A system for detecting appearance defects on the inner and outer surfaces of cylindrical, conical products

Abstract

The application relates to the technical field of automatic detection, and discloses a system for detecting appearance flaws on the inner and outer surfaces of cylindrical or conical products, which comprises: a straightening conveying module for straightening the target detection body during conveying; a visual detection module for sampling pictures of the outer side and the inner cavity of the target detection body, and outputting a judgment result according to the sampling pictures; a spiral rotating module for driving the target detection body to rotate while conveying; and a defective product removing module for executing corresponding actions according to the judgment result. The system can be used for the target detection body, such as a sweet cone, can synchronously shoot the inner and outer surfaces of the target detection body while straightening and conveying the target detection body, and can also make the target detection body rotate during conveying in the product line direction, so that 360-degree inner and outer dead-angle-free shooting sampling is realized in the uninterrupted conveying process, the flaw detection efficiency and comprehensiveness are improved, the reliability and accuracy of the flaw detection result are improved, and the occurrence of missed detection is reduced.

Description

Technical Field

[0001] This invention relates to the field of automatic inspection technology, and specifically to a system for detecting defects on the inner and outer surfaces of cylindrical and conical products. Background Technology

[0002] Ice cream cones are a delicious frozen dairy product, typically consisting of a cone-shaped crispy shell, wrapping paper around the shell, and a dessert poured into and frozen inside. With the increasing prevalence of refrigerators and freezers in China, ice cream cone sales soar during the hot summer months. Therefore, ice cream cone manufacturers urgently need to improve production efficiency to meet market demand.

[0003] In the production process of ice cream cones, the wrapping paper is usually first made into a cone shape, then the cone-shaped crispy outer shell is inserted into the wrapping paper, so that the wrapping paper covers the crispy outer shell. Finally, the dessert is poured into the crispy outer shell and cooled to solidify, thus forming an ice cream cone. However, the crispy outer shell of an ice cream cone is relatively fragile. During mass production, it may break due to bumps and knocks during transportation on the production line, or defects in the internal and external surfaces may occur due to problems in the manufacturing process, ultimately leading to product quality issues.

[0004] Currently, to improve the overall efficiency of product quality control, manufacturers are beginning to use manual sampling or automated machine inspection to detect defects in the crispy parts during the production process, removing any defective pieces. However, manual sampling requires a large workforce, is unstable, costly, and requires shift work, so automated machine inspection is currently the primary method.

[0005] Currently, most methods for automated machine inspection of cylindrical products employ multi-camera, multi-directional static or dynamic inspection, which places certain requirements on the product's placement and the number of cameras used. For example... Figure 1 As shown, this is a traditional four-camera inspection solution. To inspect the appearance of a product's side, at least four cameras need to be placed around the product, each taking pictures from different angles to obtain all the information about the product's appearance. While this method achieves the inspection objective, it has limitations in terms of cost, inspection speed, product placement, and maintenance due to the use of multiple cameras. Furthermore, with the technological innovation of vision cameras, the emergence of line scan cameras has reduced the number of cameras required, such as... Figure 2The 360° visual inspection system shown uses a combination of high-speed line scanning cameras and a rotating mechanism for product inspection. While this method reduces the number of cameras used, the product must stop when it reaches the inspection position, complete a full rotation, and take a photo before the next group of products can be inspected. In summary, both the four-camera inspection solution and the 360° visual inspection system have high requirements for product placement and positioning, and can only inspect one product at a time. The continuity of the inspection process is poor, the efficiency is low, and the cost increases significantly, necessitating urgent improvement. Summary of the Invention

[0006] The purpose of this invention is to provide a system for detecting surface defects on the inner and outer surfaces of cylindrical and conical products, thereby solving the following technical problems:

[0007] How to provide a system that can efficiently and accurately detect defects in products of rotating shape.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] A system for detecting surface defects on the inner and outer surfaces of cylindrical and conical products, comprising:

[0010] The shaping and conveying module is used to align the target inspection objects during the conveying process;

[0011] A visual inspection module, connected to the shaping and conveying module, is used to sample images of the outer surface and inner cavity surface of the target object and output a judgment result based on the obtained sampled image set.

