High-temperature alloy laser welding process parameter prediction method, terminal device and medium

Abstract

This invention provides a method, terminal equipment, and medium for predicting process parameters in high-temperature alloy laser welding. The method includes: acquiring the workpiece properties of the alloy workpiece to be welded; acquiring a real-time image of the weld pool and extracting a time series sequence of the weld pool aspect ratio from the image; inputting the workpiece properties and the time series sequence of the weld pool aspect ratio into a pre-trained process parameter prediction model, which outputs predicted values ​​of the welding process parameters; and welding the alloy workpiece according to the predicted values ​​of the welding process parameters. This invention solves the problems of inaccurate setting of process parameters and inability to adaptively adjust them in real time in existing high-temperature alloy laser welding technologies, thereby improving welding quality.

Description

Technical Field

[0001] This invention belongs to the field of laser welding technology, specifically relating to a method for predicting process parameters of high-temperature alloy laser welding, terminal equipment, and medium. Background Technology

[0002] Laser welding technology is widely used in aerospace, automotive manufacturing, and electronic equipment industries, especially for welding high-temperature alloy workpieces. The quality of high-temperature alloy laser welding is highly dependent on the precise control of process parameters (such as welding power, spot diameter, welding speed, and defocusing amount). In existing technologies, process parameters are usually set based on experience or fixed rules, which are difficult to adapt to dynamically changing welding environments (such as fluctuations in material properties and heat accumulation effects), leading to unstable welding quality and problems such as over-melting, under-melting, or poor weld formation.

[0003] Furthermore, traditional methods lack the ability to monitor and adjust the weld pool state in real time, and cannot adaptively correct process parameters during welding. Therefore, there is an urgent need for a method that can predict and optimize process parameters in real time to improve welding quality and efficiency. Summary of the Invention

[0004] This invention provides a method, terminal equipment, and medium for predicting process parameters in high-temperature alloy laser welding, aiming to solve the problems of inaccurate setting of process parameters and inability to make real-time adaptive adjustments in existing technologies, thereby improving welding quality.

[0005] In a first aspect, the present invention provides a method for predicting process parameters of high-temperature alloy laser welding, the method comprising the following steps:

[0006] Obtain the workpiece properties of the alloy workpiece to be welded; workpiece properties include high-temperature alloy type and plate thickness;

[0007] Real-time acquisition of weld pool images, and extraction of the weld pool aspect ratio time sequence from the weld pool images; the weld pool aspect ratio characterizes the state of the weld pool, which is stable, over-melted, or under-melted.

[0008] The workpiece attributes and the time series sequence of the weld pool aspect ratio are input into a pre-trained process parameter prediction model, which outputs predicted welding process parameters. The process parameter prediction model includes a first module, a second module, and a third module. The first module is used to perform high-precision interpolation prediction of the workpiece attributes based on a prior database to obtain the initial welding process parameters corresponding to the workpiece attributes. The second module is used to perform feature engineering on the time series sequence of the weld pool aspect ratio to obtain the time series features of the weld pool to characterize the trend of weld pool state changes. The third module is used to correct the initial welding process parameters based on the time series features of the weld pool to obtain the predicted process parameter values. The prior database includes a set of empirically validated optimal process parameters for the alloy workpiece to be welded under gradient plate thickness. The welding process parameters include welding power, spot diameter, welding speed, and defocusing amount.

[0009] Welding is performed on the alloy workpiece to be welded based on the predicted values ​​of the welding process parameters.

[0010] Optionally, high-precision interpolation prediction of workpiece attributes is performed based on a prior database to obtain initial welding process parameters corresponding to the workpiece attributes, including:

[0011] Through calculation formula

[0012]

[0013] Obtain initial welding process parameters ;in, Indicates the initial welding power. Indicates the initial spot diameter. Indicates the initial welding speed. Indicates the initial defocus amount. This indicates a three-layer fully connected static regression head. Indicates splicing, This indicates the category of high-temperature alloys after integer encoding. This represents the normalized thickness of the sheet metal. Represents the prior database medium and high temperature alloy categories Empirically proven optimal set of process parameters for gradient plate thicknesses This represents all parameters of the static baseline parameter regression MLP, which are learnable parameters.

