Intelligent wire feeding control method and system for laser welding

Abstract

The invention discloses an intelligent wire feeding control method and system for laser welding, and relates to the related field of the control system technology.The method comprises the steps that the resistance between a welding wire and a workpiece to be welded is measured, the distance value is indirectly calculated, initial parameters of a laser welding wire feeder are set, and in the operation process, images and point clouds of a welding pool area are collected; positioning a to-be-welded workpiece welding seam by utilizing the pre-trained welding seam recognition model, obtaining the width, gradient and depth of the welding seam, and calculating the welding seam volume and the required wire feeding amount; besides, according to the method, the shape of a plasma air mass generated by welding is recognized from an image through a classification model, the mapping relation between the plasma air mass and a wire feeding control adjustment instruction is established, and the wire feeding strategy of the laser welding wire feeder is intelligently optimized; and the wire feeding speed variable quantity is calculated in real time through a wire feeding speed matching model, and an instruction is generated to adjust parameters of the laser welding wire feeder in real time in combination with an optimized wire feeding strategy. Dynamic optimization of the laser welding wire feeding process is achieved, and the welding quality is improved.

Description

Technical Field

[0001] This application relates to the field of control system technology, and in particular to an intelligent wire feeding control method and system for laser welding. Background Technology

[0002] Laser welding is a highly efficient and precise welding method that uses a high-energy-density laser beam as a heat source. It can quickly focus energy on the weld seam, achieve rapid welding, improve production efficiency, facilitate automated control, and improve welding quality and consistency.

[0003] Wire feeding is a critical step in laser welding. Traditional wire feeding methods are mainly accomplished manually or with simple mechanical devices. This requires operators to concentrate highly on adjusting the wire feeding speed and force, which demands high operator skills, is inefficient, and is prone to inconsistent welding quality due to human factors. Mechanical wire feeding can achieve a certain degree of automation. Laser welding wire feeders are devices used to automatically or manually control the feeding of welding wire into the weld seam during laser welding. However, they suffer from problems such as unstable wire feeding, easy slippage, and wire clamping, which affect welding quality and efficiency.

[0004] Existing technologies incorporate intelligent welding systems into laser welding wire feeders to achieve automated wire feeding control. By precisely controlling the wire feeding speed and amount, wire vibration and feeding difficulties are reduced. However, existing control methods adjust relevant parameters based on pre-set strategies, which cannot adapt to dynamically changing weld conditions. The control strategies are relatively simple, and the control accuracy and efficiency need to be further improved to enhance the quality of laser welding. Summary of the Invention

[0005] To address the technical problems of the prior art, the present invention provides an intelligent wire feeding control method and system for laser welding. Based on visual perception, weld condition assessment and layered control, it realizes dynamic optimization of the laser welding process and adjusts the wire feeding parameters in real time according to the requirements of different welding tasks, adapting to welding workpieces of different materials, thicknesses and shapes.

[0006] This invention provides an intelligent wire feeding control method for laser welding, comprising: Step S10: Before welding begins, measure the resistance between the welding wire and the workpiece to be welded, then calculate the distance between them, determine the initial parameters of the laser welding wire feeder, and set the initial wire feeding length and initial wire feeding speed. In step S20, during the operation of the laser welding wire feeder with initial parameters, images and point cloud data of the welding molten pool area are collected. Using a pre-trained weld recognition model, the weld of the workpiece to be welded is located from the image, the weld width is obtained, and the weld slope and depth data are obtained from the point cloud. The weld volume and the required wire feed amount are calculated. Step S30: Based on the image of the weld pool area, the shape of the plasma cloud generated during the welding process is automatically identified using a pre-trained classification model. A mapping relationship between the shape of the plasma cloud and the wire feeding control adjustment command is established to intelligently optimize the wire feeding strategy of the laser welding wire feeder. Step S40: Establish a wire feeding speed matching model, calculate the change in wire feeding speed in real time, and generate control commands to adjust the parameters of the laser welding wire feeder in real time, combined with the optimized wire feeding strategy. Repeat steps S30 and S40 until the welding is completed.