[0012] A spiral rotation module, connected to the visual inspection module and the shaping and conveying module, is used to drive the target inspection body to rotate while conveying the target inspection body;

[0013] The defective product rejection module is connected to the spiral rotation module and is used to perform corresponding actions based on the judgment result.

[0014] The above technical solution enables simultaneous imaging of the inner and outer surfaces of rotating objects, such as ice cream cone shells, during the orderly transport of the object. Furthermore, the object continuously rotates during transport, allowing for 360° imaging without blind spots. This significantly improves the efficiency and comprehensiveness of defect detection, thereby enhancing the reliability and accuracy of the results and further reducing missed detections.

[0015] As a further embodiment of the present invention: the shaping and conveying module includes a profile frame, a variable frequency motor, a circulating chain, a rising support surface, and a push rod; the profile frame includes a rising section and a translation section; the rising support surface is disposed on the rising section;

[0016] The variable frequency motor drives the push rod to slide in a direction parallel to the rising support surface via the circulating chain;

[0017] The rising support surface includes multiple V-shaped guide grooves arranged in parallel to each other, which limit the target detection body.

[0018] As a further embodiment of the present invention: the spiral rotation module includes a rotary motor, a synchronous belt assembly, a support base plate, and several conveying screws;

[0019] The conveying screws are rotatably mounted on the supporting base plate, and the rotary motor drives all the conveying screws to rotate synchronously through the synchronous belt assembly;

[0020] Each of the conveying screws is provided with a spiral guide groove, and the adjacent conveying screws are arranged in parallel.

[0021] As a further embodiment of the present invention: the defective product rejection module includes a flipping cylinder, a mounting base, and a flip cover plate; the flipping cylinder is fixed to the bottom of the mounting base, the flip cover plate is rotatably mounted on the mounting base, and the flipping cylinder is kinetically connected to the flip cover plate.

[0022] As a further embodiment of the present invention: the visual detection module includes a light shield, a surface light source, a camera, a cooling fan, and a host computer module;

[0023] The surface light source is used to provide supplemental lighting to the target detection body, and the camera is used to shoot at a set frequency f. c Obtain a sampled image of the target detection object;

[0024] The host computer module includes a control unit and an identification unit; the surface light source, the camera, and the cooling fan are connected to the control unit, and the identification unit is used to output a judgment result on whether the sampled image is a defective product.

[0025] As a further aspect of the present invention: the conveying speed of the shaping and conveying module to the target detection body is V1, and the conveying speed of the spiral rotation module to the target detection body is V2;

[0026] The relationship between V1 and V2 satisfies:

[0027] αV1=V2

[0028] V2=p*f

[0029] Where p is the pitch of the spiral guide groove on the conveying screw, f is the rotational speed of the conveying screw, and α is an adjustment coefficient greater than 1.

[0030] As a further aspect of the present invention: the recognition unit includes an image processing unit and a recognizer;

[0031] The image processing unit is used to extract the image feature vectors of the images captured by the camera, and merge the image feature vectors into the sampled image according to a preset merging order;

[0032] For the same group of target objects, the supplementary light wavelengths provided by the surface light source are different for each other;

[0033] The shooting frequency f c In [f l ,f h The selection is randomly set between [ ], and reset every Δt time interval; Δt is the time required for the target detection body to spin once;

[0034] f l For the lowest shooting frequency, f h This is the highest shooting frequency;

[0035] The recognizer is a trained neural network model.

[0036] As a further aspect of the present invention: the preset merging order includes:

[0037] The feature vectors of m*n images are arranged in a time-order matrix to obtain the sampled images; the feature vectors of the m*n images are the feature vectors corresponding to the images acquired from the start to the sampling time of 2Δt.

[0038] As a further aspect of the present invention: for the i-th group of target detection bodies, the image feature vector is obtained as follows:

[0039] Get the captured image P i,k RGB components;

[0040]

[0041] Among them, F i,k For the captured image P i,k The image feature vector, where k is the wavelength setting parameter of the surface light source.

[0042] As a further aspect of the present invention: the number of sampled images is N, where N is an odd number and greater than 1;

[0043] The sampled images obtained by arranging the feature vectors of m*n images in a time-order matrix are used as the base sampled images.

