Miniature PCB cutter automatic welding machine system with active learning function

Abstract

The invention provides an active learning miniature PCB cutter automatic welding machine system, and relates to the technical field of miniature PCB cutter automatic welding, and the system comprises a main system which is composed of a welding machine control system, a numerical control intelligent welding machine, an information collection system and a welding path planning system. According to the welding path planning system, a Seq2Seq deep learning model is adopted to build a welding path planning deep learning model, learning is carried out by recording welding path data and welding data which are manually planned in the past, and the optimal welding path is predicted; the main system is a comprehensive system integrating high-precision control, intelligent operation, real-time data acquisition and analysis and path planning optimization, and aims at improving the welding quality, efficiency and automation level and reducing manual intervention and errors, the welding control system serves as the core of the whole system, and the welding control system is suitable for large-scale popularization and application. And the controller is responsible for receiving instructions, controlling various parameters of the welding machine and ensuring the stability and safety of the welding process.

Description

Technical Field

[0001] The invention relates to the technical field of automatic welding of micro PCB cutters, and in particular to an active learning automatic welding machine system for micro PCB cutters. Background Art

[0002] With the rapid development of the electronics manufacturing industry, the precision and efficiency requirements of PCB tools are increasing. Traditional manual welding methods can no longer meet the needs of large-scale, high-precision, and high-efficiency production. Therefore, automatic welding technology has emerged and is gradually developing in the direction of intelligence and automation. Active learning, as an advanced machine learning technology, enables the machine to continuously optimize its own performance during the welding process and improve welding quality and efficiency. Micro PCB tools are usually small in size, high in precision, and complex in structure, and have extremely high requirements for welding technology.

[0003] According to a PCB welding method based on visual positioning, the Chinese patent number CN 115049726 A includes: collecting an image of the current PCB chip to be identified, matching the collected image with each rough positioning template, and matching the image coordinates of the center of the PCB chip to be identified in the collected image; constructing a sample image based on the center of the PCB chip to be identified in the collected image, inputting the sample image into the recognition network, and identifying the category of the PCB chip to be identified in the collected image; determining the precise positioning area in the collected image based on the image coordinates of the center of the PCB chip to be identified, and determining the image coordinates of the welding point of the precise positioning template in the current collected image based on the matching between the precise positioning area and the precise positioning template; converting the image coordinates of the welding point in the collected image to the robot coordinate system, and controlling the solder gun at the end of the robot to move to the welding point position for welding operation. The autonomous recognition and automatic welding of PCB chips by industrial robots can be realized, and can be applied to mixed-line production of various types of PCB chip products.

[0004] The above patent documents and prior art have the following technical problems when used:

[0005] Problem 1: Traditional welding methods are difficult to ensure welding consistency and stability, which can easily lead to problems such as unstable welding quality and high scrap rate;

[0006] The second problem is that the collected information is not processed based on the data model. If there is a lack of support for data prediction, the detection process may only be based on traditional physical inspection and manual judgment.

[0007] Therefore, it is particularly important to develop an automatic welding machine system that can perform efficient and precise welding according to the characteristics of micro PCB tools. Summary of the invention

[0008] Technical issues solved

[0009] In view of the shortcomings of the prior art, the present invention provides an active learning micro PCB tool automatic welding machine system, which solves the following problems:

[0010] 1. Traditional welding methods are difficult to ensure the consistency and stability of welding, which can easily lead to unstable welding quality and high scrap rate;

[0011] 2. The collected information is not processed based on the data model. If there is a lack of support for data prediction, the detection process may only be based on traditional physical inspection and manual judgment.

[0012] Technical Solution

[0013] To achieve the above objectives, the present invention is implemented through the following technical solutions: an active learning micro PCB tool automatic welding machine system, including a main system, the main system is composed of a welding machine control system, a CNC intelligent welding machine, an information acquisition system and a welding path planning system, wherein:

[0014] The welding path planning system adopts Seq2Seq deep learning model to build a welding path planning deep learning model, and learns by recording the previously manually planned welding path data and welding data to predict the optimal welding path;

[0015] The welding path planning system classifies the previously recorded welding path data, takes 70% of the data as a training set and 30% as a test set, and performs data enhancement on the input data by random cropping and random flipping to expand the data set;

[0016] Before planning, the welding path planning system traverses all the workstations to be welded and obtains a weld point vector sequence. Secondly, the planned unordered weld point vector sequence B of workstation A is input into the welding path planning deep learning model. Then the model outputs the planned weld point vector sequence. The order of the weld point vectors in the sequence is the welding order. Finally, after all the workstations to be planned are traversed, the welding path planning results of all the workstations are summarized. Using this method for welding path planning avoids the complex and cumbersome constraint setting in the traditional welding path planning algorithm, and improves the efficiency and quality of welding path planning.

