PCBA circuit board defect detection and classification system based on deep learning

Abstract

A PCBA circuit board defect detection and classification system based on deep learning disclosed by the present invention comprises two groups of butt joint plates, two groups of first transmission columns are installed between the two groups of butt joint plates in a transmission manner, the outer surfaces of the two groups of first transmission columns are sleeved with a first conveyor belt, and the two sides of the other group of first transmission columns are in engaged transmission with centering pushing mechanisms. The center pushing mechanism is arranged on the first conveying belt, so that the center pushing mechanism can push the PCBA conveyed by the first conveying belt to the center of the first conveying belt, the PCBA is flush with the defect detection mechanism in the conveying process, and the defect detection mechanism is installed on the upper surface of the shrinking end of the center pushing mechanism in an overhead mode. Through the centering pushing mechanism and the defect detection mechanism, PCBA circuit boards with defects can be automatically pushed out of a production line, then automatic defect separation is achieved, manual defect product screening is avoided, and the automation degree of the production line is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of PCBA circuit board defect detection equipment, and in particular to a PCBA circuit board defect detection classification system based on deep learning. Background Art

[0002] PCBA circuit board is a printed circuit board made using electronic printing technology.

[0003] The patent with announcement number CN220603325U discloses a defect detection device for a PCBA circuit board, including a detection needle, which is arranged on a mounting plate. The defect detection device includes a main body, an adjustment device, an elastic plate and a sensor, a part of the elastic plate is arranged at the bottom of the main body, and the other part is connected to the mounting plate, the mounting plate is arranged below the elastic plate, the sensor is arranged on one side of the elastic plate, and the end of the mounting plate is provided with a sensing part, the sensing part and the sensor are arranged oppositely, the adjustment device is arranged on the main body, the adjustment device includes a fixed part, a rotating part and a spring, the rotating part is threadedly connected to the fixed part, the spring is sleeved on the rotating part, one end of the spring abuts against the rotating part, and the other end abuts against the mounting plate. A defect detection device for a PCBA circuit board of the present invention can quickly determine whether the detection needle is in contact with the circuit board, and improve the contact stability between the detection needle and the circuit board.

[0004] The above device detects defects in PCBA circuit boards by personnel picking up single pieces for inspection, and cannot perform inspection during the transportation process on the production line. An additional step is required to transfer the PCBA circuit board from the production line to the inspection device, and then send it back to the production line or for subsequent processing after the inspection is completed. The above inspection process interrupts the production process, reducing production efficiency. Summary of the invention

[0005] The purpose of the present invention is to provide a PCBA circuit board defect detection and classification system based on deep learning to solve the technical problem proposed in the above background technology that when inspecting defects on PCBA circuit boards, they can only be picked up and inspected by staff individually, but cannot be inspected while being transported on the production line.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A PCBA circuit board defect detection and classification system based on deep learning comprises two groups of docking plates, two groups of first transmission columns are transmission-installed between the two groups of docking plates, first conveyor belts are sleeved on the outer surfaces of the two groups of first transmission columns, a reduction motor is fixedly installed at one end of the docking plate, the output shaft of the reduction motor passes through the docking plate and is fixedly connected to the first transmission column, and both sides of the other group of first transmission columns are meshed with a central pushing mechanism, the pushing end of the central pushing mechanism is suspended on the upper surface of the first conveyor belt and is in a contracted shape from wide to narrow, so that the central pushing mechanism can push the PCBA circuit board conveyed by the first conveyor belt to the center of the first conveyor belt, so that the PCBA circuit board can be flush with the defect detection mechanism during the conveying process, and the defect detection mechanism is suspended on the upper surface of the contracted end of the central pushing mechanism.

[0008] As a preferred solution of the present invention, an electric push rod is fixedly installed at one end of the docking plate, the piston rod of the electric push rod passes through the docking plate and a push plate is fixedly installed at the end, so that the push plate can push the PCBA circuit board detected as having defects by the defect detection mechanism from the upper surface of the first conveyor belt into the slide, and the slide is fixedly installed at one end of the other set of docking plates and is flush with the push plate.