[0014] Optionally, feature engineering is performed on the aspect ratio time series of the weld pool to obtain time series features of the weld pool used to characterize the trend of weld pool state changes, including:

[0015] Through calculation formula The timing characteristics of the molten pool are obtained. ;in, This represents a three-layer one-dimensional residual network. This represents the time series sequence indicating the aspect ratio of the molten pool. , Indicates time The aspect ratio of the molten pool , Indicates time The length of the molten pool, Indicates time The width of the molten pool, Indicates the length of the timing window.

[0016] Optionally, the initial welding process parameters are corrected based on the molten pool timing characteristics to obtain predicted process parameter values, including:

[0017] High-temperature alloy categories after integer encoding and normalized plate thickness By performing feature stitching and nonlinear transformation, the theoretically optimal dynamic trajectory of the molten pool aspect ratio corresponding to the workpiece attributes is obtained. ;

[0018] By linearly projecting the temporal characteristics of the weld pool, the instantaneous state characteristics of the weld pool at each moment are obtained. and the deviation vector used to correct welding process parameters at that moment. ;

[0019] According to the theoretical optimal dynamic trajectory Instantaneous state characteristics and deviation vector Calculate the context offset representation vector used to characterize the anomaly in the molten pool state. ;

[0020] The theoretical optimal dynamic trajectory and context offset representation vector Input the multilayer perceptron network with a shared backbone to obtain the correction values ​​of welding process parameters;

[0021] The initial welding process parameters are corrected based on the correction values ​​of the welding process parameters to obtain the predicted values ​​of the process parameters.

[0022] Optional, theoretically optimal dynamic trajectory The expression is:

[0023]

[0024] in, Representation layer normalization, This represents the learnable parameters.

[0025] Optional, instantaneous state features The expression is Deviation vector The expression is ;in, and This represents two different projection matrices.

[0026] Optionally, based on the theoretically optimal dynamic trajectory Instantaneous state characteristics and deviation vector Calculate the context offset representation vector used to characterize the anomaly in the molten pool state. ,include:

[0027] Through calculation formula

[0028]

[0029] Obtain the attention weight vector ; where attention weight vector Used to pinpoint the moment when the molten pool condition is abnormal. Indicates the hidden feature dimension. This represents a learnable positional offset mask;

[0030] Through calculation formula

[0031]

[0032] Obtain the context offset representation vector .

[0033] Optionally, the theoretically optimal dynamic trajectory and context offset representation vector Input a multilayer perceptron network with a shared backbone to obtain welding process parameter correction values, including:

[0034] Through calculation formula

[0035]

[0036]

[0037]

[0038] Obtain welding process parameter correction values ;in, This indicates a welding power correction head, used to convert deviations in the molten pool condition into power compensation. This indicates a spot diameter correction head, used to convert molten pool state deviations into spot diameter compensation. This indicates a welding speed correction head, used to convert deviations in the molten pool state into welding speed compensation. This indicates a defocusing correction head, used to convert molten pool state deviations into defocusing position compensation. This represents the final linear output weight matrix for the four welding heads, used to determine the relative correction ratio and direction of the four parameters—welding power, spot diameter, welding speed, and defocusing amount—to the same molten pool state deviation. Represents a smooth nonlinear activation function. This represents the second-layer shared backbone weight matrix, used to capture nonlinear physical phenomena caused by abnormal molten pool conditions. These nonlinear physical phenomena include the secondary thermal accumulation effect. This represents the first-layer shared backbone weight matrix, used to nonlinearly map the physical space of molten pool state deviations to the process parameter sensitive space. This represents the first-level bias vector, used to compensate for systematic global offsets. This represents the second-level bias vector, used to correct systematic mapping biases in the shared backbone. This represents the final bias vector of the four heads, used to compensate for the zero-point mechanical error and response lag of each actuator, which includes a laser, an optical head, a motion platform, and a defocusing motor.

