Gold bonding wire welding method and system based on artificial intelligence optimization

Abstract

The present invention relates to the technical field of gold bonding wires, specifically to a gold bonding wire welding method and system based on artificial intelligence optimization, the welding method comprising: S1: collecting various parameters of the welding process; S2: constructing a grid map, using the RRT algorithm to generate an initial optimal path; using the DWA algorithm and the bidirectional Theta* algorithm to optimize the initial optimal path to obtain an optimal welding path; S3: constructing a prediction model based on a machine learning algorithm, outputting a prediction result, and then adjusting the optimal welding path; S4: welding a workpiece according to the optimal path; S5: collecting a workpiece image of a finished gold bonding wire to determine whether the appearance of the finished gold bonding wire meets a preset standard. The present invention optimizes the initial optimal path generated by the RRT algorithm by combining the DWA algorithm with the bidirectional Theta* algorithm, thereby improving the efficiency of welding path planning, ensuring the safety of gold bonding wire welding, and significantly improving welding production efficiency and production quality.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent linkage control, and in particular to a gold bonding wire welding method based on artificial intelligence optimization and a gold bonding wire welding system based on artificial intelligence optimization. Background Art

[0002] Gold bonding wire is a thin metal wire used in the microelectronic packaging process. It is mainly used as a connecting lead in integrated circuits and semiconductor devices. It connects the semiconductor chip to the outside by hot-pressed ball welding or ultrasonic hot-pressed ball welding. As an electrical connection lead between the chip and the external circuit, gold bonding wire is an essential basic raw material in the traditional packaging process of semiconductor integrated circuits, discrete devices, sensors, optoelectronics, etc. It is mainly used to connect the electrical path between the semiconductor chip and the external circuit. They are usually made of metals with good electrical conductivity such as gold, aluminum or copper. Gold bonding wire is an important material used in the manufacture of electronic products. Controlling the stability and accuracy of its production process is crucial to product quality.

[0003] However, in the technical field related to gold bonding wire production, the currently widely used modeling and analysis methods have obvious limitations. These modeling and analysis methods are inaccurate in accurately identifying the welding path planning of gold bonding wires during the production process. This technical shortcoming directly leads to fluctuations in product quality and significantly affects the improvement of production efficiency. In view of the fact that precise control of the welding path plays a vital role in ensuring the connection quality and long-term stability of gold bonding wires, this significant shortcoming of the existing modeling and analysis methods has become the core bottleneck restricting the further improvement of the production quality of gold bonding wires and the overall optimization of production efficiency. In order to overcome this challenge, the present invention proposes a gold bonding wire welding method based on artificial intelligence optimization to achieve precise control of the welding path during the production process of gold bonding wires. Summary of the invention

[0004] Based on this, it is necessary to propose a bonding wire welding method and system based on artificial intelligence optimization to address the problems of low welding path planning accuracy and low efficiency in the existing bonding wire welding process.

[0005] The present invention is implemented by the following technical solution: The artificial intelligence optimized bonding wire system welding method comprises the following steps:

[0006] S1: Collect real-time parameter data based on welding current, voltage, and temperature, obtain status data based on weld morphology and liquid splash, and record sound data generated during welding.

[0007] S2: The workpiece to be processed is modeled in three dimensions to obtain a workpiece model, and the workpiece model is spatially divided to obtain a grid map. According to the grid map, the RRT algorithm is used to generate the initial optimal path in combination with real-time parameter data, state data, and sound data. The DWA algorithm is used to locally optimize the initial optimal path to obtain a local optimized path, and the bidirectional Theta* algorithm is used to adjust the local optimized path to obtain the optimal welding path.

[0008] S3: Build a prediction model based on machine learning algorithm, input real-time parameter data into the prediction model to obtain prediction results, and adaptively adjust the optimal welding path according to the prediction results.

[0009] S4: Weld the workpiece to be processed according to the optimal path to obtain a finished bonding gold wire.

[0010] S5: collecting workpiece images of finished gold bonding wires, performing preprocessing, feature extraction and defect recognition on the workpiece images, determining whether the appearance of the finished gold bonding wires meets preset standards, and retaining the finished gold bonding wires that meet the preset standards.