[0017] Preferably, the welding path planning system obtains welding information through an information acquisition system, and the information acquisition system extracts previous weld features through a laser vision probe, wherein the previous weld features include laser stripe lines and weld feature points, and the laser stripes include weld groove height, width and position information;

[0018] When extracting the features of the welding spots, the information acquisition system will use the segmented fitting method and the welding path planning system to obtain welding information through the information acquisition system, and the Steger algorithm, cluster analysis algorithm and consistency straight line fitting method for centralized extraction. The Steger algorithm is used for sub-pixel edge detection, and the edge is located by calculating the second-order derivative of the image, and then the accuracy is improved by filtering and denoising. Assume that the image is I(x, y), the gradient and the second derivative The calculation is as follows:

[0019]

[0020] The partial derivative of the image in the x direction represents the rate of change of brightness in the horizontal direction;

[0021] : The partial derivative of the image in the y direction, indicating the rate of change of brightness in the vertical direction.

[0022] Preferably, when extracting feature points in the information acquisition system, the slope is calculated by the piecewise fitting method to determine the weld feature, the weld feature is extracted by linear fitting of the segmented points using the least squares method, the image is filtered and denoised using the Steger algorithm, the weld rectangular area is located using the Otsu algorithm, and in order to improve the computational efficiency, the two-dimensional Gaussian convolution is simplified using the separability of the Gaussian function, which can increase the speed of laser stripe extraction, the weld feature sampling points are read using the cluster analysis algorithm, a data set is created, the sampling points are screened by setting a threshold, the sampling points are continuously read and classified, the geometric information is calculated based on the screened sampling points, and when extracting using the consistent linear fitting method, the weld feature data and the linear model equation are input, the feature points are repeatedly selected by random sampling, the model is used for verification, the model is evaluated by the accuracy of the data points, and the execution is repeated 1000 times to obtain the fitted linear equation.

[0023] Preferably, the interior of the CNC intelligent welding machine includes a CPU, a digital input and output module, a CANopen module, an RS485 communication module, and a Profibus communication module, wherein the CPU is externally connected to an HMI control display panel and a weighing sensor, the CANopen module is externally connected to a servo drive unit, and the RS485 communication indirectly controls the battery pack through a battery management unit. The internal execution components of the CNC intelligent welding machine are mainly divided into a perception layer, a decision layer, and an execution layer according to functional grouping, wherein the CNC intelligent welding machine is equipped with a Webcam camera, which is installed on the front of the CNC intelligent welding machine to detect the path of the welding points to be welded in real time, and is used to identify the guide route and correct the vehicle body. The digital input and output module in the CNC intelligent welding machine is externally connected to an obstacle sensor, which is placed in the center of the side of the CNC intelligent welding machine, scanning the area with a maximum radius of 5m, and detecting obstacles in the driving direction. The decision layer uses an embedded microcomputer as a controller, and in the Visual The control program is written in C# in the Studio2012 environment to collect the perception layer information. The execution layer is composed of CPU, digital input and output module, CANopen module, RS485 communication module and Profibus communication module.

[0024] Preferably, a welding gun fixing device is provided on the outside of the CNC intelligent welding machine on the running track of the frame, which is used to install and fix the welding gun. The welding gun fixing device can reciprocate left and right along the running track to realize horizontal movement of the welding gun on the welding plane. The motor equipped with the CNC intelligent welding machine includes a longitudinal drive motor, a transverse drive motor and a drive motor for controlling the up and down movement of the welding gun.

[0025] Preferably, the built-in control chips of the CNC intelligent welding machine are ESP8266 module and MPU-6050 chip respectively. The CNC intelligent welding machine receives instructions issued by the welding machine control system through ESP8266, converts the instructions received by ESP826 into electrical signals for driving the motor through L298N, and remotely drives and controls the welding machine.

[0026] Preferably, the information acquisition system adopts the YOLOv5 network model algorithm and integrates the FPN-PAN design to ensure the full integration of multi-scale features. The input end adopts the MOSAIC algorithm for data enhancement to increase the frequency of occurrence of small targets, and organizes and constructs a data set of solder joints of the device to be processed. The training of the model is first pre-trained on the source data set, and the backbone and neck parts of the obtained pre-trained model are parameter-frozen, and then transfer learning training is performed on the external crack data set of the main hull structure.

[0027] Preferably, in order to realize the motion control and monitoring of the CNC intelligent welding machine, the software system of the welding machine control system needs to provide a friendly human-computer interaction interface and should be able to complete the following main tasks: (1) It can set the relevant parameters of the entire welding positioning robot motion control system, such as importing, editing, saving and decoding the working mode, motion starting point, motion speed, motion stroke, etc.; (2) The system has two welding modes, namely automatic control and manual control. In the automatic control mode, the system runs step by step according to the corresponding program commands and executes in a loop. The running position status of the mobile platform can be viewed during the system operation, and the system can automatically stop in case of an emergency; under normal circumstances, the operator can see the position coordinates of the three axes of X, Y, and Z and the running speed value through the display screen. The manual mode can realize the operation instructions such as jog, linear interpolation, inverse circular interpolation, program zeroing, simulation operation, etc., so as to facilitate the operator to accurately control the welding trajectory; (3) The corresponding logic control function can be selected. To ensure the normal operation of the system, conditional judgment, program jump, interruption, and delay logic can be selected.

