Intelligent mechanical arm control system and method for flying probe tester

Abstract

The present application relates to the field of mechanical arm control technology, and specifically discloses an intelligent mechanical arm control system and method for a flying probe tester, which obtains a PCB board state image collected by a high-resolution camera and transmits it to a cloud processor, and uses machine vision-based image recognition and analysis technology in the cloud processor to perform multi-scale state feature extraction on the PCB board state image, thereby realizing intelligent recommendation of the suction angle value based on the multi-scale feature joint perception representation of semantic information between the shallow features of the PCB board state and the semantic features of the PCB board state. In this way, detailed state feature extraction can be performed based on the PCB board state image collected in real time, thereby achieving more accurate suction control, ensuring that each suction can make the best choice for the current specific situation, and reducing the risk of wrong suction or multiple suction. In addition, it can flexibly adapt to PCB boards of different shapes, sizes and stacking states.

Description

Technical Field

[0001] The present application relates to the field of robot arm control technology, and more specifically, to an intelligent robot arm control system and method for a flying probe tester. Background Art

[0002] In the modern electronics manufacturing industry, the production process of PCB (printed circuit board) is highly automated to improve efficiency, reduce costs and ensure product quality. As one of the important equipment for PCB quality inspection, the performance of the flying probe tester is directly related to the quality and reliability of the finished product. Among them, the robotic arm plays a vital role. It is responsible for accurately handling and positioning the PCB board to ensure the accuracy and consistency of the test process.

[0003] Patent CN111015698B proposes a method to avoid sucking two or more pieces at a time during the board suction process of the flying probe tester. First, the robot gently presses the middle of the PCB board pile but does not activate the suction cup. Then, the suction cup contacts the surface of the PCB board at an oblique angle. Then, the robot rotates to make the target PCB board tilt forward, and uses gravity to separate the single piece, and rotates in the opposite direction to ensure full suction. Finally, the robot moves the PCB board to the laser displacement sensor detection area to measure the thickness. If it exceeds the standard thickness of a single piece, the suction process is re-executed.

[0004] In this patent, the suction cup is operated at a predetermined angle. This method may not be flexible when faced with changes in the state of the PCB stack (such as uneven spacing between boards or differences in flatness), increasing the risk of incorrectly sucking multiple PCBs. In addition, for PCBs of different sizes or shapes, especially special shapes or non-standard sizes, the fixed angle may no longer be applicable, resulting in the need for additional adjustments or reprogramming, thereby reducing the flexibility and versatility of the system.

[0005] Therefore, an optimized intelligent robotic arm control solution for a flying probe tester is desired. Summary of the invention

[0006] The present application provides an intelligent mechanical arm control system and method for a flying probe tester, which can perform detailed state feature extraction based on the real-time collected PCB board state image, thereby achieving more accurate suction control, ensuring that each suction can make the best choice for the current specific situation, reducing the risk of wrong suction or multiple suction. In addition, it can flexibly adapt to PCB boards of different shapes, sizes and stacking states, thereby improving the reliability and accuracy of the suction operation.

[0007] In a first aspect, an intelligent mechanical arm control system for a flying probe tester is provided, comprising: a pressing module, used to control the mechanical arm to move to the top of a PCB board stack and slightly press the PCB board stack so that the PCB board stack is arranged closely; an angle adjustment module, used to control a suction cup device at the end of the mechanical arm to suck the PCB board on the top of the PCB board stack, and adjust the suction angle of the suction cup device in real time to obtain the sucked PCB board; a separation module, used to control the mechanical arm to rotate to separate the sucked PCB board to obtain a separated PCB board; a moving module, used to control the mechanical arm to move the separated PCB board to a laser detection area for thickness detection; wherein the angle adjustment module comprises: A PCB board status image acquisition unit, used to acquire a PCB board status image acquired by a high-resolution camera; An image transmission unit, used to transmit the PCB board status image to a cloud processor via a 5G network; A PCB board state image multi-scale extraction unit is used to perform multi-scale state feature extraction on the PCB board state image in the cloud processor to obtain PCB board state shallow features and PCB board state semantic features; A multi-scale feature joint perception unit, used for performing semantic information-guided multi-scale feature joint perception on the shallow features of the PCB board state and the semantic features of the PCB board state to obtain a deep and shallow joint perception coding feature of the PCB board state; The suction angle recommendation unit is used to obtain a suction angle recommendation value based on the joint perception coding features of the depth and shallowness of the PCB board state.

[0008] In a second aspect, a method for controlling an intelligent mechanical arm for a flying probe tester is provided, comprising: controlling the mechanical arm to move to the top of a PCB board stack and slightly pressing the PCB board stack so that the PCB board stack is arranged closely; controlling a suction cup device at the end of the mechanical arm to suck the PCB board at the top of the PCB board stack, and adjusting the suction angle of the suction cup device in real time to obtain the sucked PCB board; controlling the mechanical arm to rotate to separate the sucked PCB board to obtain a separated PCB board; controlling the mechanical arm to move the separated PCB board to a laser detection area for thickness detection; wherein, controlling the suction cup device at the end of the mechanical arm to suck the PCB board at the top of the PCB board stack, and adjusting the suction angle of the suction cup device in real time to obtain the sucked PCB board, comprises: Get the PCB board status image collected by a high-resolution camera; Transmitting the PCB board status image to a cloud processor via a 5G network; On the cloud processor, multi-scale state feature extraction is performed on the PCB board state image to obtain shallow features of the PCB board state and semantic features of the PCB board state; Performing semantic information-guided multi-scale feature joint perception on the shallow features of the PCB board state and the semantic features of the PCB board state to obtain a deep and shallow joint perception coding feature of the PCB board state; Based on the joint perception coding features of the depth and shallowness of the PCB board state, a recommended value of the absorption angle is obtained.

