Full-automatic strip-shaped workpiece appearance detection device based on convolutional neural network

Abstract

The utility model discloses a full-automatic strip-shaped workpiece appearance detection device based on a convolutional neural network. The full-automatic strip-shaped workpiece appearance detection device comprises a mechanical arm, an edge calculation control module, a camera, a feeding and discharging conveying system and a detection platform, the device has the beneficial effects that the device can detect produced workpieces in batches, can detect and classify mixed workpieces, can realize automatic feeding of the workpieces to the detection objective table, can adopt a PaddleClas algorithm to identify pictures shot by a camera, and if the detection is qualified after the identification is completed, the device can be used for detecting and classifying the workpieces, so that the production efficiency is improved, and the production cost is reduced. If a defect exists, the edge calculation control module can output a qualified signal, if the defect exists, the equipment gives out an alarm sound, a screen on the edge calculation control module also visually displays a defect spot, and the mechanical arm automatically places qualified products and unqualified products to different areas according to a detection result; the whole detection process does not need personnel to participate in operation and realizes full-automatic detection.

Description

Technical Field

[0001] The utility model relates to the technical field of workpiece appearance detection, in particular to a full-automatic strip workpiece appearance detection device based on a convolutional neural network. Background Art

[0002] Strip-shaped workpieces are widely used in the market, such as circuit boards for Bluetooth headsets, electronic cigarette cases, and sensor components. All of these products require inspection during the production process, using mainstream methods such as visual inspection, laser inspection, and ultrasonic inspection. Currently, the inspection equipment for these products uses semi-automatic or manual loading and unloading. Furthermore, most inspection equipment is specialized for specific workpieces, with no universal testing equipment available, resulting in low efficiency. Therefore, a fully automated appearance inspection mechanism for strip-shaped workpieces was developed and designed. Utility Model Content

[0003] The purpose of this utility model is to solve the above problems. A fully automatic strip workpiece appearance inspection device based on convolutional neural network is designed to solve the problem that the loading and unloading of current product inspection equipment during the inspection process are semi-automatic or manually placed, and at the same time, most workpieces are subdivided inspection equipment, there is no universal inspection equipment, and the work efficiency is low.

[0004] To achieve the above-mentioned purpose, the technical solution of the present invention is: a fully automatic strip workpiece appearance detection device based on convolutional neural network, characterized by comprising:

[0005] Robotic arm, edge computing control module, camera, loading and unloading conveying system and inspection platform;

[0006] The edge computing control module and the camera are arranged on the robotic arm, the loading and unloading conveying system is arranged on one side of the robotic arm, and the detection platform is arranged on one side of the robotic arm;

[0007] The edge computing control module uses the PaddlePaddle image recognition suite PaddleClas and PaddleDetection algorithm or the TensorFlow or Pytorch development kit.

[0008] Preferably, the robotic arm comprises:

[0009] Robotic arm end and grasping end;

[0010] The grasping end is arranged on the end of the robotic arm, and the camera is arranged on the end of the robotic arm.

[0011] Preferably, the loading and unloading conveying system includes:

[0012] Linear feeder, feeding vibration plate, recovery container and conveyor belt;

[0013] The linear feeder is arranged on one side of the robotic arm end, the loading vibration plate is arranged at the end of the linear feeder, the recovery container is arranged on one side of the loading vibration plate, and the conveyor belt is arranged on one side of the linear feeder, the loading vibration plate and the recovery container.

[0014] Preferably, the detection platform comprises:

[0015] Motor and stage;

[0016] The motor is arranged on one side of the upper part of the recovery container, and the loading platform is arranged on the driving end of the motor. The loading platform is located above the recovery container.

[0017] Preferably, the edge computing control module is used to identify and process the images transmitted back by the camera, and the edge computing control module provides the orientation coordinate information of the workpiece to be grasped to the robotic arm end and the grasping end, and executes the motion program for the motor, robotic arm end and the grasping end of the detection platform, and the camera is used for transmitting the images.

[0018] Preferably, the robotic arm adopts a four-axis robotic arm structure, and the robotic arm end and the grabbing end are used to grab the workpiece in the loading vibration plate and place it on the detection platform, or the robotic arm end and the grabbing end grab the workpiece on the detection platform and place it on the conveyor belt in the loading and unloading conveying system.

