Method for controlling and detecting oil outlet amount of silica gel

Abstract

The invention discloses a silica gel oil output control detection method, and relates to the field of material science, and the method comprises the following steps: sampling from produced self-oil-seepage silica gel, preparing a plurality of standard sample wafers, carrying out original image acquisition on the surfaces of the standard sample wafers, the method comprises the following steps: receiving an original image, performing gray processing, binarization processing, edge detection and morphological processing on the received original image, then automatically identifying the position, size and area of oil stain in an oil stain distribution image on the surface of a sample, quantitatively calculating the oil seepage quantity and oil seepage rate of each standard sample, and automatically generating a silica gel oil production detection report. Finally, the oil outlet condition of the produced self-oil-seepage silica gel is judged, and the production technology of the self-oil-seepage silica gel is automatically adjusted. The problems that an existing detection method is low in precision, complex in operation and greatly affected by human factors are solved, accurate detection of the oil outlet amount of the self-oil-seepage silica gel and automatic production process adjustment can be achieved, and the production precision and efficiency are improved.

Description

Technical Field

[0001] The invention relates to the field of material science, and in particular to a method for controlling and detecting the oil output of silicone. Background Art

[0002] As an important automotive wiring harness sealing material, self-penetrating silicone is widely used in the process of connecting automotive connector wires and wires. It not only achieves the best sealing effect, but also significantly improves the automation level of the wiring harness process. The control of its oil output has an important impact on the performance and service life of the product. Excessive oil output may cause contamination of the product during use, affecting the appearance and performance of the product; while too little oil output may affect the sealing and adhesion of the product. Therefore, it is particularly important to design an accurate and reliable silicone oil output control detection method.

[0003] In order to control the oil seepage amount and oil seepage rate of self-oiling silicone, the existing technology uses visual inspection and weighing methods to judge the oil seepage amount. However, judging the oil seepage amount by observing the size of oil stains on the surface of silicone with the naked eye is highly subjective and has low precision, and cannot meet high-precision requirements; estimating the oil seepage amount by weighing the weight change of silicone samples in different time periods. Since the proportion of oil seepage to the total weight of the product is very small (1%~7%), this measurement method has large errors and it is difficult to obtain reliable data. Therefore, the existing detection methods generally have the problems of low precision, complex operation, and great influence of human factors. Especially when the amount of oil seepage is small, these methods are difficult to provide accurate detection results, which affects the application effect and product quality of self-oiling silicone.

[0004] Therefore, a method for controlling and detecting the oil output of silicone is needed to solve the above problems. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention discloses a method for controlling the oil output of silicone, which can solve the problems existing in the visual inspection method and the weighing method in the prior art, improve the accuracy and objectivity of the detection, simplify the operating process, reduce the interference of human factors, and provide more reliable and accurate detection results for the production and application of self-oiling silicone.

[0006] The present invention adopts the following technical solutions: A method for controlling and detecting the amount of oil discharged from a silicone rubber comprises the following steps: Step 1: Take samples from the produced self-penetrating oil-free silica gel and prepare a plurality of standard samples. Transfer the prepared plurality of standard samples to a plurality of sample storage containers within 20 minutes and leave them for 72 hours. Place the plurality of sample storage containers in different simulated use environments of the self-penetrating oil-free silica gel. Step 2: Transfer the standard sample that has reached the static period to the detection platform, and the detection platform uses a CMOS sensor camera and an autofocus microscope to collect the original image of the surface of the standard sample, and input the collected original image into the visual inspection system; Step 3: After receiving the collected original image, the visual inspection system performs grayscale processing, binarization processing, edge detection and morphological processing on the received original image to obtain the sample surface oil stain distribution image, and uses a deep image processing algorithm to automatically identify the position, shape and area of ​​the oil stain in the sample surface oil stain distribution image. The deep image processing algorithm extracts and identifies the oil stain features of the sample surface oil stain distribution image by combining a singular value decomposition method and a convolutional neural network; Step 4: quantitatively calculate the oil seepage amount and oil seepage rate of each standard sample according to the extracted oil stain characteristics and the preset oil output calculation model, and summarize the calculation results through the visualization platform to automatically generate a silicone oil output detection report; Step 5. The production management system determines the oil leakage of the produced self-seeping silicone according to the silicone oil leakage detection report and the preset oil leakage threshold and oil penetration rate threshold, and automatically adjusts the production process of the self-seeping silicone based on the determination result.

[0007] A method for controlling and detecting the oil output of silicone according to claim 1, characterized in that, in the step one, the thickness of the standard sample is 2 mm ± 0.2 mm, the length of the standard sample is 50 mm ± 0.2 mm, the width of the standard sample is 50 mm ± 0.2 mm, the sample storage container is a non-oil-absorbing PC gasket, and the detection resolution is 5MP-25MP.

[0008] Furthermore, the self-penetrating oil-free silica gel simulates the use environment including: (1) Normal temperature storage environment: ambient temperature is 23±2℃, humidity is 50±10%; (2) High temperature environment: ambient temperature is 60±2℃; (3) Low temperature environment: ambient temperature is -20±2℃; (4) Hot and humid environment: ambient temperature is 85±2℃, humidity is 85±10%; (5) Ultraviolet light irradiation environment: Use ultraviolet light to simulate sunlight.

[0009] Furthermore, in step 2, the CMOS sensor camera converts the optical signal into an electrical signal through a plurality of photodiodes integrated on the chip, and converts the electrical signal into an image signal through a digital signal processor, and the autofocus microscope automatically adjusts the focus at different observation depths through ranging method and real-time feedback mechanism.

