Method and device for inspecting surface treatment effect of injection molded part based on machine vision

Abstract

The application discloses a machine vision-based injection molding part surface treatment effect quality inspection method and device, relates to the technical field of injection molding part surface treatment effect visual quality inspection, and adopts a visual quality inspection tool capable of supporting the injection molding part to be inspected as a standard position for visual collection, adopts detection light sources with the same angle and intensity to perform visual picture shooting, and a visual analysis module performs double-display effect visual classification on the injection molding part to be inspected based on a machine learning algorithm, the double-display effect visual classification is single classification of each display effect, and the visual effect is graded based on the single classification result combination of each display effect; the uniformity of the dissimilar part is considered in display effect evaluation, the evaluation of the uniformity of the dissimilar part is added, and the accuracy of visual quality inspection can be improved; intelligent quality inspection of the surface visual effect performance of the double-display effect injection molding part is realized, the subjectivity of manual quality inspection is avoided, and full inspection matching the production line rhythm can be realized.

Description

Technical Field

[0001] This invention relates to the field of visual quality inspection technology for the surface treatment effect of injection molded parts, and in particular to a method and apparatus for quality inspection of the surface treatment effect of injection molded parts based on machine vision. Background Technology

[0002] With the diversification and personalization of automotive interior and exterior parts, automotive exterior parts have been given more variations. PVD coating technology is added to exterior parts as embellishment to increase the sense of color layering. For existing processes, generally only a single color can be presented after the material is painted. There are also some special surface treatment processes for injection molded parts that can achieve a dual-surface visual effect. For example, the surface of injection molded parts can present the effect of electroplating and high-gloss black paint, or the surface of injection molded parts can present the effect of electroplating and light transmission.

[0003] In conventional injection molding production lines, visual methods are generally used to inspect surface dimensions and defects. Quality inspection of the visual effect of the surface is often based on manual sampling and empirical judgment, such as judging whether the visual effect meets the requirements based on experience at a distance of about 0.5 meters.

[0004] Chinese invention patent application CN108830832A discloses a machine vision-based algorithm for detecting surface defects in plastic cylinders, comprising the following steps: acquiring images using a line-scan camera vision imaging system; enhancing the image using a piecewise linear transformation grayscale transformation method; denoising the region using median filtering; segmenting the image using the LOG operator to obtain the printed edges; extracting the image region ROI on the surface of the plastic cylinder; segmenting the image using a thresholding method to determine the connected regions of the segmented image; and determining defects based on the area and edge features of the connected regions.

[0005] The above-mentioned visual inspection methods are mainly for detecting impurities on the surface of plastic cylinders and cannot achieve quality inspection of visual effects;

[0006] Therefore, there is an urgent need to develop visual quality inspection methods for the complex visual effects of a large number of automotive interior and exterior parts. Summary of the Invention

[0007] To address the aforementioned technical problems in the visual quality inspection of injection molded parts, this invention provides a machine vision-based method and apparatus for inspecting the surface treatment effect of injection molded parts. The technical solution adopted is as follows:

[0008] A machine vision-based method for quality inspection of surface treatment effects of injection molded parts includes the following steps:

[0009] Step 1: The transfer robot places the injection molded parts to be inspected onto the vision inspection fixture;

[0010] Step 2: Detect the light source to be turned on, and use the vision camera to capture visual images of the injection molded part to be inspected from multiple angles. Number the visual images based on the number of the injection molded part to be inspected, and transmit the visual images with the numbered data to the vision analysis module.

[0011] Step 3: The visual analysis module preprocesses the visual image to obtain the image to be inspected.

[0012] Step 4: The visual analysis module performs dual-display visual classification of the image to be inspected based on machine learning algorithms, grades the visual effects based on the dual-display visual classification results, and stores the visual classification results, grading results, and the image to be inspected based on the number data.

[0013] Step 5: The transfer robot sorts the injection molded parts to be inspected into qualified products storage racks, unprocessed products storage racks, and unqualified products storage racks based on the visual grading results of the dual display effect.