[0009] As a preferred embodiment of the present invention, the centering pushing mechanism includes two groups of first bevel gears, which are fixedly mounted on both sides of the first transmission column and located at the outer end of the docking plate. The first bevel gear is meshed with the second bevel gear, and the second bevel gear is fixedly mounted on the lower end of the outer surface of the connecting column. The connecting column is rotatably mounted in the connecting arm, and the connecting arm is fixedly mounted on one end of the docking plate. The upper end of the connecting column is fixedly mounted with a second transmission column, and a second conveyor belt is mounted on the outer surface of the second transmission column. The other end of the second conveyor belt is mounted on the outer surface of the third transmission column, and the third transmission column is rotatably mounted on the outer surface of the L-shaped frame, and the L-shaped frame is fixedly mounted on one end of the docking plate.

[0010] As a further preferred embodiment of the present invention, the second conveyor belt is supported by the second transmission column and the third transmission column, so that the second conveyor belt is in a contracted shape from wide to narrow, so that the PCBA circuit board can be pushed toward the center by the two groups of second conveyor belts during the process of being conveyed by the first conveyor belt, so that the PCBA circuit board can be pushed to the center of the first conveyor belt, ensuring the consistency of the PCBA circuit board when being inspected by the defect detection mechanism.

[0011] As a preferred embodiment of the present invention, the defect detection mechanism includes a mounting plate, which is fixedly mounted on the upper surface of the support column, which is fixedly mounted on the upper surface of the L-shaped frame, and a first mounting opening and a second mounting opening are provided on the lower surface of the mounting plate. A rotating groove is provided on one side of the first mounting opening, and a turntable is rotatably mounted in the rotating groove. A camera is fixedly mounted on one end of the turntable so that the camera faces the center of the first conveyor belt through the first mounting opening.

[0012] As a further preferred embodiment of the present invention, a lighting lamp is fixedly installed at the other end of the first installation opening, and the illuminating end of the lighting lamp illuminates the center of the first conveyor belt, so that the lighting lamp can illuminate the surface of the PCBA circuit board conveyed by the first conveyor belt.

[0013] As a further preferred embodiment of the present invention, a torsion bolt is threadedly installed in the second mounting port, and the threaded portion of the torsion bolt can be threaded into the rotating groove, so that the threaded portion of the torsion bolt can touch the outer surface of the turntable rotatably installed in the rotating groove, thereby providing locking force for the camera after angle adjustment.

[0014] As a preferred solution of the present invention, a connecting plate is fixedly installed at one end of the mounting plate, a laser sensor is fixedly installed in the connecting plate, a signal transmitting end of the laser sensor is connected to a signal receiving end of the controller, and a control output end of the controller is electrically connected to an electric control end of the electric push rod;

[0015] The camera is connected to the controller for data transmission via Ethernet.

[0016] As a preferred solution of the present invention, the models of the laser sensor and the controller are Keyence LV-H32 and STM32 respectively.

[0017] The working method of the camera identifying PCBA circuit board defects through the controller in the present invention is divided into the following steps:

[0018] S1. For PCBA circuit boards, their normal circuits, solder joints and components have specific shapes. The controller extracts shape features through the Canny edge detection algorithm. The Canny edge detection algorithm mainly includes steps such as noise suppression, calculation of gradient amplitude and direction, non-maximum suppression and dual threshold detection. Through these steps, the edge shape of each element on the circuit board can be accurately detected, providing a basis for judging whether there are defects;

[0019] S2. Different areas of the circuit board, such as normal circuit areas and circuit areas with short circuit risks, may have different texture features. The controller uses the texture analysis algorithm gray level co-occurrence matrix GLCM to extract texture features. GLCM obtains parameters reflecting texture features by counting the frequency of occurrence of different gray level pixel pairs in a specific direction and distance in the image, such as contrast, correlation, energy, etc. These parameters are helpful to identify abnormal texture areas on the circuit board and thus find potential defects.

