Radiator surface treatment quality assessment method based on machine vision

Abstract

The present invention relates to the field of radiator inspection technology, and in particular to a method for assessing the surface treatment quality of a radiator based on machine vision. The method comprises: obtaining a grayscale image of the radiator, dividing the grayscale image into multiple subregions, and calculating the characteristic values ​​of each subregion, wherein the characteristic values ​​include grayscale uniformity, texture roughness, and defect density; calculating the heat transfer efficiency index of the subregion, wherein the heat transfer efficiency index is negatively correlated with grayscale uniformity and defect density; calculating the air contact optimization index of the subregion, wherein the air contact optimization index represents the degree of influence of texture roughness on heat dissipation efficiency; constructing a characteristic vector using the mean and variance of the heat transfer efficiency indices of the multiple subregions and the mean and variance of the air contact optimization indices; and inputting the characteristic vector into a preset inspection model to obtain the heat dissipation power of the radiator, thereby evaluating the surface quality of the radiator. The present invention improves the accuracy of the surface quality assessment results.

Description

Technical Field

[0001] The present invention relates to the technical field of radiator detection, and in particular to a radiator surface treatment quality assessment method based on machine vision. Background Art

[0002] As an essential heat dissipation component in electronic and mechanical equipment, radiators' performance directly impacts the stable operation, energy efficiency, and service life of these devices. A radiator's thermal performance depends not only on its structural design and material selection but is also significantly influenced by surface treatment. For example, the uniformity, roughness, and microscopic defects of the surface coating can alter the radiator's surface thermal emissivity, convective heat transfer coefficient, and contact area with the fluid, thereby impacting its overall heat dissipation efficiency.

[0003] The Chinese patent application document with publication number CN119131012A discloses a busbar surface quality inspection method based on machine vision. It uses an industrial camera for continuous image acquisition and combines multi-angle oblique light sources to reduce reflective interference. Subsequently, it uses adaptive threshold segmentation and connected domain analysis technology to identify the filamentous stress marks and potential defect areas of the busbar. The PCA algorithm is further used to extract the main direction of the connected domain, construct a linear neighborhood window based on the pixel extension direction, analyze the grayscale and texture features, and extract the feature vector through the LBP algorithm. Finally, the direction and modulus of the feature vector are used to comprehensively judge the defect area. This patent application document uses an industrial camera to capture images and combines adaptive threshold segmentation, PCA and LBP algorithms to analyze the grayscale and texture features of the busbar surface defect area, thereby achieving accurate detection of aluminum busbar surface defects.

[0004] Machine vision can be used to inspect the surface quality of heat sinks. Macroscopic defects such as scratches, bubbles, and foreign matter can be identified and located based on the image's geometric, grayscale, color, or texture features. However, detecting macroscopic defects cannot reveal the heat sink's actual thermal performance parameters, such as thermal resistance, heat transfer coefficient, and heat dissipation efficiency. In other words, the impact of surface defects on actual thermal performance parameters is unknown, leading to misjudgments in heat sink surface quality assessments and inaccurate evaluation results. Summary of the Invention

[0005] In order to improve the accuracy of radiator surface quality assessment, the present invention provides a radiator surface treatment quality assessment method based on machine vision.

[0006] The present invention provides a method for evaluating the surface treatment quality of a radiator based on machine vision, which adopts the following technical solutions:

[0007] The surface treatment quality assessment method of a radiator based on machine vision includes the following steps:

[0008] Obtain a grayscale image of the radiator, divide the grayscale image into multiple sub-regions, and calculate the characteristic values ​​of each sub-region, including grayscale uniformity, texture roughness, and defect density;

[0009] The heat transfer efficiency index of the sub-region is calculated. The heat transfer efficiency index is negatively correlated with grayscale uniformity and defect density. The air contact optimization index of the sub-region is calculated. The air contact optimization index represents the influence of texture roughness on heat dissipation efficiency. The feature vector is constructed using the mean and variance of the heat transfer efficiency index and the mean and variance of the air contact optimization index of multiple sub-regions.

[0010] The characteristic vector is input into the preset detection model to obtain the heat dissipation power of the radiator, thereby realizing the evaluation of the surface quality of the radiator.

[0011] The eigenvalues ​​of the sub-regions are used to further obtain the eigenvector of the grayscale image, and the heat dissipation power of the radiator is obtained based on the prediction model using the eigenvector. This can help understand the impact of the surface characteristics of the radiator on the heat dissipation power, thereby avoiding misjudgment of the surface quality of the radiator and improving the accuracy of the surface quality assessment results.