[0019] Preferably, the linear feeder is used to transfer the workpiece to the feeding vibration plate.

[0020] Preferably, the loading vibration plate is used to vibrate and disperse the workpieces to prevent stacking.

[0021] Preferably, the recovery container is used to collect unqualified products.

[0022] Preferably, the conveyor belt is used to place qualified products.

[0023] The fully automatic strip workpiece appearance inspection device based on convolutional neural network produced by the technical solution of the present invention can batch inspect the produced workpieces, and can also inspect and classify mixed workpieces. The workpieces can be automatically loaded to the inspection stage. The edge computing control module can adopt the PaddleClas algorithm to recognize the picture taken by the camera. After the recognition is completed, if the inspection is qualified, the edge computing control module can output a qualified signal. If there are defects, the equipment will sound an alarm. The screen on the edge computing control module also intuitively displays the defect points. The robotic arm automatically places qualified and unqualified products into different areas according to the inspection results. The entire inspection process does not require human participation to achieve fully automatic inspection. At the same time, it can also statistically output the inspection results of each batch, has both inspection statistics functions, and can output relevant data information of inspection statistics. Only one edge computing control module 4 is used. On the basis of the robotic arm's visual grasping, the start and stop of the loading vibration plate 7 and the conveyor belt 8 are also determined without the intervention of sensors. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a schematic diagram of the main structure of a fully automatic strip workpiece appearance detection device based on a convolutional neural network described in the present invention.

[0025] Figure 2 This is a rear view structural schematic diagram of a fully automatic strip workpiece appearance detection device based on a convolutional neural network described in the present invention.

[0026] Figure 3 This is a workflow diagram of a fully automatic strip workpiece appearance detection device based on convolutional neural network described in the present invention.

[0027] In the figure: 1. Robotic arm end, 2. Camera, 3. Motor, 4. Edge computing control module, 5. Linear feeder, 6. Recovery container, 7. Loading vibration plate, 8. Conveyor belt, 9. Loading platform, 10. Grasping end. DETAILED DESCRIPTION

[0028] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1-3 As shown in the figure, a fully automatic strip workpiece appearance detection device based on convolutional neural network.

[0029] Through the use of wires by those skilled in the art, all electrical components in this case are connected to their corresponding power supplies, and appropriate controllers should be selected according to actual conditions to meet control requirements. The specific connection and control sequence should refer to the following working principle, in which the electrical components are electrically connected in sequence. The detailed connection means are well-known technologies in this field. The following mainly introduces the working principle and process, and does not explain the electrical control.

[0030] Embodiment: A fully automatic strip workpiece appearance inspection device based on a convolutional neural network includes: a robotic arm, an edge computing control module 4, a camera 2, a loading and unloading conveying system, and a detection platform; the edge computing control module 4 and the camera 2 are arranged on the robotic arm, the loading and unloading conveying system is arranged on one side of the robotic arm, and the detection platform is arranged on one side of the robotic arm; the edge computing control module 4 adopts the PaddlePaddle image recognition suite PaddleClas and PaddleDetection algorithms or the TensorFlow or Pytorch development kit;

[0031] Specifically, the robotic arm includes: a robotic arm end 1 and a grasping end 10; the grasping end 10 is arranged at the end of the robotic arm end 1, and the camera 2 is arranged at the end of the robotic arm end 1;

[0032] Specifically, the loading and unloading conveying system includes: a linear feeder 5, a loading vibration plate 7, a recovery container 6 and a conveyor belt 8; the linear feeder 5 is arranged on one side of the robot arm end 1, the loading vibration plate 7 is arranged at the end of the linear feeder 5, the recovery container 6 is arranged on one side of the loading vibration plate 7, and the conveyor belt 8 is arranged on one side of the linear feeder 5, the loading vibration plate 7 and the recovery container 6;

[0033] Specifically, the detection platform includes: a motor 3 and a loading platform 9; the motor 3 is arranged on the upper side of the recovery container 6, and the loading platform 9 is arranged on the driving end of the motor 3, and the loading platform 9 is located above the recovery container 6.