[0020] S3. The controller compares the extracted features with the predefined normal feature templates. If there are obvious differences, there may be defects. For example, a normal solder joint should be round and full in shape. If the shape feature extraction finds that the shape of a solder joint is irregular, there is a welding defect. Once the defect is detected, the controller will send a signal to the electric push rod to push it out of the production line.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] When inspecting defects on PCBA circuit boards, the produced PCBA circuit boards can be conveyed to the surface of the first conveyor belt. As the first conveyor belt is conveyed, the PCBA circuit boards will be pushed to the center of the first conveyor belt by the central pushing mechanisms suspended on both sides of the upper surface of the first conveyor belt. The PCBA circuit boards pushed to the center of the first conveyor belt will enter directly below the defect detection mechanism as they are conveyed, enabling the camera in the defect detection mechanism to take a picture of the PCBA circuit board, and the camera uses an optical lens to focus the reflected light from the surface of the PCBA circuit board onto the image sensor CMOS. The camera uses an optical lens to focus the reflected light from the surface of the PCBA circuit board onto the image sensor CMOS, converting the light signal into an electrical signal, and then generating a digital image signal. This digital image contains various information about the PCBA circuit board, such as circuit layout, solder joints, components, etc., as well as possible defect information. The image acquisition operation is completed, and then the camera can transmit the collected image data to the controller through Ethernet. The controller compares the extracted features with the predefined normal feature template. If there is an obvious difference, there is a defect. The normal solder joint should be round and full. Through shape feature extraction, it is found that the shape of a certain solder joint is irregular, which may be a welding defect. Once a defect is detected, the controller will detect the position information of the PCBA circuit board according to the laser sensor in the defect detection mechanism. After the PCBA circuit board is transported by the first conveyor belt and detected by the laser sensor, it will send a signal to the controller, and the controller will control the electric push rod to push the push plate, and push the PCBA circuit board transported to the current position into the slide, so that the defective PCBA circuit board can be pushed out of the production line and stored centrally for the staff to repair it.

[0023] The PCBA circuit board is pushed to the center of the first conveyor belt through the centering pushing mechanism, ensuring that the position of the PCBA circuit board is relatively fixed and standard when it enters directly under the defect detection mechanism. This helps the defect detection mechanism to obtain more stable and consistent images when taking photos, reducing problems such as image deformation or missing parts of the area that may be caused by the position offset of the PCBA circuit board, thereby improving the accuracy of subsequent image comparison. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only examples of the embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 It is a structural schematic diagram of an embodiment of the present invention;

[0026] Figure 2 It is a schematic diagram of the structure of the push plate and the slide in the embodiment of the present invention;

[0027] Figure 3 It is a structural schematic diagram of the second conveyor belt in an embodiment of the present invention;

[0028] Figure 4 It is a structural schematic diagram of the centering pushing mechanism in an embodiment of the present invention;

[0029] Figure 5 A schematic diagram of the structure of a camera and an illumination lamp in an embodiment of the present invention;

[0030] Figure 6 Schematic diagram of the structure of the defect detection mechanism in an embodiment of the present invention.

[0031] 1. Docking plate; 101. First transmission column; 102. First conveyor belt; 103. Speed ​​reduction motor; 104. Slide; 105. Electric push rod; 106. Push plate; 2. Centering push mechanism; 201. First bevel gear; 202. Second bevel gear; 203. Connecting arm; 204. Second transmission column; 205. Second conveyor belt; 206. L-shaped frame; 207. Support column; 208. Third transmission column; 209. Connecting column; 3. Defect detection mechanism; 301. Mounting plate; 302. First mounting port; 303. Illuminating lamp; 304. Camera; 305. Turntable; 306. Second mounting port; 307. Connecting plate; 308. Laser sensor; 309. Rotating groove; 310. Torsion bolt. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0033] In the description of the embodiments of the present invention, it should be understood that the terms "upper", "lower", "front", "back", "inside", "outside", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the embodiments of the present invention.

[0034] In the description of the embodiments of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installation", "connection" and "connection" should be understood in a broad sense. For example, it can be a fixed connection, an integral connection, or a detachable connection; it can be the internal connection of two components; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the embodiments of the present invention can be understood according to specific circumstances.

[0035] like Figure 1-Figure 2 As shown, a PCBA circuit board defect detection and classification system based on deep learning according to an embodiment of the present invention comprises two groups of docking plates 1, two groups of first transmission columns 101 are transmission-installed between the two groups of docking plates 1, and first conveyor belts 102 are sleeved on the outer surfaces of the two groups of first transmission columns 101. A reduction motor 103 is fixedly installed at one end of the docking plate 1, and the output shaft of the reduction motor 103 passes through the docking plate 1 and is fixedly connected to the first transmission column 101, while both sides of the other group of first transmission columns 101 are meshed with a central pushing mechanism 2, and the pushing end of the central pushing mechanism 2 is suspended on the upper surface of the first conveyor belt 102 and is in a contracted shape from wide to narrow, so that the central pushing mechanism 2 can push the PCBA circuit board conveyed by the first conveyor belt 102 to the center of the first conveyor belt 102, so that the PCBA circuit board can be flush with the defect detection mechanism 3 during the conveying process, and the defect detection mechanism 3 is suspended and installed on the upper surface of the contracted end of the central pushing mechanism 2.

