Method for monitoring glue melting state of nozzle of injection molding machine based on machine vision

Abstract

The invention relates to the field of melt state monitoring, in particular to an injection molding machine nozzle melt state monitoring method based on machine vision, and the method comprises the steps: obtaining an original image of melt at an injection molding machine nozzle, carrying out the preprocessing of the original image, obtaining a gray image and a melt region, and calculating the contour complexity representing the melt form based on the melt region; mapping the melt glue area to the gray level image, and calculating the energy and entropy value of the melt glue area; based on the gray level images of the current frame and the previous frame, the flow velocity fluctuation coefficient of the melt glue area is calculated, and the flow velocity fluctuation coefficient is in positive correlation with the fluctuation of the flow velocity die length of the pixel points; and 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 state. According to the invention, various types of defects such as caking, bubbles and flow marks can be monitored.

Description

Technical Field

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

[0002] Injection molding involves injecting molten plastic into a mold cavity through a nozzle, and then cooling the plastic part in the mold cavity before forming it. During the injection molding process, the state of the molten adhesive directly affects the quality of the product, including the uniformity, fluidity, and bubble content of the molten adhesive. For example, when other incompatible polymers are mixed into the molten adhesive, the plastic film will peel; when the molten adhesive temperature is too low and the fluidity is insufficient, it will cause problems such as flow marks and under-injection. Existing molten adhesive state monitoring mainly relies on contact monitoring methods such as pressure sensors and temperature sensors. However, existing monitoring methods have the following defects: 1. Sensors are easily corroded by high-temperature molten adhesive, have a short service life, and have high maintenance costs; 2. Sensors can only indirectly reflect the physical parameters of the molten adhesive, and the molten adhesive state cannot be judged intuitively based solely on the worker's experience; 3. The response is delayed, and real-time adjustments cannot be achieved, resulting in the production of batches of scrapped products.

[0003] The Chinese patent application document with publication number CN118658116A discloses a micro-nozzle blockage monitoring system based on image processing, including: an image acquisition unit, a data processing unit, a remote control unit and an intelligent monitoring terminal; wherein, the image acquisition unit obtains a real-time image of the working status of the micro-nozzle, the data processing unit receives the real-time image and performs data processing, identifies whether the current micro-nozzle is blocked, and transmits the processing results to the remote control unit, which sends the abnormal data to the intelligent monitoring terminal to alert the staff.

[0004] Machine vision technology has the advantages of being non-contact, intuitively displayable, and highly real-time. Therefore, it is possible to monitor the melt state of the injection molding machine nozzle through machine vision technology. However, the above-mentioned related technologies can only monitor whether the nozzle is clogged. When other defects in the melt exist, such as flow marks, bubbles, and agglomerations, effective monitoring cannot be performed, and it cannot meet the needs of the injection molding machine during operation. Summary of the Invention

[0005] In order to solve the problem in the prior art that various types of defects in molten glue cannot be monitored, the present invention provides a method for monitoring the molten glue status of an injection molding machine nozzle based on machine vision.

[0006] The present invention provides a method for monitoring the melt state of an injection molding machine nozzle based on machine vision, which adopts the following technical solutions: Obtain an original image of the melt at the injection molding machine nozzle and preprocess the original image to obtain a grayscale image and the melt area. Calculate the contour complexity of the melt area. The contour complexity is the ratio of the total length of the edge lines in the melt area to the circumference of a preset circle. The area of ​​the preset circle is the area of ​​the melt area. The melt region is mapped to a grayscale image, and the energy and entropy values ​​of the melt region are calculated. Based on the grayscale images of the current frame and the previous frame, the velocity fluctuation coefficient of the melt region is calculated. The velocity fluctuation coefficient is positively correlated with the fluctuation of the velocity modulus of the pixel point. A feature vector is constructed using contour complexity, energy, entropy value and flow rate fluctuation coefficient, and the feature vector is input into a preset prediction model to obtain a prediction result of the melt state.

[0007] By constructing a multi-dimensional feature vector that includes contour complexity, energy, entropy value and flow rate fluctuation coefficient, comprehensive and integrated monitoring of the melt state of the injection molding machine nozzle is achieved, improving the accuracy of the monitoring results. Moreover, by integrating the three characteristics of morphology, texture and fluidity, it is possible to more accurately distinguish various complex defects such as agglomerations, bubbles, flow marks, etc., significantly improving the accuracy of melt state anomaly detection.

