). Among them, indicator 1) mainly evaluates the closeness between the shape of the product and the ideal shape. The smaller the shape deformation, the better the product quality. In addition, the larger the surface error and the higher the degree of deformation, the higher the optical distortion of the product; indicator 2) can measure the mechanical properties of the product. The lower the residual stress, the better the tensile, bending, impact resistance and other properties of the product, and the product life will also be improved. In addition, for transparent products, the relationship between residual stress and optical properties is close. The greater the residual stress, the worse the optical properties; indicator 3) measures the light transmittance of the product to visible light.

[0100] The planar shrinkage and warping amount is measured by a dial indicator. Fix the dial indicator so that it is placed vertically. Place the plastic part (product) on a moving platform with two degrees of freedom in the X and Y directions, with the measured plane facing up. Rotate the two handles of the moving platform to move the position, so that the dial indicator measures the height values of certain specific points on the plastic part. The relevant measuring instrument devices are as Figure 5 shown. After obtaining the values of multiple points, the planar shrinkage and warping amount can be calculated.

[0101] The light transmittance is measured by an optical transmittance measuring instrument. Select the LH-221 optical transmittance measuring instrument, which can be used to measure the light transmittance of transparent polymer products to visible light and infrared light. This measuring instrument implements the standard of GB5137.2-2002 "Test Method Standard for Optical Properties of Safety Glass for Motor Vehicles". The measuring part device consists of a light source, a lens, and a receiver, as Figure 6 shown.

[0102] The measurement of residual stress is characterized by the birefringence value. The ST150D polariscope developed and produced by Xiamen Easte Instrument Co., Ltd. is used to quantitatively measure the optical path difference to characterize the birefringence. As Figure 7 shown, three measurement points with an interval of 20 mm are selected in the product flow direction, and the measurement of the optical path difference is excluded near the gate and at the end during actual measurement. The instrument adopts the Senarmont compensation method and is tested according to standards such as GB / T7962.5-2010.

[0103] (4) Using the final features and evaluation scores of all product infrared thermal images as the training set, support vector regression (SVR) is used to train the training set, and cross-validation is used to optimize the training model parameters to obtain the evaluation model.

[0104] As Figure 8 shown, the specific operation of parameter optimization by cross-validation is as follows:

[0105] After dividing the training set into k parts (shown as 5 parts in the figure), each part of the data is used as the validation set in turn, and the remaining part is used as the training set for training to obtain k trained models; the hyperparameters (c, g) in the k trained models are traversed with a set step size, and the accuracy under different hyperparameters (c, g) is calculated using the validation set (i.e., one of the k parts); the average value of the accuracies of the k models under the same hyperparameters is obtained, the average value of the accuracies under different hyperparameters is evaluated, and the hyperparameters when the average value of the accuracies is the highest are selected as the optimal parameters for the evaluation model.

[0106] Then, the established evaluation model is used to predict the quality of the product to be evaluated:

[0107] (5) Collect the infrared thermal images of each layer of the product to be evaluated, process them according to steps (1) - (2) to obtain the final features of the product to be evaluated, use the final features as the input, and use the evaluation model to evaluate and predict the quality of the product to be evaluated.

[0108] Detection experiment

[0109] Training of the evaluation model:

[0110] In order to enable the trained model to achieve the quality evaluation and prediction of the final product to a greater extent, stable parameter conditions were set in the experiment, an on-line measurement experiment was carried out on an injection molding machine, and 29 groups of temperature field information were collected for feature parameter extraction for the model training of machine learning.

[0111] The material for experimental injection molding is PMMA (Evonik Rohm GmbH PLEXIGLAS 8N). The formed product is a thick-walled flat plate sample of 100mm×100mm×30mm, with a thickness of 30mm and divided into six layers for injection. In order to obtain products of different qualities, the same parameters are used for each layer injection of the same product, while the same molding parameters are not used for different products during injection molding. The value range of the parameters is shown in Table 2.

[0112] Table 2 Injection molding parameter setting table

[0113]

[0114]

[0115] After each layer of injection and cooling is completed, the infrared thermal imager takes the infrared thermal image of the transparent polymer product when the mold is opened as a sample for collecting feature parameters.

[0116] The collected infrared thermal images are subjected to tilt correction to obtain undeformed or non-tilted thermal images, and the thermal images of the thick-walled transparent polymer product area (i.e., ROI images) are intercepted for subsequent feature extraction. The preprocessed (grayed) thermal images are subjected to feature extraction to obtain the one-dimensional information entropy, two-dimensional information entropy, first-order color moment, and second-order color moment of the thermal images. A set of data for each product consists of six thermal images, corresponding to the temperature information at the time of mold opening for each of the six injection layers of the product. The four features (one-dimensional information entropy, two-dimensional information entropy, first-order color moment, and second-order color moment) of the six thermal images are respectively averaged and used as the final features of the thermal images of the product, that is, the independent variables in machine learning. The eigenvalue parameters of all products are extracted and counted, and the four final eigenvalue parameters corresponding to each product are calculated.

[0117] The eigenvalue extraction process of the collected infrared thermal images and the characteristic parameters (final features) of part of the training set are as Figure 1 shown.

[0118] At the same time, the shrinkage warpage, light transmittance, and residual stress values of each product are evaluated and scored, which correspond one by one to the eigenvalue of each product. The specific scoring method is as follows: The optimal and worst values of the measurement results of each quality index (light transmittance, shrinkage warpage, and residual stress) among all samples are statistically analyzed. The optimal value is given the highest score, and the worst value is given the lowest score. The remaining scores are statistically distributed linearly. The results are shown in Table 3.