[0039] In a second aspect, the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method.

[0040] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0041] The present invention has at least the following beneficial effects:

[0042] By acquiring and analyzing molten pool images in real time, the aspect ratio of the molten pool, which characterizes the welding state, is used as a dynamic feedback signal. This allows process parameters to be adaptively adjusted according to the actual welding process, effectively improving the stability of welding quality. The process parameter prediction model first provides optimized initial parameters for specific workpieces based on a knowledge base, and then performs precise fine-tuning based on the dynamic characteristics of the molten pool, ensuring high accuracy and robustness of the prediction. By comprehensively considering static workpiece attributes and dynamic process characteristics, multiple key parameters such as welding power and speed are predicted and optimized collaboratively, realizing intelligent multi-dimensional decision-making covering the entire welding process. This reduces reliance on human experience and is conducive to improving welding quality. Attached Figure Description

[0043] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0044] Figure 1 This is a flowchart of a method for predicting process parameters of high-temperature alloy laser welding in one embodiment of this application;

[0045] Figure 2 This is a schematic diagram of the structure of a process parameter prediction model in one embodiment of this application;

[0046] Figure 3 This is a molten pool morphology diagram of an alloy workpiece to be welded when welding according to the predicted values ​​of welding process parameters in one embodiment of this application.

[0047] Figure 4 This is a schematic diagram of the structure of a terminal device in one embodiment of this application. Detailed Implementation

[0048] The technical solution of the present invention will now be described in detail and completely with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0049] In the description of this invention, it should be noted that the terms "upper", "lower", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0050] To address the problems of inaccurate setting and inability to adaptively adjust process parameters in existing high-temperature alloy laser welding technologies, this invention provides a method for predicting process parameters in high-temperature alloy laser welding. This method acquires and analyzes molten pool images in real time, using the aspect ratio of the molten pool, which characterizes the welding state, as a dynamic feedback signal. This allows the process parameters to adaptively adjust according to the actual welding process, effectively improving the stability of welding quality. The process parameter prediction model first provides optimized initial parameters for a specific workpiece based on a knowledge base, and then performs precise fine-tuning based on the dynamic characteristics of the molten pool, ensuring high accuracy and robustness of the prediction. By comprehensively considering both static workpiece attributes and dynamic process characteristics, it collaboratively predicts and optimizes multiple key parameters such as welding power and speed, achieving intelligent multi-dimensional decision-making covering the entire welding process. This reduces reliance on human experience and is beneficial for improving welding quality.

[0051] It should be noted that in this embodiment of the invention, 316L (rolled state) high-temperature alloy is used as the alloy workpiece to be welded, the standard plate thickness is 2.0mm, and the joint type is a gapless I-type butt weld. The implementation method of the invention is described in detail below. 316L high-temperature alloy is widely used in hot-end components of aerospace engines due to its excellent high-temperature strength and creep resistance. Its weldability mainly faces challenges due to its sensitivity to heat input, making it prone to welding deformation and hot cracking.

[0052] like Figure 1 As shown, the method for predicting process parameters of high-temperature alloy laser welding provided by the present invention includes steps 11 to 14.

[0053] Step 11: Obtain the workpiece properties of the alloy workpiece to be welded.

[0054] In this embodiment of the invention, the workpiece properties include the high-temperature alloy category (316L high-temperature alloy) and the plate thickness (2.0 mm).

[0055] To facilitate subsequent processing, the high-temperature alloy categories are encoded using integers to obtain the high-temperature alloy category codes. At the same time, the thickness of the sheet material is normalized.

