Injection molding machine nozzle melt state monitoring method based on machine vision

Abstract

The present application relates to the field of melt glue state monitoring, and particularly relates to a melt glue state monitoring method for an injection molding machine nozzle based on machine vision, which comprises the following steps: acquiring an original image of melt glue at the nozzle of the injection molding machine, and pre-processing the original image to obtain a gray image and a melt glue region, calculating a contour complexity representing the shape of the melt glue based on the melt glue region; mapping the melt glue region to the gray image, and calculating the energy and entropy values of the melt glue region; calculating a flow velocity fluctuation coefficient of the melt glue region based on the gray images of the current frame and the previous frame, the flow velocity fluctuation coefficient being positively correlated with the fluctuation of the flow velocity module length of the pixel points; constructing a feature vector by using the contour complexity, the energy, the entropy values and the flow velocity fluctuation coefficient, and inputting the feature vector into a preset prediction model to obtain a prediction result of the melt glue state. The present application can monitor various types of defects such as caking, bubbles and flow marks.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of melt glue state monitoring, and in particular to a melt glue state monitoring method for an injection molding machine nozzle based on machine vision. BACKGROUND

[0002] Injection molding refers to injecting molten plastic raw materials into a mold cavity through a nozzle, and then forming a plastic part after cooling in the mold cavity. In the injection molding process, the state of the melt glue directly affects the quality of the product, including the uniformity, flowability and bubble content of the melt glue. For example, when incompatible other high molecular polymers are mixed into the melt glue, the plastic film will produce a peeling phenomenon; when the melt glue temperature is too low and the flowability is insufficient, it will cause flow marks and underfilling problems. The existing melt glue state monitoring mainly relies on contact type monitoring means such as pressure sensors and temperature sensors, but the existing monitoring methods have the following defects: 1. The sensor is easily corroded by high-temperature melt glue, has a short service life and high maintenance cost; 2. The sensor can only indirectly reflect the physical parameters of the melt glue, and the melt glue state can only be judged by the experience of workers, and cannot be directly displayed; 3. The response is lagging, and real-time adjustment cannot be achieved, resulting in batch waste.

[0003] The Chinese patent application file with publication number CN118658116A discloses a micro-nozzle blockage monitoring system based on image processing, which includes an image acquisition unit, a data processing unit, a remote control unit and an intelligent monitoring terminal. The image acquisition unit acquires real-time images of the working state of the micro-nozzle, the data processing unit receives the real-time images and processes the data, identifies whether the current micro-nozzle is blocked, and transmits the processing results to the remote control unit. The remote control unit sends abnormal data to the intelligent monitoring terminal to remind the workers.

[0004] Machine vision technology has the advantages of non-contact, intuitive display and strong real-time, so the melt glue state of the injection molding machine nozzle can be monitored through machine vision technology, but in the above related technology, only whether the nozzle is blocked can be monitored, and when there are other defects in the melt glue, such as flow marks, bubbles and agglomerates, effective monitoring cannot be performed, and the needs of the injection molding machine during operation cannot be met. SUMMARY

[0005] In order to solve the problem that the existing technology cannot monitor multiple types of defects in the melt glue, the present application provides a melt glue state monitoring method for an injection molding machine nozzle based on machine vision.

[0006] The present application provides a melt glue state monitoring method for an injection molding machine nozzle based on machine vision, which adopts the following technical solution:

[0007] An original image of the melt glue at a nozzle of an injection molding machine is acquired, and the original image is preprocessed to obtain a grayscale image and a melt glue region, a contour complexity of the melt glue region is calculated, the contour complexity being a ratio of a total length of edge lines in the melt glue region to a circumference of a preset circle, an area of the preset circle being an area of the melt glue region;

[0008] The melt glue region is mapped to the grayscale image, and an energy and an entropy value of the melt glue region are calculated; based on the grayscale images of a current frame and a previous frame, a flow velocity fluctuation coefficient of the melt glue region is calculated, the flow velocity fluctuation coefficient being positively correlated with fluctuation of a flow velocity module length of a pixel point;

[0009] A feature vector is constructed using the contour complexity, the energy, the entropy value, and the flow velocity fluctuation coefficient, and the feature vector is input to a preset prediction model to obtain a prediction result of the melt glue state.

[0010] By constructing a multi-dimensional feature vector including the contour complexity, the energy, the entropy value, and the flow velocity fluctuation coefficient, comprehensive and comprehensive monitoring of the melt glue state of the nozzle of the injection molding machine is realized, the accuracy of the monitoring result is improved, and by fusing three features of morphology, texture, and flowability, various complex defects such as agglomerates, bubbles, and flow marks can be more accurately distinguished, and the accuracy of the melt glue state anomaly detection is significantly improved.