[0119] Table 3 Health Score of Products

[0120]

[0121] During training, cross-validation is performed on the data of the training set. The process of cross-validation is mainly to optimize the internal parameters during the training of the SVR (Support Vector Regression) model, that is, to find the optimal parameter values suitable for the characteristics of the training set samples in the paper.

[0122] The specific operation and principle of cross-validation are as Figure 8 shown. After dividing the training set into k parts, each part of the data is used as the validation set in turn, and the remaining part is used as the training set for training to obtain k training models; for the two hyperparameters c and g of each of the k obtained training models, they are traversed with a set step size, and the accuracy under different combinations of hyperparameters c and g is calculated using the validation set (i.e., one of the k parts). The average accuracy of the k models under the same hyperparameter combination is calculated, and the average accuracy under different hyperparameter combinations is evaluated. The hyperparameter combination with the highest average accuracy is selected as the hyperparameter for the final model training, that is, the optimal parameters of the evaluation model are obtained.

[0123] Quality assessment and verification of samples to be evaluated:

[0124] The material for experimental injection molding was PMMA (Evonik Rohm GmbH PLEXIGLAS 8N). The molded product was a thick-walled flat plate sample with dimensions of 100 mm × 100 mm × 30 mm, having a thickness of 30 mm and being injection-molded in six layers. Infrared thermal images of five products (validation set samples) were collected according to the injection parameters in Table 4:

[0125] Table 4 Partial injection parameters of the validation set samples

[0126]

[0127] Among them, the data of the single-layer cooling time was the combination of the single-layer cooling time of the first three layers and the last three layers of each sample. Taking Sample 1 as an example, the single-layer cooling time of 60 + 100 s means that the single-layer cooling time of the first three layers of this sample was 60 s, and the single-layer cooling time of the last three layers was 100 s. The remaining injection parameters of each sample were the same as those of the training set injection parameters.

[0128] The average value of the six-layer thermal image characteristic parameters corresponding to each product was calculated, and the results are shown in Table 5:

[0129] Table 5 Average values of the validation set characteristic parameters

[0130]

[0131]

[0132] The health assessment of the five products was carried out, and the results are shown in Table 6:

[0133] Table 6 Quality parameters and evaluation score values of the validation set

[0134]

[0135] The health assessment scores (test scores) predicted by the evaluation model obtained through training, the actually measured health assessment scores (true scores), and the relative error between the two are shown in Table 7:

[0136] Table 7 Test results and relative errors of the validation set

[0137]

[0138] It can be found from Table 7 that different samples are in different evaluation score ranges, and the prediction accuracy of the above-mentioned trained evaluation model has good performance in each score range.

[0139] The prediction and evaluation method for product quality proposed by the present invention can assist in the process optimization of thick-walled transparent polymers in injection molding technology. According to the product thermal images obtained each time the mold is opened, the final product quality can be evaluated and predicted, so as to better control and adjust the process parameters for each layer, and then obtain products with higher quality, providing a new method and means for the quality prediction and evaluation of thick-walled transparent polymer injection molding technology, and providing ideas and directions for the process optimization of this technology.

[0140] The above is only an application example of the present invention. In fact, this method can be used for different injection molding products to associate their quality with infrared thermal images. The products that can apply the present invention to guide process optimization and product production cannot be enumerated one by one here. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for evaluating and predicting the quality of thick-walled transparent polymer products using infrared temperature measurement, characterized in that, It includes the following steps: (1) Collect the infrared thermal images of each layer of multiple thick-walled transparent polymer products, and extract the ROI images; (2) Convert each ROI image into a grayscale matrix, and calculate the eigenvalues of each ROI image; For each product, use the average value of the eigenvalues of the ROI images corresponding to each layer of the product as the final feature of the infrared thermal image of the product; (3) Evaluate the health of each product and give the evaluation score of each product; (4) Use the final features and evaluation scores of the infrared thermal images of all products as the training set, train the training set using support vector regression, and optimize the training model parameters using cross-validation to obtain the evaluation model; (5) Collect the infrared thermal images of each layer of the product to be evaluated, process them according to steps (1)-(2) to obtain the final features of the product to be evaluated, use the final features as the input, and use the evaluation model to evaluate and predict the quality of the product to be evaluated; In step (2), the eigenvalues include the first-order color moment, the second-order color moment, the one-dimensional information entropy, and the two-dimensional information entropy; Extract the two-dimensional information entropy in the directions with slope values of , , , respectively. Take the average value of the two-dimensional information entropy in all the extracted slope directions as the final two-dimensional information entropy of the infrared thermal image; In step (3), use shrinkage warping, light transmittance, and residual stress as indicators to evaluate the health of the product, and score the final quality of the product in the form of weighted scoring to obtain the evaluation score of the product.

2. The method for evaluating and predicting the quality of thick-walled transparent polymer products using infrared temperature measurement according to claim 1, characterized in that During the weighted scoring process, the weight ratios of shrinkage warping, light transmittance, and residual stress are 2:1:

1.

3. The method for evaluating and predicting the quality of thick-walled transparent polymer products using infrared temperature measurement according to claim 1, characterized in that The specific operation of optimizing the training model parameters using cross-validation is: After dividing the training set into k parts, each part of the data is taken as the validation set in turn, and the remaining part is used as the training set for training to obtain k trained models; for two hyperparameters in the k trained models , traverse with a set step size, and use the validation set to calculate the accuracy under different hyperparameter combinations; calculate the average value of the accuracies of the k models under the same hyperparameter combination, evaluate the average value of the accuracies under different hyperparameter combinations, and select the hyperparameters when the average value of the accuracies is the highest as the optimal parameters for the evaluation model.

Images