[0056] In one feasible implementation, through calculation formula The normalized plate thickness is obtained. , This indicates the thickness of the plate material of the alloy workpiece to be welded. Indicates the thinnest thickness of the board. This indicates the maximum thickness of the sheet material; the applicable scope defined in this invention is... , (To accommodate thicker structural components).

[0057] In practice, 99.99% high-purity argon is used as the protective gas to effectively prevent the oxidation of active elements such as chromium in the 316L alloy at high temperatures. The gas flow rate is precisely set to 30 L / min using a mass flow controller; this relatively high flow rate ensures sufficient coverage of the molten pool and heat-affected zone, forming an effective inert gas protective layer. Before welding, the bevel of the workpiece and a 20mm area on both sides are mechanically ground and cleaned with acetone to thoroughly remove oxide film and oil.

[0058] Step 12: Acquire images of the weld pool in real time and extract the aspect ratio time sequence of the weld pool from the images.

[0059] In this embodiment of the invention, the aspect ratio of the weld pool characterizes the state of the weld pool, which is stable, over-melted, or under-melted.

[0060] In one feasible implementation, a high-speed CMOS camera (resolution 1024×768, frame rate 2000fps) equipped with a near-infrared filter is used, positioned at approximately 30 degrees to the laser beam axis. From the angle of view, through the observation window of the protective gas nozzle, dynamic images of the weld pool are acquired in real time.

[0061] In practice, for each frame of the acquired image:

[0062] First, Gaussian filtering is used for noise reduction. Then, the image is binarized using the maximum inter-class variance method. Finally, the contour of the melt pool is extracted using a boundary tracking algorithm (such as the Canny algorithm).

[0063] Subsequently, based on the contour of the molten pool, the current time is calculated. Molten pool feature dimensions: molten pool length (Maximum dimension along the welding direction) and weld pool width (Maximum dimension perpendicular to the welding direction).

[0064] Finally, through the calculation formula Get the current time The aspect ratio of the molten pool The aspect ratio of the weld pool is a key indicator of welding stability. A weld pool aspect ratio that is too small (e.g., less than 1.2) may indicate incomplete penetration or insufficient heat input (under-melting); a weld pool aspect ratio that is too large (e.g., greater than 2.8) may indicate excessive heat input, impending over-melting or collapse.

[0065] In one feasible implementation, the timing window length is set. This means using data from the most recent 60 moments (30ms in total) to capture the dynamic trend of the molten pool. Therefore, the current moment... The aspect ratio time series of the molten pool for: .

[0066] Step 13: Input the workpiece attributes and the time sequence of the aspect ratio of the molten pool into the pre-trained process parameter prediction model, and the process parameter prediction model outputs the predicted values ​​of welding process parameters.

[0067] like Figure 2 As shown, in this embodiment of the invention, the process parameter prediction model includes a first module 201, a second module 202, and a third module 203. The first module 201 is used to perform high-precision interpolation prediction of workpiece attributes based on a prior database to obtain the initial welding process parameters corresponding to the workpiece attributes. The second module 202 is used to perform feature engineering on the aspect ratio time series of the weld pool to obtain weld pool time series features characterizing the trend of weld pool state changes. The third module 203 is used to correct the initial welding process parameters based on the weld pool time series features to obtain predicted process parameter values.

[0068] It should be noted that, in the embodiments of the present invention, the welding process parameters include welding power, spot diameter, welding speed, and defocusing amount.

[0069] The first module will be explained in detail below.

[0070] In one feasible implementation, the prior database includes a set of empirically optimal process parameters for the alloy workpiece to be welded at varying plate thicknesses. For example, the set of empirically optimal process parameters for 316L high-temperature alloy in the prior database is shown in Table 1.