[0010] Furthermore, in the step three, after receiving the collected original image, the visual inspection system converts the color original image into a grayscale original image by converting the pixel color values ​​in the original image into brightness values ​​to simplify the original image. Then, the pixel threshold is preset by the adaptive threshold method, and all pixels in the grayscale original image with grayscale values ​​higher than the pixel threshold are set to white, and all pixels in the grayscale original image with grayscale values ​​lower than the pixel threshold are set to black to complete the conversion of the grayscale image to a black and white image. The edge contour of the oil stain in the black and white image is detected by calculating the gradient of the image grayscale value. Finally, the black and white image is smoothed, filled and denoised through expansion, corrosion, opening and closing operations.

[0011] Furthermore, in step three, the working method of the depth image processing algorithm includes the following steps: S301, extracting principal component image features from the sample surface oil stain distribution image by singular value decomposition method, wherein the singular value decomposition method performs singular value decomposition on the sample surface oil stain distribution image matrix A to obtain matrix U, matrix S and matrix V, and the function expression is: (1) In formula (1), and is an orthogonal matrix, which is used to represent the feature space of the oil stain distribution image on the sample surface. is a diagonal matrix used to represent the singular values ​​of the sample surface oil stain distribution image matrix A, where the matrix , Indicates the ordinal number of singular values ​​contained in the matrix S, Represents the kth singular value contained in the matrix S, sorts the singular values ​​of the sample surface oil stain distribution image matrix A from large to small, selects the matrix U, matrix S and matrix V corresponding to the first k singular values, and combines the feature matrix U, matrix S and matrix V to obtain the principal component representation of the sample surface oil stain distribution image; S302, inputting the main component image features extracted in S301 into a convolutional neural network for multi-scale image feature extraction, wherein the convolutional neural network uses multiple convolution kernels to extract local features of the main component image features, and uses a nonlinear activation function to activate the convolution result, and then uses a pooling operation to downsample the convolved features to capture changes in the position, shape, and area of ​​the oil stains; S303, using a frequency division duplex (FDD) network to perform quality correction on the extracted multi-scale image features; S304: Input the corrected image features into the fully connected layer. The fully connected layer performs linear transformation on different features through the weight matrix and the bias term, and outputs the recognition result of the features.

[0012] Furthermore, in the step S303, the frequency division duplex FDD network uses a convolution operation to obtain the associated quality features of the multi-scale image features, and the function is expressed as: (2) In formula (2), x is the input sample surface oil stain distribution image data, t represents the offset of the extracted multi-scale image features, represents the kernel function of the frequency division duplex FDD network, represents the quality mapping result of the extracted multi-scale image features, Represents the weighted average parameter of the frequency division duplex FDD network, The weight vector representing the image data of oil stain distribution on the sample surface is used to obtain the defect feature vector in the extracted multi-scale image features by calculating the difference between the standard feature quality and the feature quality mapping result. The formula is expressed as: (3) In formula (3), Represents the difference between the standard feature quality and the feature quality mapping result, Represents the average value of all mass elements in the standard characteristic mass, and They respectively represent the length and width of the multi-scale image feature association quality feature matrix. When the difference is greater than the preset difference threshold, it is determined to be a defect feature vector and the defect feature vector is corrected.

[0013] Furthermore, the visualization platform uses interactive charts, heat maps and dashboards to display the detection results and changing trends, and uses a user authentication mechanism to verify the identity of the accessing user. The silicone oil leakage detection report includes the oil leakage amount, oil leakage rate, oil stain size and oil stain number.

[0014] The beneficial effects of the present invention are: 1. The present invention adopts a visual inspection system to realize the automatic collection and processing of the distribution of oil stains on the surface of silicone samples, and through a deep image processing algorithm, it can automatically identify the position, size and area of ​​the oil stains, thereby greatly improving the accuracy and objectivity of the detection.

[0015] 2. The present invention adopts a deep image processing algorithm, combined with a singular value decomposition method and a convolutional neural network, to accurately extract and classify oil stain features. This algorithm can more effectively identify and analyze oil stain images, further improving the accuracy and reliability of detection.

[0016] 3. The present invention automatically collects images and analyzes and processes them through a visual inspection system. The entire process can be completed quickly and efficiently, avoiding the time waste and human errors in traditional manual inspections. The visualization platform summarizes the calculation results and automatically generates a silicone oil detection report, making the detection results more intuitive and easy to understand. This not only improves work efficiency, but also makes the detection results more transparent and traceable.

[0017] 4. The present invention quantitatively calculates the oil seepage and oil seepage rate of the sample by combining the oil stain features extracted by deep image processing and the preset oil output calculation model. This quantitative analysis method makes the oil output of silicone more accurate and scientific, rather than relying on manual judgment or subjective estimation. The oil output of the self-seeping silicone produced is judged according to the preset oil output and oil seepage rate thresholds, and the production process is automatically adjusted based on the judgment results to ensure that the oil seepage of silicone in the production process meets the set standards, reduce the production of unqualified products, and improve production quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the overall method workflow of the present invention; Figure 2 Schematic diagram of the workflow of the depth image processing algorithm in the present invention; Figure 3 Schematic diagram of the architecture of the platform and system used in the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the attached embodiment of the present invention Figure 1 To Attachment Figure 3 , the technical solutions in the embodiments of the present invention are clearly and completely described. 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] A method for controlling and detecting the amount of oil discharged from a silicone rubber, as shown in the attached Figure 1 , including the following steps: Step 1: Sampling and standard sample preparation Randomly select a certain amount of samples from the self-penetrating silicone produced to ensure that the samples are representative and can reflect the oil-discharging characteristics of the entire batch of silicone. The sampling method can be cutting or pressing. Use precision cutting tools to cut the silicone samples into standard samples of specified sizes to ensure the consistency of the size and thickness of the samples to improve the accuracy and reliability of the test results. The size of the standard sample is, for example, 2mm±0.1mm in thickness, 50mm±0.5mm in length, and 50mm±0.5mm in width. After cutting, gently wipe the surface of the sample with a clean cloth or paper to remove impurities or debris that may be attached.