[0008] Preferably, the method of preprocessing the original image to obtain the melt area includes: filtering the original image using a median filtering algorithm; enhancing the filtered original image to obtain an enhanced image, and converting the enhanced image into a grayscale image; binarizing the grayscale image using the Otsu threshold algorithm to obtain a binary image, and using the white area in the binary image as the melt area.

[0009] By adopting median filtering, image enhancement and Otsu threshold method, compared with no preprocessing or using a single processing method, it can more effectively remove random noise in the original image and improve problems such as uneven lighting, thereby improving the robustness and stability of melt monitoring.

[0010] Preferably, the expression of contour complexity is:

[0011] Where C represents the contour complexity of the melt region, L represents the total length of the edge lines in the melt region, and A represents the area of ​​the melt region.

[0012] The above formula is used to accurately quantify the morphological differences, thereby improving the sensitivity and reliability of detecting specific morphological defects.

[0013] Preferably, the method for calculating the energy and entropy value of the melt area is: mapping the melt area to a grayscale image, constructing a grayscale co-occurrence matrix about the melt area, using the energy calculation formula to calculate the energy of the grayscale co-occurrence matrix, and using the entropy calculation formula to obtain the entropy value of the grayscale co-occurrence matrix.

[0014] The gray-level co-occurrence matrix can capture the spatial distribution relationship between pixels, thereby revealing the texture information inside the melt more deeply. The energy value and entropy value quantify the uniformity and complexity of the texture respectively.

[0015] Preferably, before calculating the velocity fluctuation coefficient of the melt area, it also includes: obtaining the current frame grayscale image and the previous frame grayscale image, constructing a window area in the previous frame grayscale image, and using the Lucas-Kanade sparse optical flow algorithm to calculate the velocity modulus of each pixel point in the window area.

[0016] By introducing dynamic information in the time dimension and tracking the movement of pixels between frames, direct quantification of the melt flow state is achieved, enabling the monitoring system to analyze from both static and dynamic dimensions, thereby improving the accuracy of evaluating the melt flow stability.

[0017] Preferably, the expression of the fluctuation coefficient is:

[0018] Among them, F represents the velocity fluctuation coefficient of the melt area in the current frame grayscale image, K is the number of pixels in the window area, is the velocity modulus of the k-th pixel in the window area, is the average velocity modulus of the pixels within the window area.

[0019] By calculating the discrete degree of velocity modulus at each pixel within the window area, the complex dynamic changes of the flow field are quantified into a single clear indicator, providing an objective and repeatable evaluation basis for monitoring the melt state.

[0020] Preferably, the expression of the fluctuation coefficient is:

[0021] Among them, F represents the velocity fluctuation coefficient of the melt area in the current frame grayscale image, K is the number of pixels in the window area, is the velocity modulus of the k-th pixel in the window area, is the average velocity modulus of the pixels in the window area, Indicates the difference in contour complexity of the melt area in the current frame and the previous frame of grayscale image.

[0022] The contour complexity difference is added on the original basis. Compared with considering the two separately, it can more effectively capture complex defects such as bubble mixing that cause drastic changes in flow velocity and shape at the same time. Through feature fusion, the detection ability of specific abnormal events is enhanced, making the judgment more robust.

[0023] Preferably, the method for obtaining the prediction model is: using multiple historical grayscale images to construct a historical grayscale image set, manually setting a corresponding label for each historical grayscale image, and label Where, 、 、 、 、 Indicates the probability of the melt area being normal, agglomerated, bubbling, flow-marked, or interrupted; constructs a BP neural network model, and uses a historical grayscale image set to train the BP neural network model. When the value of the loss function is less than the preset loss threshold or the number of iterations is reached, the training is stopped to obtain a prediction model.

[0024] Preferably, the BP neural network model adopts a multi-classification cross entropy loss function during training.

[0025] By adopting 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 various states such as normal, agglomeration, bubbles, flow marks, and interruption of flow, optimize the learning objectives of the model, and thus improve the classification accuracy and discrimination of various defect types.

[0026] Preferably, the filtered original image is enhanced using a Retinex algorithm to obtain an enhanced image.