[0014] By adopting the above technical solution, a visual inspection fixture capable of supporting the injection-molded parts to be inspected is used as the standard position for visual acquisition. Visual images are captured using a detection light source of the same angle and intensity. The visual analysis module performs dual-display visual classification of the images to be inspected based on a machine learning algorithm. The visual analysis module refers to a device with visual analysis and computing capabilities, such as a computer deployed with a machine learning algorithm model. The acquired visual images are preprocessed to obtain the images to be inspected, and then input into the machine learning algorithm model. The machine learning algorithm model is trained using a large number of standard display effect images corresponding to dual-display effects, such as visual classification of monochrome injection-molded parts displaying both electroplating and high-gloss black paint effects. During quality inspection, a large number of visual images containing both standard electroplating and high-gloss black paint dual-display effects are used to train the model. Finally, the machine learning algorithm performs dual-display effect visual classification on the images to be inspected. Dual-display effect visual classification involves classifying each display effect separately, and then combining the results of each individual classification to achieve a graded visual effect. Finally, the transfer robot sorts the injection molded parts to be inspected into qualified product storage racks, unprocessed product storage racks, and unqualified product storage racks according to the dual-display effect visual grading results. This realizes intelligent quality inspection of the surface visual effect of injection molded parts with dual-display effects, improves the efficiency and accuracy of quality inspection, avoids the subjectivity of manual quality inspection, and can achieve full inspection that matches the production line rhythm.

[0015] Optionally, in step 3, the method for the visual analysis module to preprocess the visual image includes the following sub-steps:

[0016] Step 31: Perform filtering, noise reduction, and contrast enhancement operations;

[0017] Step 32: Using an edge detection algorithm, first separate the background edge of the injection molded part, multiple first display effect parts, and multiple second display effect parts.

[0018] By adopting the above technical solutions and performing filtering, noise reduction, and contrast enhancement operations, the impact of environmental factors on visual effect recognition can be reduced.

[0019] Edge detection algorithms are used to separate the background edges of injection molded parts, and then further separate multiple first display effect parts and multiple second display effect parts. For example, in the visual quality inspection of monochrome injection molded parts that present both electroplating and high-gloss black paint dual display effects, multiple electroplating display effect parts and multiple high-gloss black paint display effect parts can be separated, and the similarity of each individual display effect with the corresponding standard display effect can be compared.

[0020] Optionally, color segmentation technology can be used to distinguish the specific display colors of multiple first display effect parts and multiple second display effect parts.

[0021] By adopting the above technical solution, there should usually be a color difference between the two display effects in a dual display effect. Color segmentation technology can be used to distinguish between the first display effect and the second display effect.

[0022] Optionally, in step 4, the visual analysis module classifies the first display effect based on machine learning algorithms, including Aa, Ab, Ac, and Ad. Let X be the similarity between the first display effect and the standard first display effect, where Aa represents X ≥ 98%, Ab represents 95% ≤ X < 98%, Ac represents 90% ≤ X < 95%, and Ad represents X < 90%.

[0023] The machine learning algorithm classifies the second display effect into Ba, Bb, Bc, and Bd. Let Y be the similarity between the second display effect and the standard second display effect, where Ba represents Y≥98%, Bb represents 95%≤Y<98%, Bc represents 90%≤Y<95%, and Bd represents Y<90%.

[0024] If the combination of the first and second display effects in the dual display effect visual classification is classified as AaBa, AaBb, or AbBa, then the visual effect grading result is qualified.

[0025] If the first and second display effects in the dual-display effect visual classification are combined and classified as AbBb, AaBc, or AcBa, then the visual effect classification result is pending processing.

[0026] If the dual-display effect is classified as other combinations, the visual effect rating result is unqualified.