[0011] Preferably, the method for preprocessing the original image to obtain the melt glue region comprises: filtering the original image using a median filtering algorithm; performing enhancement processing on the filtered original image to obtain an enhanced image, and converting the enhanced image into a grayscale image; performing binarization processing on the grayscale image using an Otsu threshold algorithm to obtain a binary image, and taking a white region in the binary image as the melt glue region.

[0012] By using median filtering, image enhancement, and Otsu thresholding, compared with not preprocessing or using a single processing method, random noise in the original image can be more effectively removed, uneven illumination problems can be improved, and the robustness and stability of the melt glue monitoring are improved.

[0013] Preferably, the expression of the contour complexity is:

[0014]

[0015] In the formula, C represents the contour complexity of the melt glue region, L represents the total length of the edge lines in the melt glue region, and A represents the area of the melt glue region.

[0016] The morphology difference is accurately quantified by the above formula, and the sensitivity and reliability of the detection of specific morphological defects are improved.

[0017] Preferably, the energy and entropy value of the melt glue region is calculated by: mapping the melt glue region to a gray scale image, constructing a gray scale co-occurrence matrix about the melt glue region, calculating the energy of the gray scale co-occurrence matrix by using an energy calculation formula, and calculating the entropy value of the gray scale co-occurrence matrix by using an entropy value calculation formula.

[0018] The gray scale co-occurrence matrix can capture the spatial distribution relationship between pixels, thereby more deeply revealing the texture information inside the melt glue, and the energy value and the entropy value respectively quantify the uniformity and complexity of the texture.

[0019] Preferably, before calculating the flow velocity fluctuation coefficient of the melt glue region, the method further comprises: obtaining a current frame gray scale image and a previous frame gray scale image, constructing a window region in the previous frame gray scale image, and calculating the flow velocity module length of each pixel point in the window region by using a Lucas-Kanade sparse optical flow algorithm.

[0020] By introducing dynamic information in the time dimension, by tracking the motion of pixels between frames, the flow state of the melt glue is directly quantified, so that the monitoring system can analyze from two dimensions of static and dynamic, and the accuracy of evaluating the flow stability of the melt glue is improved.

[0021] Preferably, the expression of the fluctuation coefficient is:

[0022]

[0023] Wherein, F represents the flow velocity fluctuation coefficient of the melt glue region of the current frame gray scale image, K is the number of pixel points in the window region, is the flow velocity module length of the kth pixel point in the window region, is the average flow velocity module length of the pixel points in the window region.

[0024] By calculating the dispersion degree of the flow velocity module length of each pixel point in the window region, the complex flow field dynamic change is quantified into a single and clear index. An objective and repeatable evaluation basis is provided for monitoring the state of the melt glue.

[0025] Preferably, the expression of the fluctuation coefficient is:

[0026]

[0027] Wherein, F represents the flow velocity fluctuation coefficient of the melt glue region of the current frame gray scale image, K is the number of pixel points in the window region, is the flow velocity module length of the kth pixel point in the window region, is the average flow velocity module length of the pixel points in the window region, represents the difference value of the contour complexity of the melt glue region in the current frame and the previous frame gray scale image.

[0028] The profile complexity difference is added on the basis of the original, compared with considering the two alone, the complex defects such as bubble mixing which can cause flow rate and morphology to change sharply at the same time can be captured more effectively, the detection ability of specific abnormal events is enhanced through feature fusion, and the judgment is more stable.

[0029] Preferably, the acquisition method of the prediction model is: constructing a historical gray image set by using multiple historical gray images, setting a corresponding label for each historical gray image, the label ; in the formula, 、 、 、 、 indicate the probability of the molten adhesive area being normal, lump, bubble, flow mark and flow break; a BP neural network model is constructed, the BP neural network model is trained by using the historical gray image set, the training is stopped when the value of the loss function is less than a preset loss threshold or the iteration number is reached, and the prediction model is obtained.

[0030] Preferably, the BP neural network model adopts a multi-classification cross-entropy loss function during training.

[0031] By using the multi-classification cross-entropy loss function for model training, the multi-classification cross-entropy loss function can more effectively guide the model to learn how to distinguish the normal, lump, bubble, flow mark and flow break and the like, optimize the learning goal of the model, and thus improve the classification accuracy and distinguish degree of various defect types.

[0032] Preferably, the original image after filtering is subjected to enhancement processing by using a Retinex algorithm to obtain an enhanced image.