[0027] The present invention has the following technical effects: By combining the contour complexity that characterizes the molten plastic morphology, the energy and entropy values ​​that characterize the internal texture, and the flow rate fluctuation coefficient that characterizes dynamic stability, a comprehensive state description is formed. Then, a neural network is used to intelligently analyze and accurately classify these comprehensive features, thereby achieving comprehensive and accurate monitoring of various types of defects such as agglomerates, bubbles, and flow marks. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 The present invention is a flow chart of a method for monitoring the melt state of an injection molding machine nozzle based on machine vision.

[0029] Figure 2 It is a schematic diagram of the principle of calculating the perimeter of the edge line of the melt area of ​​the present invention. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0031] The embodiment of the present invention discloses a method for monitoring the melt state of an injection molding machine nozzle based on machine vision, referring to Figure 1 , including the following steps, as follows: S1: Obtain the original image of the molten glue at the injection molding machine nozzle and preprocess the original image.

[0032] An industrial camera was used to capture raw images of the melt through a heat-resistant observation window at a 20ms interval. The raw images were cropped to identify the region of interest of the melt flow, removing redundant background. The raw images were then filtered using a median filter to denoise them. The filtered raw images were then enhanced using the Retinex algorithm to reduce interference from reflected light. The enhanced images were then converted to grayscale and binarized using the Otsu threshold algorithm to create a binary image. It is understood that the white areas in the binary image represent the melt flow.

[0033] S2: Calculate the contour complexity of the melt area.

[0034] In a binary image, a rectangular coordinate system is constructed with any pixel as the origin, with the horizontal axis being the x-axis and the vertical axis being the y-axis. Then, the Canny operator is used to extract the edge lines of the melt region. The edge lines are one or more. It can be understood that when the melt region contains edge lines, there are multiple edge lines, and when the melt region does not contain edge lines, there is only one edge line. The perimeter of the edge line of the melt region is calculated as follows:

[0035] Where q represents the perimeter of the edge line of the melt area, 、 Represents the horizontal and vertical coordinates of edge pixel point i, 、 represents the horizontal and vertical coordinates of edge pixel i+1, i represents the index of edge pixel, and N represents the total number of edge pixels. is the starting point of the edge line of the melt area, is the end point of the edge line of the melt area. Since the edge line of the melt area is a closed curve, point with dot Therefore, when calculating the perimeter of the edge line of the melt area, let , , see Figure 2 shown.

[0036] According to the above formula, the perimeter of one edge line can be calculated, and the perimeters of all edge lines in the melt area can be further obtained. The perimeters of all edge lines are summed to obtain the total length of the edge lines.

[0037] The contourArea function is used to calculate the area A of the melt region. It should be noted here that if the melt region contains bubbles, the area A includes the area of ​​the bubbles.

[0038] The expression of contour complexity is:

[0039] Where C represents the contour complexity of the melt region, L represents the total length of the edge lines within the melt region, and A represents the area of ​​the melt region. The contour complexity is normalized using a dynamic normalization algorithm.

[0040] Construct a circle so that its area is equal to the area of ​​the melt region. Substituting the circumference and area of ​​the circle into the above formula, we can see that the contour complexity of the circle is 1. The contour complexity of the melt region can also be understood as the ratio of the total length of the edge lines within the melt region to the circumference of the circle.

[0041] The shape of the melt area is approximately circular. When there are bubbles in the melt area, more edge lines are extracted, which increases the total length L of the edge lines and further significantly increases the contour complexity. Therefore, the contour complexity can be used to preliminarily determine whether there are bubbles in the melt area. That is, when the contour complexity is small, it indicates that the melt area is normal and there are no bubbles; when the contour complexity is large, it indicates that there is a greater possibility of bubbles in the melt area.

[0042] S3: Calculate the energy and entropy of the melt region.

[0043] The melt region is mapped to the grayscale image, and a grayscale co-occurrence matrix for the melt region is constructed. The energy and entropy of the grayscale co-occurrence matrix are calculated. The energy reflects the uniformity of the grayscale distribution of the image. A larger energy indicates a uniform grayscale distribution in the corresponding melt region in the grayscale image, while a smaller energy indicates a more chaotic grayscale distribution in the corresponding melt region in the grayscale image. The entropy reflects the non-uniformity or complexity of the texture in the corresponding melt region in the grayscale image. A larger entropy indicates a more complex and disordered texture, while a smaller entropy indicates a simple and regular texture. The method for calculating the energy and entropy of the grayscale co-occurrence matrix is ​​prior art, and the specific calculation steps are not repeated here.