[0027] By adopting the above technical solution, the visual effect classification is achieved by single display effect classification. The classification is based on the similarity analysis results. For a display effect, if its similarity with the standard display effect is greater than 98%, it means that it is very close to the standard display effect. If the similarity is between 95% and 98%, it means that although there are slight differences, it can still meet the requirements in terms of visual presentation. And so on. If the similarity is less than 90%, it is considered that the visual effect is significantly different from the standard visual effect, which is unqualified for monochrome.

[0028] In the final grading, the combination of the two display effects is analyzed. For the AaBa classification combination, the dual display effect is the best, while AaBb or AbBa is visually acceptable and is therefore judged as a qualified product.

[0029] For combinations classified as AbBb, AaBc, or AcBa, if minor visual defects are found during manual observation, manual re-inspection is required. If minor defects can be repaired, such as by wiping away paint spots or stains, they can be inspected manually and judged as qualified products.

[0030] For other combinations and categories, they can be directly judged as unqualified products.

[0031] This quantitatively achieves automated visual grading of dual display effects.

[0032] Optionally, in step 4, the similarity calculation between the displayed effect and the standard displayed effect is performed using the following method:

[0033] Features are extracted from the image to be inspected and the standard surface image to form feature vectors AA and BB;

[0034] Calculate the dot product of eigenvectors AA and BB, calculate the Euclidean norm of eigenvectors AA and BB respectively, and divide the dot product by the product of the two norms to obtain the cosine similarity value.

[0035] The cosine similarity value ranges from [0%, 100%], where 100% means that the two vectors are exactly the same, and 0 means that the two vectors are orthogonal, i.e., not similar.

[0036] The cosine similarity value is the similarity between the displayed effect and the standard displayed effect.

[0037] Optionally, the cosine similarity between the display effect and the standard display effect is calculated using the following formula:

[0038] ;

[0039] Where A represents the display effect feature vector, and B represents the standard display effect feature vector. The cosine similarity value represents the difference between the displayed effect vector and the standard displayed effect vector. This represents the dot product of vectors A and B. Let A and B represent the Euclidean norms of vectors A and B, respectively.

[0040] By employing the above technical solution, the cosine similarity value ranges between [-1, 1], where 1 indicates that the two vectors are identical, 0 indicates that the two vectors are orthogonal (i.e., dissimilar), and -1 indicates that the two vectors are completely opposite. In practical applications, only non-negative similarity is usually considered, so the cosine similarity value is typically in the range of [0, 1]. This method allows for the quantitative evaluation of the similarity between the photographed injection-molded part surface and a standard surface, thereby enabling quality inspection.

[0041] Optionally, after calculating the cosine similarity value, the uniformity U of the dissimilar parts is calculated, and a first uniformity threshold Ua and a second uniformity threshold Ub are set. If U≥Ua, the similarity value between the display effect and the standard display effect is output based on the cosine similarity value. If Ub≤U<Ua, the cosine similarity value is reduced by 3% and the similarity value between the display effect and the standard display effect is output directly as less than 90%. If U<Ub, the similarity value between the display effect and the standard display effect is directly output as less than 90%.

[0042] By adopting the above technical solution, the uniformity of dissimilar parts needs to be considered in the evaluation of display effect. In simple terms, if the dissimilar parts are uniform, they will not have much impact on the visual effect. If the dissimilar parts are clustered, even if the dissimilar parts only account for 2%, they may still cause visual defects that are visible to the naked eye. Therefore, a two-stage uniformity threshold is set. For example, the uniformity is in the range of [0, 1], with 1 being the most uniform. Let the first uniformity threshold Ua = 0.9 and the second uniformity threshold Ub = 0.85. If the uniformity of the dissimilar parts U ≥ 0.9, it is considered that the uniformity of the dissimilar parts meets the requirements and does not affect the visual display effect. If 0.85 ≤ U < 0.9, it is considered that there is an impact. At this time, the cosine similarity value is subtracted by 3% and the output is the similarity value between the display effect and the standard display effect. The display effect is downgraded and evaluated. If U < 0.85, it is considered that there are clustered defects. At this time, it is necessary to directly judge that the visual effect is unqualified. Adding the evaluation of the uniformity of dissimilar parts can improve the accuracy of visual quality inspection.