[0021] Quickly transfer multiple prepared standard samples to multiple sample storage containers within 20 minutes. Each sample storage container should have good sealing and proper climate control, and the sample storage container is a non-oil-absorbing PC gasket. Each sample storage container is placed in different simulated use environments, such as high temperature, low temperature, high humidity and other environments, to simulate possible external conditions in actual use. At least 3 samples should be prepared as parallel samples for each environment to improve the reliability of the test results. Let it stand for 72 hours. Do not wipe or clean the sample during the observation period to allow the silicone sample to be fully stable in different environments, ensure that the silicone surface can fully penetrate the oil and stably form oil stains, and simulate its oil discharge in actual use.

[0022] The simulated use environment of self-penetrating oil-free silicone includes: (1) Normal temperature storage environment: ambient temperature is 23±2℃, humidity is 50±10%; (2) High temperature environment: ambient temperature is 60±2℃; (3) Low temperature environment: ambient temperature is -20±2℃; (4) Hot and humid environment: ambient temperature is 85±2℃, humidity is 85±10%; (5) Ultraviolet light irradiation environment: Use ultraviolet light to simulate sunlight.

[0023] Step 2: Collect the original image of the sample surface The standard samples that have reached the static period are transferred to the inspection platform. The inspection platform is equipped with a CMOS sensor camera and an autofocus microscope to ensure that the tiny oil stains on the surface of the sample can be captured. The autofocus function ensures that the surface of the sample is clear during image acquisition and reduces human errors. The original image of the sample surface is acquired using a CMOS sensor camera and an autofocus microscope. During acquisition, the light should be uniform to avoid shadows or reflections that affect the image quality. The acquired original image is input into the visual inspection system for subsequent processing.

[0024] A large number of micro photodiodes are integrated inside the CMOS sensor camera, which are distributed in the pixel array of the camera. When light shines on the photodiode, the photons excite the electrons in the diode, thereby generating an electrical signal. The intensity of this electrical signal is proportional to the intensity of the incident light. The generated electrical signal is converted into a digital signal through the steps of integration, amplification and analog-to-digital conversion in the internal circuit of the camera. These digital signals are then sent to the digital signal processor DSP for further processing. The DSP will perform image processing algorithms such as denoising, white balance adjustment, and color interpolation on the digital signal to ultimately generate a high-quality image signal. The processed image signal is transmitted to a computer or visual inspection system through the camera's interface (such as USB, CSI, etc.). In the visual inspection system, the image is further analyzed and processed, such as grayscale processing, binarization processing, edge detection, and morphological processing, to extract the characteristics of oil stains.

[0025] The autofocus microscope has a built-in optical sensor for real-time monitoring of the focus status of the standard sample surface. The optical sensor captures the image of the sample surface and analyzes indicators such as image clarity or contrast to determine whether the current focus is accurate. The microscope measures the distance between the objective lens and the sample surface by distance measurement and adjusts the position of the objective lens according to this distance to achieve automatic focus. The real-time feedback mechanism ensures that during the adjustment process, feedback signals from the optical sensor are continuously received and the position of the objective lens is adjusted according to the feedback signal until the best focus is achieved. At different observation depths, the autofocus microscope automatically adjusts the focus to ensure that the image of the sample surface is always clear. This is usually achieved by driving the motor to fine-tune the objective lens, which moves the objective lens precisely under the command of the control system to achieve the best imaging effect. When the focus adjustment is completed, the microscope captures the image of the sample surface and transmits the image to the computer or visual inspection system for further analysis and processing. In the visual inspection system, image processing algorithms are used to extract the characteristics of oil stains and calculate indicators such as oil penetration and oil penetration rate.

[0026] In summary, CMOS sensor cameras and autofocus microscopes play a key role in the silicone oil outflow control detection method. They provide a reliable data basis for subsequent image processing and analysis through high-precision image acquisition and autofocus technology.

[0027] Step 3: Image preprocessing and feature classification and recognition After the visual inspection system receives the collected original image, it performs grayscale processing on the received original image in the visual inspection system to simplify the pixel value of the image into grayscale. Then it performs binarization processing to separate the oil stain area in the image from the background to improve the accuracy of subsequent analysis. The edge detection algorithm is used to extract the oil stain boundary in the image to make the shape of the oil stain more obvious. The image is subjected to morphological processing such as dilation, corrosion, opening and closing operations to enhance the features of the image, so as to remove noise, fill holes and segment independent oil stain areas.