[0012] Preferably, the expression of grayscale uniformity is: ;

[0013] Where U represents the grayscale uniformity of the sub-region, represents the grayscale value of the pth pixel in the sub-region, p represents the index of the pixel in the sub-region, u represents the mean grayscale value of the pixels in the sub-region, and q represents the number of pixels in the sub-region.

[0014] Grayscale uniformity can be used to quantify the variation characteristics of the grayscale values ​​of pixels in a sub-region, thereby reflecting the variation of the grayscale values ​​of pixels in the sub-region.

[0015] Preferably, the method for calculating the texture roughness is: constructing a gray level co-occurrence matrix of the sub-region, calculating the texture contrast of the sub-region using the gray level co-occurrence matrix, and taking the texture contrast of the sub-region as the texture roughness of the sub-region.

[0016] Texture roughness can reflect the texture changes of the sub-region and provide a theoretical basis for calculating the texture roughness of the sub-region.

[0017] Preferably, the defect density is calculated by using a convolutional neural network model to detect the defect area of ​​the sub-region, and taking the ratio of the defect area to the sub-region area as the defect density of the sub-region.

[0018] Preferably, the expression of heat transfer efficiency index is:

[0019]

[0020] Where E represents the heat transfer efficiency index of the sub-region, U represents the grayscale uniformity of the sub-region, and D represents the defect density of the sub-region. Represents the influence weight of grayscale uniformity, Represents the impact weight of defect density.

[0021] The heat transfer efficiency index is calculated through grayscale uniformity and defect density, which can further reflect the heat dissipation efficiency of the radiator surface.

[0022] Preferably, the expression of the air contact optimization index is:

[0023]

[0024] Where C represents the air contact optimization index of the sub-region, represents the texture roughness of the sub-region, Represents the standard texture roughness of the preset sub-area, exp represents the exponential function with e as the base, and σ represents the exponential function attenuation control parameter.

[0025] The air contact optimization index is obtained by the difference between the texture roughness of the sub-region and the standard texture roughness, which reflects the heat transfer performance of the heat sink.

[0026] Preferably, the quality assessment method further comprises: constructing a heat transfer efficiency index set using the heat transfer efficiency indices of multiple sub-regions in the grayscale image, and constructing an air contact optimization index set using the air contact optimization indices of multiple sub-regions in the grayscale image.

[0027] Preferably, the expression of the eigenvector is:

[0028]

[0029] Where F represents the feature vector of the grayscale image, represents the mean of the heat transfer efficiency index set, represents the variance of the heat transfer efficiency index set in the grayscale image, represents the mean of the air contact optimization index set, Represents the variance of the set of air contact optimization indices.

[0030] Preferably, the method for obtaining the detection model is:

[0031] A feature vector set is constructed using the feature vectors of multiple grayscale images, and a label is set for each feature vector in the feature vector set, where the label is heat dissipation power. A gradient boosting regression tree model is constructed, and the boosting regression tree model is trained using the feature vector set to obtain a detection model.

[0032] Preferably, before calculating the heat transfer efficiency index, a step of normalizing the grayscale uniformity and the defect density is also included.

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

[0034] 1. The eigenvalues ​​of the sub-regions are used to further obtain the eigenvectors of the grayscale image. The heat dissipation power of the radiator is obtained based on the prediction model using the eigenvectors. The surface quality of the radiator is further evaluated based on the heat dissipation power. Compared with the traditional method of evaluating the surface quality of the radiator based on the surface quality defects of the radiator, this method avoids misjudgment of the surface quality of the radiator and improves the accuracy of the evaluation results.

[0035] 2. The heat dissipation power of the radiator is obtained based on the prediction model using the characteristic vector, which can understand the influence of the surface characteristics of the radiator on the heat dissipation power and improve the accuracy of surface quality evaluation.

[0036] 3. In actual application, if there are defects such as bubbles, particles, scratches, etc. on the surface of the radiator, and the heat dissipation power result is qualified, it means that the surface defects of the radiator are negligible defects and no rework is required, which improves the actual production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The present invention is a flow chart of a method for evaluating the surface treatment quality of a radiator based on machine vision. DETAILED DESCRIPTION

[0038] 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.

[0039] The embodiment of the present invention discloses a method for evaluating the surface treatment quality of a radiator based on machine vision, referring to Figure 1 , including the following steps, as follows:

[0040] S1: Get a grayscale image of the radiator.

[0041] An industrial camera is used to collect the initial image of the radiator surface. A ring light is used as the light source. The initial image is converted into a grayscale image. The grayscale image is filtered using a Gaussian filtering algorithm to remove random noise in the image. The grayscale image is enhanced using a histogram equalization algorithm to enhance the visual effect and distinguishability of the grayscale image.