[0007] Furthermore, before laser welding begins, a dedicated wire condition detection component detects the resistance between the welding wire and the workpiece. Based on the law of resistance, a specific mathematical model is used to calculate the distance between the welding wire and the workpiece. The preset distance is set to 0. When the calculated distance is greater than the preset distance, it indicates that the welding wire is not in contact with the workpiece, and a wire feeding operation is required. Based on the calculated distance L, combined with welding process requirements and experience, the initial wire feeding length of the wire feeder is set. The initial wire feeding speed is reasonably selected according to the material thickness, welding type, and welding requirements.

[0008] Furthermore, during the welding process, images of the weld pool area are acquired in real time using a camera. The acquisition frequency is determined based on the welding speed and process requirements. The acquired images are stored in a computer for image processing and analysis to adjust and optimize the welding wire feeder parameters and monitor the welding status in real time, thereby better optimizing the wire feeding strategy. A pre-trained weld recognition model is introduced. From a machine vision perspective, the powerful feature extraction capability of neural networks is used to learn the feature information of welds of different workpieces from the images. The model outputs a binary mask image of the weld pool area, highlighting the weld position of the workpiece, which facilitates the calculation of the weld width, slope, and depth, and obtains the weld volume and wire feeding amount.

[0009] Furthermore, the weld recognition model adopts a convolution-based neural network structure and applies Haar wavelet downsampling to replace pooling operations, preserving more weld edge contours to achieve accurate calculations. After downsampling, a scale-aware module is used to extract weld features at different scales, adapting to the diversity of different weld shapes on the workpiece to be welded, and an attention calibration mechanism is introduced to give higher weights to important weld features. Detailed steps for obtaining a pre-trained weld recognition model include: Collect images of different types of welds on specific workpieces, label them with pixel-level information, distinguish between weld and non-weld areas, and use bounding boxes to delineate the weld locations, labeling the coordinates of the four vertices of the bounding boxes. Combine the workpiece weld seam image data and the corresponding label information as the dataset for model training, and divide the dataset into training set, validation set and test set according to a set ratio; A weld seam recognition model was built and the model parameters were randomly initialized. The workpiece weld seam image was used as the input feature. The loss function was set as a weighted sum of classification loss and regression loss. The error between the model output feature and the label was calculated. The ADAM optimizer was selected, and the initial learning rate, training period and batch size were set. The loss function change curve was recorded during the training process, and the model parameters were updated by backpropagation through the calculation of the loss function gradient. Cross-validation is performed during training, and the model performance is evaluated using error rate, recall rate, and F1 score. When the loss function converges, the optimal model parameters are retained as the pre-trained weld recognition model.

[0010] Furthermore, to monitor welding quality in real time, a classification model based on a simple neural network structure is adopted to automatically identify the shape of the plasma cloud generated during welding. A real-time adjustment mechanism based on visual feedback is formulated, classifying the plasma cloud shape into three categories: slender, suspended, and spattered. The plasma cloud shape and corresponding wire feeding parameters under different welding conditions are collected to establish a mapping relationship between the plasma cloud shape and the wire feeding control adjustment command, and the wire feeding parameters are dynamically adjusted to ensure the stability of the welding process and the welding quality.

[0011] The present invention also provides an intelligent wire feeding control system for laser welding, comprising: Wire feeding parameter initialization module: used to measure the resistance value between the welding wire and the workpiece to be welded before welding begins, and then calculate the distance between them to determine the initial parameters of the laser welding wire feeder, and set the initial wire feeding length and initial wire feeding speed; Intelligent wire feeding calculation module: During the operation of the laser welding wire feeder with initial parameters, it collects images and point cloud data of the welding molten pool area, uses a pre-trained weld recognition model to locate the weld of the workpiece to be welded from the image, obtains the weld width, and obtains the weld slope and depth data from the point cloud to calculate the weld volume and the required wire feeding amount. Wire feeding strategy optimization module: Based on the image of the weld pool area, it uses a pre-trained classification model to automatically identify the shape of the plasma cloud generated during the welding process, establish a mapping relationship between the shape of the plasma cloud and the wire feeding control adjustment command, and intelligently optimize the wire feeding strategy of the laser welding wire feeder. Wire feeding control command execution module: used to establish a wire feeding speed matching model, calculate the change in wire feeding speed in real time, and generate control commands to adjust the parameters of the laser welding wire feeder in real time, combined with the optimized wire feeding strategy.