[0033] The present application has the following technical effects:

[0034] By combining the profile complexity representing the molten adhesive morphology, the energy and entropy value representing the internal texture, and the flow rate fluctuation coefficient representing the dynamic stability, a comprehensive state description is formed, and then the neural network is used for intelligent analysis and accurate classification of these comprehensive features, so that the comprehensive and accurate monitoring of various types of defects such as lumps, bubbles and flow marks is realized. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a flow chart of the molten adhesive state monitoring method of the injection molding machine nozzle of the present application based on machine vision.

[0036] Figure 2 is a principle diagram for calculating the perimeter of the edge line of the molten adhesive area. DETAILED DESCRIPTION

[0037] Clearly, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort are within the protection scope of the present application.

[0038] The embodiment of the present application discloses a machine vision-based injection molding machine nozzle melt glue state monitoring method, referring to Figure 1 , comprising the following steps, specifically as follows:

[0039] S1: Obtain the original image of the melt glue at the injection molding machine nozzle, and pre-process the original image.

[0040] The original image of the melt glue is collected by the industrial camera through the heat-resistant observation window, the collection interval is 20ms, the region of interest of the melt glue flow is obtained by cropping the original image, the redundant background is eliminated, and the median filter algorithm is used to filter the original image to achieve the effect of denoising the original image. The original image after filtering is enhanced by using the Retinex algorithm to obtain an enhanced image, which reduces the interference caused by reflected light, converts the enhanced image into a gray-scale image, and uses the Otsu threshold algorithm to process the gray-scale image to obtain a binary image. It can be understood that the white area in the binary image is the melt glue area.

[0041] S2: Calculate the contour complexity of the melt glue area.

[0042] In the binary image, a rectangular coordinate system is constructed with any pixel point as the origin, the horizontal axis as the x-axis, and the vertical axis as the y-axis. Then, the Canny operator is used to extract the edge line of the melt glue area, which is one or more. It can be understood that when the melt glue area contains an edge line, the edge line is multiple, and when the melt glue area does not contain an edge line, the edge line is one. The perimeter of the melt glue area edge line is calculated, and the expression is:

[0043]

[0044] In the formula, q represents the perimeter of the melt glue area edge line, , represents the horizontal coordinate and vertical coordinate of the edge pixel point i, , represents the horizontal coordinate and vertical coordinate of the edge pixel point i+1, i represents the index of the edge pixel point, and N represents the total number of edge pixel points. It should be noted here that is the starting point of the melt glue area edge line, is the end point of the melt glue area edge line, and since the melt glue area edge line is a closed curve, the point is the starting point of the melt glue area edge line, The perimeters of the edge lines in the molten adhesive region are calculated by using the following formula: , , the specific principle is shown in Figure 2 .

[0045] The perimeter of one of the edge lines can be calculated according to the above formula, and the perimeters of all the edge lines in the molten adhesive region are further calculated, and the total length of the edge lines is obtained by summing the perimeters of all the edge lines.

[0046] The area A of the molten adhesive region is calculated by using the contourArea function, and it is noted that if the molten adhesive region contains bubbles, the area A includes the area of the bubbles.

[0047] The expression of the contour complexity is as follows:

[0048]

[0049] In the formula, C represents the contour complexity of the molten adhesive region, L represents the total length of the edge lines in the molten adhesive region, and A represents the area of the molten adhesive region. The contour complexity is normalized by using a dynamic normalization algorithm.

[0050] A circle is constructed such that the area of the circle is equal to the area of the molten adhesive region, and the perimeter and area of the circle are substituted into the above formula. It can be known that the contour complexity of the circle is 1, and the contour complexity of the molten adhesive region can also be understood as the ratio of the total length of the edge lines in the molten adhesive region to the perimeter of the circle.

[0051] The shape of the molten adhesive region is approximately circular. When the molten adhesive region contains bubbles, the extracted edge lines are more, which leads to the increase of the total length L of the edge lines, and further leads to the significant increase of the contour complexity. Therefore, the contour complexity can be used to preliminarily determine whether the molten adhesive region contains bubbles, that is, when the contour complexity is small, it indicates that the molten adhesive region is normal and does not contain bubbles; when the contour complexity is large, it indicates that the molten adhesive region is more likely to contain bubbles.

[0052] S3: Calculate the energy and entropy values of the molten adhesive region.