[0036] An electric push rod 105 is fixedly installed at one end of the docking plate 1. The piston rod of the electric push rod 105 passes through the docking plate 1 and a push plate 106 is fixedly installed at the end, so that the push plate 106 can push the PCBA circuit board detected as having defects by the defect detection mechanism 3 from the upper surface of the first conveyor belt 102 into the slide 104. The slide 104 is fixedly installed at one end of another group of docking plates 1 and is flush with the push plate 106.

[0037] When inspecting defects on PCBA circuit boards, the produced PCBA circuit boards are conveyed to the surface of the first conveyor belt 102. As the first conveyor belt 102 conveys the PCBA circuit boards, they are pushed to the center of the first conveyor belt 102 by the central pushing mechanisms 2 suspended on both sides of the upper surface of the first conveyor belt 102. The PCBA circuit boards pushed to the center of the first conveyor belt 102 enter directly below the defect detection mechanism 3 as they are conveyed, so that the camera 304 in the defect detection mechanism 3 can take pictures of the PCBA circuit boards, and the camera 304 uses an optical lens to focus the reflected light on the surface of the PCBA circuit boards onto the image sensor CMOS, and converts the light signals into electrical signals, thereby generating digital image signals. The digital images contain various information about the PCBA circuit boards, such as circuit layout, solder joints, components, etc., and also contain possible defect information, thus completing the inspection. After the image acquisition operation is completed, the camera 304 can then transmit the acquired image data to the controller via Ethernet. The controller compares the extracted features with the predefined normal feature template. If there is an obvious difference, there may be a defect. For example, a normal solder joint should be a round and full shape. If the shape of a solder joint is found to be irregular through shape feature extraction, there may be a welding defect. Once a defect is detected, the controller will detect the position information of the PCBA circuit board according to the laser sensor 308 in the defect detection mechanism 3, until the PCBA circuit board is transported by the first conveyor belt 102 to the laser sensor 308, it will send a signal to the controller, and the controller will control the electric push rod 105 to push the push plate 106, and push the PCBA circuit board transported to the current position into the slide 104, so that the defective PCBA circuit board can be pushed out of the production line and stored centrally for the staff to repair it.

[0038] The PCBA circuit board is pushed to the center of the first conveyor belt 102 by the centering pushing mechanism 2, ensuring that the position of the PCBA circuit board is relatively fixed and standard when it enters directly under the defect detection mechanism 3. This helps the defect detection mechanism 3 to obtain more stable and consistent images when taking photos, reducing problems such as image deformation or missing parts of the area that may be caused by the position offset of the PCBA circuit board, thereby improving the accuracy of subsequent image comparison.

[0039] like Figure 3-Figure 4As shown, the centering pushing mechanism 2 includes two groups of first bevel gears 201, which are fixedly mounted on both sides of the first transmission column 101 and located at the outer end of the docking plate 1, the first bevel gear 201 is meshed with the second bevel gear 202, the second bevel gear 202 is fixedly mounted on the lower end of the outer surface of the connecting column 209, the connecting column 209 is rotatably mounted in the connecting arm 203, the connecting arm 203 is fixedly mounted on one end of the docking plate 1, the upper end of the connecting column 209 is fixedly mounted with a second transmission column 204, the outer surface of the second transmission column 204 is sleeved with a second conveyor belt 205, the other end of the second conveyor belt 205 is sleeved on the outer surface of the third transmission column 208, the third transmission column 208 is rotatably mounted on the outer surface of the L-shaped frame 206, and the L-shaped frame 206 is fixedly mounted on one end of the docking plate 1.

[0040] The second conveyor belt 205 is supported by the second transmission column 204 and the third transmission column 208, so that the second conveyor belt 205 is in a contracted shape from wide to narrow, so that the PCBA circuit board can be pushed toward the center by the two groups of second conveyor belts 205 during the process of being conveyed by the first conveyor belt 102, so that the PCBA circuit board can be pushed to the center of the first conveyor belt 102, ensuring the consistency of the PCBA circuit board when being inspected by the defect detection mechanism 3.