[0043] Optionally, the uniformity U of dissimilar regions can be calculated using the following method:

[0044] Calculate the difference vector of two vectors in dissimilar directions. ;

[0045] ;

[0046] Calculate the standard deviation of the difference vector :

[0047] ;

[0048] in It is the i-th component of the difference vector, where n is the dimension of the vector. It is the mean of the difference vector;

[0049] Calculate the uniformity U of the dissimilar regions:

[0050] ;

[0051] in It is a constant.

[0052] By adopting the above technical solution, the uniformity U of dissimilar parts can be quantitatively calculated.

[0053] The machine vision-based surface treatment effect inspection device for injection molded parts includes a visual inspection fixture, a visual camera, a visual analysis module, and a storage rack assembly. The visual inspection fixture is equipped with multiple adjustable support blocks and a detection light source. When the transfer robot places the mounting surface of the injection molded part to be inspected on the upper surface of the multiple adjustable support blocks, the surface of the injection molded part to be inspected faces upwards. After the transfer robot moves away, the detection light source is turned on, and the visual camera captures a visual image of the surface of the injection molded part to be inspected. The visual camera then communicates and interacts with the visual analysis module to display the visual image of the surface. The visual analysis module runs an inspection program designed using a machine vision-based surface treatment effect inspection method for injection molded parts, outputs the inspection results, and stores the inspection results according to the number of the injection molded part to be inspected.

[0054] Optionally, the storage rack group includes a qualified product storage rack, a pending product storage rack, and a non-qualified product storage rack. The transfer robot communicates and interacts with the vision analysis module to exchange quality inspection results, and based on the quality inspection results, transfers the injection molded parts that have completed quality inspection to one of the qualified product storage rack, the pending product storage rack, and the non-qualified product storage rack.

[0055] In summary, the present invention has at least one of the following beneficial technical effects:

[0056] This invention provides a machine vision-based method and apparatus for quality inspection of the surface treatment effect of injection molded parts. It uses a vision inspection fixture that can support the injection molded parts to be inspected as the standard position for visual acquisition. It uses a detection light source with the same angle and intensity to capture visual images. The vision analysis module performs dual-display effect visual classification on the images to be inspected based on machine learning algorithms. The dual-display effect visual classification classifies each display effect separately, and then combines the results of each individual classification to achieve a grade of the visual effect. Finally, the transfer robot sorts the injection molded parts to be inspected into qualified product storage racks, unprocessed product storage racks, and unqualified product storage racks according to the dual-display effect visual grading results.

[0057] In evaluating display performance, the uniformity of dissimilar parts should be considered. Incorporating the evaluation of the uniformity of dissimilar parts can improve the accuracy of visual quality inspection.

[0058] It enables intelligent quality inspection of the surface visual effect of injection molded parts with dual display effects, improving the efficiency and accuracy of quality inspection, avoiding the subjectivity of manual quality inspection, and achieving full inspection that matches the production line rhythm. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating the machine vision-based quality inspection method for surface treatment effects of injection molded parts according to the present invention.

[0060] Figure 2 This is a visual image to be inspected, showing the dual display effects of electroplating and high-gloss black paint of this invention. Detailed Implementation

[0061] The present invention will be further described in detail below with reference to the accompanying drawings.

[0062] This invention discloses a machine vision-based method and apparatus for quality inspection of the surface treatment effect of injection molded parts.

[0063] Reference Figure 1 and Figure 2 Example 1: A machine vision-based method for quality inspection of surface treatment effects of injection molded parts, comprising the following steps:

[0064] Step 1: The transfer robot places the injection molded parts to be inspected onto the vision inspection fixture;

[0065] Step 2: Detect the light source to be turned on, and use the vision camera to capture visual images of the injection molded part to be inspected from multiple angles. Number the visual images based on the number of the injection molded part to be inspected, and transmit the visual images with the numbered data to the vision analysis module.