[0071] Table 1

[0072]

[0073] The process of obtaining the initial welding process parameters corresponding to the workpiece attributes by performing high-precision interpolation prediction of workpiece attributes based on a prior database includes:

[0074] Through calculation formula

[0075]

[0076] Obtain initial welding process parameters ;in, Indicates the initial welding power. Indicates the initial spot diameter. Indicates the initial welding speed. This indicates the initial defocus amount. This represents a three-layer fully connected static regression head with input dimension . and The concatenation dimension is 4, and the output dimension is 4 (i.e., ), Indicates splicing, This indicates the type of high-temperature alloy after integer encoding (e.g., 316L (rolled state) integer code). ), This represents the normalized thickness of the sheet metal. Represents the prior database medium and high temperature alloy categories The empirically optimal set of process parameters for gradient plate thicknesses (see Table 1). This represents all parameters of the static baseline parameter regression MLP, which are learnable parameters obtained through training with historical data.

[0077] For example, in one feasible implementation, the initial welding process parameters are: , , , .

[0078] The second module will be explained in detail below.

[0079] Specifically, the process of performing feature engineering on the aspect ratio time series of the weld pool to obtain the time series features of the weld pool used to characterize the trend of weld pool state changes includes:

[0080] Through calculation formula The timing characteristics of the molten pool are obtained. ;in, This represents a three-layer one-dimensional residual network with an input dimension of 50 and an output dimension of 128, which can effectively capture long-term dependencies and local mutation patterns in time series. (Molten pool time series features) With a value of 128, this vector encapsulates the trend of the molten pool state over the past 30ms (e.g., whether it is stabilizing, rising rapidly, or oscillating). This represents the network parameters, obtained through pre-training.

[0081] The third module will be explained in detail below.

[0082] Specifically, the process of correcting the initial welding process parameters based on the molten pool timing characteristics to obtain the predicted values ​​of the process parameters includes steps I to V.

[0083] Step I, through the high-temperature alloy category after integer encoding and normalized plate thickness By performing feature stitching and nonlinear transformation, the theoretically optimal dynamic trajectory of the molten pool aspect ratio corresponding to the workpiece attributes is obtained. .

[0084] Specifically, the theoretically optimal dynamic trajectory The expression is:

[0085]

[0086] in, Representation layer normalization, Indicates learnable parameters, .

[0087] It should be noted that the optimal dynamic trajectory An embedded characterization of the dynamic trajectory of the molten pool aspect ratio that should be exhibited in an "ideal" welding process under the current high-temperature alloy type and plate thickness.

[0088] Step II: By linearly projecting the temporal characteristics of the weld pool, the instantaneous state characteristics of the weld pool at each moment are obtained. and the deviation vector used to correct welding process parameters at that moment. .

[0089] Specifically, instantaneous state characteristics The expression is ;

[0090] Deviation vector The expression is ;in, and This represents two different projection matrices.

[0091] Instantaneous state characteristics Used with optimal dynamic trajectory Comparison, deviation vector It contains information on the specific process parameter adjustments required to address the current state of the molten pool.

[0092] Step III, based on the theoretical optimal dynamic trajectory Instantaneous state characteristics and deviation vector Calculate the context offset representation vector used to characterize the anomaly in the molten pool state. .

[0093] Specifically, this includes steps A through B.

[0094] Step A, through calculation formula

[0095]

[0096] Obtain the attention weight vector Among them, the attention weight vector Used to pinpoint the moment when the molten pool condition is abnormal. Indicates the hidden feature dimension. This represents a learnable positional offset mask.

[0097] It should be noted that, in this embodiment of the invention, an attention mechanism is employed, through calculation... and The similarity is used to obtain the attention weight. Attention weights It is a probability distribution, and its peak corresponds to the abnormal moment in the molten pool state sequence with the largest deviation from the optimal dynamic trajectory (for example, in this embodiment, two instantaneous overheating caused by minor inhomogeneities of the underlying material were detected at two different times).