[0009] The present application provides a kind of intelligent mechanical arm control system and method for flying probe tester, which obtains the PCB board state image collected by the high-resolution camera and transmits it to the cloud processor, and uses the image recognition and analysis technology based on machine vision to extract the multi-scale state features of the PCB board state image in the cloud processor, so as to realize the intelligent recommendation of the suction angle value according to the multi-scale feature joint perception representation of the semantic information between the shallow features of the PCB board state and the semantic features of the PCB board state. The present application can perform detailed state feature extraction according to the PCB board state image collected in real time, so as to realize more accurate suction control, ensure that each suction can make the best choice for the current specific situation, and reduce the risk of wrong suction or multi-piece suction. In addition, it can flexibly adapt to PCB boards of different shapes, sizes and stacking states, thereby improving the reliability and accuracy of the suction operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0011] Figure 1 A schematic block diagram of an intelligent robotic arm control system for a flying probe tester according to an embodiment of the present application.

[0012] Figure 2 A schematic block diagram of an angle adjustment module in an intelligent robotic arm control system for a flying probe tester according to an embodiment of the present application.

[0013] Figure 3 This is a schematic diagram of data flow of an angle adjustment module in an intelligent robotic arm control system for a flying probe tester according to an embodiment of the present application.

[0014] Figure 4 This is a schematic block diagram of a multi-scale feature joint perception unit in an intelligent robotic arm control system for a flying probe tester according to an embodiment of the present application.

[0015] Figure 5This is a schematic flowchart of a method for controlling an intelligent robotic arm of a flying probe tester according to an embodiment of the present application.

[0016] Figure 6 This is a schematic flowchart of step S2 in the intelligent robotic arm control method for a flying probe tester according to an embodiment of the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0018] It should be understood that the flying probe tester includes: a manipulator, a suction cup device, a loading storage rack, a laser detection area, and a control system. First of all, the manipulator is the core execution component of the entire system. It is installed on the frame of the flying probe tester and realizes complex movements through a series of drive motors and rotating wheels. These drive motors are responsible for the first to fourth level of motion control, so that the manipulator can move flexibly in space, and the end can make subtle angle adjustments. A loading storage rack is provided on one side of the manipulator, which is used to store the PCB boards to be tested. When it is necessary to grab the PCB board, the manipulator will first move above the loading storage rack and perform a light pressure action on the PCB board stack to arrange the PCB boards slightly compactly, but will not start the suction cup to prevent the vacuum from being too large and causing multiple pieces to be sucked at one time. Next, in the process of sucking the PCB board, the suction cup device plays an important role. The suction cup is fixed at the end of the suction cup fixing plate, and can contact and suck the top PCB board by forming a certain oblique angle with the surface of the PCB board. This not only reduces the possibility of adhesion with the next PCB board, but also through the subsequent rotation action, the sucked PCB board can be clearly separated from other PCB boards. Once the separation is successful, the suction cup will rotate in the opposite direction around its own axis to ensure full contact and steady suction of the target PCB board. In addition, in order to verify that only one PCB board is sucked each time, the system is also equipped with a laser detection area. This area is located above the loading storage rack and is supported by the laser displacement sensor bracket. The laser displacement sensor is equipped with a transmitter and receiver lens, in which a CCD linear camera and a digital signal processor are used to process the laser beam information reflected from the PCB board. When the robot moves the sucked PCB board to this laser detection area, if the detected board thickness meets the standard thickness of a single PCB board, it will continue to the next step; if the detected board thickness is greater than the set value, it indicates that multiple PCB boards may be sucked at the same time, and the robot will re-execute the grabbing process until the conditions are met. Among them, the control system is the brain of the flying probe tester, responsible for coordinating and controlling the workflow of all key components to ensure the accuracy and reliability of the PCB board suction process.

[0019] This application proposes an intelligent mechanical arm control system for a flying probe tester, such as Figure 1As shown, the intelligent robot arm control system for the flying probe tester includes: a pressing module 10, used to control the robot arm to move to the top of the PCB board pile and slightly press the PCB board pile so that the PCB board pile is arranged tightly; an angle adjustment module 20, used to control the suction cup device at the end of the robot arm to suck the PCB board on the top of the PCB board pile, and adjust the suction angle of the suction cup device in real time to obtain the sucked PCB board; a separation module 30, used to control the robot arm to rotate to separate the sucked PCB board to obtain a separated PCB board; a moving module 40, used to control the robot arm to move the separated PCB board to the laser detection area for thickness detection.

[0020] Exemplarily, in the pressing module 10, the robot arm is controlled to move above the PCB board pile and slightly press the PCB board pile so that the PCB board pile is arranged closely. It should be understood that the task of the pressing module is to control the robot arm to move above the PCB board pile and slightly press the PCB board pile to ensure that the PCB boards are arranged more compactly. The purpose of this operation is to reduce the risk of sucking multiple pieces at one time due to excessive vacuum during suction. In a specific embodiment, the robot arm is mounted on the frame of the flying probe tester by screws, and one side of the robot arm is equipped with a loading storage rack, which is used to place the PCB boards to be tested. When it is necessary to grab the PCB board, the control system first commands the robot arm to move above the loading storage rack and position it directly above the PCB board pile to be grabbed. At this time, the robot arm does not immediately start the suction cup for suction, but first allows the end effector (usually a flat surface or a device with a slight pressure) to gently press the middle position of the PCB board pile. The force of this action is accurately calculated, which can make the PCB boards slightly tighter without causing deformation or damage to the PCB boards. The reason for this slight pressing is that it helps eliminate the problem of excessive vacuum when the suction cup contacts the PCB. If the suction cup is used directly to suck up loosely arranged PCBs, a strong negative pressure may be generated due to the large sealing area formed between the suction cup and the PCB, which may cause the next one or more PCBs to be sucked at the same time. By slightly pressing, the air circulation space between the PCBs can be reduced, and the vacuum strength formed when the suction cup contacts the first PCB can be reduced, thereby reducing the possibility of sucking up multiple PCBs. In addition, this method can also help correct slightly misaligned PCBs and restore them to a more regular position, so that a single PCB can be sucked more accurately later.