[0066] Step 3: The visual analysis module preprocesses the visual image to obtain the image to be inspected.

[0067] Step 4: The visual analysis module performs dual-display visual classification of the image to be inspected based on machine learning algorithms, grades the visual effects based on the dual-display visual classification results, and stores the visual classification results, grading results, and the image to be inspected based on the number data.

[0068] Step 5: The transfer robot sorts the injection molded parts to be inspected into qualified products storage racks, unprocessed products storage racks, and unqualified products storage racks based on the visual grading results of the dual display effect.

[0069] A visual inspection fixture capable of supporting the injection-molded parts to be inspected is used as the standard position for visual acquisition. Visual images are captured using a detection light source with the same angle and intensity. The visual analysis module, based on a machine learning algorithm, performs dual-display visual classification on the images to be inspected. The visual analysis module refers to a device with visual analysis and computing capabilities, such as a computer deployed with a machine learning algorithm model. The acquired visual images are preprocessed to obtain the images to be inspected, which are then input into the machine learning algorithm model. The machine learning algorithm model is trained using a large number of standard display effect images corresponding to dual-display effects. For example, in the visual inspection of monochrome injection-molded parts exhibiting dual-display effects of electroplating and high-gloss black paint, see [reference needed]. Figure 2 The model is trained using a large number of visual images containing both standard electroplating and high-gloss black paint dual-display effects. The machine learning algorithm then performs dual-display effect visual classification on the images to be inspected. This classification involves classifying each display effect separately and then combining the results of each classification to achieve a grading of the visual effects. Finally, the transfer robot sorts the injection molded parts to be inspected into qualified product storage racks, unprocessed product storage racks, and unqualified product storage racks based on the dual-display effect visual grading results. This achieves intelligent quality inspection of the surface visual effects of injection molded parts with dual-display effects, improving inspection efficiency and accuracy, avoiding the subjectivity of manual quality inspection, and enabling full inspection that matches the production line rhythm.

[0070] In Example 2, step 3, the method for the visual analysis module to preprocess the visual image includes the following sub-steps:

[0071] Step 31: Perform filtering, noise reduction, and contrast enhancement operations;

[0072] Step 32: Using an edge detection algorithm, first separate the background edge of the injection molded part, multiple first display effect parts, and multiple second display effect parts.

[0073] After performing filtering, noise reduction, and contrast enhancement operations, the impact of environmental factors on visual recognition can be reduced.

[0074] Edge detection algorithms are used to separate the background edges of injection molded parts, and then further separate multiple first display effect parts and multiple second display effect parts. For example, in the visual quality inspection of monochrome injection molded parts that present both electroplating and high-gloss black paint dual display effects, multiple electroplating display effect parts and multiple high-gloss black paint display effect parts can be separated, and the similarity of each individual display effect with the corresponding standard display effect can be compared.

[0075] Example 3 uses color segmentation technology to distinguish the specific display colors of multiple first display effect parts and multiple second display effect parts.

[0076] Typically, there should be a color difference between the two display effects in a dual-display effect. Color segmentation technology can be used to distinguish between the first display effect and the second display effect.

[0077] In Example 4, in step 4, the visual analysis module classifies the first display effect based on a machine learning algorithm, resulting in Aa, Ab, Ac, and Ad. Let X be the similarity between the first display effect and the standard first display effect, where Aa represents X ≥ 98%, Ab represents 95% ≤ X < 98%, Ac represents 90% ≤ X < 95%, and Ad represents X < 90%.

[0078] The machine learning algorithm classifies the second display effect into Ba, Bb, Bc, and Bd. Let Y be the similarity between the second display effect and the standard second display effect, where Ba represents Y≥98%, Bb represents 95%≤Y<98%, Bc represents 90%≤Y<95%, and Bd represents Y<90%.

[0079] If the combination of the first and second display effects in the dual display effect visual classification is classified as AaBa, AaBb, or AbBa, then the visual effect grading result is qualified.