[0098] Step B, through calculation formula

[0099]

[0100] Obtain the context offset representation vector Context offset representation vector It is the weighted average deviation vector, which reflects the comprehensive correction instructions needed to correct these abnormal states.

[0101] Step IV, apply the theoretically optimal dynamic trajectory and context offset representation vector Input a multilayer perceptron network with a shared backbone to obtain the correction values ​​of welding process parameters.

[0102] Specifically, through calculation formula

[0103]

[0104]

[0105]

[0106] Obtain welding process parameter correction values ;in, This indicates a welding power correction head, used to convert deviations in the molten pool condition into power compensation. This indicates a spot diameter correction head, used to convert molten pool state deviations into spot diameter compensation. This indicates a welding speed correction head, used to convert deviations in the molten pool state into welding speed compensation. This indicates a defocusing correction head, used to convert molten pool state deviations into defocusing position compensation. This represents the final linear output weight matrix for the four welding heads, used to determine the relative correction ratio and direction of the four parameters—welding power, spot diameter, welding speed, and defocusing amount—to the same molten pool state deviation. Represents a smooth nonlinear activation function. This represents the second-layer shared backbone weight matrix, used to capture nonlinear physical phenomena caused by abnormal molten pool conditions. These nonlinear physical phenomena include the secondary thermal accumulation effect. This represents the first-layer shared backbone weight matrix, used to nonlinearly map the physical space of molten pool state deviations to the process parameter sensitive space. This represents the first-level bias vector, used to compensate for systematic global offsets. This represents the second-level bias vector, used to correct systematic mapping biases in the shared backbone. This represents the final bias vector of the four heads, used to compensate for the zero-point mechanical error and response lag of each actuator, which includes a laser, an optical head, a motion platform, and a defocusing motor.

[0107] For example, (Reduce welding power to suppress overheating tendency) (Slightly increase the beam diameter to disperse the energy density) (Increase welding speed and reduce heat input to the welding area) (Adjust the defocus amount and fine-tune the energy distribution).

[0108] Step V: Correct the initial welding process parameters based on the welding process parameter correction values ​​to obtain the predicted process parameter values.

[0109] Specifically, through calculation formula

[0110]

[0111]

[0112]

[0113]

[0114] Obtain predicted values ​​of process parameters (Welding power prediction) (Predicted spot diameter) (Predicted welding speed) (Predicted value of defocus).

[0115] It should be noted that, in this embodiment of the invention, the process parameter prediction model minimizes the composite loss function. The system is trained to make its predicted values ​​accurately fit the true values ​​in Table 1. The loss function is defined as:

[0116]

[0117] in, The actual values ​​are those in Table 1. These are the model's predicted values. This is an L2 regularization term.

[0118] It should be noted that in actual implementation, the welding process continues, and the third module executes steps II and III in a 10ms cycle to form a closed-loop control.

[0119] Step 14: Weld the alloy workpiece to be welded according to the predicted values ​​of the welding process parameters.

[0120] Specifically, the final predicted set of process parameters The data is sent to the laser welding system in real time. The laser welding system then adjusts the execution parameters of each actuator.

[0121] Figure 3 This illustrates a feasible implementation method, showing the molten pool morphology during welding of the alloy workpiece based on predicted welding process parameters. Figure 3It is evident that the weld has a uniform weld width, sufficient weld depth, appropriate reinforcement height, smooth and defect-free surface, and a smooth transition with the base material, meeting the standards for high-quality welding.

[0122] In one feasible implementation, a tensile test was also performed on the welded joint after welding, and the test results are shown in Table 2.

[0123] Table 2

[0124]

[0125] As shown in Table 2, the tensile strength of all samples is much higher than the standard of 500 MPa for 316L laser welding strength.