[0021] Exemplarily, in the angle adjustment module 20, the suction cup device at the end of the robot arm is controlled to suck the PCB board on the top of the PCB board pile, and the suction angle of the suction cup device is adjusted in real time to obtain the sucked PCB board. It should be understood that when the robot moves to the top of the PCB board pile and completes the slight pressing, the angle adjustment module starts to work. First, the robot does not directly press the suction cup vertically down to contact the PCB board, but allows the suction cup to form a certain oblique angle with the surface of the PCB board. The design of this oblique angle is to avoid excessive vacuum when the suction cup contacts the PCB board vertically, thereby causing one or more PCB boards behind to be sucked at the same time. In this way, the problem of joint suction caused by excessive vacuum adsorption force can be significantly reduced. Next, the control system starts the four-stage drive motor to drive the suction cup fixing plate to rotate, so that the suction cup continues to tilt forward and forms an acute angle (for example, less than 45 degrees) with the surface of the PCB board, and then moves forward to use the top row or the top two rows of suction nozzles to suck the PCB board. Such an operation allows the suction cup to first contact one side edge of the PCB board instead of the entire plane, thereby reducing the large-area negative pressure effect that may be generated during initial contact. As the suction cup gradually slides toward the center of the PCB board, it will gradually build up enough adsorption force to stably grasp the target PCB board, but at this time the other PCB boards will still remain relatively free.

[0022] Exemplarily, in the separation module 30, the robot arm is controlled to rotate to separate the sucked PCB board to obtain the separated PCB board. It should be understood that when the robot arm completes the sucking action of the top PCB board of the PCB board stack, it contacts and initially absorbs the target PCB board at a certain angle, and then enters the separation stage. Specifically, the robot arm first rotates with the bottom edge of the PCB board in contact with the bottom surface of the loading storage rack as the axis. This rotation action is a key step, which utilizes the physical lever principle and gravity. As the robot arm rotates, the sucked PCB board will fall forward under the combined action of the pulling force provided by the suction cup and its own gravity. This forward tilting action causes a clear separation between the target PCB board and other PCB boards that have not been sucked. For example, in a specific implementation example, it is assumed that the robot arm starts to rotate slowly at an angle of about 30 degrees, which can ensure that the PCB board slides forward smoothly without violent swinging or falling. At the same time, since the suction cup has been attached to one side edge of the target PCB board, as the rotation angle increases, the suction cup gradually moves toward the center of the PCB board, and finally achieves full contact and firmly grasps the target PCB board. Once the target PCB is successfully separated from the previous PCB, the control system instructs the suction cup to rotate in the opposite direction around its own axis of rotation. The purpose of this reverse rotation is to make the contact between the suction cup and the PCB closer and to ensure uniform adsorption over the entire surface. In this way, the PCB will not fall off even if slight vibrations or other interference factors are encountered during subsequent movement. In addition, the reverse rotation helps to eliminate the small amount of vacuum effect that may remain due to the initial tilted contact, further reducing the risk of multiple PCBs sticking together.

[0023] Exemplarily, in the moving module 40, the robot arm is controlled to move the separated PCB board to the laser detection area for thickness detection. It should be understood that after the separation of the PCB board is completed, the control system will command the robot to smoothly move the sucked PCB board to the laser detection area. This area is located above the loading storage rack, supported by the laser displacement sensor bracket, and equipped with a laser displacement sensor for thickness measurement. Specifically, the laser displacement sensor is equipped with a transmitter, a receiver lens, a CCD linear camera for processing reflected signals, and a digital signal processor. In a specific implementation example, it is assumed that the robot has successfully sucked and separated a PCB board. Next, the control system will issue an instruction to make the robot slowly and steadily lift up along a predetermined path while keeping the adsorption force between the suction cup and the PCB board unchanged. As the robot rises, it will gradually move away from the PCB board pile, and move horizontally after reaching a certain height until it is aligned with the entrance position of the laser detection area. At this time, the robot will adjust its posture again to ensure that the PCB board enters the detection area in the correct direction. Once the robot arrives at the specified position of the laser detection area with the PCB board, the laser displacement sensor will start working. The sensor emits a visible red laser beam to illuminate the surface of the PCB board. The light reflected by the object is captured by the receiver lens and input to the CCD linear camera to convert into an electrical signal. The digital signal processor is responsible for processing these signals and calculating the specific thickness value of the PCB board. If the test result shows that the PCB board thickness meets the single-piece standard, the system will proceed to the next step; if the board thickness is detected to be abnormal (i.e. greater than the set value), it indicates that multiple PCB boards may be sucked at the same time. At this time, the robot will re-execute the grabbing process until the conditions are met. The reason for this implementation is that by moving the separated PCB board to a specially set laser detection area for thickness detection, it can be ensured that each suction is only for a single PCB board. This method utilizes non-contact laser measurement technology to avoid the error or damage risk that may be caused by traditional physical measurement methods. In addition, due to the high precision and fast response characteristics of laser detection, a complete thickness measurement can be completed in a very short time, greatly improving the detection efficiency. More importantly, this design makes the entire suction process form a closed-loop control system: from suction, separation to final verification, each step can be monitored and fed back in real time, thereby ensuring the accuracy, reliability and consistency of the operation.