[0053] The molten adhesive region is mapped into a gray-scale image, and a gray-scale co-occurrence matrix about the molten adhesive region is constructed. The energy and entropy values of the gray-scale co-occurrence matrix are calculated. The energy reflects the uniformity of the gray-scale distribution of the image, and the greater the energy, the more uniform the gray-scale distribution of the corresponding molten adhesive region in the gray-scale image, and the smaller the energy, the more chaotic the gray-scale distribution of the corresponding molten adhesive region in the gray-scale image. The entropy value reflects the non-uniformity or complexity of the texture in the corresponding molten adhesive region in the gray-scale image, and the greater the entropy value, the greater the complexity of the texture, and the more complex and disordered the texture, and the smaller the entropy value, the simpler and more regular the texture. The calculation method of the energy and entropy values of the gray-scale co-occurrence matrix is a prior art, and the specific calculation steps are not described herein.

[0054] When the energy value is small, it indicates that the flow mark appears in the melt adhesive area. If the profile complexity is large at this time, it is initially indicated that there are bubbles or stratification in the melt adhesive area, resulting in uneven melt adhesive texture. When the entropy value is large and the energy value is small, it is initially indicated that there are agglomerates in the melt adhesive area.

[0055] S4: Calculate the flow velocity fluctuation coefficient of the melt adhesive area.

[0056] In one embodiment, the current frame gray image and the last frame gray image are obtained, a 5x5 window region is constructed in the last frame gray image, the flow velocity module length of each pixel point in the window region is calculated by using the Lucas-Kanade sparse optical flow algorithm, and the expression of the fluctuation coefficient is:

[0057]

[0058] Wherein, F represents the flow velocity fluctuation coefficient of the melt adhesive area of the current frame gray image, K is the number of pixel points in the window region, is the flow velocity module length of the kth pixel point in the window region, is the average flow velocity module length of the pixel points in the window region.

[0059] When F is small, it indicates that the flow velocity of the pixel points in the window region is stable. When F is large, it indicates that the flow velocity of the pixel points in the window region is large, indicating that the melt adhesive area has a large possibility of agglomeration, that is, the melt adhesive area has agglomeration, resulting in poor flowability of part of the pixel point area. This calculation method is suitable for scenes with low requirements for melt adhesive defect monitoring. This calculation method is simple, fast in calculation rate and good in stability.

[0060] In another embodiment, the current frame gray image and the last frame gray image are obtained, a 5x5 window region is constructed in the last frame gray image, the flow velocity module length of each pixel point in the window region is calculated by using the Lucas-Kanade sparse optical flow algorithm, and the expression of the fluctuation coefficient is:

[0061]

[0062] Wherein, F represents the flow velocity fluctuation coefficient of the melt adhesive area of the current frame gray image, K is the number of pixel points in the window region, is the flow velocity module length of the kth pixel point in the window region, is the average flow velocity module length of the pixel points in the window region, represents the difference value of the profile complexity of the melt adhesive area in the current frame and the last frame gray image.

[0063] When F is small, it indicates that the flow speed of the pixel points in the window region is relatively stable; when F is large, it indicates that the flow speed of the pixel points in the window region is greatly different, indicating that the possibility of the occurrence of agglomeration in the molten adhesive region is large, that is, the occurrence of agglomeration in the molten adhesive region leads to poor fluidity of part of the pixel point region, and when F is large, air may also be mixed into the injection molding machine nozzle, and the air expands due to heating, leading to an increase in the flow speed of part of the molten adhesive, that is, the performance of The values of the two items increase at the same time, at which time it indicates that there are agglomeration and / or bubble defects in the molten adhesive. This calculation method is suitable for scenarios with high requirements for molten adhesive defect monitoring, and the calculation method improves the accuracy of the molten adhesive monitoring result.

[0064] S5: Construct a feature vector to determine whether the molten adhesive state is abnormal.

[0065] A feature vector is constructed by using the calculated contour complexity, energy, entropy value and fluctuation coefficient, the feature vector G=[C, E, H, F], C represents the contour complexity of the molten adhesive region, E represents the energy of the molten adhesive region, H represents the entropy value of the molten adhesive region, and F represents the flow speed fluctuation coefficient of the molten adhesive region. It can be understood that each gray image corresponds to a feature vector.

[0066] A historical gray image set is constructed by using a plurality of historical gray images, a corresponding label is artificially set for each historical gray image, the label ; in the formula, , , , , represent the probabilities that the molten adhesive region is normal, agglomeration, bubble, flow mark and flow interruption, a BP neural network model is constructed, the BP neural network model is trained by using the historical gray image set to obtain a prediction model, the feature vector of the current frame image is obtained in real time and input into the prediction model, a prediction result is obtained, and it is determined whether the molten adhesive state is abnormal. When training the BP neural network model, a multi-classification cross-entropy loss function is used, and when the value of the loss function is less than a preset loss threshold value or the number of iterations is reached, the training is stopped.