[0011] Furthermore, in step S2, the specific steps of dividing the grid map include:

[0012] S211: Divide the three-dimensional area where the workpiece model is located into a grid.

[0013] S212: Each grid is defined as a workpiece position node, and each workpiece position node is recorded on the grid map to obtain a grid map.

[0014] Furthermore, in step S2, the method for generating the initial optimal path includes:

[0015] S221: Select multiple workpiece position nodes as passing nodes according to the processing technology.

[0016] S222: Using the RRT algorithm to generate a path passing through all the path nodes as the initial optimal path. Further, in step S222, the method for generating a local initial optimal path between two adjacent path nodes includes:

[0017] S2221: Set the next path node to q goal , divide the space range between the current path node and the next path node according to the preset space shape, and randomly generate a temporary node q within the space range random , select the current path node as the parent node of the temporary node.

[0018] S2222: Check temporary node q randomWhether the line connecting the current node passes through an obstacle, if yes, the temporary node is discarded and a new temporary node is generated, otherwise, the temporary node q is used to generate a new temporary node. random The angle θ between the line connecting the current node and the x-axis and the step length s determine the new node q new The position of the new node q new Put it into the node list P.

[0019] Among them, the coordinate formula of the new node is expressed as:

[0020] ;

[0021] ;

[0022] In the formula, x new is the x-axis coordinate of the new node, y new is the y-axis coordinate of the new node, is the x-axis coordinate of the current path node, The y-axis coordinate of the current path node.

[0023] Furthermore, in step S2, using the DWA algorithm to locally optimize the local initial optimal path includes:

[0024] S231: Form multiple predicted trajectories Ci according to the local initial optimal path and real-time parameter data.

[0025] S232: Take the velocity v at the end of the predicted trajectory to extend the predicted trajectory curve, and remove unqualified predicted trajectories according to the angular velocity w at the end of the predicted trajectory.

[0026] S233: Determine a temporary target point according to the new node in the node linked list P.

[0027] S234: Determine whether there is an obstacle between the current path node and the temporary target point, and if so, mark the obstacle. And discard the corresponding cluster of predicted trajectories C i .

[0028] S235: Selecting the next temporary target point according to the position and yaw angle of the current temporary target point.

[0029] S236: Determine whether the next temporary target point is the next path node, if yes, output the local initial optimal path passing through each temporary target point, otherwise continue to search for the next temporary target point.

[0030] Furthermore, in step S232, the evaluation function of the DWA algorithm is used to eliminate unqualified prediction trajectories, and the formula is expressed as:

[0031] ;

[0032] Where α, β, and γ are the weights corresponding to the various indicators of the evaluation function. head(v, w) represents the evaluation function of the azimuth angle, and its value is π-θ, ​​where θ is the angle difference. vel(v, w) represents the linear velocity of the current robot at the moment of motion. dist(v, w) represents the shortest distance between its corresponding motion trajectory and the obstacle when the robot is at the speed (v, w). σ is the normalization coefficient.

[0033] Further, in step S2, the step of using the bidirectional Theta* algorithm to adjust the local optimal path between the current path node and the next path node to obtain the optimal welding path includes:

[0034] S241: Initialize the current path node and the next path node, create an OPEN table and a CLOSE table in the positive and negative search directions respectively, and add the current path node and the next path node to the corresponding OPEN table, the actual path cost G value from the current path node to the temporary target point, and the heuristic estimated cost F value from the temporary target point to the next path node.

[0035] S242: Compare the heuristic function F value from the current path node to each temporary target point in the Open list, and update the temporary target point with the smallest F value as the optimal temporary node.

[0036] S243: Check whether the optimal temporary node is the next path node, if yes, generate a forward path passing through all the optimal temporary nodes, otherwise move the optimal temporary node into the Close list and remove it from the Open list.

[0037] S244: Search for the reverse optimal temporary node from the next path node in reverse order until the current path node, generate a reverse path passing through all the reverse optimal temporary nodes, and determine whether the forward path is consistent with the reverse path. If so, retain the forward path as the optimal welding path.