[0044] A new sampled image is obtained by randomly selecting from the image feature vector obtained after sampling for 2Δt time and randomly replacing it in the base sampled image.

[0045] The judgment result output by the recognizer is the judgment result with the highest probability among N judgment results.

[0046] The beneficial effects of this invention are as follows: This invention can target rotating objects such as ice cream cone shells or cylindrical shapes, allowing for simultaneous imaging of their inner and outer surfaces while they are being transported in an orderly fashion. Furthermore, the target object can continuously rotate during the orderly transport process, achieving 360° imaging without blind spots during continuous transport. This significantly improves the efficiency and comprehensiveness of defect detection, thereby indirectly enhancing the reliability and accuracy of defect detection results and further reducing the occurrence of missed detections. Attached Figure Description

[0047] The invention will now be further described with reference to the accompanying drawings.

[0048] Figure 1 This is a schematic diagram of a four-camera detection scheme in the prior art;

[0049] Figure 2 This is a schematic diagram of a 360° visual inspection system in the prior art;

[0050] Figure 3 This is a schematic diagram of the overall appearance of the appearance defect detection system in this invention;

[0051] Figure 4 This is a schematic diagram of the overall structure of the appearance defect detection system in this invention;

[0052] Figure 5 These are schematic diagrams of the spiral rotation module in the appearance defect detection system of this invention.

[0053] Figure 6 This is a schematic diagram of the structure of the visual inspection module in the appearance defect detection system of the present invention;

[0054] Figure 7 This is a schematic diagram of the defective product rejection module in the appearance defect detection system of the present invention.

[0055] Figure Descriptions: 100, Shaping and Conveying Module; 101, Variable Frequency Motor; 102, Profile Frame; 103, Circulating Chain; 104, Lifting Support Surface; 105, Push Rod; 200, Spiral Rotation Module; 201, Rotary Motor; 202, Synchronous Belt Assembly; 203, Support Base Plate; 204, Conveying Screw; 300, Vision Inspection Module; 301, Light Shield; 302, Surface Light Source; 303, Camera; 304, Cooling Fan; 305, Host Computer Module; 400, Defective Product Rejection Module; 401, Tilting Cylinder; 402, Mounting Base; 403, Flip-Top Plate. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Please see Figure 3 As shown, the present invention is a system for detecting surface defects on the inner and outer surfaces of cylindrical and conical products, comprising:

[0058] The shaping and conveying module 100 is used to align the target inspection objects during the conveying process;

[0059] The visual inspection module 300 is connected to the shaping and conveying module 100 and is used to sample images of the outer surface and inner cavity surface of the target object and output the judgment result based on the obtained sampled image set.

[0060] The spiral rotation module 200 is connected to the vision inspection module 300 and the shaping and conveying module 100, and is used to drive the target inspection body to rotate while conveying the movement.

[0061] The defective product rejection module 400 is connected to the spiral rotation module 200 and is used to perform corresponding actions based on the judgment result.

[0062] The above technical solution enables simultaneous imaging of the inner and outer surfaces of rotating objects, such as ice cream cone shells, during the orderly transport of the object. Furthermore, the object continuously rotates during transport, allowing for 360° imaging without blind spots. This significantly improves the efficiency and comprehensiveness of defect detection, thereby enhancing the reliability and accuracy of the results and further reducing missed detections.

[0063] Combination Figure 3 and Figure 4As shown, the profile conveying module 100 includes a profile frame 102, a variable frequency motor 101, a circulating chain 103, a rising support surface 104, and a push rod 105; the profile frame 102 includes a rising section and a translation section; the rising support surface 104 is disposed on the rising section;

[0064] The variable frequency motor 101 drives the push rod 105 to slide in a direction parallel to the rising support surface 104 via the circulating chain 103;

[0065] The rising support surface 104 includes multiple V-shaped guide grooves arranged in parallel to each other, which limit the target detection body.