[0042] S2: Divide the grayscale image into multiple sub-regions and calculate the feature value of each sub-region.

[0043] The grayscale image is evenly divided into multiple sub-regions, each of which has a size of M×N. The characteristic values ​​include grayscale uniformity, texture roughness, and defect density.

[0044] S21: Calculate the grayscale uniformity of the sub-region.

[0045] The expression of grayscale uniformity is:

[0046] Where U represents the grayscale uniformity of the sub-region, represents the grayscale value of the pth pixel in a subregion, p represents the index of the pixel in the subregion, u represents the mean grayscale value of the pixels in the subregion, and q represents the number of pixels in the subregion. The greater the fluctuation in grayscale values ​​within a subregion, the greater the grayscale uniformity, indicating that the coating uniformity in the corresponding subregion is poorer. Poor coating uniformity leads to uneven heat distribution in the subregion, which in turn affects heat dissipation efficiency.

[0047] S22: Calculate the texture roughness of the sub-region.

[0048] A gray-level co-occurrence matrix is ​​constructed for each subregion. The gray-level co-occurrence matrix is ​​used to calculate the texture contrast of the subregion. This contrast is used as the texture roughness of the subregion. A greater texture contrast indicates a more dramatic texture change in the subregion. The methods for constructing the gray-level co-occurrence matrix and calculating the texture contrast are known techniques, and the detailed process will not be repeated here.

[0049] Increasing the surface roughness of a heat sink creates a microscopic concave-convex structure, increasing the surface area actually exposed to air and improving heat exchange efficiency. However, if the surface roughness is too high, it can cause localized airflow stagnation, reducing convection efficiency and, consequently, heat exchange efficiency. Therefore, the surface roughness of a heat sink has a certain impact on heat dissipation efficiency.

[0050] S23: Calculate the defect density of the sub-region.

[0051] The convolutional neural network model is used to identify the defect area of ​​the sub-region, and the ratio of the defect area to the sub-region area is used as the defect density of the sub-region. The area of ​​the sub-region is the number of pixels in the sub-region.

[0052] The convolutional neural network model can adopt a semantic segmentation network such as UNet and FCN to classify each pixel point in the sub-region, and the classification results are defective pixels and non-defective pixels. The number of defective pixels classified as defective pixels in the sub-region is counted, and the number of defective pixels is used as the defect area of ​​the sub-region.

[0053] When the defect area in the sub-region is larger, the defect density is higher, indicating that the integrity of the sub-region is worse, which will lead to a decrease in the heat exchange capacity between the corresponding area of ​​the radiator and the air, thereby affecting the heat dissipation performance of the radiator.

[0054] S3: Construct the feature vector of the grayscale image.

[0055] S31: Calculate the heat transfer efficiency index of the sub-region.

[0056] Both coating uniformity and defect density affect the heat dissipation efficiency of the heat sink. Therefore, the heat transfer efficiency index of the corresponding sub-region is calculated based on the grayscale uniformity and defect density of the sub-region. The grayscale uniformity and defect density are normalized using the minimum-maximum normalization algorithm. The expression of the heat transfer efficiency index is:

[0057]

[0058] Where E represents the heat transfer efficiency index of the sub-region, U represents the grayscale uniformity of the sub-region, and D represents the defect density of the sub-region. Represents the influence weight of grayscale uniformity, represents the impact weight of defect density, 、 This value is used to balance the effects of defect density and grayscale uniformity on the heat transfer efficiency index. Its value is determined experimentally. When grayscale uniformity or defect density increases, the heat transfer efficiency index decreases, indicating that the effective heat transfer efficiency of the coating in the corresponding sub-region decreases, negatively impacting the heat dissipation performance of the heat sink.

[0059] S32: Calculate the air contact optimization index of the sub-region.

[0060] The expression of air contact optimization index is:

[0061]

[0062] Where C represents the air contact optimization index of the sub-region, represents the texture roughness of the sub-region, It represents the standard texture roughness of the preset sub-region. Its value is obtained according to the experiment. exp represents the exponential function with e as the base. σ represents the attenuation control parameter of the exponential function, which reflects the tolerance for deviation from the standard texture roughness. Its value is set manually according to the actual situation.

[0063] When the texture roughness of the sub-region is closer to the standard texture roughness, the air contact optimization index is larger, indicating that the contact effect between the sub-region and the air is better, which is more conducive to the heat dissipation of the radiator and the heat dissipation efficiency of the radiator is higher.

[0064] S33: Constructing a feature vector of the grayscale image according to the heat transfer efficiency index and the air contact optimization index.