[0038] Furthermore, in step S3, a prediction model based on a machine learning algorithm is constructed, and the welding parameters are put into the prediction model, and the prediction results are output to determine whether the prediction results exceed the preset parameter range. If so, the optimal welding path is adjusted according to the RRT algorithm.

[0039] Further, in step S5, the preprocessing includes digitization, normalization, grayscale conversion, and histogram equalization processing. Digitization is used to convert the workpiece image into a digital image. Normalization is used to convert the pixel value of the digital image to between 0 and 1. Grayscale conversion is used to adjust the grayscale distribution of the digital image. Histogram equalization widens the grayscale of the image with a large number of pixels and reduces the grayscale of the image with a small number of pixels.

[0040] The present invention also provides a gold bonding wire welding system based on artificial intelligence optimization, comprising:

[0041] The data collection module is used to collect real-time data of welding current, voltage, and temperature, obtain the status data of weld morphology and liquid splash, and record the sound data generated during the welding process.

[0042] The path planning module includes a grid division unit, an initial path unit, an optimization path unit, and a fine-tuning path unit. The grid division unit is used to divide the three-dimensional workpiece model of the workpiece to be processed into a grid to form a grid map. The initial path unit is used to generate an initial optimal path using the RRT algorithm based on the grid map, real-time parameter data, status data, and sound data. The optimization path unit is used to use the DWA algorithm to locally optimize the initial path to obtain a local optimized path. The fine-tuning path unit is used to use the bidirectional Theta* algorithm to fine-tune the local optimized path to obtain the optimal welding path.

[0043] The adaptive adjustment module is used to determine whether the actual welding path passes through an obstacle during the welding process, and if so, the RRT algorithm is used to adaptively adjust the optimal welding path.

[0044] The welding module is used to weld the workpiece to be processed according to the optimal welding path to obtain a finished bonding gold wire.

[0045] The welding detection module is used to detect the finished gold bonding wire product, determine whether the appearance of the finished gold bonding wire product meets the preset standard, and retain the finished gold bonding wire product that meets the preset standard.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention uses the RRT algorithm to quickly generate the initial optimal path, providing a good foundation for subsequent local optimization and adjustment. The bidirectional Theta* algorithm searches from both ends simultaneously, further improving the efficiency of path planning. The DWA algorithm fully considers the real-time obstacle information and the dynamic constraints of the robot during the local optimization process to ensure the safety of the path. The bidirectional Theta* algorithm can find a more natural and smooth path, avoid unnecessary turns and pauses, and improve the efficiency and quality of welding. The entire algorithm process combines the advantages of multiple algorithms and can be flexibly adjusted and optimized according to different application scenarios and requirements. The DWA algorithm has the characteristics of strong real-time performance and can dynamically adjust and optimize the path according to real-time parameter data and status data.

[0048] By optimizing the welding path and reducing unnecessary movement and pauses, the welding production efficiency is significantly improved. The intelligent path planning function can ensure that the welding line is seamless and smooth at the joint, thereby improving the quality and reliability of the welded product. By improving welding efficiency and quality, intelligence can reduce manufacturing costs and scrap rates, creating greater economic benefits for the company. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a step diagram of a gold bonding wire welding method based on artificial intelligence optimization in Example 1 of the present invention;

[0050] Figure 2 This is a flow chart of a gold bonding wire welding method based on artificial intelligence optimization in Example 1 of the present invention;

[0051] Figure 3 This is a module diagram of the gold bonding wire welding system based on artificial intelligence optimization in Example 2 of the present invention. DETAILED DESCRIPTION

[0052] 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.

[0053] It should be noted that when a component is said to be "mounted on" another component, it is directly on the other component or there may be a component centered therein. When a component is considered to be "set on" another component, it is directly set on the other component or there may be a component centered therein. When a component is considered to be "fixed to" another component, it is directly fixed to the other component or there may be a component centered therein.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "or / and" used herein includes any and all combinations of one or more of the related listed items.

[0055] Example 1: Please refer to Figure 1-Figure 2 , this embodiment provides an artificial intelligence optimized bonding wire system welding method, comprising the following steps:

[0056] S1: Collect real-time data of welding current, voltage, and temperature. Capture visual information during the welding process, such as weld shape, spatter, etc. Record the sound generated during the welding process to analyze the stability of the welding process. Remove noise and outliers to ensure the accuracy and reliability of the data.