[0066] Through the above technical solution, the entire system uses a profile frame 102 as support, and a variable frequency motor 101 provides power for driving. The circulating chain 103 drives the push rod 105 to move along the rising support surface 104, which can push the rows of target detection objects upward. Taking the cone-shaped ice cream cone as an example, each V-shaped guide groove can hold a row of nested ice cream cones. During the pushing process, they will be automatically shaped and arranged under the action of gravity, and neatly reach the junction of the rising section and the translation section. Then, the spiral rotation module 200 with the translation section is set to continue to transport them.

[0067] In this embodiment of the invention, as Figure 5 As shown, the spiral rotation module 200 includes a rotary motor 201, a synchronous belt assembly 202, a support base plate 203, and several conveying screws 204;

[0068] The conveying screw 204 is rotatably mounted on the support base plate 203, and the rotary motor 201 drives all the conveying screws 204 to rotate synchronously through the synchronous belt assembly 202.

[0069] Each conveying screw 204 is provided with a spiral guide groove, and adjacent conveying screws 204 are arranged in parallel.

[0070] Through the above technical solution, when the target detection body reaches the spiral rotation module 200, the target detection body will generate spin under the rotation of the conveying screw 204, and the spiral guide groove on the conveying screw 204 can provide the target detection body with a forward conveying force.

[0071] Specifically, the conveying speed of the shaping conveying module 100 to the target detection body can be set to V1, and the conveying speed of the spiral rotation module 200 to the target detection body can be set to V2;

[0072] The relationship between V1 and V2 satisfies:

[0073] αV1=V2

[0074] V2=p*f

[0075] Wherein, is the pitch of the spiral guide groove on the conveying screw 204, is the rotational speed of the conveying screw 204, and is an adjustment coefficient greater than 1. This ensures that the target detection body at the junction of the rising section and the translation section can be quickly separated from the target detection body still in the rising section, ensuring that the target detection body spins more smoothly and stably.

[0076] As a further aspect of the present invention: such as Figure 6 As shown, the visual inspection module 300 includes a light shield 301, a surface light source 302, a camera 303, a cooling fan 304, and a host computer module 305;

[0077] The surface light source 302 is used to provide supplemental lighting to the target object, and the camera 303 is used to shoot at a set frequency f. c Acquire sampled images of the target object;

[0078] The host computer module 305 includes a control unit and an identification unit; the surface light source 302, the camera 303, and the cooling fan 304 are connected to the control unit, and the identification unit is used to output a judgment result on whether the sampled image is a defective product.

[0079] In this embodiment of the invention, the camera 303 can be positioned near the junction of the ascending segment and the translating segment, and the shooting direction can be selected as either diagonally downward or straight downward. Therefore, when the target object reaches the junction of the ascending segment and the translating segment, the inner cavity of the target object can be more fully displayed under the action of the slope of the ascending segment, and thus be captured by the camera 303. When the target object exceeds the junction of the ascending segment and the translating segment, the camera 303, under the setting of the shooting direction and the action of the surface light source 302, can not only still capture the inner cavity of the target object, but also capture the outer surface of the target object in a spinning state.

[0080] As a further aspect of the present invention: the recognition unit includes an image processing unit and a recognizer;

[0081] The image processing unit is used to extract the image feature vectors of the images captured by the camera 303, and merge the image feature vectors into sampled images according to a preset merging order;

[0082] For the same group of target objects, the supplementary light wavelengths provided by the surface light source 302 are different;

[0083] Shooting frequency f c In [f l ,f h The setting is randomly selected between [] and reset every Δt time interval; Δt is the time required for the target object to spin once.

[0084] f lFor the lowest shooting frequency, f h This is the highest shooting frequency;

[0085] The recognizer is a trained neural network model.

[0086] The preset merging order includes:

[0087] The feature vectors of m*n images are arranged in chronological order to obtain the sampled images; the feature vectors of m*n images are the feature vectors corresponding to the images acquired from the start to the sampling time of 2Δt.

[0088] In the above technical solution, the recognizer can be obtained by training a Convolutional Neural Network (CNN) as a prototype. A CNN is a type of feedforward neural network that includes convolutional computation and has a deep structure; it is one of the representative algorithms of deep learning. During the training phase, labeled training samples are used to train the CNN. The method for obtaining these training samples is the same as the method for obtaining the sampled images.