[0044] When the energy value is small, it indicates that flow marks appear in the melt area. If the contour complexity is large at this time, it preliminarily indicates that there are bubbles or stratification in the melt area, resulting in uneven melt texture. When the entropy value is large and the energy value is small, it preliminarily indicates that there is agglomeration in the melt area.

[0045] S4: Calculate the velocity fluctuation coefficient of the melt area.

[0046] In one embodiment, a current frame grayscale image and a previous frame grayscale image are obtained, a 5×5 window area is constructed in the previous frame grayscale image, and the Lucas-Kanade sparse optical flow algorithm is used to calculate the velocity modulus of each pixel in the window area. The expression of the fluctuation coefficient is:

[0047] Among them, F represents the velocity fluctuation coefficient of the melt area in the current frame grayscale image, K is the number of pixels in the window area, is the velocity modulus of the k-th pixel in the window area, is the average velocity modulus of the pixels within the window area.

[0048] When F is small, it indicates that the flow velocity of the pixels in the window area is relatively stable. When F is large, it indicates that the flow velocity of the pixels in the window area varies greatly, indicating that there is a high possibility of agglomeration in the molten adhesive area. That is, agglomeration in the molten adhesive area causes poor fluidity in some pixel areas. This calculation method is suitable for scenarios with low requirements for molten adhesive defect monitoring. This calculation method is simple, has a fast calculation rate, and good stability.

[0049] In another embodiment, a current grayscale image and a previous grayscale image are obtained, a 5×5 window area is constructed in the previous grayscale image, and the Lucas-Kanade sparse optical flow algorithm is used to calculate the velocity modulus of each pixel in the window area. The expression of the fluctuation coefficient is:

[0050] Among them, F represents the velocity fluctuation coefficient of the melt area in the current frame grayscale image, K is the number of pixels in the window area, is the velocity modulus of the k-th pixel in the window area, is the average velocity modulus of the pixels in the window area, Indicates the difference in contour complexity of the melt area in the current frame and the previous frame of grayscale image.

[0051] When F is small, it indicates that the flow velocity of the pixels in the window area is relatively stable; when F is large, it indicates that the flow velocity of the pixels in the window area varies greatly, indicating that the possibility of agglomeration in the molten glue area is high, that is, the agglomeration in the molten glue area causes the fluidity of some pixel areas to be poor. Moreover, when F is large, it is possible that air is mixed into the nozzle of the injection molding machine, and the air expands due to heat, causing the flow velocity of some molten glue to increase, which is manifested as If the values ​​of both indicators increase simultaneously, it indicates the presence of agglomerates and / or bubbles in the melt. This calculation method is suitable for scenarios with high requirements for melt defect monitoring and improves the accuracy of melt monitoring results.

[0052] S5: Construct a feature vector to determine whether the melt state is abnormal.

[0053] The calculated contour complexity, energy, entropy, and fluctuation coefficient are used to construct a feature vector: G = [C, E, H, F], where C represents the contour complexity of the melt region, E represents the energy of the melt region, H represents the entropy of the melt region, and F represents the velocity fluctuation coefficient of the melt region. It can be understood that each grayscale image corresponds to a feature vector.

[0054] Use multiple historical grayscale images to construct a historical grayscale image set, and set corresponding labels for each historical grayscale image. Where, 、 、 、 、 The probability of the melt being normal, agglomerated, bubbling, with flow marks, or interrupted is represented by constructing a BP neural network model. This model is trained using a historical grayscale image set to generate a prediction model. The feature vector of the current frame image is acquired in real time and input into the prediction model to obtain a prediction result, thereby determining whether the melt is abnormal. The BP neural network model uses a multi-classification cross-entropy loss function during training. Training stops when the loss function value falls below a preset loss threshold or when the number of iterations is reached.

[0055] For example, if the prediction result is , where 0.8 is greater than the preset standard threshold of 0.75, indicating that there are flow mark defects in the melt area; if the prediction result is , where 0.8 is greater than the preset standard threshold of 0.75, indicating that there are both bubble defects and flow mark defects in the melt area.