[0057] S2: The workpiece to be processed is modeled in three dimensions to obtain a workpiece model, and the workpiece model is spatially divided to obtain a grid map. According to the grid map, the RRT algorithm is used to generate the initial optimal path in combination with real-time parameter data, state data, and sound data. The DWA algorithm is used to locally optimize the initial optimal path to obtain a local optimized path, and the bidirectional Theta* algorithm is used to adjust the local optimized path to obtain the optimal welding path.

[0058] S21: Divide the welding workpiece into a grid map, including:

[0059] S211: The environment map is a grid map, a geometric element map and a topological map. The three-dimensional area where the bonding wire model is located is divided into a grid according to a certain ratio, and the divided grid is further subdivided into free grids and obstacles. The free grid represents an area without obstacles and can be passed through without obstacles. The map is divided into 100 grids of equal scale, starting from grid No. 1, numbered in sequence, and ending at 100.

[0060] S212: Define each grid as a workpiece location node, and each workpiece location node is recorded on the grid map to obtain a grid map. Start the path planning from the initial grid to search for at least one feasible path to achieve the optimal solution. The grid map describes the real environment and corresponds one-to-one to a specific environment. The grid map has the advantages of being easy to establish, easy to change, and able to retain more environmental information. The grid map method uses the quadtree method and the octree method, and then uses the corresponding algorithm to complete the purpose of searching the path. Taking the quadtree as an example, this improved method will effectively compress environmental information, improve search efficiency, and save time. The grid map method is simple and practical, easy to implement, and the higher the accuracy of the grid.

[0061] S22: The method of generating an initial optimal path using the RRT algorithm includes:

[0062] S221: selecting a plurality of workpiece position nodes as passing nodes according to the processing technology;

[0063] S222: Using the RRT algorithm to generate a path passing through all the nodes along the way as an initial optimal path.

[0064] S2221: Set the target node to q goal , divide the space range between the current path node and the next path node according to the preset space shape, and randomly generate a temporary node q within the space rangerandom , select the current path node as the parent node of the temporary node;

[0065] S2222: Check temporary node q random Whether the line connecting the current node passes through an obstacle, if yes, the temporary node is discarded and a new temporary node is generated, otherwise, the temporary node q is used to generate a new temporary node. random The angle θ between the line connecting the current node and the x-axis and the step length s determine the new node q new The position of the new node q new Put it into the node list P;

[0066] Among them, the new node coordinate formula and the distance formula between two points are expressed as:

[0067] ;

[0068] ;

[0069] ;

[0070] Where d is the Euclidean distance between two nodes, x new is the x coordinate of the new node, y new is the y coordinate of the new node, is the x-axis coordinate of the neighboring node, is the y-axis coordinate of the neighboring node.

[0071] S23: Use DWA to locally optimize the initial path generated by the RRT algorithm.

[0072] The DWA algorithm, also known as the dynamic window method, converts position change information into linear velocity and angular velocity information, converts the obstacle avoidance problem into a kinematic constraint problem, and finds the best path under its own information analysis and motion constraints. The DWA algorithm mainly samples in the velocity space. After sampling multiple groups of velocities, it simulates the next path for these velocity groups, sets the path evaluation criteria, evaluates and analyzes the simulated path, and selects the optimal path.

[0073] Since the DWA algorithm is a local path planning algorithm, it often deviates from the optimal path during the movement process. Therefore, the global path planning RRT algorithm is used to find an optimal path as the guide of the DWA algorithm, and it always tends to move along the optimal path. After the RRT algorithm performs the reselection of parent nodes and rewiring, it will store the nodes in the optimal path in a linked list S in the order from the target point to the starting point, and propose a method to achieve the optimal path based on this.