[0080] If the first and second display effects in the dual-display effect visual classification are combined and classified as AbBb, AaBc, or AcBa, then the visual effect classification result is pending processing.

[0081] If the dual-display effect is classified as other combinations, the visual effect rating result is unqualified.

[0082] Visual effect classification is achieved using a single display effect classification system. The classification is based on similarity analysis results. For a display effect, if its similarity to the standard display effect is greater than 98%, it means that it is very close to the standard display effect. If the similarity is between 95% and 98%, it means that although there are slight differences, it can still meet the requirements in terms of visual presentation. And so on. If the similarity is less than 90%, the visual effect is considered to be significantly different from the standard visual effect, which is unacceptable for monochrome.

[0083] In the final grading, the combination of the two display effects is analyzed. For the AaBa classification combination, the dual display effect is the best, while AaBb or AbBa is visually acceptable and is therefore judged as a qualified product.

[0084] For combinations classified as AbBb, AaBc, or AcBa, if minor visual defects are found during manual observation, manual re-inspection is required. If minor defects can be repaired, such as by wiping away paint spots or stains, they can be inspected manually and judged as qualified products.

[0085] For other combinations and categories, they can be directly judged as unqualified products.

[0086] This quantitatively achieves automated visual grading of dual display effects.

[0087] In Example 5, step 4, the similarity calculation between the display effect and the standard display effect is performed using the following method:

[0088] Features are extracted from the image to be inspected and the standard surface image to form feature vectors AA and BB;

[0089] Calculate the dot product of eigenvectors AA and BB, calculate the Euclidean norm of eigenvectors AA and BB respectively, and divide the dot product by the product of the two norms to obtain the cosine similarity value.

[0090] The cosine similarity value ranges from [0%, 100%], where 100% means that the two vectors are exactly the same, and 0 means that the two vectors are orthogonal, i.e., not similar.

[0091] The cosine similarity value is the similarity between the displayed effect and the standard displayed effect.

[0092] Example 6: The cosine similarity between the display effect and the standard display effect is calculated using the following formula:

[0093] ;

[0094] Where A represents the display effect feature vector, and B represents the standard display effect feature vector. The cosine similarity value represents the difference between the displayed effect vector and the standard displayed effect vector. This represents the dot product of vectors A and B. Let A and B represent the Euclidean norms of vectors A and B, respectively.

[0095] The cosine similarity value ranges from [-1, 1], where 1 indicates that the two vectors are identical, 0 indicates that the two vectors are orthogonal (i.e., dissimilar), and -1 indicates that the two vectors are completely opposite. In practical applications, only non-negative similarity is usually considered, so the cosine similarity value is typically in the range of [0, 1]. This method allows for the quantitative evaluation of the similarity between the photographed surface of an injection-molded part and a standard surface, thus enabling quality inspection.

[0096] In Example 7, after calculating the cosine similarity value, the uniformity U of the dissimilar parts is calculated. A first uniformity threshold Ua and a second uniformity threshold Ub are set. If U≥Ua, the similarity value between the display effect and the standard display effect is output based on the cosine similarity value. If Ub≤U<Ua, the cosine similarity value is reduced by 3% and the similarity value between the display effect and the standard display effect is output as the similarity value between the display effect and the standard display effect. If U<Ub, the similarity value between the display effect and the standard display effect is directly output as less than 90%.

[0097] In evaluating display effects, the uniformity of dissimilar parts needs to be considered. Simply put, if the dissimilar parts are uniform, they won't significantly impact the visual effect. However, if the dissimilar parts are clustered, even if they only account for 2%, they can still cause visual defects visible to the naked eye. Therefore, a two-stage uniformity threshold is set, for example, with uniformity within the range of [0, 1], where 1 represents the most uniform. Let the first uniformity threshold Ua = 0.9 and the second uniformity threshold Ub = 0.85. If the uniformity of the dissimilar parts U ≥ 0.9, then the uniformity of the dissimilar parts is considered to meet the requirements and does not affect the visual display effect. If 0.85 ≤ U < 0.9, then there is an impact. In this case, the cosine similarity value is subtracted by 3% to output the similarity value between the display effect and the standard display effect, and the display effect is downgraded. If U < 0.85, then there are clustered defects, and the visual effect needs to be directly judged as unqualified. Adding the evaluation of the uniformity of dissimilar parts can improve the accuracy of visual quality inspection.