[0034] It should be noted that:

[0035] As a preferred embodiment, further, the robot arm end 1 and the grabbing end 10 are mainly used to grab the workpiece in the loading vibration plate 7 and place it on the detection platform, or grab the workpiece on the detection platform and place it on the conveyor belt 8 in the loading and unloading conveying system;

[0036] As a preferred embodiment, further, the edge computing control module 4 is mainly used to identify and process the images sent back by the camera 2. The edge computing control module 4 can provide the robot arm end 1 and the gripping end 10 with the orientation coordinate information of the workpiece to be grasped, and execute the motion program for the motor 3 of the detection platform and the robot arm end 1;

[0037] As a preference, further, the camera 2 is installed at the end of the robotic arm end 1, mainly used for image transmission. When grabbing the workpiece, the camera 2 and the grabbing end 10 are both above the loading vibration plate 7, and the camera 2 transmits the image of the vibration plate below to the edge computing control module 4. When the robotic arm end 1 and the grabbing end 10 place the workpiece on the inspection platform, the camera 2 is located above the inspection platform, and at this time the camera 2 transmits the image to the edge computing control module 4;

[0038] As a preferred embodiment, further, the linear feeder 5 mainly transfers the workpiece to the loading vibration plate 7, and starts and stops according to the judgment result of the edge computing control module 4 on the number of workpieces in the loading vibration plate 7;

[0039] As a preferred embodiment, further, the loading vibration plate 7 mainly spreads out the workpieces to prevent stacking, and the edge computing control module 4 starts and stops according to the judgment result of whether the workpieces in the loading vibration plate 7 meet the grasping conditions;

[0040] As a preferred embodiment, further, the recovery container 6 is used to collect unqualified products, and the conveyor belt 8 is used to place qualified products;

[0041] The loading platform 9 is used to place the workpiece grasped by the robot arm end 1 and the grasping end 10. The start and stop of the motor 3 are controlled by the edge computing control module 4 based on the detection results of the workpiece;

[0042] It should be noted that the robotic arm end 1 and the gripping end 10 edit the movement trajectory program according to the position of the loading and unloading conveying system and the detection platform; when the motor 3 of the detection platform receives the unqualified signal from the edge computing control module 4, it will rotate, causing the workpiece to fall from the detection platform 9 to the recovery container 6. If a qualified signal is received, the motor 3 will not move, and the robotic arm end 1 and the gripping end 10 will grab the workpiece and place it on the conveyor belt 8; when a certain amount of products is placed on the conveyor belt 8, the conveyor belt 8 rotates forward or reversely to change the position of the workpiece on the belt so that the next batch of qualified workpieces can be placed;

[0043] It should be noted that the edge computing control module 4, as the information processing core of the entire recognition system, controls the program startup of the robotic arm end 1 and the grasping end 10 through image discrimination before loading, and controls the program startup of the motor 3 or the robotic arm through the image during detection; the convolutional neural network algorithm is used to train and learn the content that needs to be detected, such as workpiece recognition and classification, scratches, sand holes, bubbles, etc., mainly using the PaddlePaddle image recognition suite Paddl eC l as and Paddl eDetection algorithm, and the TensorFlow or Pytorch development kit can also be used; Paddl eC l as is a tool set for image recognition and image classification tasks prepared by PaddlePaddle for industry and academia, helping users train better visual models and application implementation; Paddl eDetection is an end-to-end target detection development kit based on Paddl ePaddle, which focuses on end-to-end industrial application while providing rich model components and test benchmarks, realizing the full process of data preparation, model selection, model training, and model deployment; the latest version of its open source algorithm update can be queried in the PP PaddlePaddle product panorama;

[0044] It should be noted that the camera 2 is fixed at the end of the robotic arm end 1. The camera 2, the robotic arm end 1 and the gripping end 10 are all perpendicular to the feeding vibration plate 7. They are connected to the edge computing control module via a USB cable for signal transmission. Their main function is to transmit the images in the feeding vibration plate 7 and the images of the detection stage 9 to the edge computing control module 4 for recognition.