[0041] Through the design of the first bevel gear 201, the second bevel gear 202, the second transmission column 204, the second conveyor belt 205 and the third transmission column 208, the first conveyor belt 102 will drive the first bevel gears 201 on both sides to rotate together during the process of conveying the PCBA circuit board by the first conveyor belt 102, and the first bevel gear 201 will mesh with the second bevel gear 202 to rotate, so that the second bevel gear 202 drives the second transmission column 204 on the upper surface of the connecting column 209 to rotate, and the second conveyor belt 205 set on the outer surface of the second transmission column 204 can be conveyed between the third transmission column 208, thereby realizing that the first conveyor belt 102 conveys the PCBA circuit board while driving the second conveyor belt 205 for transmission, and the PCBA circuit board As the first conveyor belt 102 is conveyed, it will be conveyed between the two groups of second conveyor belts 205 by the first conveyor belt 102, and the two groups of second conveyor belts 205 are in a contracted shape from wide to narrow, so that the two groups of second conveyor belts 205 can push the PCBA circuit board to the center of the first conveyor belt 102, and at the same time, the two groups of second conveyor belts 205 can be clamped on both sides of the PCBA circuit board for the defect detection mechanism 3 to shoot it to detect defects, and the pushing and clamping effects provide a stable shooting position for the defect detection mechanism 3, which makes the position deviation of the PCBA circuit board extremely small during each shooting, ensuring the consistency of image acquisition. For example, when detecting the welding condition of tiny components, accurate shooting position can ensure that the component position in the image is fixed, which is convenient for accurate identification of welding defects.

[0042] like Figure 5-Figure 6 As shown, the defect detection mechanism 3 includes a mounting plate 301, which is fixedly mounted on the upper surface of the support column 207, and the support column 207 is fixedly mounted on the upper surface of the L-shaped frame 206. A first mounting port 302 and a second mounting port 306 are provided on the lower surface of the mounting plate 301. A rotating groove 309 is provided on one side of the first mounting port 302, and a turntable 305 is rotatably mounted in the rotating groove 309. A camera 304 is fixedly mounted on one end of the turntable 305 so that the camera 304 faces the center of the first conveyor belt 102 through the first mounting port 302.

[0043] The other end of the first installation opening 302 is fixedly installed with a lighting lamp 303 , and the irradiation end of the lighting lamp 303 illuminates the center of the first conveyor belt 102 , so that the lighting lamp 303 can illuminate the surface of the PCBA circuit board conveyed by the first conveyor belt 102 .

[0044] A torsion bolt 310 is installed in the inner thread of the second installation port 306, and the threaded portion of the torsion bolt 310 can be threaded into the rotating groove 309, so that the threaded portion of the torsion bolt 310 can touch the outer surface of the turntable 305 rotatably installed in the rotating groove 309, thereby providing a locking force for the camera 304 after the angle is adjusted.

[0045] A connecting plate 307 is fixedly installed at one end of the mounting plate 301 , and a laser sensor 308 is fixedly installed inside the connecting plate 307 . The signal transmitting end of the laser sensor 308 is connected to the signal receiving end of the controller, and the control output end of the controller is electrically connected to the electric control end of the electric push rod 105 .

[0046] The camera 304 is connected to the controller for data transmission via Ethernet.

[0047] The models of the laser sensor 308 and the controller are Keyence LV-H32 and STM32 respectively.

[0048] The working method of the camera 304 identifying PCBA circuit board defects through the controller is divided into the following steps:

[0049] S1. For PCBA circuit boards, their normal circuits, solder joints and components have specific shapes. The controller extracts shape features through the Canny edge detection algorithm. The Canny edge detection algorithm mainly includes noise suppression, calculation of gradient amplitude and direction, non-maximum suppression and dual threshold detection. Through these steps, the edge shape of each element on the circuit board can be accurately detected, providing a basis for judging whether there are defects.

[0050] S2. Different areas of the circuit board, such as normal line areas and line areas with short circuit risks, may have different texture features. The controller uses the texture analysis algorithm gray level co-occurrence matrix GLCM to extract texture features. GLCM obtains parameters reflecting texture features by counting the frequency of occurrence of different gray level pixel pairs in the image in a specific direction and distance, such as contrast, correlation, energy, etc. These parameters help to identify abnormal texture areas on the circuit board and thus discover potential defects.