[0126] In summary, the high-temperature alloy laser welding process parameter prediction method provided by this invention, by acquiring and analyzing molten pool images in real time, uses the aspect ratio of the molten pool, which characterizes the welding state, as a dynamic feedback signal, enabling the process parameters to be adaptively adjusted according to the actual welding process, effectively improving the stability of welding quality. The process parameter prediction model first provides optimized initial parameters for specific workpieces based on a knowledge base, and then performs precise fine-tuning based on the dynamic characteristics of the molten pool, ensuring high accuracy and strong robustness of the prediction. By comprehensively considering the static workpiece attributes and dynamic process characteristics, it performs collaborative prediction and optimization of multiple key parameters such as welding power and speed, realizing intelligent multi-dimensional decision-making covering the entire welding process, reducing reliance on human experience, and contributing to improved welding quality.

[0127] like Figure 4 As shown, embodiments of the present invention provide a terminal device, such as... Figure 4 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 4 The diagram shows only one processor, a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 executes the computer program D102 to implement the steps in any of the above method embodiments.

[0128] Specifically, when the processor D100 executes the computer program D102, it acquires the workpiece attributes of the alloy workpiece to be welded; it acquires welding pool images in real time and extracts the aspect ratio time series from the welding pool images; it inputs the workpiece attributes and the aspect ratio time series into a pre-trained process parameter prediction model, which outputs predicted welding process parameters; and it welds the alloy workpiece to be welded according to the predicted welding process parameters. Specifically, by acquiring and analyzing the welding pool images in real time, the aspect ratio of the welding pool, which characterizes the welding state, is used as a dynamic feedback signal, enabling the process parameters to be adaptively adjusted according to the actual welding process, effectively improving the stability of welding quality. The process parameter prediction model first provides optimized initial parameters for a specific workpiece based on a knowledge base, and then performs precise fine-tuning based on the dynamic characteristics of the welding pool, ensuring high accuracy and robustness of the prediction. By comprehensively considering static workpiece attributes and dynamic process characteristics, it collaboratively predicts and optimizes multiple key parameters such as welding power and speed, achieving intelligent multi-dimensional decision-making covering the entire welding process, reducing reliance on human experience, and contributing to improved welding quality.