[0045] It should be noted that the edge computing control module 4 is connected to the robot arm through a network cable to communicate and control the operation of the robot arm end 1 and the grasping trajectory program of the grasping end 10;

[0046] It should be noted that the linear feeder 5, the feeding vibration plate 7, the detection platform and the conveyor belt 8 are connected to the edge computing control module 4 via USB to transmit the control signal;

[0047] It should be noted that the start and stop of the linear feeder 5 is determined by the edge computing control module 4 receiving the camera 2 image to determine whether the number of workpieces in the feeding vibration plate 7 is less than the set number. If it is less than the set number, the feeding operation is started;

[0048] It should be noted that the loading vibration plate 7 receives the camera 2 image through the edge computing control module 4 to determine whether there is a workpiece that meets the requirements of the inspection platform. If not, the vibration plate works to adjust the workpiece to a state suitable for the inspection requirements by vibration;

[0049] It should be noted that the detection platform controls whether the motor 3 rotates based on the judgment result transmitted by the camera 2 to the edge computing control module 4; the conveyor belt 8 controls its forward and reverse rotation based on the number of times the robotic arm end 1 and the grasping end 10 grasp and place qualified workpieces.

[0050] Working principle: The operator puts the workpiece to be inspected into the linear feeder 5. When the equipment is started, the camera 2 above the loading vibration plate 7 will transmit the picture in the loading vibration plate 7 to the edge computing control module 4. The edge computing control module 4 determines whether the number of workpieces in the picture is less than the set number. If it is less than the set number, the edge computing control module 4 sends a signal to the linear vibration feeder to start working, and feeds the workpiece into the loading vibration plate 7 through vibration. If it is greater than or equal to the set number, the edge computing control module 4 determines whether there is a workpiece that meets the grasping conditions by identifying the picture information sent back by the camera 2 installed at the end of the robot arm 1. If there is no workpiece that meets the conditions, the loading vibration plate 7 will work and adjust the state of the workpiece by vibration. If If the conditions for grasping are met, the coordinate information of the grasped workpiece is transmitted to the robot arm end 1 and the grasping end 10 to execute the grasping program; the robot arm end 1 and the grasping end 10 grasp and then move to the detection platform, and the workpiece image on the detection stage 9 is transmitted to the edge computing control module 4 to judge whether it is qualified. If qualified, the robot arm end 1 and the grasping end 10 grasp the workpiece and place it on the conveyor belt 8. If unqualified, the motor 3 receives the signal and rotates to pour the workpiece into the recovery container 6 below to complete the entire detection process of a workpiece. Only one edge computing control module 4 is used. On the basis of the robot arm grasping through vision, the start and stop of the loading vibration plate 7 and the conveyor belt 8 are also determined without the intervention of sensors.

[0051] The inspection results for each batch are recorded in the edge computing module. Statistics can be exported if needed, such as the daily inspection pass rate, the total number of inspected parts, and the specific number of unqualified parts.

[0052] Example 1: Detection of sensor glass probe

[0053] Place the glass probe of the sensor in the linear feeder 5 and start the machine; first, the robot arm end 1 and the gripping end 10 will move to the top of the loading vibration plate 7 by default, and the camera 2 at the end of the robot arm end 1 will transmit the image of the loading vibration plate 7 to the edge computing control module 4 to identify whether there is material in the vibration plate and whether there is a workpiece that meets the requirements of the robot arm end 1 to be gripped. When all conditions are met, the robot arm end 1 and the gripping end 10 execute a fixed gripping trajectory program, and the robot arm end 1 and the gripping end 10 will open the appropriate gripping width according to the workpiece category feedback from the edge computing control module 4 to implement clamping and placement on the detection platform.

[0054] After the workpiece is placed on the inspection platform, the camera 2 at the end of the robotic arm 1 will execute the corresponding program according to the workpiece category feedback from the edge computing control module 4 in the previous step and descend to the appropriate height to inspect the appearance of the workpiece. At this time, the camera 2 transmits the picture to the edge computing control module 4 for appearance inspection. The glass probe is mainly composed of a glass ball with two magnesium-plated wires built in. Therefore, when there are bubbles in the glass ball, an unqualified signal will be output to control the motor 3 of the inspection platform to rotate, causing the workpiece to fall into the unqualified product recovery container 6. If the workpiece is qualified, the robotic arm 1 and the grasping end 10 execute the grasping program to grab the workpiece and place it on the conveyor belt 8 for placing qualified workpieces.

[0055] After testing, the original production inspection speed was 600 to 800 workpieces per hour. After replacing the equipment, the production inspection speed is about 1000 to 1200 workpieces per hour, and there is basically no missed inspection.