[0024] It should be understood that during the stacking process, PCB boards may experience positional displacement, warping, etc. For example, when a PCB board is warped, the edge of the board is not on the same plane as the middle part. This difference in physical state will directly affect the optimal angle at which the suction cup device at the end of the robot arm absorbs the PCB board. In addition, for thicker or harder PCB boards, a larger suction angle may need to be adjusted to ensure safe and effective picking. In addition, due to manufacturing differences, component layout and potential physical deformation, each PCB board has its own unique requirements, so the suction angle needs to be customized according to the specific situation.

[0025] Based on this, in the angle adjustment module, the technical concept of the present application is to obtain the PCB board state image collected by the high-resolution camera, and transmit it to the cloud processor, and use the image recognition and analysis technology based on machine vision to extract the multi-scale state features of the PCB board state image in the cloud processor, so as to realize the intelligent recommendation of the suction angle value according to the multi-scale feature joint perception representation of the semantic information between the shallow features of the PCB board state and the semantic features of the PCB board state. The present application can perform detailed state feature extraction according to the PCB board state image collected in real time, so as to realize more accurate suction control, ensure that each suction can make the best choice for the current specific situation, and reduce the risk of wrong suction or multi-piece suction. In addition, it can flexibly adapt to PCB boards of different shapes, sizes and stacking states, thereby improving the reliability and accuracy of the suction operation.

[0026] In one embodiment, Figure 2 and Figure 3 As shown, the angle adjustment module 20 includes: a PCB board state image acquisition unit 21, used to obtain a PCB board state image acquired by a high-resolution camera; an image transmission unit 22, used to transmit the PCB board state image to a cloud processor through a 5G network; a PCB board state image multi-scale extraction unit 23, used to perform multi-scale state feature extraction on the PCB board state image in the cloud processor to obtain a shallow feature of the PCB board state and a semantic feature of the PCB board state; a multi-scale feature joint perception unit 24, used to perform semantic information-guided multi-scale feature joint perception on the shallow feature of the PCB board state and the semantic feature of the PCB board state to obtain a PCB board state deep and shallow joint perception coding feature; an absorption angle recommendation unit 25, used to obtain an absorption angle recommendation value based on the PCB board state deep and shallow joint perception coding feature.

[0027] Exemplarily, in the PCB board state image acquisition unit 21, a PCB board state image acquired by a high-resolution camera is obtained. It should be understood that the high-resolution image provides detailed visual data, so that the control system can identify the specific morphological features of the PCB board, such as warping, offset or component layout, etc., thereby providing a basis for subsequent angle adjustment. Different PCB boards may have different sizes, shapes and even surface characteristics (such as with or without special markings). High-resolution images can help the system quickly adapt to various types of PCB boards without having to program each type separately. In addition, by analyzing the feature points in the image, the system can determine the most suitable suction method in the current state, reducing the possibility of multiple suction or multiple adhesion caused by improper suction angles. Using machine vision technology, the intelligent mechanical arm can handle a wider range of task requirements without changing the hardware structure, improving the application range and efficiency of the equipment, and reducing the risk of incorrect suction.

[0028] Exemplarily, in the image transmission unit 22, the PCB board status image is transmitted to the cloud processor via the 5G network. It should be understood that the high-resolution image data volume is large and complex, and local processing may require powerful computing resources and time. With the help of the 5G network, these images can be quickly uploaded to the cloud, and the powerful computing power and advanced algorithms of the cloud can be used for efficient processing. In this way, not only can the burden on local equipment be reduced, but also accurate analysis results can be obtained in a timely manner to guide the robot arm to make the best operational decisions. In addition, the stability and rapid response characteristics provided by the 5G network enable the entire system to maintain efficient operation in an industrial environment and adapt to dynamically changing work needs.

[0029] Exemplarily, in the multi-scale extraction unit 23 of the PCB board state image, in the cloud processor, the multi-scale state feature extraction is performed on the PCB board state image to obtain the shallow state feature of the PCB board and the semantic state feature of the PCB board. It should be understood that considering that the PCB board state image contains a wealth of feature representations, such as the edge contour of the PCB board can determine the direction and position of the PCB board; different functional areas and specific component distributions have different requirements for the absorption angle, and for the PCB board state image, these features may exist in different scale ranges. Therefore, in order to effectively capture these diverse features and dig deep into those features that are of great value to the subsequent absorption angle determination, in the technical solution of the present application, the multi-scale extraction unit of the PCB board state image is used to: pass the PCB board state image through a state feature extractor based on a hollow pyramid network to obtain a shallow feature map of the PCB board state as the shallow state feature of the PCB board state and a semantic feature map of the PCB board state as the semantic feature of the PCB board state. The state feature extractor based on the hollow pyramid network extracts the PCB board state image to comprehensively and meticulously extract features of different levels, and obtains a shallow feature map of the PCB board state and a semantic feature map of the PCB board state. It can be understood that the dilated pyramid network is a deep learning network structure. It expands the receptive field by using dilated convolutions in the convolutional layer. This network organizes these layers with dilated convolutions in a pyramid structure. The pyramid structure means that it can extract features from images at multiple scales. In the lower layers, local detail features of the image may be extracted, while in the higher layers, more abstract and semantic features can be obtained. For example, for a PCB board status image, the lower layers may capture details such as the edge of the line and the shape of the solder joint to help locate the position and shape of the PCB board, while the higher layers may involve semantic information such as the distribution of the entire functional area, so that the status of the PCB board can be fully understood from multiple angles, providing sufficient information for subsequent precise operations.