[0129] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0130] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0131] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0132] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0133] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0134] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A method for predicting process parameters in high-temperature alloy laser welding, characterized in that, include: Obtain the workpiece properties of the alloy workpiece to be welded; the workpiece properties include the high-temperature alloy type and the plate thickness; Real-time acquisition of weld pool images, and extraction of the weld pool aspect ratio time sequence from the weld pool images; the weld pool aspect ratio characterizes the state of the weld pool, which is stable, over-melted, or under-melted. The workpiece attributes and the aspect ratio time sequence of the molten pool are input into a pre-trained process parameter prediction model, which outputs the predicted values ​​of welding process parameters. The process parameter prediction model includes a first module, a second module, and a third module. The first module is used to perform high-precision interpolation prediction of the workpiece attributes based on a prior database to obtain the initial welding process parameters corresponding to the workpiece attributes. The step of performing high-precision interpolation prediction on the workpiece attributes based on a prior database to obtain the initial welding process parameters corresponding to the workpiece attributes includes: using calculation formulas... Obtain initial welding process parameters ;in, Indicates the initial welding power. Indicates the initial spot diameter. Indicates the initial welding speed. Indicates the initial defocus amount. This indicates a three-layer fully connected static regression head. Indicates splicing, This indicates the category of high-temperature alloys after integer encoding. This represents the normalized thickness of the sheet metal. Represents the prior database medium and high temperature alloy categories Empirically proven optimal set of process parameters for gradient plate thicknesses This represents all parameters of the static baseline parameter regression MLP, which are learnable parameters; The second module is used to perform feature engineering on the aspect ratio time series of the weld pool to obtain weld pool time series features characterizing the trend of weld pool state changes; the feature engineering on the aspect ratio time series of the weld pool to obtain weld pool time series features characterizing the trend of weld pool state changes includes: using calculation formulas The timing characteristics of the molten pool are obtained. ;in, This represents a three-layer one-dimensional residual network. This represents the time series sequence indicating the aspect ratio of the molten pool. , Indicates time The aspect ratio of the molten pool , Indicates time The length of the molten pool, Indicates time The width of the molten pool, Indicates the timing window length; The third module is used to correct the initial welding process parameters based on the molten pool timing characteristics to obtain predicted process parameter values; the prior database includes a set of empirically validated optimal process parameters for the alloy workpiece to be welded under gradient plate thickness; the welding process parameters include welding power, spot diameter, welding speed, and defocusing amount; the step of correcting the initial welding process parameters based on the molten pool timing characteristics to obtain predicted process parameter values ​​includes: High-temperature alloy categories after integer encoding and normalized plate thickness By performing feature stitching and nonlinear transformation, the theoretically optimal dynamic trajectory of the molten pool aspect ratio corresponding to the workpiece attributes is obtained. The theoretical optimal dynamic trajectory The expression is: ,in, Representation layer normalization, Indicates learnable parameters; By performing a linear projection on the temporal characteristics of the weld pool, the instantaneous state characteristics of the weld pool at each moment are obtained. and the deviation vector used to correct welding process parameters at that moment. The instantaneous state characteristics The expression is The deviation vector The expression is ;in, and This represents two different projection matrices; According to the theoretical optimal dynamic trajectory Instantaneous state characteristics and deviation vector Calculate the context offset representation vector used to characterize the abnormal state of the molten pool. The optimal dynamic trajectory based on the theory... Instantaneous state characteristics and deviation vector Calculate the context offset representation vector used to characterize the abnormal state of the molten pool. This includes: through calculation formulas Obtain the attention weight vector ; where attention weight vector Used to pinpoint the moment when the molten pool condition is abnormal. Indicates the hidden feature dimension. Represents a learnable positional offset mask; calculated using the formula Obtain the context offset representation vector ; The theoretical optimal dynamic trajectory and the context offset representation vector The input is a multilayer perceptron network with a shared backbone, and the corrected values ​​of the welding process parameters are obtained; the theoretically optimal dynamic trajectory is then used. and the context offset representation vector The input is a multilayer perceptron network with a shared backbone, which obtains the correction values ​​for welding process parameters, including: through calculation formulas. Obtain welding process parameter correction values ;in, This indicates a welding power correction head, used to convert deviations in the molten pool condition into power compensation. This indicates a spot diameter correction head, used to convert molten pool state deviations into spot diameter compensation. This indicates a welding speed correction head, used to convert deviations in the molten pool state into welding speed compensation. This indicates a defocusing correction head, used to convert molten pool state deviations into defocusing position compensation. This represents the final linear output weight matrix for the four welding heads, used to determine the relative correction ratio and direction of the four parameters—welding power, spot diameter, welding speed, and defocusing amount—to the same molten pool state deviation. Represents a smooth nonlinear activation function. This represents the second-layer shared backbone weight matrix, used to capture nonlinear physical phenomena caused by abnormal molten pool conditions, including the secondary thermal accumulation effect. This represents the first-layer shared backbone weight matrix, used to nonlinearly map the physical space of molten pool state deviations to the process parameter sensitive space. This represents the first-level bias vector, used to compensate for systematic global offsets. This represents the second-level bias vector, used to correct systematic mapping biases in the shared backbone. This represents the final bias vector of the four heads, used to compensate for the zero-point mechanical error and response hysteresis of each actuator, which includes a laser, an optical head, a motion platform, and a defocusing motor. The initial welding process parameters are corrected based on the corrected welding process parameter values ​​to obtain the predicted process parameter values; this includes: using calculation formulas. Obtain welding power prediction value Predicted spot diameter Predicted welding speed Defocusing amount prediction value ; The alloy workpiece to be welded is welded according to the predicted values ​​of the welding process parameters.

2. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in claim 1.

3. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in claim 1.

Images