[0028] Preferably, the welding path planning system selects the ant colony algorithm when selecting the welding path. The ant colony algorithm has two positive and negative feedback mechanisms, which are as follows: (1) Positive feedback: When ants walk on a certain path, they will release information to increase the pheromone concentration of the path; (2) Negative feedback: The residual pheromone concentration of the path is in a state of continuous decay. When using this algorithm to complete path planning, its basic workflow is as follows: (1) Initialize the parameters used; (2) Randomly obtain the initial position of the ant, and then repeat the experiment; (3) Update the global pheromone, record the ant's route, compare it, obtain the best working route, determine the optimal or suboptimal path of the welding robot from the starting point to the end point, and consider factors such as robot kinematic constraints, collision avoidance, and welding process constraints. Establish a set of spatial points that can be reached by the welding robot as the search space of the algorithm. Establish a set of spatial points that can be reached by the welding robot as the search space of the algorithm. The ant colony algorithm is used for path planning and combinatorial optimization problems, and works as follows: Parameter initialization: Set the initial pheromone concentration τ0 and the volatility factor ρ parameter, and randomly select the next step path from the starting point with a probability of:

[0029]

[0030] τ ij is the pheromone concentration from path i to path j, η ij For heuristic information, usually the feasibility of the path or the reverse distance, pheromone update:

[0031] τ ij =(1-ρ)τ ij +Δτ ij

[0032] Δτ ij is the amount of pheromones released by ants as they walk on the path.

[0033] Beneficial Effects

[0034] The present invention provides an active learning micro PCB tool automatic welding machine system.

[0035] Beneficial effects:

[0036] 1. The main system of the present invention is a comprehensive system integrating high-precision control, intelligent operation, real-time data acquisition and analysis, and optimized path planning, which aims to improve welding quality, efficiency and automation level, reduce manual intervention and errors. The welding control system, as the core of the whole system, is responsible for receiving instructions, controlling various parameters of the welding machine (such as current, voltage, speed, etc.), and ensuring the stability and safety of the welding process. It is connected to the CNC intelligent welding machine and the information acquisition system through a high-speed communication interface (such as Ethernet, USB, etc.) to realize data exchange and the issuance of control instructions.

[0037] The data intelligent welding machine performs precise welding actions according to the instructions of the welding machine control system. It has built-in high-precision sensors and controllers, which can monitor various parameters in the welding process in real time and make necessary adjustments. It is directly connected to the welding machine control system, receives control instructions and feeds back real-time data. At the same time, it is connected to the information acquisition system through a specific interface to provide key data in the welding process.

[0038] The information acquisition system is responsible for collecting various data during the welding process (such as current, voltage, temperature, speed, etc.), preprocessing and storing them. These data are the basis for subsequent analysis and optimization of the welding path. They are installed on the welding machine in an external or built-in way, and are connected to the welding machine control system and welding path planning system through wired or wireless means.

[0039] 2. Through the laser vision probe combined with a variety of algorithms (such as segmented fitting method, Steger algorithm, cluster analysis algorithm, and consistent straight line fitting method), the weld features can be accurately captured. The groove height, width, position and weld point features of the weld can be accurately captured. These high-precision data provide a solid foundation for subsequent welding path planning. The Otsu algorithm is used to locate the weld rectangular area, and the separability of the Gaussian function is used to simplify the two-dimensional Gaussian convolution, which significantly improves the extraction speed of laser stripes. This not only reduces the calculation time, but also improves the real-time performance of the entire system. The cluster analysis algorithm and the consistent straight line fitting method perform well in processing large amounts of data. They can quickly and accurately screen out key sampling points and calculate geometric information to ensure the accuracy of the welding path. The welding path planning system uses the Seq2Seq deep learning model, which can dynamically adjust and optimize the welding path based on historical data and real-time collected information. This intelligent path planning avoids the complex and cumbersome constraint setting in traditional algorithms, improves planning efficiency and welding quality, and the entire system has achieved a high degree of automation and intelligence from welding path planning, parameter setting to welding execution. The user only needs to input basic information through the human-machine interface, and the system can automatically complete the welding task, greatly reducing manual intervention and errors.

[0040] 3. The system supports both automatic control and manual control modes, meeting the operational requirements in different scenarios. In automatic control mode, the system can automatically execute preset programs, improving work efficiency; in manual mode, the operator can achieve precise control of the welding trajectory, providing a wealth of logical control functions, such as conditional judgment, program jump, interruption, delay, etc., to ensure the stable operation of the system under different working conditions. The decision-making layer uses an embedded microcomputer as a controller, and uses C# to write control programs in the Visual Studio 2012 environment, which can efficiently collect information from the perception layer and make quick decisions. The information acquisition system uses advanced algorithms for data processing and feature extraction, providing accurate data support for welding path planning.