[0089] In this invention, m can be selected as 4 and n can be selected as 6. Therefore, the sampled image contains the image feature vectors corresponding to 24 captured images. These 24 captured images are the images acquired by the camera from the start of shooting to the sampling time of 2Δt. Considering that the spin state of the target detection objects may not be completely consistent, some target detection objects have rotated one revolution during the Δt time, while others have not rotated one revolution. Therefore, before ending a round of shooting, it is necessary to ensure that all target detection objects have completed one revolution of shooting.

[0090] Furthermore, if the shooting frequency of camera 303 is constant and the spin speed of the target object is also constant, although the target object is moving, the display surface of the target object in the images captured by camera 303 in a later period may be roughly the same as that in the images captured in the earlier period. Considering the importance of the display surface of the target object for defect detection, it is necessary to change the shooting frequency of camera 303 to make its shooting frequency f c In [f l ,f h The settings are randomly selected between [] and reset every Δt time interval to maximize the display area of ​​the target detection object during shooting.

[0091] As a further aspect of the present invention: the number of sampled images is N, where N is an odd number and greater than 1;

[0092] The sampled images obtained by arranging the feature vectors of m*n images in a time-order matrix are used as the base sampled images.

[0093] Randomly extract from the image feature vector obtained after sampling for 2Δt time, and randomly replace the base sampled image to obtain a new sampled image;

[0094] The judgment result output by the recognizer is the judgment result with the highest probability among N judgment results.

[0095] In this embodiment of the invention, random samples are extracted from the image feature vector obtained after sampling for 2Δt time, and randomly replaced in the base sampled image, combined with the shooting frequency f. c The changes can provide more diverse sampling images because the sampling images and training samples are obtained in the same way. As long as the order of the image feature vectors in the sampling images is changed, or new image feature vectors are replaced in different positions, new training samples can be obtained. This means that the difficulty of obtaining training samples is extremely small and it is very easy to multiply them. Similarly, the complexity of manual annotation can be greatly reduced, and the training process can be completed very quickly, efficiently and accurately.

[0096] As a further aspect of the present invention: for the i-th group of target detection objects, the image feature vector is obtained as follows:

[0097] Get the captured image P i,k RGB components;

[0098]

[0099] Among them, F i,k Editing the photos i,k The image feature vector, where k is the wavelength setting parameter of the surface light source.

[0100] By using the above technical solution, the supplementary lighting parameters used by the camera 303 are adjusted for each shot, which is equivalent to introducing a more fluctuating environmental variable. When training the convolutional neural network based on the image feature vector obtained in this way, the robustness of the convolutional neural network can be increased. Since changes in ambient light may not be noticed by the human eye, but will definitely be captured by the machine, it is necessary to incorporate changes in ambient light into the generation of training samples and the training process during the training phase.

[0101] like Figure 7As shown, the defective product rejection module 400 includes a tilting cylinder 401, a mounting base 402, and a flip cover 403. The tilting cylinder 401 is fixed to the bottom of the mounting base, and the flip cover 403 is rotatably mounted on the mounting base 402. The tilting cylinder 401 and the flip cover 403 are connected by a transmission. Thus, based on the judgment result of the visual inspection module 300, the corresponding tilting cylinder 401 can be activated to tilt the flip cover 403, thereby rejecting the defective target.