[0028] A deep learning algorithm is used to automatically identify oil stains in the oil stain distribution image and calculate their size, number and area. The deep learning algorithm uses singular value decomposition (SVD) combined with convolutional neural network (CNN) for feature extraction and classification, accurately extracting the features of oil stains, such as the size, distribution and shape of oil stains. The singular value decomposition method is used to extract the feature vector of the oil stain area. The convolutional neural network is used to classify and identify the feature vector to automatically identify the location, size and area of ​​the oil stains in the oil stain distribution image on the sample surface.

[0029] Step 4: Quantitative calculation of oil seepage and generation of test report Analyze various factors that affect the oil output of silicone, such as silicone formula, production process, environmental conditions, etc. Determine the key influencing factors and collect relevant data. Based on the collected data, establish a linear regression mathematical model between oil output and influencing factors. Use silicone samples with known oil output to verify the model. According to the verification results, adjust the model parameters to improve the accuracy and reliability of the model. Use the verified and optimized model as the preset oil output calculation model for subsequent standard sample oil seepage and oil seepage rate calculations.

[0030] Use image processing algorithms (singular value decomposition method and convolutional neural network) to extract features from the collected standard sample images. Features may include the shape, size, distribution, etc. of the oil stain. Based on the extracted features, calculate the relevant parameters of the oil stain, such as area, perimeter, color depth, etc. These parameters will be used for subsequent oil output calculations.

[0031] Substitute the calculated oil stain parameters into the preset oil output calculation model. The model calculates the oil seepage of each standard sample based on the input parameters. The oil seepage rate is the ratio of the oil seepage to the area of ​​the silicone sample. Calculate the oil seepage rate based on the oil seepage and the sample area. Summarize the oil seepage and oil seepage rate data of each standard sample into the visualization platform. The platform supports real-time data update and query functions. Based on the summarized data, automatically generate a silicone oil output detection report. The report contains information such as the oil seepage and oil seepage rate of each standard sample, the average oil seepage and oil seepage rate, and comparison with other samples. The report format should be clear and easy to read, and data can be presented in the form of charts, tables, and images.

[0032] Through the implementation of the above steps, the calculation and visualization of silicone oil output can be realized, providing strong support for production process adjustment and product quality control.

[0033] Step 5: Determination of oil leakage of self-penetrating silicone and adjustment of production process The user logs in to the production management system with an account and password. The production management system displays the corresponding functional modules and operation interface according to the user's role and authority. The oil output threshold and oil permeation rate threshold are preset in the production management system. These thresholds should be set according to product quality standards and production process requirements. The silicone oil output test report is imported into the production management system in the form of an electronic file. The production management system should support multiple file formats, such as PDF, Excel, etc. The production management system automatically parses the data in the report and extracts the oil permeation and oil permeation rate of each standard sample. The parsed data is compared with the preset threshold.

[0034] The production management system determines the oil output of each standard sample according to the preset oil output threshold and oil penetration rate threshold. If the oil penetration and oil penetration rate are both lower than the threshold, it is determined to be qualified; otherwise, it is determined to be unqualified. For standard samples that are determined to be unqualified, the system should automatically trigger the production process adjustment process. The adjustment content may include lubricant formula, lubricant addition amount, production process parameters, etc. The production management system recommends appropriate adjustment plans based on historical data and process knowledge base. The operator adjusts the production process according to the adjustment plan recommended by the production management system. The production management system monitors the adjusted production situation in real time to ensure that the adjustment effect meets expectations. The production management system records the records and results of each production process adjustment. When quality problems occur, the system can be traced back to specific production batches, processes and operators.

[0035] As attached Figure 3 As shown, the present invention uses a detection platform, a visual detection system, a visualization platform and a production management system to complete the above steps.

[0036] Through the implementation of the above steps, the production management system can accurately determine the oil leakage of self-penetrating silicone rubber based on the silicone oil leakage detection report and the preset threshold, and automatically adjust the production process to ensure product quality and production efficiency. At the same time, the system also supports quality traceability and continuous improvement functions, providing strong support for the sustainable development of the enterprise.

[0037] After the visual inspection system receives the captured original color image, in order to simplify image processing and reduce the amount of calculation, the color image needs to be converted into a grayscale image. Color images are usually based on the RGB color model, which contains three color channels: red (R), green (G), and blue (B). The information in the color channel can help better capture the brightness distribution in the scene during the image conversion process, thereby improving the quality of the grayscale image. For example, the red, green, and blue channels have different brightness contributions, so more accurate grayscale conversion can be achieved by processing these channels by weighted averaging (such as weighted summation). By obtaining a color image, different color channels can be selected as needed during post-processing, and even other color spaces (such as HSV, LAB, etc.) can be used to extract specific visual information, perform more accurate grayscale conversion, or perform other image processing tasks.

[0038] Grayscale images convert the color value of each pixel in a color image into a single brightness value. The main conversion methods are weighted averaging and simple averaging: Weighted average method: Since the human eye is most sensitive to green, followed by red, and least sensitive to blue, different weights are usually used to calculate the grayscale value. The specific grayscale value calculation formula is: Grayscale value = 0.299×R+0.587×G+0.114×B. This method can more accurately reflect the human eye's perception of color.

[0039] Simple average method: The average value of the RGB value of each pixel in the color image is taken as the gray value, that is, gray value = (R+G+B) / 3. This method does not take into account the sensitivity of the human eye to different colors, but the calculation is simpler.