[0041] 4. The present invention uses an ant colony algorithm to simulate the pheromone transmission and path selection behavior of ants in the foraging process, and can find the optimal or suboptimal path from the starting point to the end point in a complex welding environment. This intelligent optimization capability helps to reduce unnecessary movement during the welding process and improve welding efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a system structure diagram of the present invention;

[0043] Figure 2 It is a planning flow chart of the present invention;

[0044] Figure 3 It is a schematic diagram of feature extraction of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Specific embodiment one:

[0047] like Figure 1-3 As shown, an active learning micro PCB tool automatic welding machine system includes a main system, which is composed of a welding machine control system, a CNC intelligent welding machine, an information acquisition system and a welding path planning system, wherein:

[0048] The welding path planning system uses the Seq2Seq deep learning model to build a welding path planning deep learning model. It learns by recording the previously manually planned welding path data and welding data to predict the optimal welding path.

[0049] The welding path planning system classifies the previously recorded welding path data, takes 70% of the data as the training set and 30% as the test set, and performs data enhancement on the input data by random cropping and random flipping to expand the data set;

[0050] Before planning, the welding path planning system traverses all the workstations to be planned for welding path planning to obtain a sequence of weld point vectors. Secondly, the planned unordered weld point vector sequence B of workstation A is input into the welding path planning deep learning model, and then the model outputs the planned weld point vector sequence. The order of the weld point vectors in the sequence is the welding order. Finally, after all the workstations to be planned are traversed, the welding path planning results of all workstations are summarized. Using this method for welding path planning avoids the complex and cumbersome constraint setting in the traditional welding path planning algorithm, and improves the efficiency and quality of welding path planning.

[0051] In this system, the main system is a comprehensive system that integrates high-precision control, intelligent operation, real-time data acquisition and analysis, and optimized path planning. It aims to improve welding quality, efficiency and automation level, and reduce human intervention and errors. The welding control system, as the core of the entire system, is responsible for receiving instructions, controlling various parameters of the welding machine (such as current, voltage, speed, etc.), and ensuring the stability and safety of the welding process. It is connected to the CNC intelligent welding machine and information acquisition system through high-speed communication interfaces (such as Ethernet, USB, etc.) to realize data exchange and the issuance of control instructions.

[0052] The data intelligent welding machine performs precise welding actions according to the instructions of the welding machine control system. It has built-in high-precision sensors and controllers, which can monitor various parameters in the welding process in real time and make necessary adjustments. It is directly connected to the welding machine control system, receives control instructions and feeds back real-time data. At the same time, it is connected to the information acquisition system through a specific interface to provide key data in the welding process.

[0053] The information acquisition system is responsible for collecting various data during the welding process (such as current, voltage, temperature, speed, etc.), preprocessing and storing them. These data are the basis for subsequent analysis and optimization of the welding path. They are installed on the welding machine in an external or built-in way, and are connected to the welding machine control system and welding path planning system through wired or wireless means.

[0054] The welding path planning system automatically generates the optimal welding path according to the shape, size and welding requirements of the welding workpiece. At the same time, it dynamically adjusts and optimizes the welding path in combination with the real-time collected data, connects with the welding machine control system and information acquisition system, receives welding requirements and real-time data, and outputs the optimized welding path to the CNC intelligent welding machine for execution.

[0055] During specific use, the user inputs basic information such as the shape, size, material and welding requirements of the welding workpiece through the human-machine interface. The welding path planning system automatically generates the optimal welding path based on the input information, combined with the built-in optimization algorithm and real-time collected data. The welding machine control system sets the various parameters of the CNC intelligent welding machine (such as current, voltage, speed, etc.) according to the output of the welding path planning system. The CNC intelligent welding machine starts the welding operation according to the set parameters and planned path. At the same time, the information acquisition system collects various data during the welding process in real time. During the welding process, the welding path planning system dynamically adjusts and optimizes the welding path according to the real-time collected data. The welding machine control system also adjusts parameters as needed to ensure welding quality. After the welding operation is completed, the system automatically stops and outputs the welding results. Users can view the various data of the welding process and the quality assessment of the welding results through the human-machine interface.