[0012] The present invention also proposes an intelligent wire feeding control device for laser welding. The device includes: a memory, a processor, and programs such as an intelligent wire feeding control algorithm for laser welding stored in the memory and executable on the processor. The intelligent wire feeding control algorithm for laser welding comprises the steps for implementing the intelligent wire feeding control method for laser welding as described above.

[0013] The present invention also provides a computer program product, which includes programs such as an intelligent wire feeding control algorithm for laser welding. When the intelligent wire feeding control algorithm for laser welding is executed by a processor, it implements an intelligent wire feeding control method for laser welding as described above.

[0014] The present invention discloses the following technical effects: This invention proposes an intelligent wire feeding control method and system for laser welding. Combining machine vision, mathematical models, and deep learning principles, it dynamically optimizes the wire feeding strategy and adjusts the wire feeding parameters of the laser welding wire feeder. To improve the automation and intelligence of the wire feeding process in laser welding, multiple deep learning models are embedded in the intelligent welding system to collect images and point cloud data of the weld pool area. A pre-trained weld recognition model automatically locates the weld position on the workpiece to be welded, retaining more accurate and comprehensive weld edge information for easy weld width calculation. Combined with image processing algorithms, the required wire feeding amount is accurately calculated based on the image and point cloud data, preventing excessive waste of welding wire material and saving production costs. Furthermore, the system utilizes a pre-trained classification model, requiring only lightweight model parameters to achieve intelligent classification of plasma cloud shapes. Through experience, a mapping relationship is established between the classification results and wire feeding parameters, automatically optimizing and adjusting the parameter configuration of the laser welding wire feeder without manual operation, saving time. This invention, based on visual perception, weld condition assessment, and hierarchical control, achieves dynamic optimization of the laser welding process, improves wire feeding efficiency, and ensures the quality stability of the welded product. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0016] Figure 1 This is a flowchart illustrating an intelligent wire feeding control method for laser welding, provided as an embodiment of this application.

[0017] Figure 2 A detailed structural diagram of the pre-trained weld recognition model provided in the embodiments of this application.

[0018] Figure label: R represents the dilation rate of dilated convolution.

[0019] Figure 3 This is a schematic diagram of an intelligent wire feeding control system for laser welding, provided as an embodiment of this application. Detailed Implementation

[0020] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the following description, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0023] Example 1: This application provides an intelligent wire feeding control method for laser welding, such as... Figure 1 As shown, the method includes: Step S10: Before welding begins, measure the resistance between the welding wire and the workpiece to be welded, then calculate the distance between them, determine the initial parameters of the laser welding wire feeder, and set the initial wire feeding length and initial wire feeding speed.

[0024] In this embodiment, before laser welding begins, a dedicated welding wire condition detection component detects the resistance information between the welding wire and the workpiece to be welded. Based on the law of resistance, the distance between the welding wire and the workpiece to be welded is calculated using a specific mathematical model, including the following steps: A resistance sampling circuit is built to measure the loop resistance and inversely determine the distance between the welding wire and the workpiece to be welded. The basic formula of the resistance law is expressed as:

[0025] in, It is the measured total resistance between the welding wire and the workpiece to be welded, in ohms; This indicates the length of the conductive path through which the current flows, i.e., the distance between the welding wire and the workpiece to be welded; It refers to the resistivity of the welding wire material, measured in ohms. rice; This indicates the cross-sectional area of ​​the welding wire, based on the wire diameter. The calculation yielded:

[0026] During welding, the temperature of the welding wire rises due to the current flowing through it and the nearby heat source. If the resistivity at room temperature is still used for calculation, the distance calculation will be inaccurate. The resistivity should be corrected according to the actual temperature of the welding wire.