[0051] S3. The controller compares the extracted features with the predefined normal feature template. If there are obvious differences, there may be defects. For example, a normal solder joint should be round and full in shape. If the shape feature extraction finds that the shape of a solder joint is irregular, there is a welding defect. Once the defect is detected, the controller sends a signal to the electric push rod 105 to push it out of the production line.

[0052] When the PCBA circuit board is pushed to the center of the first conveyor belt 102, it enters the bottom of the installation plate 301 and is within the illumination range of the camera 304 and the lighting lamp 303 at the same time, so that the camera 304 takes a picture of the illuminated PCBA circuit board, and uses an optical lens to focus the reflected light on the surface of the PCBA circuit board onto the image sensor CMOS, converts the light signal into an electrical signal, and then generates a digital image signal. This digital image contains various information of the PCBA circuit board, such as circuit layout, solder joints, components, etc., and also contains possible defect information, which completes the image acquisition operation. Then, the camera 304 can transmit the collected image data to the controller through Ethernet, so that the controller extracts shape features through the edge detection algorithm Canny, calculates gradient amplitude and direction, non-maximum suppression and dual threshold detection, etc. Through these steps, the edge shape of each element on the circuit board can be accurately detected, providing a basis for judging whether there is a defect, and at the same time, the texture analysis algorithm gray level co-occurrence matrix GLCM is used to extract texture features. GLCM obtains parameters reflecting texture features, such as contrast, correlation, energy, etc., by counting the frequency of occurrence of different grayscale pixel pairs in a specific direction and distance in an image. These parameters are helpful to identify abnormal texture areas on the circuit board and then find potential defects. The extracted features are compared with the predefined normal feature template. If there is a significant difference, there may be a defect. For example, a normal solder joint should be a round and full shape. If the shape of a solder joint is found to be irregular through shape feature extraction, there is a welding defect. Once the defect is detected, the controller will detect the position where the PCBA circuit board is conveyed through the laser sensor 308. After the PCBA circuit board is conveyed by the first conveyor belt 102 and detected by the laser sensor 308, a signal will be sent to the controller, and the controller will control the electric push rod 105 to push the push plate 106, so that the push plate 106 pushes the PCBA circuit board conveyed to the current position into the slide 104, and can push the defective PCBA circuit board out of the production line, thereby realizing automated defect separation, which avoids the tedious process of manually screening defective products, improves the automation level of the entire production line, and makes the production process smoother.

[0053] According to the above technical solution, the working steps of the embodiment of the present invention are sorted out: when detecting defects in PCBA circuit boards, the produced PCBA circuit boards can be conveyed to the surface of the first conveyor belt 102. During the process of the first conveyor belt 102 conveying the PCBA circuit boards, the first transmission column 101 will drive the first bevel gears 201 on both sides to rotate together, and the first bevel gear 201 will mesh with the second bevel gear 202 to rotate, so that the second bevel gear 202 drives the second transmission column 204 on the upper surface of the connecting column 209 to rotate, so that the second transmission column 204 can drive the second conveyor belt 205 on the outer surface to be transported between the third transmission column 208, thereby achieving the first conveyor belt 102 to convey the PCBA circuit The second conveyor belt 205 will be driven together with the PCBA circuit board when it is conveyed by the first conveyor belt 102, and the PCBA circuit board will be conveyed between the two sets of second conveyor belts 205 as it is conveyed by the first conveyor belt 102. The two sets of second conveyor belts 205 are in a contracted shape from wide to narrow, so that the two sets of second conveyor belts 205 can push the PCBA circuit board to the center of the first conveyor belt 102, and at the same time, the two sets of second conveyor belts 205 can be clamped on both sides of the PCBA circuit board. When the PCBA circuit board is pushed to the center of the first conveyor belt 102, it enters the lower part of the mounting plate 301 and is within the irradiation range of the camera 304 and the lighting lamp 303 at the same time, so that the camera 304 can take a picture of the illuminated PCBA circuit board. The camera 304 uses an optical lens to focus the reflected light from the surface of the PCBA circuit board onto the image sensor CMOS, converts the optical signal into an electrical signal, and then generates a digital image signal. This digital image contains various information about the PCBA circuit board, such as circuit layout, solder joints, components, etc., and also contains possible defect information. The image acquisition operation is completed, and then the camera 304 can transmit the collected image data to the controller via Ethernet, so that the controller extracts shape features through the edge detection algorithm Canny, calculates gradient amplitude and direction, non-maximum suppression and dual threshold detection, etc., through These steps can accurately detect the edge shape of each element on the circuit board, providing a basis for judging whether there are defects, and at the same time use the texture analysis algorithm gray level co-occurrence matrix GLCM to extract texture features. GLCM obtains parameters reflecting texture features by counting the frequency of occurrence of different gray level pixel pairs in the image in a specific direction and distance, such as contrast, correlation, energy, etc. These parameters help to identify abnormal texture areas on the circuit board, and then find potential defects, and compare the extracted features with the predefined normal feature templates. If there are obvious differences, there may be defects. For example, a normal solder joint should be a round and full shape. If the shape feature extraction finds that the shape of a solder joint is irregular, there is a welding defect.Once a defect is detected, the controller will detect the position where the PCBA circuit board is conveyed through the laser sensor 308. When the PCBA circuit board is conveyed by the first conveyor belt 102 and detected by the laser sensor 308, a signal will be sent to the controller, and the controller will control the electric push rod 105 to push the push plate 106, so that the push plate 106 pushes the PCBA circuit board conveyed to the current position into the slide 104, and the defective PCBA circuit board can be pushed out of the production line, thereby realizing automatic defect separation.