[0056] By integrating high-precision control, intelligent operation, real-time data collection and analysis, and optimized path planning, the system significantly improves the automation level of the welding process, reduces manual intervention and human errors, and improves production efficiency and consistency. The Seq2Seq deep learning model is used for welding path planning, which not only avoids the complex and cumbersome constraint setting in traditional algorithms, but also predicts the optimal welding path by learning historical data. At the same time, the data set is expanded using data enhancement technology to improve the generalization ability and prediction accuracy of the model, thereby optimizing the efficiency and quality of welding path planning. The information acquisition system can collect key data in the welding process in real time, such as current, voltage, temperature, speed, etc., providing a reliable basis for the dynamic adjustment and optimization of the welding path. The welding path planning system can make instant adjustments to the welding path in combination with real-time data to ensure the stability of the welding process and the welding quality. Specific embodiment 2:

[0058] like Figure 1-3 As shown, the welding path planning system obtains welding information through the information acquisition system, and the information acquisition system extracts previous weld features through the laser vision probe, wherein the previous weld features include laser stripe lines and weld feature points, and the laser stripe includes the weld groove height, width and position information;

[0059] When extracting the features of the weld points, the information acquisition system will use the segmented fitting method and the welding path planning system to obtain the welding information through the information acquisition system, and the Steger algorithm, cluster analysis algorithm and consistency straight line fitting method for centralized extraction.

[0060] When extracting feature points in the information acquisition system, the slope is calculated using the segmented fitting method to determine the weld features. The weld features are extracted by fitting the segmented points with a straight line using the least squares method. The Steger algorithm is used to filter and denoise the image. The Otsu algorithm is used to locate the rectangular area of ​​the weld. In order to improve the calculation efficiency, the separability of the Gaussian function is used to simplify the two-dimensional Gaussian convolution, which can improve the extraction speed of the laser stripes. The cluster analysis algorithm is used to read the weld feature sampling points, create a data set, filter the sampling points by setting a threshold, read the sampling points continuously and classify them, calculate the geometric information based on the filtered sampling points, and use the consistent straight line fitting method to extract. The weld feature data and the straight line model equation are input, and the feature points are repeatedly selected by random sampling. The model is used for verification, and the model is evaluated by the accuracy of the data points. The execution is repeated 1000 times to obtain the fitting straight line equation.

[0061] The Steger algorithm is used for sub-pixel edge detection. It locates the edge by calculating the second-order derivative of the image, and then improves the accuracy by filtering and denoising. Suppose the image is I(x, y), the gradient and the second-order derivative The calculation is as follows:

[0062]

[0063] The partial derivative of the image in the x direction represents the rate of change of brightness in the horizontal direction;

[0064] : The partial derivative of the image in the y direction, indicating the rate of change of brightness in the vertical direction.

[0065] By combining the laser vision probe with a variety of algorithms (such as the segmented fitting method, Steger algorithm, cluster analysis algorithm, and consistent straight line fitting method) to extract weld features, the groove height, width, position, and weld point features of the weld can be accurately captured. These high-precision data provide a solid foundation for subsequent welding path planning. The Otsu algorithm is used to locate the weld rectangular area, and the separability of the Gaussian function is used to simplify the two-dimensional Gaussian convolution, which significantly improves the extraction speed of laser stripes. This not only reduces the calculation time, but also improves the real-time performance of the entire system. The cluster analysis algorithm and the consistent straight line fitting method perform well in processing large amounts of data, and can quickly and accurately screen out key sampling points and calculate geometric information to ensure the accuracy of the welding path. The welding path planning system uses the Seq2Seq deep learning model, which can dynamically adjust and optimize the welding path based on historical data and real-time collected information. This intelligent path planning avoids the complex and cumbersome constraint setting in traditional algorithms, improves planning efficiency and welding quality, and the entire system has achieved a high degree of automation and intelligence from welding path planning, parameter setting to welding execution. The user only needs to input basic information through the human-machine interface, and the system can automatically complete the welding task, greatly reducing manual intervention and errors. Specific embodiment three:

[0067] like Figure 1-3As shown in the figure, the internal structure of the CNC intelligent welding machine includes a CPU, a digital input and output module, a CANopen module, an RS485 communication module, and a Profibus communication module. The CPU is connected to an HMI control display panel and a weighing sensor, the CANopen module is connected to a servo drive unit, and the RS485 communication indirectly controls the battery pack through a battery management unit. The internal execution components of the CNC intelligent welding machine are mainly divided into a perception layer, a decision layer, and an execution layer according to the functional grouping. The CNC intelligent welding machine is equipped with a Webcam camera, which is installed on the front of the CNC intelligent welding machine to detect the path of the welding points to be welded in real time, and is used to identify the guide route and correct the vehicle body. The digital input and output module in the CNC intelligent welding machine is connected to an obstacle sensor, which is placed in the center of the side of the CNC intelligent welding machine, scanning the area with a maximum radius of 5m, and detecting obstacles in the driving direction. The decision layer uses an embedded microcomputer as a controller, and uses C# to write a control program in the Visual Studio 2012 environment to collect perception layer information. The execution layer is composed of a CPU, a digital input and output module, a CANopen module, an RS485 communication module, and a Profibus communication module.

[0068] The outside of the CNC intelligent welding machine is equipped with a welding gun fixing device on the running track of the frame, which is used to install and fix the welding gun. The welding gun fixing device can reciprocate left and right along the running track to realize the horizontal movement of the welding gun on the welding plane. The motors equipped with the CNC intelligent welding machine include a longitudinal drive motor, a transverse drive motor and a drive motor that controls the up and down movement of the welding gun.