[0030] Exemplarily, in the multi-scale feature joint perception unit 24, the PCB board state shallow feature and the PCB board state semantic feature are subjected to semantic information-guided multi-scale feature joint perception to obtain the PCB board state depth joint perception coding feature. It should be understood that the PCB board state shallow feature map contains relatively intuitive information such as color, contour, simple texture, etc., which can help quickly locate the physical form of the PCB board, such as determining the boundary, general shape and surface flatness of the PCB board. The PCB board state semantic feature map focuses on mining the semantic information of each part of the PCB board, including the type of components, the division of functional areas, etc. In actual operation, when it is necessary to determine the suction angle, not only the physical shape of the PCB board (shallow feature) but also the position and functional area (semantic feature) of the key components on the PCB board should be considered to more comprehensively characterize the feature information. Based on this, in the technical solution of the present application, the semantic information-guided multi-scale feature joint perception is introduced to fuse the PCB board state shallow feature and the PCB board state semantic feature to obtain the PCB board state depth joint perception coding feature. That is, the mechanism can establish a semantic association between different features, and it can guide the feature fusion process based on semantic information, so that the shallow features and semantic features can better follow the semantic logic contained in the actual state of the PCB board when fused. For example, if the edge of the sensor installation area is irregular and has a close position relationship with other key components, then when determining the suction angle, the edge shape should be used to avoid interfering with other components, while ensuring the accuracy of the sensor's position during suction and subsequent movement to meet its functional requirements.

[0031] In one embodiment, Figure 4 As shown, the multi-scale feature joint perception unit 24 includes: a PCB board state semantic information field construction subunit 241, which is used to construct a semantic information field between the PCB board state shallow feature map and the PCB board state semantic feature map to obtain a PCB board state feature semantic information field; a PCB board state semantic shallow attention joint subunit 242, which is used to perform attention joint perception on the PCB board state shallow feature map and the PCB board state semantic feature map based on the PCB board state feature semantic information field to obtain a PCB board state deep and shallow joint perception coding feature map as the PCB board state deep and shallow joint perception coding feature.

[0032] In one embodiment, the PCB board state semantic information field construction subunit includes: a PCB board state upsampling secondary subunit, which is used to upsample the PCB board state semantic feature map to obtain an upsampled PCB board state semantic feature map, wherein the upsampled PCB board state semantic feature map has the same size as the PCB board state shallow feature map. Specifically, the process can be expressed by a formula: ; in, is the semantic feature graph of the PCB board state, is the upsampling operation, It is the upsampled PCB board state semantic feature map.

[0033] The semantic shallow joint semantic field calculation secondary subunit is used to perform feature connection on the upsampled PCB board state semantic feature map and the PCB board state shallow feature map, and then input the feature into the semantic information field predictor based on gated convolution to obtain the PCB board state feature semantic information field. In one embodiment, the semantic shallow joint semantic field calculation secondary subunit is used to: perform feature connection on the upsampled PCB board state semantic feature map and the PCB board state shallow feature map to obtain the PCB board state semantic shallow connection feature map; use the convolution kernel-based feature joint interaction module to process the PCB board state semantic shallow connection feature map to obtain the PCB board state feature semantic information field. Specifically, the process can be expressed as follows: ; in, It is the shallow characteristic diagram of the PCB board status. is the feature connection operation, is a convolutional code with a convolution kernel of 3×3. is a convolutional code with a convolution kernel of 1×1, It is the semantic information field of PCB board status features.

[0034] It should be understood that when the PCB board state semantic feature map is upsampled, some detail information that may be lost during the deep network processing is actually restored. Since the deep feature map has been downsampled multiple times, although the high-level semantic information of the image is retained, a certain degree of spatial resolution and fine-grained structural features are also lost. Through the upsampling operation, these fine-grained features can be reconstructed so that the upsampled semantic feature map can have the same size as the original shallow feature map. This size consistency allows the features of two different levels to interact directly, which helps to capture multi-scale information, which is crucial for subsequent operations. Next, the upsampled PCB board state semantic feature map is feature-connected with the shallow feature map. Doing so not only increases the model's understanding of features of different scales, but also allows the two to share information, further enriching the model's understanding of the PCB board state. Then, this fused feature map is fed into the semantic information field predictor based on gated convolution. As the core component of the predictor, gated convolution adaptively adjusts the importance of feature channels by introducing additional learning parameters, thereby constructing a representation that can reflect the global and local semantic structure of the image - the PCB board state feature semantic information field. Such a semantic information field plays a key role in guiding subsequent feature modulation. It is not just a simple feature combination, but integrates shallow features (such as intuitive information such as color and contour) and deep semantic features (such as component type, functional area division, etc.) at a higher level. This deep understanding can help the robot arm perceive the unique needs of each PCB board more intelligently. For example, when determining the suction angle, it is necessary to consider not only the physical shape of the PCB board, but also the location and functional area of ​​the key components on it to fully characterize the feature information.

[0035] In one embodiment, the PCB board state semantic shallow attention joint subunit is used to: map the upsampled PCB board state semantic feature map to the PCB board state feature semantic information field to obtain the PCB board state field domain modulated deep feature map. Specifically, the process can be expressed by the formula: ; in, It is the point multiplication by position. It is a deep feature map of PCB board state field modulation.

[0036] The shallow feature map of the PCB board state and the deep feature map of the PCB board state field modulation are input into the feature joint perception module based on the multiple attention structure to obtain the shallow and deep joint perception encoding feature map of the PCB board state. Specifically, the process can be expressed by the formula: ; ; ; in, is a deep point-by-point convolution operation, is the batch normalization operation, It is the deep enhancement feature map of PCB board state field modulation. is the activation function, is the global average pooling operation, It is a shallow point-by-point convolution operation. It is a shallow enhanced feature map of the PCB board status. Click on the location to add. It is the joint perception coding feature map of the depth and shallowness of the PCB board state.