[0027] in, This indicates the corrected resistivity of the welding wire. Represents resistivity at room temperature. It is the temperature coefficient of resistance. Indicates the measured real-time temperature; Substituting the corrected resistivity into the basic formula of the resistance law, the distance between the welding wire and the workpiece to be welded is derived and calculated using the following formula:

[0028] Therefore, by simply finding the resistivity based on the welding wire material, obtaining the welding wire diameter, and reading the resistance value between the welding wire and the workpiece to be welded using a digital multimeter, the distance between the welding wire and the workpiece to be welded can be obtained, solving the problems of low safety and insufficient accuracy caused by the reliance on manual measurement in traditional methods.

[0029] The preset distance value is set to 0. When the calculated distance value is greater than the preset distance, it indicates that the welding wire is not in contact with the workpiece, and a wire feeding operation needs to be performed. Based on the calculated distance between the welding wire and the workpiece to be welded, combined with welding process requirements and experience, the initial wire feeding length is set to be slightly greater than the calculated distance. This ensures that the welding wire can reach the weld smoothly and has sufficient allowance for welding during the welding process; the initial wire feeding speed should be reasonably selected according to the material thickness, welding type and welding requirements, and the wire feeding speed needs to match the welding speed.

[0030] In step S20, during the operation of the laser welding wire feeder with initial parameters, images and point cloud data of the welding molten pool area are collected. Using a pre-trained weld recognition model, the weld seam of the workpiece to be welded is located from the image, the weld seam width is obtained, and the weld seam slope and depth data are obtained from the point cloud. The weld seam volume and the required wire feed amount are calculated.

[0031] In this embodiment, an industrial CCD camera and a laser vision sensor are used to acquire two-dimensional images and three-dimensional point cloud data of the welding pool area, respectively. The coordinates of the image pixels and the three-dimensional point cloud are transformed to the world coordinate system through camera calibration parameters.

[0032] To obtain more accurate weld edge information of the workpiece to be welded, a pre-trained weld recognition model is used to segment the weld position image from the weld pool area image. The detailed steps for obtaining the pre-trained weld recognition model include: Collect images of different types of welds on specific workpieces, label them with pixel-level information, distinguish between weld and non-weld areas, and use bounding boxes to delineate the weld locations, labeling the coordinates of the four vertices of the bounding boxes. Combine the workpiece weld seam image data and the corresponding label information as the dataset for model training, and divide the dataset into training set, validation set and test set according to a set ratio; A weld seam recognition model was built and the model parameters were randomly initialized. The workpiece weld seam image was used as the input feature. The loss function was set as a weighted sum of classification loss and regression loss. The error between the model output feature and the label was calculated. The ADAM optimizer was selected, and the initial learning rate, training period and batch size were set. The loss function change curve was recorded during the training process, and the model parameters were updated by backpropagation through the calculation of the loss function gradient. Cross-validation is performed during training, and the model performance is evaluated using error rate, recall rate, and F1 score. When the loss function converges, the optimal model parameters are retained as the pre-trained weld recognition model.

[0033] The pre-trained weld recognition model outputs a binary mask image of the weld pool area, with the pixel value of the weld position being 0 and the pixel value of the non-weld position being 1. The binary image is further cropped to retain the weld position in the image, the left and right edges of the weld are located, the Euclidean distance between them is calculated, and point-by-point and segment-by-segment calculations are performed along the length of the weld to obtain the width variation curve. For V-groove and U-groove welds, the groove slope is fitted by a straight line on the cross section of the three-dimensional point cloud data, and the angle between the surfaces is calculated to obtain the weld slope data; and the vertical distance between the bottom point of the groove and the reference plane of the workpiece surface is calculated, or the weld depth data is obtained directly through the z-axis coordinate difference. For V-groove welds, the weld cross-section is simplified to a regular geometric shape, considered as an isosceles trapezoid, and the cross-sectional area of ​​the weld is calculated. :

[0034] in, and These represent the width of the upper surface of the weld groove and the width of the weld root (which may be 0), respectively. Indicates the depth of the V-groove weld bevel; For U-shaped weld bevels, the weld cross-sectional area It is calculated using the following formula:

[0035] in, Indicates the thickness of the workpiece to be welded. This indicates the root radius of the U-shaped weld bevel. Indicates the weld slope. Indicates the depth of the U-shaped weld groove; Calculate the cross-sectional area along the length of the weld at fixed intervals; the weld volume. By integrating the cross-sectional area of ​​all segments along the weld path, we obtain:

[0036] in, Indicates the weld length. Indicates position The cross-sectional area of ​​the V-groove or U-groove weld. The number of segments in the weld length. Indicates the first The cross-sectional area of ​​the V-groove or U-groove weld segment. Indicates the first The segment lengths are determined; further calculations are performed to obtain the total wire feed required for the workpiece to be welded. :

[0037] in, Indicates the diameter of the welding wire. This is the deposition loss coefficient; the total wire feed is used to adjust the wire feed speed and other wire feed parameters of the laser welding wire feeder.

[0038] Step S30: Based on the image of the weld pool area, a pre-trained classification model is used to automatically identify the shape of the plasma cloud generated during the welding process, establish a mapping relationship between the shape of the plasma cloud and the wire feeding control adjustment command, and intelligently optimize the wire feeding strategy of the laser welding wire feeder.

[0039] In this embodiment, to monitor welding quality in real time, a classification model based on a simple neural network structure is used to automatically identify the shape of the plasma cloud generated during welding. A real-time adjustment mechanism based on visual feedback is established to classify the plasma cloud shape into three categories: elongated, suspended, and spattered. A classification model was trained using a large amount of plasma cloud shape image data. The model consisted of three convolutional layers and one multilayer perceptron. The convolutional operation extracted feature information from the image data and gradually reduced the feature size while increasing the number of feature channels. The multilayer perceptron, as a linear layer, mapped the feature dimension to the number of plasma cloud shape categories, obtaining the probability distribution matrix of the categories. This matrix was used to calculate the cross-entropy loss with the label data and update the classification model parameters. The best-performing model parameters were used as the pre-trained classification model and embedded into the intelligent welding system to process images of the weld pool area in real time and output the classification results of the plasma cloud shape.

[0040] Collect plasma cloud shapes and corresponding wire feeding parameters under different welding conditions, establish a mapping relationship between plasma cloud shapes and wire feeding control adjustment commands, and dynamically adjust wire feeding parameters to ensure the stability of the welding process and welding quality. When the plasma cloud shape is identified as elongated, the welding is determined to be stable, and the current wire feeding parameters are maintained. When the plasma cloud shape is identified as suspended, it is determined that the welding wire may not be effectively contacting the molten pool. The instruction is to increase the wire feed speed by 5% and the wire feed length by 5% to make the welding wire re-contact the workpiece. When the plasma cloud shape is identified as spatter, it is determined that there may be severe spatter leading to insufficient wire filling. The instruction is to increase the wire feeding speed by 10% and the wire feeding length by 10% for rapid compensation; and to reduce the wire feeding motor current and torque to reduce the wire feeding thrust and reduce the degree of spatter.

[0041] Step S40: Establish a wire feeding speed matching model, calculate the change in wire feeding speed in real time, and generate control commands to adjust the parameters of the laser welding wire feeder in real time, combined with the optimized wire feeding strategy. Repeat steps S30 and S40 until the welding is completed.

[0042] In this example, the wire feed speed matching model describes the real-time welding speed. welding wire diameter workpiece gap Workpiece thickness The mathematical relationship between the wire feed rate and the wire feed speed, based on the mathematical relationship expressed by this model, is the change in wire feed speed. The calculation formula is expressed as:

[0043] in, For correction factor, The original wire feeding speed is used; based on the wire feeding strategy generated in step S30 and the calculated change in wire feeding speed, the parameters of the laser welding wire feeder are adjusted.

[0044] Example 2: This invention application provides a detailed structure of a weld seam recognition model for identifying the weld seam position of a workpiece to be welded, as follows: Figure 2 As shown: The weld seam recognition model uses the workpiece weld seam image as input features. First, a shallow feature extraction module obtains shallow features, reducing noise interference and mapping the image data to a high-dimensional feature space for easier convolution processing. Then, the shallow features are fed into a Haar wavelet downsampling module to reduce the spatial resolution of the feature mapping. This module replaces pooling, preserving more weld seam edge contours. Haar wavelet transform is applied, and through low-pass and high-pass decomposition filters, each channel feature in the shallow features is mapped into four components, each with a resolution half that of the original channel features. The wavelet basis functions of the Haar wavelet transform are... and scaling function Represented as:

[0045] in, This represents the feature value at each position in the channel features. , Represents the step function. and This indicates the scale and orientation of the image data contained in the channel features; To enrich the feature information, a scale-aware module is used to extract multi-scale information from the downsampled features to adapt to different weld shapes of different types of workpieces to be welded, thereby improving the accuracy of the model in locating welds. The downsampled features are processed in parallel by the residual mapping layer, the scale-aware module, and the global average pooling layer to obtain comprehensive weld feature information from local to global perspectives. The residual mapping layer uses a kernel size of The standard convolutional method captures local detail features. The scale-aware module contains three parallel branches with dilation rates of 1, 3, and 5, and a kernel size of [missing information]. The dilated convolutions are used to expand the receptive field and capture contextual information at different scales. After each dilated convolution, an attention calibration module is applied to calculate channel attention to suppress redundant features and enhance effective information. The features output by the three branches are summed element-wise. A pixel-level attention map is generated by convolutional layers and Softmax. The attention map is used to perform weighted fusion of the multi-scale features output by the three branches. A global average pooling layer is used to obtain global contextual information.

[0046] The fusion features of the scale-aware module, the output features of the standard convolutional branch, and the global pooling features are concatenated along the channel dimension to supplement local details and global context information. Finally, a feature map that integrates multi-scale and multi-view information is output. This feature map is then mapped through a fully connected layer and, guided by the label information, a binarized mask image of the workpiece weld is generated.

[0047] Example 3: The intelligent wire feeding control system for laser welding provided in this embodiment of the invention can execute the intelligent wire feeding control method for laser welding provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method, such as... Figure 3 As shown, it includes the following modules: Wire feeding parameter initialization module: used to measure the resistance value between the welding wire and the workpiece to be welded before welding begins, and then calculate the distance between them to determine the initial parameters of the laser welding wire feeder, and set the initial wire feeding length and initial wire feeding speed; Intelligent wire feeding calculation module: During the operation of the laser welding wire feeder with initial parameters, it collects images and point cloud data of the welding molten pool area, uses a pre-trained weld recognition model to locate the weld of the workpiece to be welded from the image, obtains the weld width, and obtains the weld slope and depth data from the point cloud to calculate the weld volume and the required wire feeding amount. Wire feeding strategy optimization module: Based on the image of the weld pool area, it uses a pre-trained classification model to automatically identify the shape of the plasma cloud generated during the welding process, establish a mapping relationship between the shape of the plasma cloud and the wire feeding control adjustment command, and intelligently optimize the wire feeding strategy of the laser welding wire feeder. Wire feeding control command execution module: used to establish a wire feeding speed matching model, calculate the change in wire feeding speed in real time, and generate control commands to adjust the parameters of the laser welding wire feeder in real time, combined with the optimized wire feeding strategy.

[0048] Example 4: This application provides an intelligent wire feeding control device for laser welding. The intelligent wire feeding control device for laser welding includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute an intelligent wire feeding control method for laser welding as described in the above embodiment.

[0049] Example 5: This application provides a computer program product including a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication system, or installed from a storage system. When the computer program is executed by a processing system, it performs the functions defined in the methods of Examples 1 and 2 of this application.

[0050] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A smart wire feeding control method for laser welding, characterized in that, The method includes: Step S10: Before welding begins, measure the resistance between the welding wire and the workpiece to be welded, then calculate the distance between them, determine the initial parameters of the laser welding wire feeder, and set the initial wire feeding length and initial wire feeding speed. In step S20, during the operation of the laser welding wire feeder with initial parameters, images and point cloud data of the welding molten pool area are collected. Using a pre-trained weld recognition model, the weld of the workpiece to be welded is located from the image, the weld width is obtained, and the weld slope and depth data are obtained from the point cloud. The weld volume and the required wire feed amount are calculated. Step S30: Based on the image of the weld pool area, the shape of the plasma cloud generated during the welding process is automatically identified using a pre-trained classification model. A mapping relationship between the shape of the plasma cloud and the wire feeding control adjustment command is established to intelligently optimize the wire feeding strategy of the laser welding wire feeder. Step S40: Establish a wire feeding speed matching model, calculate the change in wire feeding speed in real time, and generate control commands to adjust the parameters of the laser welding wire feeder in real time, combined with the optimized wire feeding strategy. Repeat steps S30 and S40 until the welding is completed.