[0074] The local optimization of the initial path using the DWA algorithm includes:

[0075] S231: forming multiple predicted trajectories Ci according to the local initial optimal path and real-time parameter data;

[0076] S232: Taking the velocity v at the end of the predicted trajectory to extend the predicted trajectory curve, and eliminating unqualified predicted trajectories according to the angular velocity w at the end of the predicted trajectory;

[0077] S233: Determine a temporary target point according to the new node in the node list P;

[0078] S234: Determine whether there is an obstacle between the current path node and the temporary target point, if yes, mark the obstacle; and discard the corresponding cluster of predicted trajectories C i ;

[0079] S235: selecting the next temporary target point according to the position and yaw angle of the current temporary target point;

[0080] S236: Determine whether the next temporary target point is the next path node, if yes, output the local initial optimal path passing through each temporary target point; otherwise, continue to search for the next temporary target point.

[0081] After all trajectory samples are collected, these trajectories are all feasible, but only the best trajectory is needed. Therefore, it is necessary to evaluate the pros and cons of these trajectories. The criterion for selecting the best trajectory should be to avoid obstacles safely while moving towards the destination in the shortest path.

[0082] The evaluation function of the DWA algorithm is used to eliminate unqualified prediction trajectories. The formula is expressed as:

[0083] ;

[0084] In the formula, α, β, γ are the weights corresponding to each indicator of the evaluation function, head(v, w) represents the evaluation function of the azimuth angle, and its value is π-θ, ​​θ is the difference in angles, that is, the difference between the azimuth angle of the last position of the trajectory and the azimuth angle of the target point. The evaluation standard is that the smaller the angle, the higher the function score. vel(v, w) represents the linear velocity of the current robot at the moment of motion. The larger its value, the faster the robot is and the higher the evaluation. dist(v, w) represents the shortest distance between its corresponding motion trajectory and the obstacle when the robot is at the speed (v, w). This indicator determines the safety of the path trajectory. It penalizes the sampling points close to the obstacle to ensure that the distance between the robot and the obstacle is always at a safe distance. The σ value is the normalization coefficient.

[0085] The formula after normalization is:

[0086] ;

[0087] ;

[0088] ;

[0089] Where n represents the total number of trajectories simulated by the sampled velocity group, i represents the ith simulated trajectory, and v i represents the absolute value of the linear velocity at the end point when simulating the i-th trajectory, d i It represents the shortest distance between the ith trajectory and the obstacle in the simulated trajectory. After normalization, the trajectory with the highest evaluation function value is used as the next motion trajectory, and its corresponding speed parameter will be retained.

[0090] S24: The step of using the bidirectional Theta* algorithm to adjust the local optimal path between the current path node and the next path node to obtain the optimal welding path includes:

[0091] S241: Initialize the current path node and the next path node, create an OPEN table and a CLOSE table in the forward and reverse search directions respectively, and add the current path node and the next path node to the corresponding OPEN table, the actual path cost G value from the current path node to the temporary target point, and the heuristic estimated cost F value from the temporary target point to the next path node;

[0092] S242: Compare the heuristic function F value from the current path node to each temporary target point in the Open list, and update the temporary target point with the smallest F value as the optimal temporary node;

[0093] S243: Check whether the optimal temporary node is the next path node, if yes, generate a forward path passing through all the optimal temporary nodes; otherwise, move the optimal temporary node into the Close list and remove it from the Open list;

[0094] S244: Search for the reverse optimal temporary node from the next path node in reverse order until the current path node, generate a reverse path passing through all the reverse optimal temporary nodes, and determine whether the forward path is consistent with the reverse path. If so, retain the forward path as the optimal welding path.

[0095] S3: During the welding process, the artificial intelligence system collects parameters such as welding current, voltage, welding speed, temperature distribution, etc. in real time through sensors. It uses a prediction model built from historical welding data based on a machine learning algorithm and puts the real-time monitored welding parameters into the constructed prediction model to make accurate predictions. Once a potential problem is predicted, the artificial intelligence will immediately adjust the initial welding path of the RRT algorithm to reduce the occurrence of welding problems.

[0096] Predictive adjustment not only improves the stability of the welding process, but also significantly reduces the incidence of welding defects. By adjusting the welding path in advance, the intelligent algorithm can ensure the consistency and reliability of welding quality, thereby improving overall welding efficiency. The real-time feedback and adjustment mechanism can also improve the flexibility and adaptability of the welding process. When the welding environment or material changes, the intelligent system can quickly identify and adjust the welding path and parameters to adapt to the new welding conditions. This flexibility ensures the stability and reliability of the welding process in various complex environments.