[0040] After obtaining the grayscale image, the adaptive threshold method is used to convert the grayscale image into a black and white image. The adaptive threshold method calculates the local threshold according to the brightness distribution of different areas of the image, so that it can better handle images with uneven light and dark distribution. For each pixel in the grayscale image, take the n×n area around it (n is usually an odd number, such as 3, 5, 7, etc.), and calculate the mean or Gaussian weighted value of the area as the local threshold. Compare the grayscale value of the current pixel with the local threshold. If the grayscale value is higher than the threshold, the pixel is set to white (grayscale value is 255); if the grayscale value is lower than the threshold, it is set to black (grayscale value is 0).

[0041] After obtaining the black and white image, the edge contour of the oil stain can be detected by calculating the gradient of the image grayscale value. The gradient is the rate of change of the image brightness, and the edge usually has a larger gradient value. Commonly used edge detection algorithms include Canny edge detector, Sobel operator, Prewitt operator, etc. Among them, Canny edge detector is an edge detection algorithm with excellent performance. It uses Gaussian filter to smooth the image, calculate the gradient amplitude and direction, and apply non-maximum suppression and double threshold detection to obtain the edge.

[0042] Finally, the black and white image is smoothed, filled and denoised through morphological operations such as dilation, erosion, opening and closing. Dilation: Enlarging the outline of the object in the image can be used to fill the holes between objects and strengthen discrete points. Erosion: Shrinking the outline of the object in the image can be used to disconnect the connection between objects and eliminate discrete points. Opening: Erosion followed by dilation can smooth the outline of the object, disconnect the narrow connection between objects, and eliminate the sharp protrusions on the edge of the object. Closing: Dilation followed by erosion can be used to bridge narrow discontinuities and elongated gullies, eliminate small holes between objects, and fill the breaks in the contour line.

[0043] In summary, through the above steps, the visual inspection system can realize grayscale, binarization, edge detection and morphological processing of the original color image, so as to accurately detect the edge contour of the oil stain.

[0044] The deep image processing algorithm extracts and identifies the oil stain features of the sample surface oil stain distribution image by combining the singular value decomposition method and the convolutional neural network. Figure 2 As shown, the following steps are included: S301, extracting the principal component image features from the sample surface oil stain distribution image by using the singular value decomposition method. The singular value decomposition method performs singular value decomposition on the sample surface oil stain distribution image matrix A to obtain the matrix U, the matrix S and the matrix V. The function expression is: (1) In formula (1), and is an orthogonal matrix used to represent the feature space of the sample surface oil stain distribution image. The matrix S is a diagonal matrix, in which the diagonal elements are singular values, representing the amount of information in the matrix A. Singular values ​​are usually sorted by size, and the features corresponding to larger singular values ​​have a stronger influence in the image. The matrix A is the original sample surface oil stain distribution image, usually a two-dimensional matrix, in which each element represents a pixel value or feature of the image, where the matrix , Indicates the ordinal number of singular values ​​contained in the matrix S, Represents the kth singular value contained in the matrix S, sorts the singular values ​​of the sample surface oil stain distribution image matrix A from large to small, selects the first k singular values ​​and the corresponding matrices U, S and V, and combines the feature matrix U, S and V to obtain the principal component representation of the sample surface oil stain distribution image; by selecting the first k largest singular values ​​and their corresponding columns and rows, the principal component features of the image can be extracted, the redundant information can be reduced, and the core features of the image can be retained, and a matrix of principal component representation can be obtained.

[0045] S302: Input the principal component image features extracted in S301 into a convolutional neural network for multi-scale image feature extraction. Input the principal component image features obtained from singular value decomposition into a convolutional neural network (CNN) to extract more advanced image features. CNN uses multiple convolution kernels to perform convolution operations on the principal component image to extract local features of the image. For example, information such as edges, textures, and the morphology of oil stains in the image can be extracted through convolution kernels. The convolution kernel usually slides on the image to generate a feature map. After each convolution operation, a nonlinear activation function (such as ReLU) is usually used to activate the convolution result. This step can help the network learn complex nonlinear features and improve the expressive power of the model. In order to reduce the amount of calculation and capture features at different scales, CNN usually adds a pooling layer (such as maximum pooling) after the convolution layer. The pooling operation helps to extract the most significant information from the feature map and reduce the spatial resolution of the feature map, thereby reducing the amount of calculation and avoiding overfitting.

[0046] S303, using a frequency division duplex (FDD) network to perform quality correction on the extracted multi-scale image features; In this step, a frequency division duplex (FDD) network is used to perform quality correction on the extracted multi-scale image features. A frequency division duplex (FDD) network is a network architecture that processes features by dividing them into different frequency bands. In image processing, different types of details are processed by separating features of different scales or frequency bands. The purpose of quality correction is to enhance important features and reduce the impact of noise or irrelevant information, thereby improving the accuracy of subsequent classification.

[0047] The frequency division duplex (FDD) network uses convolution operations to obtain the associated quality features of multi-scale image features. The function is expressed as: (2) In formula (2), x is the input sample surface oil stain distribution image data, t represents the offset of the extracted multi-scale image features, represents the kernel function of the frequency division duplex FDD network, represents the quality mapping result of the extracted multi-scale image features, Represents the weighted average parameter of the frequency division duplex FDD network, The weight vector representing the image data of oil stain distribution on the sample surface, the quality mapping result of multi-scale image features is calculated by the weighted average parameter inside the network Control, which represents the quality of the features calculated by convolution. These mapping results need to be compared with the standard feature quality to form a feature quality difference metric.