[0054] In summary, the present invention can automatically push defective PCBA circuit boards out of the production line, thereby realizing automated defect separation, avoiding the tedious process of manually screening defective products, improving the degree of automation of the entire production line, and making the production process smoother.

[0055] The above shows and describes the basic principles of the present invention. The above are only preferred embodiments of the present invention and are not intended to limit the present invention. The above embodiments and descriptions in the specification are only intended to illustrate the principles of the present invention. Without departing from the scope of the present invention, any modifications, equivalent substitutions and improvements made within the spirit and scope of the present invention should be included in the protection scope of the present invention.

Claims

1. A PCBA circuit board defect detection and classification system based on deep learning, characterized by: The invention comprises two groups of docking plates (1), two groups of first transmission columns (101) are installed in a transmission manner between the two groups of docking plates (1), the outer surfaces of the two groups of first transmission columns (101) are sleeved with first conveyor belts (102), a reduction motor (103) is fixedly installed at one end of the docking plate (1), the output shaft of the reduction motor (103) passes through the docking plate (1) and is fixedly connected to the first transmission column (101), and both sides of the other group of first transmission columns (101) are meshed with a central pushing mechanism (2), the pushing end of the central pushing mechanism (2) is suspended on the upper surface of the first conveyor belt (102) and is in a narrow shape from wide to narrow, so that the central pushing mechanism (2) can push the PCBA circuit board conveyed by the first conveyor belt (102) to the center of the first conveyor belt (102), so that the PCBA circuit board can be flush with the defect detection mechanism (3) during the conveying process, and the defect detection mechanism (3) is suspended on the upper surface of the narrow end of the central pushing mechanism (2).

2. According to a PCBA circuit board defect detection and classification system based on deep learning according to claim 1, it is characterized in that: An electric push rod (105) is fixedly mounted on one end of the docking plate (1); a piston rod of the electric push rod (105) passes through the docking plate (1) and a push plate (106) is fixedly mounted on the end thereof, so that the push plate (106) can push a PCBA circuit board detected as having a defect by the defect detection mechanism (3) from the upper surface of the first conveyor belt (102) into a slide (104); the slide (104) is fixedly mounted on one end of another set of docking plates (1) and is flush with the push plate (106).

3. According to a PCBA circuit board defect detection and classification system based on deep learning according to claim 1, it is characterized in that: The centering pushing mechanism (2) comprises two groups of first bevel gears (201), the two groups of first bevel gears (201) are fixedly mounted on both sides of the first transmission column (101) and are located at the outer end of the docking plate (1), the first bevel gears (201) are meshed with the second bevel gears (202), the second bevel gears (202) are fixedly mounted on the lower end of the outer surface of the connecting column (209), the connecting column (209) is rotatably mounted in the connecting arm (203), the connecting arm (203) is fixedly mounted on one end of the docking plate (1), the upper end of the connecting column (209) is fixedly mounted with a second transmission column (204), the outer surface of the second transmission column (204) is sleeved with a second conveyor belt (205), the other end of the second conveyor belt (205) is sleeved on the outer surface of a third transmission column (208), the third transmission column (208) is rotatably mounted on the outer surface of an L-shaped frame (206), and the L-shaped frame (206) is fixedly mounted on one end of the docking plate (1).