[0097] S4: Perform accurate and efficient welding operations according to the optimal path planned by the RRT algorithm. The RRT algorithm explores feasible paths by randomly sampling the space and constructing an expanding tree, allowing the robot to quickly find a collision-free path from the starting point to the target point in a complex environment. During the welding process, the robot strictly follows this path to ensure the precise alignment of the welding points and the uniformity of the weld, thereby improving the welding quality and production efficiency, and providing new technical support for intelligent manufacturing and precision machining.

[0098] S5: Use high-resolution industrial cameras and precision optical systems to capture images of the bonded gold wire after welding. Then use image processing technology to pre-process, extract features, and identify defects on the captured images to determine whether the bonded gold wire is within the preset standard data range. If it is, the bonded workpiece is qualified. Otherwise, it does not meet the standard.

[0099] Preprocessing includes digitization, normalization, grayscale transformation, and histogram equalization. Preprocessing improves image quality and facilitates subsequent feature extraction and defect identification. Digitization converts analog images into digital images. Normalization changes the pixel values ​​of 0-255 to between 0-1, reducing their distribution distance. Grayscale transformation changes the grayscale distribution of the image. Grayscale transformation enhances the visualization of the image and improves the quality of the image. Histogram equalization widens the grayscale of the digital image with a large number of pixels and reduces the grayscale of the digital image with a small number of pixels. Histogram equalization makes the grayscale distribution of the image more uniform and enhances the contrast of the image.

[0100] Image processing algorithms are used to extract characteristic information such as shape, size, and texture of the welding area.

[0101] Defect recognition is the process of judging whether there are defects in the welding area based on the extracted feature information and the preset defect recognition model. Welding defects include cracks, pores, slag inclusions, etc. Cracks appear as lines or mesh structures. By extracting the length, width, and direction of the welding area, the preset defect recognition model is used to automatically identify cracks. Porosity is a cavity formed due to incomplete gas discharge during welding. Porosity is automatically identified by extracting the texture features of the welding area and the preset defect recognition model. Slag inclusion is a defect formed by the mixing of impurities during welding. The shape and texture features of the welding area are extracted and combined with the preset defect recognition model to realize automatic identification of slag inclusions.

[0102] Example 2: Please refer to Figure 3 This embodiment provides an artificial intelligence optimized gold bonding wire system welding system, which can be controlled by the artificial intelligence optimized gold bonding wire system welding method of embodiment 1, including:

[0103] Data collection module, used to collect real-time data of welding current, voltage, and temperature, obtain data on weld shape and liquid splash status, and record sound data generated during welding;

[0104] The path planning module includes a grid division unit, an initial path unit, an optimization path unit, and a fine-tuning path unit; the grid division unit is used to divide the three-dimensional workpiece model of the workpiece to be processed into a grid to form a grid map; the initial path unit is used to generate an initial optimal path using the RRT algorithm according to the grid map, real-time parameter data, state data, and sound data; the optimization path unit is used to use the DWA algorithm to locally optimize the initial path to obtain a local optimization path; the fine-tuning path unit is used to use the bidirectional Theta* algorithm to fine-tune the local optimization path to obtain the optimal welding path;

[0105] The adaptive adjustment module is used to determine whether the actual welding path passes through obstacles during the welding process. If so, the RRT algorithm is used to adaptively adjust the optimal welding path.

[0106] A welding module is used to weld the workpiece to be processed according to the optimal welding path to obtain a finished bonding gold wire;

[0107] The welding detection module is used to detect the finished gold bonding wires, determine whether the appearance of the finished gold bonding wires meets the preset standards, and retain the finished gold bonding wires that meet the preset standards.

[0108] Compared with the prior art, the present invention has the following beneficial effects:

[0109] The RRT algorithm can quickly generate the initial optimal path, providing a good foundation for subsequent local optimization and adjustment. The bidirectional Theta* algorithm searches from both ends simultaneously, further improving the efficiency of path planning. The DWA algorithm fully considers the real-time obstacle information and the robot's dynamic constraints during the local optimization process to ensure the safety of the path. The bidirectional Theta* algorithm can find a more natural and smooth path, avoid unnecessary turns and pauses, and improve the efficiency and quality of welding. The entire algorithm process combines the advantages of multiple algorithms and can be flexibly adjusted and optimized according to different application scenarios and requirements. The DWA algorithm has the characteristics of strong real-time performance and can dynamically adjust and optimize the path according to real-time parameter data and status data.