[0037] It should be understood that when the upsampled PCB board state semantic feature map is mapped to the PCB board state feature semantic information field, a deeper semantic association is actually established. In this process, each deep feature element is connected to the corresponding position on the semantic information field, and the feature response is adjusted according to the global and local semantic context. In this way, the deep features gain stronger semantic expression, and they are more easily interpreted as part of a specific object or scene. For example, when facing a PCB board with a complex component layout, the system can better understand which areas are key functional areas and where sensitive components may exist through this mapping, which helps to improve the model's understanding of complex scenes, such as distinguishing similar objects in a crowded environment or identifying objects under occlusion. With the mapping completed, the PCB board state field modulated deep feature map is obtained. This feature map not only inherits the spatial distribution characteristics of the original deep features, but also incorporates the high-level understanding provided by the semantic information field. It enables the system to more keenly capture the unique needs of PCB boards, such as different shapes, sizes, and changes in stacking status. This enhanced perception capability provides a solid foundation for subsequent operations. Next, the shallow feature map of the PCB board state is combined with the semantically modulated deep feature map Figure 1At the same time, they are fed into the feature joint perception module based on the multiple attention structure. The design of this module is inspired by the processing method of the human visual system - flexibly allocating attention resources according to different task requirements. In this module, the shallow feature map and the semantically modulated deep feature map complement each other to form a multi-level feature representation system. The multiple attention mechanism allows the model to dynamically adjust the focus according to the requirements of different tasks and automatically learn the feature combination that is most beneficial to the current task. This means that the system can respond to various different PCB board situations more flexibly, ensuring that each absorption can make the best choice for the specific situation. The final generated PCB board state deep and shallow joint perception encoding feature map contains not only rich low-level visual information, but also high-level semantic understanding. Such a feature representation system provides strong support for achieving accurate target detection, segmentation or other computer vision tasks.

[0038] Exemplarily, in the suction angle recommendation unit 25, based on the PCB board state depth joint perception coding feature, the suction angle recommendation value is obtained. In one embodiment, the suction angle recommendation unit is used to: pass the PCB board state depth joint perception coding feature map through a decoder-based robotic arm intelligent controller to obtain a suction angle recommendation decoding value as the suction angle recommendation value. That is, the PCB board state depth joint perception coding feature obtained by joint perception using the PCB board state shallow feature map and the PCB board state semantic feature map is decoded, so as to use the decoder to parse and process this complex feature map, and convert the PCB board state information contained therein into a suction angle recommendation decoding value that can be understood and executed by the robotic arm, thereby building a bridge from image feature analysis to actual action control, so as to minimize the problems of PCB board damage, suction failure and the possibility of multiple suction caused by improper suction angles.

[0039] In one example, the depth-shallowness joint perception coding feature map of the PCB board state is passed through a decoder-based robotic arm intelligent controller to obtain a recommended decoding value of the absorption angle as the recommended absorption angle value, including: expanding the depth-shallowness joint perception coding feature map of the PCB board state into a depth-shallowness joint perception coding feature vector of the PCB board state; using a decoder-based robotic arm intelligent controller to decode and regress the depth-shallowness joint perception coding feature vector of the PCB board state using the following formula to obtain a recommended decoding value of the absorption angle; wherein the formula is: ,in, is the PCB board state depth joint perception encoding feature vector, is the recommended decoding value of the absorption angle, is the weight matrix of the decoder-based robotic arm intelligent controller, Represents matrix multiplication.

[0040] Here, in the case where the shallow feature map of the PCB board state and the semantic feature map of the PCB board state respectively represent the shallow image semantic coding features and the deep image semantic coding features of the PCB board state image, when performing feature joint perception based on semantic information field guided migration, the receptive field and feature order differences of the image features will cause an imbalance in the fine-grainedness of the semantic information field, thereby leading to the richness of the joint perceptual distribution of the shallow and deep joint perceptual coding feature map of the PCB board state, resulting in repetition of unstructured probability mapping, affecting the accuracy of the decoding result.

[0041] Based on this, in a preferred example, in the process of passing the PCB board state depth joint perception coding feature map through a decoder-based robotic arm intelligent controller to obtain the recommended decoding value of the suction angle, the PCB board state depth joint perception coding feature map is feature modulated, including: Quantify the number of valid response elements in the PCB board state depth joint perception coding feature map, and deduct the number of valid response elements from the rank parameter of the PCB board state depth joint perception coding feature map to obtain the PCB board state depth joint perception coding empty rank parameter, wherein the number of valid response elements in the PCB board state depth joint perception coding feature map is the number of non-zero eigenvalues ​​in the PCB board state depth joint perception coding feature map, expressed as: ; ; in, Indicates the number of valid response elements in the PCB board state depth-shallowness joint perception coding feature map, , and Respectively represent the width, length and number of channels of the PCB board state depth joint perception coding feature map, Indicates the number of valid response elements in the PCB board state depth-shallowness joint perception coding feature map, Indicates the rank parameter of the joint perception coding of the PCB board status; Based on the PCB board state depth-shallow joint perception coding empty rank parameter, the first PCB board state depth-shallow joint perception coding field shaping amount and the second PCB board state depth-shallow joint perception coding field shaping amount are derived, which are expressed as: ; ;

[0042] in, Indicates the depth-shallowness joint perception coding field shaping quantity of the first PCB board state, Indicates the depth-shallowness joint perception coding field shaping quantity of the second PCB board state; The norm regularization values ​​of all the response elements of the PCB board state depth-shallow joint perception coding feature map are extracted as the PCB board state depth-shallow joint perception coding configuration identification quantity, which is expressed as: ;