[0069] The built-in control chips of the CNC intelligent welding machine are ESP8266 module and MPU-6050 chip. The CNC intelligent welding machine receives the instructions from the welding machine control system through ESP8266, and converts the instructions received by ESP826 into electrical signals for driving the motor through L298N, so as to remotely drive and control the welding machine.

[0070] The YOLOv5 network model algorithm is adopted in the information collection system, and the FPN-PAN design is integrated to ensure the full integration of multi-scale features. The MOSAIC algorithm is used for data enhancement at the input end to increase the frequency of small targets and organize and construct a data set of solder joints of the devices to be processed. The training of the model is first pre-trained on the source data set, and the backbone and neck parts of the obtained pre-trained model are parameter-frozen, and then transfer learning training is performed on the external crack data set of the main hull structure.

[0071] In order to realize the motion control and monitoring of the CNC intelligent welding machine, the software system of the welding machine control system needs to provide a friendly human-computer interaction interface and should be able to complete the following main tasks: (1) It can set the relevant parameters of the entire welding positioning robot motion control system, such as working mode, motion starting point, motion speed, motion stroke, etc., and import, edit, save and decode; (2) The system has two welding modes, namely automatic control and manual control. In the automatic control mode, the system runs step by step according to the corresponding program commands and executes in a loop. The running position status of the mobile platform can be viewed during the system operation, and the system can automatically stop in case of emergencies; under normal circumstances, the operator can see the position coordinates of the X, Y, and Z axes and the running speed values ​​through the display screen. The manual mode can realize operation instructions such as jog, linear interpolation, inverse circular interpolation, program zeroing, simulation operation, etc., which is convenient for the operator to accurately control the welding trajectory; (3) The corresponding logic control function can be selected. To ensure the normal operation of the system, conditional judgment, program jump, interruption, and delay logic can be selected.

[0072] The system supports both automatic control and manual control modes, meeting the operational requirements in different scenarios. In automatic control mode, the system can automatically execute preset programs, improving work efficiency; in manual mode, the operator can achieve precise control of the welding trajectory, providing a wealth of logical control functions, such as conditional judgment, program jump, interruption, delay, etc., to ensure the stable operation of the system under different working conditions. The decision-making layer uses an embedded microcomputer as a controller, and uses C# to write control programs in the Visual Studio 2012 environment, which can efficiently collect information from the perception layer and make quick decisions. The information acquisition system uses advanced algorithms for data processing and feature extraction, providing accurate data support for welding path planning. Specific embodiment four:

[0074] like Figure 1-3As shown in the figure, when selecting the welding path, the welding path planning system selects the ant colony algorithm. The ant colony algorithm has two positive and negative feedback mechanisms, as follows: (1) Positive feedback: When ants walk on a certain path, they will release information to increase the pheromone concentration of the path; (2) Negative feedback: The residual pheromone concentration of the path is in a state of continuous decay. When using this algorithm to complete path planning, its basic workflow is as follows: (1) Initialize the parameters used; (2) Randomly obtain the initial position of the ant, and then repeat the experiment; (3) Update the global pheromone, record the ant's route, compare it, obtain the best working route, and determine the optimal or suboptimal path of the welding robot from the starting point to the end point. At the same time, consider factors such as robot kinematic constraints, collision avoidance, and welding process constraints, establish a set of spatial points that the welding robot can reach as the algorithm's search space, and establish a set of spatial points that the welding robot can reach as the algorithm's search space. The ant colony algorithm is used for path planning and combinatorial optimization problems, and works as follows: Parameter initialization: Set the initial pheromone concentration τ0 and the volatility factor ρ parameter, and randomly select the next step path from the starting point with a probability of:

[0075]

[0076] τ ij is the pheromone concentration from path i to path j, η ij For heuristic information, usually the feasibility of the path or the reverse distance, pheromone update:

[0077] τ ij =(1-ρ)τ ij +Δτ ij

[0078] Δτ ij is the amount of pheromones released by ants as they walk on the path.

[0079] The ant colony algorithm is used for welding path planning, and the scheme combining positive feedback, negative feedback mechanism and considering factors such as robot kinematic constraints, collision avoidance, and welding process constraints has many advantages, which are summarized as follows:

[0080] 1. Intelligent optimization capability: The ant colony algorithm can find the optimal or suboptimal path from the starting point to the end point in a complex welding environment by simulating the pheromone transmission and path selection behavior of ants during foraging. This intelligent optimization capability helps reduce unnecessary movement during welding and improve welding efficiency.

[0081] 2. Adaptability and robustness: The positive feedback mechanism gradually increases the pheromone concentration on the high-quality path, attracting more ants to choose this path, thereby enhancing the algorithm's adaptability. At the same time, the negative feedback mechanism prevents the algorithm from falling into a local optimal solution through the attenuation of pheromones, thereby improving the algorithm's robustness.