[0110] By optimizing the welding path and reducing unnecessary movement and pauses, the welding production efficiency is significantly improved. The intelligent path planning function can ensure that the welding line is seamless and smooth at the joint, thereby improving the quality and reliability of the welded product. By improving welding efficiency and quality, intelligence can reduce manufacturing costs and scrap rates, creating greater economic benefits for the company.

[0111] The technical features of the above-described embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0112] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A gold bonding wire welding method based on artificial intelligence optimization, characterized in that: The steps include: S1: Collect real-time parameter data based on welding current, voltage, and temperature, obtain status data based on weld morphology and liquid splash, and record sound data generated during welding; S2: Performing three-dimensional modeling on the workpiece to be processed to obtain a workpiece model, and spatially dividing the workpiece model to obtain a grid map; generating an initial optimal path using the RRT algorithm based on the grid map in combination with the real-time parameter data, the state data, and the sound data; locally optimizing the initial optimal path using the DWA algorithm to obtain a local optimized path, and adjusting the local optimized path using the bidirectional Theta* algorithm to obtain an optimal welding path; S3: constructing a prediction model based on a machine learning algorithm, inputting the real-time parameter data into the prediction model to obtain a prediction result, and adaptively adjusting the optimal welding path according to the prediction result; S4: welding the workpiece to be processed according to the optimal path to obtain a finished bonding gold wire; S5: collecting a workpiece image of the finished gold bonding wire, performing preprocessing, feature extraction and defect recognition on the workpiece image, determining whether the appearance of the finished gold bonding wire meets a preset standard, and retaining the finished gold bonding wire that meets the preset standard.

2. The bonding wire welding method based on artificial intelligence optimization according to claim 1 is characterized in that: In step S2, the specific steps of dividing the grid map include: S211: dividing the three-dimensional area where the workpiece model is located into grids; S212: Each grid is defined as a workpiece position node, and each of the workpiece position nodes is recorded on the grid map to obtain a grid map.

3. The bonding wire welding method based on artificial intelligence optimization according to claim 2 is characterized in that: In step S2, the method for generating the initial optimal path includes: S221: Selecting a plurality of workpiece position nodes as passing nodes according to the processing technology; S222: Using the RRT algorithm to generate a path passing through all the nodes along the way as an initial optimal path.

4. The bonding wire welding method based on artificial intelligence optimization according to claim 3 is characterized in that: In step S222, the method for generating a local initial optimal path between two adjacent path nodes includes: S2221: Set the next path node to q goal , divide the space range between the current path node and the next path node according to the preset space shape, and randomly generate a temporary node q within the space range random , select the current path node as the parent node of the temporary node; S2222: Check the temporary node q random Whether the line connecting the current path node passes through an obstacle, if yes, the temporary node is discarded and a new temporary node is generated, otherwise, the temporary node q is generated according to the temporary node q random The angle θ between the line connecting the current path node and the x-axis and the step length s determine the new node q new The new node q new Put it into the node list P; The coordinate formula of the new node is expressed as: ； ； In the formula, x new is the x-axis coordinate of the new node, y new is the y-axis coordinate of the new node, is the x-axis coordinate of the current path node, The y-axis coordinate of the current path node.

5. The bonding wire welding method based on artificial intelligence optimization according to claim 4 is characterized in that: In step S2, using the DWA algorithm to locally optimize the local initial optimal path includes: S231: forming a plurality of predicted trajectories Ci according to the local initial optimal path and the real-time parameter data; S232: Taking the velocity v of the end of the predicted trajectory to extend the predicted trajectory curve, and eliminating unqualified predicted trajectories according to the angular velocity w of the end of the predicted trajectory; S233: Determine a temporary target point according to the new node in the node list P; S234: Determine whether an obstacle appears between the current path node and the temporary target point, and if so, mark the obstacle; and discard the corresponding cluster of predicted trajectories C i ; S235: selecting the next temporary target point according to the position and yaw angle of the current temporary target point; S236: Determine whether the next temporary target point is the next path node, if yes, output the local initial optimal path passing through each temporary target point; otherwise, continue to search for the next temporary target point.