[0043] in, Representing the PCB board state depth joint perception coding feature map The eigenvalues ​​of the positions, Indicates the identification quantity of the joint perception coding configuration of the PCB board status; Based on the PCB board state depth and shallow joint perception coding empty rank parameter, the response element index mapping of the PCB board state depth and shallow joint perception coding feature map is constructed, and the Hadamard product is performed with the first PCB board state depth and shallow joint perception coding field shaping amount to generate the PCB board state depth and shallow joint perception coding neighborhood representation amount, which is expressed as: ;

[0044] in, Represents the neighborhood representation of the joint perception coding of the depth of the PCB board state; Solve the Kronecker product of each response element of the PCB board state depth and shallow joint perception coding feature map, the second PCB board state depth and shallow joint perception coding field shaping quantity and the PCB board state depth and shallow joint perception coding configuration identification quantity to obtain the PCB board state depth and shallow joint perception coding core locator; ;

[0045] in, Indicates the core vector of the joint perception encoding of the depth and shallowness of the PCB board status; The neighborhood representation of the PCB board state depth-shallow joint perception coding and the core directional vector of the PCB board state depth-shallow joint perception coding are subjected to a standard weight difference operation to synthesize an optimized PCB board state depth-shallow joint perception coding feature map, which is expressed as: ;

[0046] in, and represents the weight hyperparameter, Represents the optimized PCB board state depth joint perception encoding feature map Finally, the optimized PCB board state depth joint perception coding feature map is passed through a decoder-based robotic arm intelligent controller to generate the recommended decoding value of the suction angle.

[0047] Accordingly, relying on the field quantity shaping principle of multi-dimensional topological structure, the discrete singular points in the PCB board state depth and shallow joint perception coding feature map are regarded as the differentiable dimensional parameters of the multi-dimensional topological structure, and then the orthogonal reference direction pointing to the distribution core is locked in the neighborhood coordinate system represented by the eigenvalue of the PCB board state depth and shallow joint perception coding feature map, thereby achieving the precise registration of the probability configuration of the PCB board state depth and shallow joint perception coding feature map with the probability distribution smooth field, avoiding the redundant iteration in the probability conversion process caused by the disordered property of the diversity distribution of the PCB board state depth and shallow joint perception coding feature map. In this way, the accuracy of the analysis result obtained by the decoder-based robot arm intelligent controller is enhanced.

[0048] In summary, the intelligent mechanical arm control system for a flying probe tester according to the embodiment of the present application is explained, which obtains a PCB board state image collected by a high-resolution camera and transmits it to a cloud processor, and uses machine vision-based image recognition and analysis technology in the cloud processor to perform multi-scale state feature extraction on the PCB board state image, thereby realizing intelligent recommendation of the suction angle value according to the multi-scale feature joint perception representation of the semantic information between the shallow features of the PCB board state and the semantic features of the PCB board state. The present application can perform detailed state feature extraction according to the PCB board state image collected in real time, so as to achieve more accurate suction control, ensure that each suction can make the best choice for the current specific situation, and reduce the risk of wrong suction or multiple suction. In addition, it can flexibly adapt to PCB boards of different shapes, sizes and stacking states, thereby improving the reliability and accuracy of the suction operation.

[0049] Figure 5 Schematic flow chart of the intelligent mechanical arm control method for a flying probe tester according to an embodiment of the present application. Figure 5 As shown, the intelligent mechanical arm control method for a flying probe tester includes: S1, controlling the mechanical arm to move to the top of the PCB board stack and slightly pressing the PCB board stack so that the PCB board stack is arranged tightly; S2, controlling the suction cup device at the end of the mechanical arm to suck the PCB board on the top of the PCB board stack, and adjusting the suction angle of the suction cup device in real time to obtain the sucked PCB board; S3, controlling the mechanical arm to rotate to separate the sucked PCB board to obtain a separated PCB board; S4, controlling the mechanical arm to move the separated PCB board to a laser detection area for thickness detection.

[0050] In one embodiment, Figure 6As shown, the suction cup device at the end of the robot arm is controlled to suck the PCB board on the top of the PCB board pile, and the suction angle of the suction cup device is adjusted in real time to obtain the sucked PCB board, including: S21, acquiring a PCB board state image collected by a high-resolution camera; S22, transmitting the PCB board state image to a cloud processor through a 5G network; S23, in the cloud processor, performing multi-scale state feature extraction on the PCB board state image to obtain a shallow feature of the PCB board state and a semantic feature of the PCB board state; S24, performing semantic information-guided multi-scale feature joint perception on the shallow feature of the PCB board state and the semantic feature of the PCB board state to obtain a deep and shallow joint perception coding feature of the PCB board state; S25, based on the deep and shallow joint perception coding feature of the PCB board state, obtaining a recommended value of the suction angle.

[0051] In one embodiment, in the cloud processor, multi-scale state feature extraction is performed on the PCB board state image to obtain PCB board state shallow features and PCB board state semantic features, including: the PCB board state image is passed through a state feature extractor based on a hollow pyramid network to obtain a PCB board state shallow feature map as the PCB board state shallow features and a PCB board state semantic feature map as the PCB board state semantic features.

[0052] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned intelligent mechanical arm control method for flying probe tester have been described in the above reference. Figures 1 to 4 The invention has been described in detail in the description of the intelligent robotic arm control system for a flying probe tester, and therefore, its repeated description will be omitted.