[0048] The defect feature vector in the extracted multi-scale image features is obtained by calculating the difference between the standard feature quality and the feature quality mapping result. The formula is expressed as: (3) In formula (3), Represents the difference between the standard feature quality and the feature quality mapping result, Represents the average value of all mass elements in the standard characteristic mass, and They represent the length and width of the multi-scale image feature-associated quality feature matrix respectively. When the difference is greater than the preset difference threshold, it is determined to be a defect feature vector and the defect feature vector is corrected. After being determined to be a defect feature vector, the next step is to correct the defect feature vector. The correction process usually includes further training or optimizing the adjustment of network parameters to reduce the difference between defect features and standard features, thereby improving the accuracy and robustness of image feature extraction.

[0049] This process combines the convolutional computing capabilities of the frequency division duplex (FDD) network to achieve defect detection of oil stain distribution in the image by extracting multi-scale features of the image and calculating feature quality differences. By comparing the standard feature quality with the extraction results, the abnormal areas (defects) in the image can be effectively identified, and by correcting the defect feature vector, the quality of the image features can be further optimized, thereby improving the accuracy and robustness of defect detection.

[0050] S304, the corrected multi-scale image features will be input into the fully connected layer for classification and recognition. The fully connected layer linearly transforms the input features through the weight matrix and the bias term, and finally outputs a classification feature vector. The fully connected layer uses the weight matrix to weight the input features, and adjusts the results through the bias term to calculate the output value. The weights and biases are learned through training. The output of the fully connected layer is usually a multi-scale vector, which represents the probability that the image belongs to each category. By performing Softmax processing on this output, the probability value of each category is obtained, so that the category to which the image belongs can be determined. By comparing the difference between the predicted classification result and the true label (such as cross entropy loss), the network will update the weights and biases through the back propagation algorithm to optimize the performance of the model.

[0051] Before performing singular value decomposition, the oil stain distribution image needs to be standardized or noise removed. When training CNN, it is necessary to select appropriate hyperparameters such as convolution kernel size, pooling layer configuration, and learning rate. The model needs to be trained on a large amount of labeled data, and methods such as cross-validation are used to evaluate the generalization ability of the model.

[0052] Combining the advantages of singular value decomposition and convolutional neural network, through the following key steps: first, use singular value decomposition to reduce the dimension of the oil stain distribution image and extract the main component features, then use the convolutional neural network to further extract multi-scale features and perform quality correction, and finally output the classification results through the fully connected layer, which can effectively extract and classify the features of the oil stain image. This method not only takes advantage of the mathematical advantages of singular value decomposition, but also gives full play to the powerful ability of convolutional neural network in image feature extraction and classification.

[0053] Deep image processing algorithms, which combine singular value decomposition methods and convolutional neural network algorithms, have a series of specific requirements on the hardware working environment. These requirements usually involve aspects such as computing performance, storage capacity, and data transmission speed to ensure that the algorithm can be executed efficiently and accurately. The hardware working environment of deep image processing algorithms includes the following: CPU: Although deep learning calculations mainly rely on GPUs, CPUs still play an important role, especially in data preprocessing, algorithm scheduling, and management. Multi-core CPUs can process more operations in parallel, increase the speed of data preprocessing, and ensure the stable operation of algorithms. The frequency and cache size of the CPU are also key factors that affect the speed and efficiency of data processing.

[0054] GPU: Graphics processing unit (GPU) is the core hardware of deep image processing algorithms. Compared with CPU, GPU can handle more parallel calculations and is suitable for performing matrix operations and convolution operations in deep learning. GPU with high performance and high video memory can accelerate the training and reasoning process of the algorithm and improve the accuracy and speed of oil stain feature extraction and recognition.

[0055] Memory (RAM): A large amount of image data needs to be loaded into memory for processing. Larger memory can increase the speed of data loading and reduce bottlenecks in the processing process. For deep image processing algorithms, sufficient memory is the key to ensure smooth operation of the algorithm.

[0056] Storage devices: Using solid-state drives (SSDs) instead of traditional mechanical hard disks (HDDs) can significantly increase data reading speeds. Fast storage devices are essential for loading large image datasets and algorithm models, which can shorten processing time and improve overall efficiency.

[0057] Interfaces and buses: High-speed data interfaces and buses are essential to ensure fast transmission of image data between the CPU, GPU, and storage devices. High-speed interfaces such as PCIe interfaces can provide sufficient data bandwidth to meet the data transmission speed requirements of deep image processing algorithms.

[0058] Specialized hardware: Specialized hardware such as tensor processing units (TPUs) are also an important part of deep image processing algorithms. These hardware are designed to accelerate deep learning calculations and can further improve the execution efficiency of the algorithm.

[0059] Cooling system: High-performance CPUs and GPUs generate a lot of heat during operation. Therefore, an effective cooling system is essential to ensure stable operation and extend the service life of the hardware.

[0060] In summary, the hardware working environment of the deep image processing algorithm includes high-performance CPU and GPU, sufficient memory and storage devices, high-speed data transmission interface and bus, and effective heat dissipation system. These hardware requirements together form the basis for the efficient and accurate execution of the algorithm.

[0061] In a specific embodiment, in order to verify the effectiveness of the method for controlling the oil output of silicone, an experimental test platform was built for experiments. The experiment was conducted on the Ubuntu 16.04 operating system, and the Label Image software was used to annotate the manually input oil stain feature data set on the surface of the sample to generate the label data required for training and verification. There were 33,250 images in the training set and 14,812 images in the test set. The training set was used to train the deep image processing algorithm, and the specific parameter settings of the model are shown in Table 1.