[0102] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A system for detecting cosmetic defects on the inner and outer surfaces of cylindrical, conical products, characterized in that, The application relates to a whole-type conveying module (100) for aligning a target detection body during conveying; the whole-type conveying module (100) comprises a profile rack (102) which comprises a lifting section and a translation section; an ascending support surface (104) is arranged on the lifting section; a visual detection module (300) connected with the whole-type conveying module (100) and used for sampling pictures of the outer side and the inner cavity of the target detection body and outputting a judgment result according to the obtained sample picture group; a spiral rotation module (200) connected with the visual detection module (300) and the whole-type conveying module (100) and used for driving the target detection body to rotate while conveying; the conveying speed of the target detection body by the whole-type conveying module (100) is V1, and the conveying speed of the target detection body by the spiral rotation module (200) is V2; the relationship between V1 and V2 satisfies: the visual detection module (300) comprises a light shield (301), a surface light source (302), a camera (303), a heat dissipation fan (304) and an upper computer module (305); ； ； wherein, is the pitch of the helical channel on the conveying screw (204), is the rotational speed of the conveying screw (204), is a regulation factor greater than 1; a defective product removing module (400) connected with the spiral rotation module (200) and used for executing corresponding actions according to the judgment result; The area light source (302) is used to provide light compensation for the target detection body, and the camera (303) is used to take pictures of the target detection body at a set shooting frequency A sample picture of the target detection body is acquired. The host computer module (305) comprises a control unit and an identification unit; the identification unit comprises a picture processing unit and an identifier; the shooting frequency is randomly set between and is reset every time; time required for one spin of the target detection body; is the lowest shooting frequency, is the highest shooting frequency; the spiral rotation module (200) comprises a rotating motor (201), a synchronous belt assembly (202), a support bottom plate (203) and a plurality of conveying screws (204); the conveying screws (204) are rotationally arranged on the support bottom plate (203), and the rotating motor (201) drives all the conveying screws (204) to synchronously rotate through the synchronous belt assembly (202); each conveying screw (204) is provided with a spiral guide groove, and adjacent conveying screws (204) are arranged in parallel. the whole-type conveying module (100) comprises a variable frequency motor (101), a circulating chain (103), an ascending support surface (104) and a pushing rod (105); the variable frequency motor (101) drives the pushing rod (105) to slide in a direction parallel to the ascending support surface (104) through the circulating chain (103); 2. The system for inspecting the external appearance of the inner and outer surfaces of cylindrical or conical products according to claim 1, characterized in that, the ascending support surface (104) comprises a plurality of V-shaped guide grooves arranged in parallel; the V-shaped guide grooves limit the target detection body. the defective product removing module (400) comprises a turnover air cylinder (401), a mounting seat (402) and a turnover cover plate (403); the turnover air cylinder (401) is fixed at the bottom of the mounting seat (402), the turnover cover plate (403) is rotationally arranged on the mounting seat (402), and the turnover air cylinder (401) is in transmission connection with the turnover cover plate (403).

3. The system for inspecting the external and internal surface appearance flaws of cylindrical and conical products according to claim 1, characterized in that, the surface light source (302), the camera (303) and the heat dissipation fan (304) are connected with the control unit, and the recognition unit is used for outputting a judgment result of whether the sample picture is a defective product.

4. The system for inspecting the external and internal surface appearance flaws of cylindrical and conical products according to claim 2, characterized in that, ​ 5. The system for inspecting the external and internal surface appearance flaws of cylindrical and conical products according to claim 4, characterized in that, The picture processing unit is configured to extract picture feature vectors of pictures captured by the camera (303), and merge the picture feature vectors into the sample pictures according to a preset merging order. The light supplement wavelengths provided by the area light source (302) are different for the same group of target detection bodies. The identifier is a trained neural network model.

6. The system for inspecting the external and internal surface appearance flaws of cylindrical, conical products according to Claim 5, characterized in that, The preset merging order includes: arranging m*n picture feature vectors in time sequence matrix to obtain a sample picture; the m*n picture feature vectors are feature vectors corresponding to pictures acquired from the beginning to the sampling 2 seconds.

7. The system for inspecting the external and internal surface appearance flaws of cylindrical and conical products according to claim 5, characterized in that, For the ith group of target detection bodies, the picture feature vectors are obtained in the following manner: Acquiring a captured picture of RGB components; ； wherein, is the picture feature vector of the photographed picture is the picture feature vector of the photographed picture is a wavelength setting parameter of the area light source (302).

8. The system for inspecting the external and internal surface appearance flaws of cylindrical and conical products according to claim 6, characterized in that, The number of the sample pictures is N, N is an odd number and greater than 1; The sample picture obtained by arranging m*n picture feature vectors in a time sequence matrix is taken as a basic sample picture; From sample 2 Randomly sampling in the picture feature vectors obtained after the time, randomly replacing in the base sample picture, obtaining a new sample picture; The judgment result output by the identifier is a judgment result with the highest probability in N judgment results.

Images