[0082] 3. Comprehensive consideration of multiple constraints: During the path planning process, this solution not only considers the shortest path from the starting point to the end point of the welding robot, but also comprehensively considers the robot's kinematic constraints (such as joint angle limitations, speed limitations, etc.), collision avoidance (ensuring that the robot does not collide with the surrounding environment or workpiece) and welding process constraints (such as welding speed, welding angle, welding quality, etc.), which makes the planned path more in line with actual production needs.

[0083] 4. Efficient search capability: By establishing a set of spatial points that can be reached by the welding robot as the algorithm's search space, the search scope is greatly narrowed and the search efficiency is improved. At the same time, by utilizing the parallel search characteristics of the ant colony algorithm, a better solution can be found in a shorter time.

[0084] 5. Flexibility and scalability: The scheme has good flexibility and can adjust algorithm parameters and constraints according to different welding tasks and environmental conditions. In addition, since the ant colony algorithm itself is a heuristic algorithm, it is easy to combine with other algorithms (such as genetic algorithm, particle swarm algorithm, etc.) to form a hybrid optimization algorithm to further improve the effect of path planning.

[0085] 6. Real-time and dynamic adaptability: In actual applications, the welding environment may change (such as workpiece position offset, new obstacles appearing, etc.). This solution can update the global pheromone and ant travel route in real time, and dynamically adjust the welding path to adapt to environmental changes.

[0086] 7. Easy to implement and integrate: The ant colony algorithm has a relatively simple algorithm structure and clear implementation steps, which is easy to implement and integrate in the welding robot control system. At the same time, the algorithm can work closely with the hardware system of the welding robot (such as sensors, actuators, etc.) to realize the automation and intelligence of the welding process.

[0087] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprising a reference structure" do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0088] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An active learning micro PCB tool automatic welding machine system, including a main system, characterized in that: The main system consists of a welding machine control system, a CNC intelligent welding machine, an information acquisition system and a welding path planning system, wherein: The welding path planning system adopts Seq2Seq deep learning model to build a welding path planning deep learning model, and learns by recording the previously manually planned welding path data and welding data to predict the optimal welding path; The welding path planning system classifies the previously recorded welding path data, takes 70% of the data as a training set and 30% as a test set, and performs data enhancement on the input data by random cropping and random flipping to expand the data set; Before planning, the welding path planning system traverses all the workstations to be planned for welding path to obtain a weld point vector sequence. Secondly, the planned unordered weld point vector sequence B of workstation A is input into the welding path planning deep learning model. Then the model outputs the planned weld point vector sequence. The order of the weld point vectors in the sequence is the welding order. Finally, after all the workstations to be planned are traversed, the welding path planning results of all workstations are summarized.

2. The active learning micro PCB tool automatic welding machine system according to claim 1 is characterized in that: The welding path planning system obtains welding information through an information acquisition system, and the information acquisition system extracts previous weld features through a laser vision probe, wherein the previous weld features include laser stripe lines and weld feature points, and the laser stripe includes weld groove height, width and position information; When extracting the features of the welding points, the information acquisition system uses the segmented fitting method and the welding path planning system to obtain welding information through the information acquisition system, and uses the Steger algorithm for centralized extraction. The Steger algorithm is used for sub-pixel edge detection, and the edge is located by calculating the second-order derivative of the image, and then the accuracy is improved by filtering and denoising. Assume that the image is I(x, y), the gradient and the second derivative The calculation is as follows: The partial derivative of the image in the x direction represents the rate of change of brightness in the horizontal direction; The partial derivative of the image in the y direction represents the rate of change of brightness in the vertical direction.

3. The active learning micro PCB tool automatic welding machine system according to claim 2 is characterized in that: When extracting feature points in the information acquisition system, the slope is calculated by the segmented fitting method to determine the weld feature, the weld feature is extracted by linear fitting of the segmented points by the least squares method, the image is filtered and denoised by the Steger algorithm, the weld rectangular area is located by the Otsu algorithm, in order to improve the calculation efficiency, the separability of the Gaussian function is used to simplify the two-dimensional Gaussian convolution, which can improve the extraction speed of the laser stripes, the weld feature sampling points are read by the cluster analysis algorithm, a data set is created, the sampling points are screened by setting a threshold, the sampling points are continuously read and classified, and the geometric information is calculated according to the screened sampling points. At the same time, when extracting by the consistent straight line fitting method, the weld feature data and the straight line model equation are input, the feature points are repeatedly selected by random sampling, the model is used for verification, the model is evaluated by the accuracy of the data points, and the execution is repeated 1000 times to obtain the fitted straight line equation.