[0056] Example 2: Circuit board welding quality inspection

[0057] This implementation scheme is not suitable for placement on a rotating inspection platform because the circuit boards are large and different specifications or models of circuit boards may be mixed in during the production process. During implementation, the linear feeder 5 and the loading vibration plate 7 are replaced with a conveyor belt 8. When the workpiece to be inspected is received, the camera 2 at the end of the robotic arm end 1 performs fixed-point inspection or full inspection on the circuit board. The inspection results include whether there is tin deficiency, tin connection, tin accumulation, etc. If there are unqualified products, the robotic arm end 1 and the grasping end 10 will grab them according to the model of the circuit board and place them on another conveyor belt 8 for classified stacking to achieve product inspection of different specifications.

[0058] Example 3: Sensor probe packaging inspection

[0059] Because this implementation involves inspecting a cylindrical workpiece, the entire workpiece must be inspected. Therefore, the inspection platform's loading platform 9 is replaced with a fixture that is transparent or hollowed out from top to bottom, and a camera is installed underneath for a 360° appearance inspection. The implementation process is similar to that of Case 1.

[0060] The above technical solutions only reflect the preferred technical solutions of the present utility model. Any changes that may be made to certain parts thereof by technicians in this technical field all reflect the principles of the present utility model and fall within the scope of protection of the present utility model.

Claims

1. A fully automatic strip workpiece appearance detection device based on convolutional neural network, characterized in that: include: Grasping system, edge computing control module (1), camera (2), loading and unloading conveying system and detection platform; The edge computing control module (1) and the camera (2) are arranged on the gripping system, the loading and unloading conveying system is arranged on one side of the gripping system, and the detection platform is arranged on one side of the gripping system; The edge computing control module (1) adopts the PaddleClas and PaddleDetection algorithms of the PaddlePaddle image recognition suite or the development suite of TensorFlow or Pytorch; The loading and unloading conveying system includes: A linear feeder (5), a feeding vibration plate (6), a recovery container (7) and a conveyor belt (8); The linear feeder (5) is arranged on one side of the robot arm (3), the loading vibration plate (6) is arranged at the end of the linear feeder (5), the recovery container (7) is arranged on one side of the loading vibration plate (6), and the conveyor belt (8) is arranged on one side of the linear feeder (5) and the loading vibration plate (6) and the recovery container (7); The detection platform comprises: Motor (9) and stage (10); The motor (9) is arranged on one side of the upper portion of the recovery container (7), and the loading platform (10) is arranged on the driving end of the motor (9). The loading platform (10) is located above the recovery container (7); The edge computing control module (1) is used to identify and process the images sent back by the camera (2), and the edge computing control module (1) provides the robot arm (3) with the orientation coordinate information of the workpiece to be grasped, and provides the motor (9) and the robot arm (3) of the detection platform with the detection result information, and the camera (2) is used for transmitting the images.

2. The fully automatic strip workpiece appearance detection device based on convolutional neural network according to claim 1 is characterized in that: The gripping system comprises: A robotic arm (3) and a gripping end (4); The gripping end (4) is arranged on the end of the robotic arm (3), and the camera (2) is arranged on the end of the robotic arm (3).

3. The fully automatic strip workpiece appearance detection device based on convolutional neural network according to claim 1 is characterized in that: The robotic arm (3) adopts a four-axis robotic arm structure, and is used to grab a workpiece in a loading vibration plate (6) and place it on a detection platform, or the robotic arm (3) grabs the workpiece on the detection platform and places it on a conveyor belt (8) in a loading and unloading conveying system.

4. The fully automatic strip workpiece appearance detection device based on convolutional neural network according to claim 3 is characterized in that: The linear feeder (5) is used to transfer the workpiece to the feeding vibration plate (6).

5. The fully automatic strip workpiece appearance detection device based on convolutional neural network according to claim 3 is characterized in that: The loading vibration plate (6) is used to vibrate and disperse the workpieces to prevent them from stacking.

6. The fully automatic strip workpiece appearance detection device based on convolutional neural network according to claim 1, characterized in that: The recovery container (7) is used to collect unqualified products.

7. The fully automatic strip workpiece appearance detection device based on convolutional neural network according to claim 3 is characterized in that: The conveyor belt (8) is used for placing qualified products.

Images