[0062] Table 1 Algorithm parameter settings

[0063] The data was simulated using matlab2018a, and the test sets were used to compare the effects of the deep image processing algorithm, the FAST algorithm, and the SURF algorithm. The results are shown in Table 2.

[0064] Table 2 Effect statistics

[0065] It can be seen from the data in Table 2 that in all five groups of data, the accuracy and robustness of the deep image processing algorithm are higher than those of the FAST algorithm and the SURF algorithm, which shows that the use of the deep image processing algorithm for silicone oil output control detection can improve the accuracy and reliability of detection, proving the feasibility of this method.

[0066] In the visualization platform, first of all, a data collection and processing mechanism needs to be established to capture the relevant data of silicone oil outflow: Oil seepage: The amount of oil seeping out per unit area of ​​the silicone surface can be measured by experimental methods.

[0067] Oil seepage rate: the change in oil volume per unit time, usually the ratio of the oil seepage volume to the initial oil volume.

[0068] Oil stain size: Measure the average size of oil stains on the silicone surface using a microscope or image processing method.

[0069] Number of oil stains: Calculating the number of oil stains per unit area can usually be done through image processing algorithms.

[0070] This data can be obtained through sensors, experimental equipment (such as microscopes, imaging instruments, etc.) or user input. Data needs to be stored and managed, and you can choose a suitable database system to store this information: Relational database: such as MySQL, PostgreSQL, etc., which can manage test data and support report generation through SQL queries. NoSQL database: such as MongoDB, which is suitable for storing unstructured or semi-structured data, such as image data.

[0071] The platform requires an authentication mechanism to ensure that only authorized users can access sensitive information: User registration and login: Users need to provide a username and password to log in, and the login information is stored in an encrypted database. Two-factor authentication: To enhance security, in addition to passwords, SMS verification codes, email verification, or application-based authentication (such as Google Authenticator) can also be used. Role permission management: Different access rights are authorized based on the user's identity (such as administrator, ordinary user, etc.).

[0072] The design of the visualization platform should have a good user experience and be able to clearly display the test results and trends: Interactive charts: Use JavaScript chart libraries (such as Chart.js, D3.js or Plotly) to display the time trend of data such as oil seepage, oil seepage rate, and oil stain size. Users can select different time ranges to view according to their needs. Example: Oil seepage trend chart, X-axis is time, Y-axis is oil seepage (unit: mg / cm²). Heat map: Use heat map technology to show the distribution of oil stains in different areas of the silicone surface. Tools such as heatmap.js or Google MapsAPI can be used to generate it. Example: Display different colors according to the size or number of oil stains in different areas of the silicone surface to help users intuitively identify the areas with the most serious oil situation. Dashboard: A dashboard is a view used to display multiple key indicators in a centralized manner. It can display real-time data, warnings, overall trends, etc. Dashboards are usually implemented using commercial tools such as Grafana or Tableau, or a simple dashboard is custom developed to display real-time data such as oil seepage, oil stain size, and oil stain number.

[0073] The user interaction part needs to provide a convenient operation method: allowing users to filter and view data according to time period, area, oil stain size and other conditions. Users can choose to generate a test report in PDF or Excel format, and the report content includes a detailed description of data such as oil seepage volume, oil seepage rate, oil stain size, and oil stain quantity. If the test result reaches a certain threshold, the system can automatically trigger an alarm to remind the user to take action.

[0074] Through the above steps, an efficient, interactive, safe and fully functional silicone oil detection platform can be realized, providing users with real-time and accurate detection data and trend analysis.

[0075] In a specific embodiment, the oil output of the produced self-seeping silicone is judged according to the preset oil output and oil seepage rate thresholds, and the production process is automatically adjusted based on the judgment result to ensure that the oil seepage of the silicone during the production process meets the set standards, thereby reducing the production of unqualified products and improving production quality. MATLAB is used to simulate and verify the effectiveness of the method, and ten groups of simulated product data are manually input. The detection method and the traditional weighing detection method are used for product detection, and the verification effect is shown in Table 3.

[0076] Table 3 Detection results

[0077] It can be seen from the experimental comparison table that in the same amount of time, the detection by this method consumes less time and has a higher accuracy rate, which proves the feasibility of this method.

[0078] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for controlling and detecting the oil output of silicone, characterized in that: The following steps are involved: Step 1: Take samples from the produced self-penetrating oil-free silica gel and prepare a plurality of standard samples. Transfer the prepared plurality of standard samples to a plurality of sample storage containers within 20 minutes and leave them for 72 hours. Place the plurality of sample storage containers in different simulated use environments of the self-penetrating oil-free silica gel. Step 2: Transfer the standard sample that has reached the static period to the detection platform, and the detection platform uses a CMOS sensor camera and an autofocus microscope to collect the original image of the surface of the standard sample, and input the collected original image into the visual inspection system; Step 3: After receiving the collected original image, the visual inspection system performs grayscale processing, binarization processing, edge detection and morphological processing on the received original image to obtain the sample surface oil stain distribution image, and uses a deep image processing algorithm to automatically identify the position, shape and area of ​​the oil stain in the sample surface oil stain distribution image. The deep image processing algorithm extracts and identifies the oil stain features of the sample surface oil stain distribution image by combining a singular value decomposition method and a convolutional neural network; Step 4: quantitatively calculate the oil seepage amount and oil seepage rate of each standard sample according to the extracted oil stain characteristics and the preset oil output calculation model, and summarize the calculation results through the visualization platform to automatically generate a silicone oil output detection report; Step 5. The production management system determines the oil leakage of the produced self-seeping silicone according to the silicone oil leakage detection report and the preset oil leakage threshold and oil penetration rate threshold, and automatically adjusts the production process of the self-seeping silicone based on the determination result.