[0056] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for monitoring the melt state of an injection molding machine nozzle based on machine vision, characterized in that: Including steps: Obtain an original image of the melt at the injection molding machine nozzle and preprocess the original image to obtain a grayscale image and the melt area. Calculate the contour complexity of the melt area. The contour complexity is the ratio of the total length of the edge lines in the melt area to the circumference of a preset circle. The area of ​​the preset circle is the area of ​​the melt area. The melt region is mapped to a grayscale image, and the energy and entropy values ​​of the melt region are calculated. Based on the grayscale images of the current frame and the previous frame, the velocity fluctuation coefficient of the melt region is calculated. The velocity fluctuation coefficient is positively correlated with the fluctuation of the velocity modulus of the pixel point. A feature vector is constructed using contour complexity, energy, entropy value and flow rate fluctuation coefficient, and the feature vector is input into a preset prediction model to obtain a prediction result of the melt state.

2. The method for monitoring the melt state of an injection molding machine nozzle based on machine vision according to claim 1, characterized in that: The method for preprocessing the original image to obtain the melt area includes: filtering the original image using a median filtering algorithm; enhancing the filtered original image to obtain an enhanced image, and converting the enhanced image into a grayscale image; binarizing the grayscale image using an Otsu threshold algorithm to obtain a binary image, and using the white area in the binary image as the melt area.

3. The method for monitoring the melt state of an injection molding machine nozzle based on machine vision according to claim 1, characterized in that: The expression of contour complexity is: ； Where C represents the contour complexity of the melt region, L represents the total length of the edge lines in the melt region, and A represents the area of ​​the melt region.

4. The method for monitoring the melt state of an injection molding machine nozzle based on machine vision according to claim 1, characterized in that: The energy and entropy values ​​of the melt area are calculated as follows: the melt area is mapped to the grayscale image, a grayscale co-occurrence matrix of the melt area is constructed, the energy of the grayscale co-occurrence matrix is ​​calculated using the energy calculation formula, and the entropy value of the grayscale co-occurrence matrix is ​​obtained using the entropy calculation formula.

5. The method for monitoring the melt state of an injection molding machine nozzle based on machine vision according to claim 1, characterized in that: Before calculating the velocity fluctuation coefficient of the melt area, the method also includes: obtaining a current frame grayscale image and a previous frame grayscale image, constructing a window area in the previous frame grayscale image, and using the Lucas-Kanade sparse optical flow algorithm to calculate the velocity modulus of each pixel point in the window area.

6. The method for monitoring the melt state of an injection molding machine nozzle based on machine vision according to claim 5, characterized in that: The expression of the volatility coefficient is: ； Among them, F represents the velocity fluctuation coefficient of the melt area in the current frame grayscale image, K is the number of pixels in the window area, is the velocity modulus of the k-th pixel in the window area, is the average velocity modulus of the pixels within the window area.

7. The method for monitoring the melt state of an injection molding machine nozzle based on machine vision according to claim 5, characterized in that: The expression of the volatility coefficient is: ； Among them, F represents the velocity fluctuation coefficient of the melt area in the current frame grayscale image, K is the number of pixels in the window area, is the velocity modulus of the k-th pixel in the window area, is the average velocity modulus of the pixels in the window area, Indicates the difference in contour complexity of the melt area in the current frame and the previous frame of grayscale image.

8. The method for monitoring the melt state of an injection molding machine nozzle based on machine vision according to claim 1, characterized in that: The method for obtaining the prediction model is to use multiple historical grayscale images to construct a historical grayscale image set, and manually set a corresponding label for each historical grayscale image. Where, 、 、 、 、 Indicates the probability of the melt area being normal, agglomerated, bubbling, flow-marked, or interrupted; constructs a BP neural network model, and uses a historical grayscale image set to train the BP neural network model. When the value of the loss function is less than the preset loss threshold or the number of iterations is reached, the training is stopped to obtain a prediction model.

9. The method for monitoring the melt state of an injection molding machine nozzle based on machine vision according to claim 8, characterized in that: The BP neural network model uses a multi-classification cross entropy loss function during training.

10. The method for monitoring the melt state of an injection molding machine nozzle based on machine vision according to claim 2, characterized in that: The Retinex algorithm is used to enhance the filtered original image to obtain an enhanced image.

Images