[0067] For example, if the prediction result is , wherein 0.8 is greater than a preset standard threshold value 0.75, it indicates that there is a flow mark defect in the molten adhesive region; if the prediction result is , wherein 0.8 is greater than a preset standard threshold value 0.75, it indicates that there are a bubble defect and a flow mark defect in the molten adhesive region.

[0068] The above are preferred embodiments of the present application, and do not limit the protection scope of the present application, so that: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for monitoring the melt condition of an injection molding machine nozzle based on machine vision, characterized in that, The method comprises the steps of: obtaining an original image of the melt glue at a nozzle of an injection molding machine, and pre-processing the original image to obtain a gray image and a melt glue region, calculating a contour complexity of the melt glue region, the contour complexity being a ratio of a total length of edge lines in the melt glue region to a circumference of a preset circle, and an area of the preset circle being an area of the melt glue region; mapping the melt glue region to the gray image, calculating an energy and an entropy value of the melt glue region, and calculating a flow velocity fluctuation coefficient of the melt glue region based on the gray images of the current frame and a previous frame, the flow velocity fluctuation coefficient being positively correlated with a fluctuation of a flow velocity module length of a pixel point; constructing a feature vector by using the contour complexity, the energy, the entropy value and the flow velocity fluctuation coefficient, and inputting the feature vector into a preset prediction model to obtain a prediction result of the melt glue state; The acquisition method of the prediction model is: a plurality of historical gray images are used to construct a historical gray image set, a corresponding label is set for each historical gray image, the label ; in the formula, , , , , represent the probabilities that the melt adhesive area is normal, lump, bubble, flow mark and flow break; a BP neural network model is constructed, the historical gray image set is used to train the BP neural network model, the training is stopped when the value of the loss function is less than a preset loss threshold or the number of iterations is reached, and the prediction model is obtained.

2. The machine vision-based monitoring method of a glue state of a nozzle of an injection molding machine according to claim 1, characterized in that, the method for pre-processing the original image to obtain the melt glue region comprises: filtering the original image by using a median filtering algorithm; performing enhancement processing on the filtered original image to obtain an enhanced image, and converting the enhanced image into a gray image; performing binarization processing on the gray image by using an Otsu threshold algorithm to obtain a binary image, and taking a white region in the binary image as the melt glue region.

3. The machine vision-based monitoring of a glue-up condition of a nozzle of an injection molding machine method according to claim 1, wherein, The expression of the contour complexity is: ； In the formula, C represents the contour complexity of the melt glue region, L represents the total length of the edge lines in the melt glue region, and A represents the area of the melt glue region.

4. The machine vision-based monitoring of a glue-up condition of a nozzle of an injection molding machine method according to claim 1, wherein, The calculation method of the energy and the entropy value of the melt glue region is: mapping the melt glue region to the gray image, constructing a gray level co-occurrence matrix about the melt glue region, calculating the energy of the gray level co-occurrence matrix by using an energy calculation formula, and obtaining the entropy value of the gray level co-occurrence matrix by using an entropy value calculation formula.

5. The machine vision-based monitoring of a glue-up condition of a nozzle of an injection molding machine method according to claim 1, wherein, Before calculating the flow velocity fluctuation coefficient of the melt glue region, the method further comprises: obtaining a current frame gray image and a previous frame gray image, constructing a window region in the previous frame gray image, and calculating a flow velocity module length of each pixel point in the window region by using a Lucas-Kanade sparse optical flow algorithm.

6. The machine vision-based monitoring method of an injection molding machine nozzle melt condition of claim 5, wherein, The expression of the fluctuation coefficient is: ； Wherein, F represents the flow velocity fluctuation coefficient of the current frame gray image glue area, K is the number of pixel points in the window area, is the flow velocity module length of the kth pixel point in the window area, is the average flow velocity module length of the pixel points in the window area.

7. The machine vision-based monitoring method of an injection molding machine nozzle melt condition of claim 5, wherein, The expression of the fluctuation coefficient is: ； wherein F represents the flow velocity fluctuation coefficient of the current frame of the glue area of the gray image, K is the number of pixel points in the window region, is the flow velocity module length of the kth pixel point in the window region, is the average flow velocity module length of the pixel points in the window region, represents the difference value of the outline complexity of the glue area in the current frame and the previous frame of the gray image.

8. The machine vision-based monitoring method of a glue state of an injection molding machine nozzle according to claim 1, characterized in that, The BP neural network model adopts a multi-classification cross-entropy loss function during training.

9. The machine vision-based monitoring method of an injection molding machine nozzle's molten gel state as defined in claim 2, wherein, The Retinex algorithm is used to perform enhancement processing on the filtered original image to obtain the enhanced image.

Images