2. The intelligent wire feeding control method for laser welding as described in claim 1, characterized in that, In step S10, before laser welding begins, a dedicated welding wire condition detection component detects the resistance between the welding wire and the workpiece to be welded. Based on the law of resistance, the distance between the welding wire and the workpiece to be welded is calculated using a specific mathematical model. in, This indicates the length of the conductive path through which the current flows, i.e., the distance between the welding wire and the workpiece to be welded; It is the measured total resistance between the welding wire and the workpiece to be welded; This indicates the cross-sectional area of ​​the welding wire, based on the wire diameter. Calculated; The resistivity of the welding wire material after temperature correction is calculated using the following formula: in, This indicates the resistivity of a specific type of welding wire at room temperature, obtained from the query. It is the temperature coefficient of resistance. This indicates the measured real-time temperature.

3. The intelligent wire feeding control method for laser welding as described in claim 1, characterized in that, Step S20, the detailed steps for obtaining the pre-trained weld recognition model, include: Collect images of different types of welds on specific workpieces, label them with pixel-level information, distinguish between weld and non-weld areas, and use bounding boxes to delineate the weld locations, labeling the coordinates of the four vertices of the bounding boxes. Combine the workpiece weld seam image data and the corresponding label information as the dataset for model training, and divide the dataset into training set, validation set and test set according to a set ratio; A weld seam recognition model was built and the model parameters were randomly initialized. The workpiece weld seam image was used as the input feature. The loss function was set as a weighted sum of classification loss and regression loss. The error between the model output feature and the label was calculated. The ADAM optimizer was selected, and the initial learning rate, training period and batch size were set. The loss function change curve was recorded during the training process, and the model parameters were updated by backpropagation through the calculation of the loss function gradient. Cross-validation is performed during training, and the model performance is evaluated using error rate, recall rate, and F1 score. When the loss function converges, the optimal model parameters are retained as the pre-trained weld recognition model.

4. The intelligent wire feeding control method for laser welding as described in claim 3, characterized in that, The weld recognition model takes the workpiece weld image as input feature. First, it obtains shallow features through a shallow feature extraction module to reduce noise interference in the features and maps the image data to a high-dimensional feature space. Then, the shallow features are fed into the Haar wavelet downsampling module to reduce the spatial resolution of the feature mapping. This module replaces the pooling operation, retaining more weld edge contours. The Haar wavelet transform is applied, and each channel feature in the shallow features is mapped into four components through low-pass and high-pass decomposition filters. The resolution of each component is half that of the original channel feature. The downsampled features are then processed in parallel by the residual mapping layer, the scale-aware module, and the global average pooling layer to obtain comprehensive weld feature information from local to global.

5. The intelligent wire feeding control method for laser welding as described in claim 4, characterized in that, The scale-aware module comprises three parallel branches, employing expansion rates of 1, 3, and 5 respectively, with a kernel size of [missing value]. The dilated convolutions are used to expand the receptive field and capture contextual information at different scales. An attention calibration module is applied after each dilated convolution to calculate channel attention to suppress redundant features and enhance effective information. The features output by the three branches are summed element-wise. A pixel-level attention map is generated by convolutional layers and Softmax. The attention map is then used to perform weighted fusion of the multi-scale features output by the three branches.

6. The intelligent wire feeding control method for laser welding as described in claim 1, characterized in that, In step S20, the pre-trained weld recognition model outputs a binarized mask image of the weld pool area image, with the pixel value of the weld position being 0 and the pixel value of the non-weld position being 1. The binarized image is further cropped to retain the weld position in the image, the left and right edges of the weld are located, the Euclidean distance between them is calculated, and point-by-point and segmented calculations are performed along the length of the weld to obtain the width variation curve. For V-shaped and U-shaped groove welds, the groove slope is obtained by fitting a straight line on the cross section of the three-dimensional point cloud data, calculating the angle between the surfaces, and obtaining the weld slope data; and the vertical distance between the bottom point of the groove and the reference plane of the workpiece surface is calculated, or the weld depth data is obtained directly through the z-axis coordinate difference.