4. A PCBA circuit board defect detection and classification system based on deep learning according to claim 3, characterized in that: The second conveyor belt (205) is supported by the second transmission column (204) and the third transmission column (208), so that the second conveyor belt (205) is in a contracted shape from wide to narrow, so that the PCBA circuit board can be pushed toward the center by the two groups of second conveyor belts (205) during the process of being conveyed by the first conveyor belt (102), so that the PCBA circuit board can be pushed to the center of the first conveyor belt (102), thereby ensuring the consistency of the PCBA circuit board when being inspected by the defect inspection mechanism (3).

5. According to a PCBA circuit board defect detection and classification system based on deep learning according to claim 1, it is characterized in that: The defect detection mechanism (3) comprises a mounting plate (301), wherein the mounting plate (301) is fixedly mounted on the upper surface of a support column (207), wherein the support column (207) is fixedly mounted on the upper surface of an L-shaped frame (206), wherein a first mounting opening (302) and a second mounting opening (306) are provided on the lower surface of the mounting plate (301), wherein a rotation groove (309) is provided on one side of the first mounting opening (302), wherein a rotating disk (305) is rotatably mounted in the rotating groove (309), and a camera (304) is fixedly mounted on one end of the rotating disk (305), such that the camera (304) faces the center of the first conveyor belt (102) through the first mounting opening (302).

6. A PCBA circuit board defect detection and classification system based on deep learning according to claim 5, characterized in that: A lighting lamp (303) is fixedly mounted at the other end of the first mounting opening (302), and an illuminating end of the lighting lamp (303) illuminates the center of the first conveyor belt (102), so that the lighting lamp (303) can illuminate the surface of the PCBA circuit board conveyed by the first conveyor belt (102).

7. A PCBA circuit board defect detection and classification system based on deep learning according to claim 5, characterized in that: A torsion bolt (310) is threadedly installed in the second installation port (306), and the threaded portion of the torsion bolt (310) can be threadedly penetrated into the rotation groove (309), so that the threaded portion of the torsion bolt (310) can contact the outer surface of the turntable (305) rotatably installed in the rotation groove (309), thereby providing a locking force for the camera (304) after the angle is adjusted.

8. A PCBA circuit board defect detection and classification system based on deep learning according to claim 5, characterized in that: A connecting plate (307) is fixedly mounted on one end of the mounting plate (301), a laser sensor (308) is fixedly mounted inside the connecting plate (307), a signal transmitting end of the laser sensor (308) is connected to a signal receiving end of the controller, and a control output end of the controller is electrically connected to an electric control end of the electric push rod (105); The camera (304) is connected to the controller for data transmission via an Ethernet connection.

9. A PCBA circuit board defect detection and classification system based on deep learning according to claim 8, characterized in that: The models of the laser sensor (308) and the controller are Keyence LV-H32 and STM32 respectively.

10. A PCBA circuit board defect detection and classification system based on deep learning according to claim 8, characterized in that: The working method of the camera (304) identifying PCBA circuit board defects through the controller is divided into the following steps: S1. For PCBA circuit boards, their normal circuits, solder joints and components have specific shapes. The controller extracts shape features through the Canny edge detection algorithm. The Canny edge detection algorithm mainly includes noise suppression, calculation of gradient amplitude and direction, non-maximum suppression and dual threshold detection steps. Through these steps, the edge shape of each element on the circuit board is accurately detected, providing a basis for judging whether there are defects; S2. Different areas of the circuit board, such as normal circuit areas and circuit areas with short circuit risks, may have different texture features. The controller uses the texture analysis algorithm gray level co-occurrence matrix GLCM to extract texture features. GLCM obtains parameters reflecting texture features by counting the frequency of occurrence of different gray level pixel pairs in a specific direction and distance in the image. These parameters are helpful to identify abnormal texture areas on the circuit board and thus find potential defects. S3. The controller compares the extracted features with the predefined normal feature template. If there is a significant difference, there may be a defect. A normal solder joint is a round and full shape. If the shape of a solder joint is found to be irregular through shape feature extraction, there is a welding defect. Once the defect is detected, the controller sends a signal to the electric push rod (105) to push it out of the production line.

Images