[0065] In the grayscale image, each sub-region corresponds to a heat transfer efficiency index and an air contact optimization index. The heat transfer efficiency indices of all sub-regions in the grayscale image constitute a heat transfer efficiency index set, and the air contact optimization indices of all sub-regions constitute an air contact optimization index set.

[0066] The mean and variance of the heat transfer efficiency index set and the mean and variance of the air contact optimization index set are calculated. The expression of the eigenvector is:

[0067]

[0068] Where F represents the feature vector of the grayscale image, represents the mean of the heat transfer efficiency index set, It represents the variance of the heat transfer efficiency index set in the grayscale image, reflecting the uniformity of the heat transfer efficiency index distribution in the grayscale image. represents the mean of the set of air contact optimization indices, The variance of the air contact optimization index set is expressed as , which reflects the uniformity of the air contact optimization index distribution.,It can be understood that each grayscale image corresponds to a feature vector, and the feature vector set is constructed using the feature vectors of multiple grayscale images.

[0069] S4: Use the detection model to detect the heat dissipation power of the radiator and evaluate the surface quality of the radiator.

[0070] For each feature vector in the feature vector set, a label is set, the label is the heat dissipation power, and a gradient boosting regression tree model is constructed. The feature vector set is used to train the boosting regression tree model to obtain a detection model. The obtained grayscale image of the radiator surface is input into the detection model to obtain the heat dissipation power. The training method of the boosting regression tree model is an existing technology, and the specific steps are not repeated here. When the heat dissipation power is less than the preset heat dissipation threshold, it indicates that the surface treatment quality of the corresponding radiator is unqualified.

[0071] 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 radiator surface treatment quality assessment method based on machine vision, characterized in that: Including steps: Obtain a grayscale image of the radiator, divide the grayscale image into multiple sub-regions, and calculate the characteristic values ​​of each sub-region, including grayscale uniformity, texture roughness, and defect density; Calculate the heat transfer efficiency index of the sub-region using the expression: , E represents the heat transfer efficiency index of the sub-region, U represents the grayscale uniformity of the sub-region, and D represents the defect density of the sub-region. Represents the influence weight of grayscale uniformity, Indicates the impact weight of defect density; Calculate the air contact optimization index of the sub-area using the expression: , C represents the air contact optimization index of the sub-region, represents the texture roughness of the sub-region, represents the standard texture roughness of the preset sub-region, exp represents the exponential function with e as the base, and σ represents the exponential function attenuation control parameter; The air contact optimization index represents the influence of texture roughness on heat dissipation efficiency; The quality assessment method further includes: constructing a heat transfer efficiency index set using heat transfer efficiency indices of multiple sub-regions in the grayscale image, and constructing an air contact optimization index set using air contact optimization indices of multiple sub-regions in the grayscale image; The eigenvector is constructed using the mean and variance of the heat transfer efficiency index of multiple sub-regions and the mean and variance of the air contact optimization index. The expression is: , F represents the feature vector of the grayscale image, represents the mean of the heat transfer efficiency index set, represents the variance of the heat transfer efficiency index set in the grayscale image, represents the mean of the set of air contact optimization indices, represents the variance of the set of air contact optimization indices; Input the feature vector into the preset detection model to obtain the heat dissipation power of the radiator and evaluate the surface quality of the radiator; The method for obtaining the detection model is as follows: construct a feature vector set using the feature vectors of multiple grayscale images, set a label for each feature vector in the feature vector set, and the label is the heat dissipation power; construct a gradient boosting regression tree model, and use the feature vector set to train the boosting regression tree model to obtain the detection model.

2. The method for evaluating the surface treatment quality of a heat sink based on machine vision according to claim 1, characterized in that: The expression of grayscale uniformity is: ; Where U represents the grayscale uniformity of the sub-region, represents the grayscale value of the pth pixel in the sub-region, p represents the index of the pixel in the sub-region, u represents the mean grayscale value of the pixels in the sub-region, and q represents the number of pixels in the sub-region.

3. The method for evaluating the surface treatment quality of a heat sink based on machine vision according to claim 1, wherein: The calculation method of texture roughness is as follows: construct a gray level co-occurrence matrix of the sub-region, calculate the texture contrast of the sub-region using the gray level co-occurrence matrix, and use the texture contrast of the sub-region as the texture roughness of the sub-region.

4. The method for evaluating the surface treatment quality of a heat sink based on machine vision according to claim 1, wherein: The defect density is calculated by using a convolutional neural network model to detect the defect area of ​​the sub-region, and the ratio of the defect area to the sub-region area is used as the defect density of the sub-region.

5. The method for evaluating the surface treatment quality of a heat sink based on machine vision according to claim 1, wherein: Before calculating the heat transfer efficiency index, the steps of normalizing the gray uniformity and defect density are also included.

Images