2. A method for controlling and detecting the amount of silicone oil output according to claim 1, characterized in that: In the step 1, the thickness of the standard sample is 2 mm ± 0.2 mm, the length of the standard sample is 50 mm ± 0.2 mm, the width of the standard sample is 50 mm ± 0.2 mm, the sample storage container is an oil-free PC gasket, and the detection resolution is 5MP-25MP.

3. A method for controlling and detecting the amount of silicone oil output according to claim 2, characterized in that: The simulated use environment of the self-penetrating oil-free silica gel includes: (1) Normal temperature storage environment: ambient temperature is 23±2℃, humidity is 50±10%; (2) High temperature environment: ambient temperature is 60±2℃; (3) Low temperature environment: ambient temperature is -20±2℃; (4) Hot and humid environment: ambient temperature is 85±2℃, humidity is 85±10%; (5) Ultraviolet light irradiation environment: Use ultraviolet light to simulate sunlight.

4. A method for controlling and detecting the amount of silicone oil output according to claim 1, characterized in that: In the step 2, the CMOS sensor camera converts the optical signal into an electrical signal through a plurality of photodiodes integrated on the chip, and converts the electrical signal into an image signal through a digital signal processor. The autofocus microscope automatically adjusts the focus at different observation depths through a ranging method and a real-time feedback mechanism.

5. A method for controlling and detecting the amount of silicone oil output according to claim 1, characterized in that: In the step three, after receiving the collected original image, the visual inspection system converts the color original image into a grayscale original image by converting the pixel color values ​​in the original image into brightness values ​​to simplify the original image. Then, the pixel threshold is preset by the adaptive threshold method, and all pixels in the grayscale original image with grayscale values ​​higher than the pixel threshold are set to white, and all pixels in the grayscale original image with grayscale values ​​lower than the pixel threshold are set to black to complete the conversion of the grayscale image to a black and white image. The edge contour of the oil stain in the black and white image is detected by calculating the gradient of the image grayscale value. Finally, the black and white image is smoothed, filled and denoised through expansion, corrosion, opening and closing operations.

6. A method for controlling and detecting the amount of silicone oil output according to claim 1, characterized in that: In step three, the working method of the depth image processing algorithm includes the following steps: S301, extracting principal component image features from the sample surface oil stain distribution image by singular value decomposition method, wherein the singular value decomposition method performs singular value decomposition on the sample surface oil stain distribution image matrix A to obtain matrix U, matrix S and matrix V, and the function expression is: (1) In formula (1), and is an orthogonal matrix, which is used to represent the feature space of the oil stain distribution image on the sample surface. is a diagonal matrix used to represent the singular values ​​of the sample surface oil stain distribution image matrix A, where the matrix , Indicates the ordinal number of singular values ​​contained in the matrix S, Represents the kth singular value contained in the matrix S, sorts the singular values ​​of the sample surface oil stain distribution image matrix A from large to small, selects the matrix U, matrix S and matrix V corresponding to the first k singular values, and combines the feature matrix U, matrix S and matrix V to obtain the principal component representation of the sample surface oil stain distribution image; S302, inputting the main component image features extracted in S301 into a convolutional neural network for multi-scale image feature extraction, wherein the convolutional neural network uses multiple convolution kernels to extract local features of the main component image features, and uses a nonlinear activation function to activate the convolution result, and then uses a pooling operation to downsample the convolved features to capture changes in the position, shape, and area of ​​the oil stains; S303, using a frequency division duplex (FDD) network to perform quality correction on the extracted multi-scale image features; S304: Input the corrected image features into the fully connected layer. The fully connected layer performs linear transformation on different features through the weight matrix and the bias term, and outputs the recognition result of the features.

7. A method for controlling and detecting the amount of silicone oil output according to claim 6, characterized in that: In step S303, the frequency division duplex FDD network uses convolution operation to obtain the associated quality features of the multi-scale image features, and the function is expressed as: （2） In formula (2), x is the input sample surface oil stain distribution image data, t represents the offset of the extracted multi-scale image features, represents the kernel function of the frequency division duplex FDD network, represents the quality mapping result of the extracted multi-scale image features, Represents the weighted average parameter of the frequency division duplex FDD network, The weight vector representing the image data of oil stain distribution on the sample surface is used to obtain the defect feature vector in the extracted multi-scale image features by calculating the difference between the standard feature quality and the feature quality mapping result. The formula is expressed as: （3） In formula (3), Represents the difference between the standard feature quality and the feature quality mapping result, Represents the average value of all mass elements in the standard characteristic mass, and They respectively represent the length and width of the multi-scale image feature association quality feature matrix. When the difference is greater than the preset difference threshold, it is determined to be a defect feature vector and the defect feature vector is corrected.

8. A method for controlling and detecting the amount of silicone oil output according to claim 1, characterized in that: The visualization platform uses interactive charts, heat maps and dashboards to display the test results and changing trends, and uses a user identity authentication mechanism to verify the identity of the accessing user. The silicone oil leakage test report includes the oil leakage amount, oil leakage rate, oil stain size and oil stain number.

Images