[0098] Example 8: The uniformity U of dissimilar parts was calculated using the following method:

[0099] Calculate the difference vector of two vectors in dissimilar directions. ;

[0100] ;

[0101] Calculate the standard deviation of the difference vector :

[0102] ;

[0103] in It is the i-th component of the difference vector, where n is the dimension of the vector. It is the mean of the difference vector;

[0104] Calculate the uniformity U of the dissimilar regions:

[0105] ;

[0106] in It is a constant.

[0107] The uniformity U of dissimilar parts can be quantitatively calculated.

[0108] Example 9: A machine vision-based surface treatment effect inspection device for injection molded parts includes a visual inspection fixture, a visual camera, a visual analysis module, and a storage rack assembly. The visual inspection fixture is equipped with multiple adjustable support blocks and a detection light source. When the transfer robot places the mounting surface of the injection molded part to be inspected on the upper surface of the multiple adjustable support blocks, the surface of the injection molded part to be inspected faces upwards. After the transfer robot moves away, the detection light source is turned on, the visual camera captures a visual image of the surface of the injection molded part to be inspected, and communicates and interacts with the visual analysis module to view the surface visual image. The visual analysis module runs an inspection program designed using a machine vision-based surface treatment effect inspection method for injection molded parts, outputs the inspection results, and stores the inspection results according to the number of the injection molded part to be inspected.

[0109] Example 10: The storage rack group includes a qualified product storage rack, a pending product storage rack, and a non-qualified product storage rack. The transfer robot communicates and interacts with the visual analysis module to exchange quality inspection results, and based on the quality inspection results, transfers the injection molded parts that have completed quality inspection to one of the qualified product storage rack, the pending product storage rack, and the non-qualified product storage rack.