4. The active learning micro PCB tool automatic welding machine system according to claim 1 is characterized in that: The interior of the CNC intelligent welding machine includes a CPU, a digital input and output module, a CANopen module, an RS485 communication module, and a Profibus communication module, wherein the CPU is externally connected to an HMI control display panel and a weighing sensor, the CANopen module is externally connected to a servo drive unit, and the RS485 communication indirectly controls the battery pack through a battery management unit. The internal execution components of the CNC intelligent welding machine are mainly divided into a perception layer, a decision layer, and an execution layer according to functional grouping, wherein the CNC intelligent welding machine is equipped with a Webcam camera, which is installed on the front of the CNC intelligent welding machine to detect the path of the welding points to be welded in real time, and is used to identify the guide route and correct the vehicle body. The digital input and output module in the CNC intelligent welding machine is externally connected to an obstacle sensor, which is placed in the center of the side of the CNC intelligent welding machine, scanning the area with a maximum radius of 5m, and detecting obstacles in the driving direction. The decision layer uses an embedded microcomputer as a controller, and uses C# to write a control program in the Visual Studio 2012 environment to collect perception layer information. The execution layer is composed of a CPU, a digital input and output module, a CANopen module, an RS485 communication module, and a Profibus communication module.

5. The active learning micro PCB tool automatic welding machine system according to claim 4 is characterized in that: The outside of the CNC intelligent welding machine is provided with a welding gun fixing device on the running track of the frame, which is used to install and fix the welding gun. The welding gun fixing device can reciprocate left and right along the running track to realize the horizontal movement of the welding gun on the welding plane. The motor equipped with the CNC intelligent welding machine includes a longitudinal drive motor, a transverse drive motor and a drive motor for controlling the up and down movement of the welding gun.

6. The active learning micro PCB tool automatic welding machine system according to claim 5 is characterized in that: The built-in control chips of the CNC intelligent welding machine are respectively an ESP8266 module and an MPU-6050 chip. The CNC intelligent welding machine receives instructions from the welding machine control system through ESP8266, converts the instructions received by ESP826 into electrical signals for driving the motor through L298N, and remotely drives and controls the welding machine.

7. The active learning micro PCB tool automatic welding machine system according to claim 6 is characterized in that: The information acquisition system adopts the YOLOv5 network model algorithm and integrates the FPN-PAN design to ensure the full integration of multi-scale features. The MOSAIC algorithm is used at the input end to enhance the data, increase the frequency of small targets, and organize and construct a data set of solder joints of the device to be processed. The training of the model is first pre-trained on the source data set, and the backbone and neck parts of the obtained pre-trained model are parameter-frozen, and then transfer learning training is performed on the external crack data set of the main hull structure.

8. The active learning micro PCB tool automatic welding machine system according to claim 7, characterized in that: In order to realize the motion control and monitoring of the CNC intelligent welding machine, the software system of the welding machine control system needs to provide a friendly human-computer interaction interface and should be able to complete the following main tasks: (1) It can set the relevant parameters of the entire welding positioning robot motion control system, including the parameters such as working mode, motion starting point, motion speed, motion stroke, etc., and import, edit, save and decode them; (2) The system has two welding modes, namely automatic control and manual control. In the automatic control mode, the system runs step by step according to the corresponding program commands and executes in a loop. The running position status of the mobile platform can be viewed during the system operation, and the system can automatically stop in case of an emergency; under normal circumstances, the operator can see the position coordinates of the three axes of X, Y, and Z and the running speed value through the display screen. The manual mode can realize the operation instructions such as jog, linear interpolation, inverse circular interpolation, program zeroing, simulation operation, etc., so as to facilitate the operator to accurately control the welding trajectory; (3) The corresponding logic control function can be selected. To ensure the normal operation of the system, conditional judgment, program jump, interruption, and delay logic can be selected.

9. The active learning micro PCB tool automatic welding machine system according to claim 8, characterized in that: When selecting a welding path, the welding path planning system selects the ant colony algorithm. The ant colony algorithm has two positive and negative feedback mechanisms, which are as follows: (1) Positive feedback: When ants walk on a certain path, they will release information to increase the pheromone concentration of the path; (2) Negative feedback: The residual pheromone concentration of the path is in a state of continuous decay. When using this algorithm to complete path planning, its basic workflow is as follows: (1) Initialize the parameters used; (2) Randomly obtain the initial position of the ant, and then repeat the experiment; (3) Update the global pheromone, record the ant's route, compare it, obtain the best working route, determine the optimal or suboptimal path of the welding robot from the starting point to the end point, and consider factors such as robot kinematic constraints, collision avoidance, and welding process constraints. Establish a set of spatial points that can be reached by the welding robot as the search space of the algorithm. Establish a set of spatial points that can be reached by the welding robot as the search space of the algorithm. The ant colony algorithm is used for path planning and combinatorial optimization problems, and works as follows: Parameter initialization: Set the initial pheromone concentration τ0 and the volatility factor ρ parameter, and randomly select the next step path from the starting point with a probability of: τ ij is the pheromone concentration from path i to path j, η ij For heuristic information, usually the feasibility of the path or the reverse distance, pheromone update: t ij =(1-ρ)τ ij +Δt ij Δτ ij is the amount of pheromones released by ants as they walk on the path.

Images