6. The bonding wire welding method based on artificial intelligence optimization according to claim 5 is characterized in that: In step S232, the evaluation function of the DWA algorithm is used to eliminate unqualified prediction trajectories, and the formula is expressed as: ； Where α, β, and γ are the weights corresponding to the various indicators of the evaluation function. head(v, w) represents the evaluation function of the azimuth angle, and its value is π-θ, ​​where θ is the angle difference. vel(v, w) represents the linear velocity of the current robot at the moment of motion. dist(v, w) represents the shortest distance between its corresponding motion trajectory and the obstacle when the robot is at the speed (v, w). σ is the normalization coefficient.

7. The bonding wire welding method based on artificial intelligence optimization according to claim 1 is characterized in that: In step S2, the step of using the bidirectional Theta* algorithm to adjust the local optimal path between the current path node and the next path node to obtain the optimal welding path includes: S241: Initialize the current path node and the next path node, create an OPEN table and a CLOSE table in the forward and reverse search directions respectively, and add the current path node and the next path node to the corresponding OPEN table, the actual path cost G value from the current path node to the temporary target point, and the heuristic estimated cost F value from the temporary target point to the next path node; S242: Compare the heuristic function F value from the current path node to each temporary target point in the Open list, and update the temporary target point with the smallest F value as the optimal temporary node; S243: Check whether the optimal temporary node is the next path node, if yes, generate a forward path passing through all the optimal temporary nodes; otherwise, move the optimal temporary node into the Close list and remove it from the Open list; S244: reversely search for the optimal temporary node from the next path node to the current path node, generate a reverse path passing through all reverse optimal temporary nodes, and determine whether the forward path is consistent with the reverse path. If so, retain the forward path as the optimal welding path.

8. The bonding wire welding method based on artificial intelligence optimization according to claim 1 is characterized in that: In step S3, a prediction model based on a machine learning algorithm is constructed, and the welding parameters are placed in the prediction model, and the prediction results are output to determine whether the prediction results exceed the preset parameter range. If so, the optimal welding path is adjusted according to the RRT algorithm.

9. The bonding wire welding method based on artificial intelligence optimization according to claim 1 is characterized in that: In step S5, the preprocessing includes digitization, normalization, grayscale conversion, and histogram equalization processing, wherein the digitization is used to convert the workpiece image into a digital image; the normalization is used to convert the pixel value of the digital image to between 0 and 1; and the grayscale conversion is used to adjust the grayscale distribution of the digital image; The histogram equalization expands the grayscales of the digital image with a large number of pixels and reduces the grayscales of the digital image with a small number of pixels.

10. A gold bonding wire welding system based on artificial intelligence optimization, which adopts the gold bonding wire system welding method optimized by artificial intelligence as claimed in any one of claims 1 to 9, characterized in that: Data collection module, used to collect real-time data of welding current, voltage, and temperature, obtain data on weld shape and liquid splash status, and record sound data generated during welding; The path planning module includes a grid division unit, an initial path unit, an optimization path unit, and a fine-tuning path unit; the grid division unit is used to divide the three-dimensional workpiece model of the workpiece to be processed into a grid shape to form a grid map; the initial path unit is used to generate an initial optimal path using an RRT algorithm according to the grid map, the real-time parameter data, the state data, and the sound data; the optimization path unit is used to use a DWA algorithm to locally optimize the initial path to obtain a local optimization path; the fine-tuning path unit is used to use a bidirectional Theta* algorithm to fine-tune the local optimization path to obtain an optimal welding path; An adaptive adjustment module is used to determine whether the actual welding path passes through an obstacle during welding, and if so, the RRT algorithm is used to adaptively adjust the optimal welding path; A welding module, used for welding the workpiece to be processed according to the optimal welding path to obtain a finished bonding gold wire; The welding detection module is used to detect the finished gold bonding wire product, determine whether the appearance of the finished gold bonding wire product meets the preset standard, and retain the finished gold bonding wire product that meets the preset standard.

Images