7. The intelligent wire feeding control method for laser welding as described in claim 6, characterized in that, The width variation curve, weld slope data, and weld depth data are used to calculate the weld volume and wire feed amount. For V-groove welds, the weld cross-section is simplified to a regular geometric shape, considered as an isosceles trapezoid, and the cross-sectional area of ​​the weld is calculated. : in, and These represent the width of the upper surface of the weld groove and the width of the weld root, respectively. Indicates the depth of the V-groove weld bevel; For U-shaped weld bevels, the weld cross-sectional area It is calculated using the following formula: in, Indicates the thickness of the workpiece to be welded. This indicates the root radius of the U-shaped weld bevel. Indicates the weld slope. Indicates the depth of the U-shaped weld groove; Calculate the cross-sectional area along the length of the weld at fixed intervals; the weld volume. By integrating the cross-sectional area of ​​all segments along the weld path, we obtain: in, Indicates the weld length. Indicates position The cross-sectional area of ​​the V-groove or U-groove weld; further calculations are performed to obtain the total wire feed required to complete the welding of the workpiece. : in, Indicates the diameter of the welding wire. This is the deposition loss coefficient; the total wire feed is used to adjust the wire feed speed and other wire feed parameters of the laser welding wire feeder.

8. The intelligent wire feeding control method for laser welding as described in claim 1, characterized in that, In step S30, a classification model is trained using a large amount of plasma cloud shape image data. The model consists of three convolutional layers and one multilayer perceptron. The convolutional operation extracts feature information from the image data and gradually reduces the feature size while increasing the number of feature channels. The multilayer perceptron, as a linear layer, maps the feature dimension to the number of plasma cloud shape categories, obtaining the probability distribution matrix of the categories. This matrix is ​​used to calculate the cross-entropy loss with the label data, update the classification model parameters, and use the best-performing model parameters as the pre-trained classification model. This model is then embedded into the intelligent welding system to process images of the weld pool area in real time and output the classification results of the plasma cloud shape.

9. The intelligent wire feeding control method for laser welding as described in claim 1, characterized in that, In step S40, the wire feed speed matching model describes the real-time welding speed. welding wire diameter workpiece gap Workpiece thickness The mathematical relationship between the wire feed rate and the wire feed speed, based on the mathematical relationship expressed by this model, is the change in wire feed speed. The calculation formula is expressed as: in, For correction factor, The original wire feeding speed is used; based on the wire feeding strategy generated in step S30 and the calculated change in wire feeding speed, the configuration parameters of the laser welding wire feeder are adjusted in real time.

10. An intelligent wire feeding control system for laser welding, characterized in that, The system is used to implement the intelligent wire feeding control method for laser welding according to any one of claims 1-9, the system comprising: Wire feeding parameter initialization module: used to measure the resistance value between the welding wire and the workpiece to be welded before welding begins, and then calculate the distance between them to determine the initial parameters of the laser welding wire feeder, and set the initial wire feeding length and initial wire feeding speed; Intelligent wire feeding calculation module: During the operation of the laser welding wire feeder with initial parameters, it collects images and point cloud data of the welding molten pool area, uses a pre-trained weld recognition model to locate the weld of the workpiece to be welded from the image, obtains the weld width, and obtains the weld slope and depth data from the point cloud to calculate the weld volume and the required wire feeding amount. Wire feeding strategy optimization module: Based on the image of the weld pool area, it uses a pre-trained classification model to automatically identify the shape of the plasma cloud generated during the welding process, establish a mapping relationship between the shape of the plasma cloud and the wire feeding control adjustment command, and intelligently optimize the wire feeding strategy of the laser welding wire feeder. Wire feeding control command execution module: used to establish a wire feeding speed matching model, calculate the change in wire feeding speed in real time, and generate control commands to adjust the parameters of the laser welding wire feeder in real time, combined with the optimized wire feeding strategy.

Images