[0110] 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, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A machine vision-based method for quality inspection of surface treatment effects of injection molded parts, characterized in that: Includes the following steps: Step 1: The transfer robot places the injection molded parts to be inspected onto the vision inspection fixture; Step 2: Detect the light source to be turned on, and use the vision camera to capture visual images of the injection molded part to be inspected from multiple angles. Number the visual images based on the number of the injection molded part to be inspected, and transmit the visual images with the numbered data to the vision analysis module. Step 3: The visual analysis module preprocesses the visual image to obtain the image to be inspected. Step 4: The visual analysis module performs dual-display visual classification of the image to be inspected based on machine learning algorithms, grades the visual effects based on the dual-display visual classification results, and stores the visual classification results, grading results, and the image to be inspected based on the number data. Step 5: The transfer robot sorts the injection molded parts to be inspected into qualified products storage rack, unprocessed products storage rack and unqualified products storage rack according to the visual classification results of the dual display effect. In step 3, the visual analysis module preprocesses the visual images using the following sub-steps: Step 31: Perform filtering, noise reduction, and contrast enhancement operations; Step 32: Using an edge detection algorithm, first separate the background edge of the injection molded part, multiple first display effect parts, and multiple second display effect parts; Color segmentation technology is used to distinguish the specific display colors of multiple first display effect parts and multiple second display effect parts; In step 4, the visual analysis module classifies the first display effect based on machine learning algorithms, resulting in Aa, Ab, Ac, and Ad. Let X be the similarity between the first display effect and the standard first display effect, where Aa represents X ≥ 98%, Ab represents 95% ≤ X < 98%, Ac represents 90% ≤ X < 95%, and Ad represents X < 90%. The machine learning algorithm classifies the second display effect into Ba, Bb, Bc, and Bd. Let Y be the similarity between the second display effect and the standard second display effect, where Ba represents Y≥98%, Bb represents 95%≤Y<98%, Bc represents 90%≤Y<95%, and Bd represents Y<90%. If the combination of the first and second display effects in the dual display effect visual classification is classified as AaBa, AaBb, or AbBa, then the visual effect grading result is qualified. If the first and second display effects in the dual-display effect visual classification are combined and classified as AbBb, AaBc, or AcBa, then the visual effect classification result is pending processing. If the visual classification of the dual display effect is other combinations, the visual effect rating result is unqualified. In step 4, the similarity between the displayed effect and the standard displayed effect is calculated using the following method: Features are extracted from the image to be inspected and the standard surface image to form feature vectors AA and BB; Calculate the dot product of eigenvectors AA and BB, calculate the Euclidean norm of eigenvectors AA and BB respectively, and divide the dot product by the product of the two norms to obtain the cosine similarity value. The cosine similarity value ranges from [0%, 100%], where 100% means that the two vectors are exactly the same, and 0 means that the two vectors are orthogonal, i.e., not similar. The cosine similarity value is the similarity between the displayed effect and the standard displayed effect; After calculating the cosine similarity value, the uniformity U of the dissimilar parts is calculated. A first uniformity threshold Ua and a second uniformity threshold Ub are set. If U≥Ua, the similarity value between the displayed effect and the standard displayed effect is output based on the cosine similarity value. If Ub≤U<Ua, the cosine similarity value is reduced by 3% and the similarity value between the displayed effect and the standard displayed effect is output directly as less than 90%. If U<Ub, the similarity value between the displayed effect and the standard displayed effect is directly output as less than 90%.

2. The machine vision-based surface treatment quality inspection method for injection molded parts according to claim 1, characterized in that: The cosine similarity between the displayed effect and the standard displayed effect is calculated using the following formula: ； Where A represents the display effect feature vector, and B represents the standard display effect feature vector. The cosine similarity value represents the difference between the displayed effect vector and the standard displayed effect vector. This represents the dot product of vectors A and B. Let A and B represent the Euclidean norms of vectors A and B, respectively.

3. The machine vision-based surface treatment effect inspection method for injection molded parts according to claim 2, characterized in that: The uniformity U of dissimilar regions is calculated using the following method: Calculate the difference vector of two vectors in dissimilar directions. ; ； Calculate the standard deviation of the difference vector : ； in It is the i-th component of the difference vector, where n is the dimension of the vector. It is the mean of the difference vector; Calculate the uniformity U of the dissimilar regions: ； in It is a constant.

4. A machine vision-based quality inspection device for the surface treatment effect of injection molded parts, characterized in that: The system includes a visual inspection fixture, a visual camera, a visual analysis module, and a storage rack assembly. The visual inspection fixture is equipped with multiple adjustable support blocks and a detection light source. When the transfer robot places the mounting surface of the injection molded part to be inspected on the upper surface of the multiple adjustable support blocks, the surface of the injection molded part to be inspected faces upwards. After the transfer robot moves away, the detection light source is turned on, and the visual camera captures a visual image of the surface of the injection molded part to be inspected. The visual camera then communicates and interacts with the visual analysis module to display the visual image of the surface. The visual analysis module runs an inspection program designed using the machine vision-based injection molded part surface treatment effect inspection method described in claim 3, outputs the inspection results, and stores the inspection results according to the number of the injection molded part to be inspected.

5. The machine vision-based surface treatment effect inspection device for injection molded parts according to claim 4, characterized in that: The storage rack group includes a qualified product storage rack, a pending product storage rack, and a non-qualified product storage rack. The transfer robot communicates and interacts with the vision analysis module to exchange quality inspection results, and based on the quality inspection results, it transfers the injection molded parts that have completed quality inspection to one of the qualified product storage rack, the pending product storage rack, and the non-qualified product storage rack.

Images