[0053] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0054] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the module division is only a logical function division, and there may be other division methods in actual implementation. The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0055] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0056] Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. An intelligent mechanical arm control system for a flying probe tester, comprising: A pressing module is used to control the robot arm to move to the top of the PCB board stack and slightly press the PCB board stack so that the PCB board stack is arranged tightly; an angle adjustment module is used to control the suction cup device at the end of the robot arm to suck the PCB board on the top of the PCB board stack, and adjust the suction angle of the suction cup device in real time to obtain the sucked PCB board; a separation module is used to control the robot arm to rotate to separate the sucked PCB board to obtain a separated PCB board; The moving module is used to control the mechanical arm to move the separated PCB board to the laser detection area for thickness detection; characterized in that the angle adjustment module includes: A PCB board status image acquisition unit, used to acquire a PCB board status image acquired by a high-resolution camera; An image transmission unit, used to transmit the PCB board status image to a cloud processor via a 5G network; A PCB board state image multi-scale extraction unit is used to perform multi-scale state feature extraction on the PCB board state image in the cloud processor to obtain PCB board state shallow features and PCB board state semantic features; A multi-scale feature joint perception unit, used for performing semantic information-guided multi-scale feature joint perception on the shallow features of the PCB board state and the semantic features of the PCB board state to obtain a deep and shallow joint perception coding feature of the PCB board state; The suction angle recommendation unit is used to obtain a suction angle recommendation value based on the joint perception coding features of the depth and shallowness of the PCB board state.

2. The intelligent mechanical arm control system for a flying probe tester according to claim 1, characterized in that: The PCB board state image multi-scale extraction unit is used to: pass the PCB board state image through a state feature extractor based on a hollow pyramid network to obtain a PCB board state shallow feature map as the PCB board state shallow feature and a PCB board state semantic feature map as the PCB board state semantic feature.

3. The intelligent mechanical arm control system for a flying probe tester according to claim 2, characterized in that: The multi-scale feature joint perception unit includes: A PCB board state semantic information field construction subunit is used to construct a semantic information field between the PCB board state shallow feature map and the PCB board state semantic feature map to obtain a PCB board state feature semantic information field; The PCB board state semantic shallow attention joint subunit is used to perform attention joint perception on the PCB board state shallow feature map and the PCB board state semantic feature map based on the PCB board state feature semantic information field to obtain the PCB board state deep and shallow joint perception coding feature map as the PCB board state deep and shallow joint perception coding feature.

4. The intelligent mechanical arm control system for a flying probe tester according to claim 3, characterized in that: The PCB board status semantic information field construction subunit includes: A PCB board state upsampling secondary subunit is used to upsample the PCB board state semantic feature map to obtain an upsampled PCB board state semantic feature map; The semantic shallow joint semantic field calculation secondary subunit is used to perform feature connection on the upsampled PCB board state semantic feature map and the PCB board state shallow feature map, and then input the feature into the gated convolution-based semantic information field predictor to obtain the PCB board state feature semantic information field.

5. The intelligent mechanical arm control system for a flying probe tester according to claim 4, characterized in that: The upsampled PCB board state semantic feature map has the same size as the PCB board state shallow feature map.

6. The intelligent mechanical arm control system for a flying probe tester according to claim 5, characterized in that: The semantic shallow layer combined with the semantic field calculation secondary sub-unit is used to: Performing feature connection on the upsampled PCB board state semantic feature map and the PCB board state shallow feature map to obtain a PCB board state semantic shallow connection feature map; The PCB board state semantic shallow connection feature map is processed using a feature joint interaction module based on a convolution kernel to obtain the PCB board state feature semantic information field.

7. The intelligent mechanical arm control system for a flying probe tester according to claim 6, characterized in that: The PCB state semantic shallow attention joint sub-unit is used to: Mapping the upsampled PCB board state semantic feature map to the PCB board state feature semantic information field to obtain a PCB board state field domain modulation deep feature map; The shallow feature map of the PCB board state and the deep feature map of the PCB board state field modulation are input into a feature joint perception module based on a multiple attention structure to obtain the shallow and deep joint perception encoding feature map of the PCB board state.

8. The intelligent mechanical arm control system for a flying probe tester according to claim 7, characterized in that: The suction angle recommendation unit is used to: pass the PCB board state depth joint perception coding feature map through a decoder-based robotic arm intelligent controller to obtain a suction angle recommended decoding value as the suction angle recommendation value.

9. A method for controlling an intelligent mechanical arm of a flying probe tester, comprising: Controlling the mechanical arm to move to the top of the PCB board stack and slightly pressing the PCB board stack so that the PCB board stack is arranged closely; controlling the suction cup device at the end of the mechanical arm to suck the PCB board at the top of the PCB board stack, and adjusting the suction angle of the suction cup device in real time to obtain the sucked PCB board; controlling the mechanical arm to rotate to separate the sucked PCB board to obtain the separated PCB board; controlling the mechanical arm to move the separated PCB board to the laser detection area for thickness detection; characterized in that controlling the suction cup device at the end of the mechanical arm to suck the PCB board at the top of the PCB board stack, and adjusting the suction angle of the suction cup device in real time to obtain the sucked PCB board, comprises: Get the PCB board status image collected by a high-resolution camera; Transmitting the PCB board status image to a cloud processor via a 5G network; On the cloud processor, multi-scale state feature extraction is performed on the PCB board state image to obtain shallow features of the PCB board state and semantic features of the PCB board state; Performing semantic information-guided multi-scale feature joint perception on the shallow features of the PCB board state and the semantic features of the PCB board state to obtain a deep and shallow joint perception coding feature of the PCB board state; Based on the joint perception coding features of the depth and shallowness of the PCB board state, a recommended value of the absorption angle is obtained.

10. The intelligent mechanical arm control method for a flying probe tester according to claim 9, characterized in that: In the cloud processor, multi-scale state feature extraction is performed on the PCB board state image to obtain PCB board state shallow features and PCB board state semantic features, including: the PCB board state image is passed through a state feature extractor based on a hollow pyramid network to obtain a PCB board state shallow feature map as the PCB board state shallow feature and a PCB board state semantic feature map as the PCB board state semantic feature.

Images