A method and apparatus for controlling the performance of laser-welded electrothermal alloy joints

Abstract

This invention relates to the field of laser welding technology, specifically disclosing a method and apparatus for controlling the performance of laser-welded joints of electrothermal alloys. The apparatus includes a laser welding head, a single multifunctional induction coil, a programmable digital induction power supply, an infrared vision sensing system, and a computer control system. The computer control system has a pre-set time-domain convolutional neural network model and a constant effective value compensation algorithm. It acquires images of the molten pool through the infrared vision sensing system and extracts a one-dimensional grayscale time-series signal, inputting it into the model to analyze the fluid dynamics characteristics of the molten pool and outputting the modulation depth. Simultaneously, it adjusts the carrier reference amplitude in real time according to the constant effective value compensation algorithm to ensure that the effective value of the total current output by the induction coil remains constant. This invention achieves the synergistic effect of induction heating heat treatment and electromagnetic stirring through a single coil, realizing decoupled control of thermal and mechanical effects, refining grains, reducing grain tilt, and decreasing continuous grain boundaries, thus significantly improving the mechanical properties of the welded joint.

Description

Technical Field

[0001] This invention relates to the field of laser welding technology, and more specifically to a method and apparatus for controlling the performance of laser-welded electrothermal alloy joints. Background Technology

[0002] Currently, laser welding technology is widely used in mechanical manufacturing due to its advantages such as high energy density and fast welding speed. During laser welding, different zones of the weld joint experience different thermal cycles, resulting in significant differences in their microstructure and mechanical properties. Tensile tests on laser-welded samples of electrothermal alloys revealed that fracture occurred in the fusion zone of the weld joint, failing to guarantee the required mechanical properties after welding. Specifically, the fusion zone has a large heat input, resulting in coarse columnar grains. Because temperature gradient orientation changes lead to changes in grain growth direction, laser welding creates a continuous grain boundary at the center of the fusion zone that penetrates the thickness of the plate, easily causing brittle intergranular fracture at this point. Simultaneously, the top and bottom of the joint, due to direct contact with air, experience higher temperature gradients, causing the grains to grow more slender and obliquely, also easily leading to brittle transgranular fracture.

[0003] Existing electromagnetic induction control technologies are mostly static closed-loop systems based on temperature, which cannot sense the fluid dynamics state inside the molten pool. The lack of real-time feedback on the dynamic oscillation characteristics of the molten pool (such as frequency and amplitude) makes it difficult to adapt to complex welding conditions, easily leading to fine-grain failure due to insufficient stirring or spatter defects caused by over-stirring.

[0004] Therefore, how to provide a control method that can decouple thermo-mechanical effects through a single coil is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide a method and apparatus for improving the performance of laser-welded joints of electrothermal alloys to overcome or at least partially solve the above problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention relates to a method for controlling the performance of laser-welded electrothermal alloy joints, specifically comprising the following steps: S1. Collect image sequences of the molten pool area during the laser welding process, extract the data of the average gray value of the central area of ​​the molten pool changing over time, and construct a one-dimensional gray-scale time sequence signal.

[0008] S2. Set the target effective value of the induced current according to the target temperature at which the grain tilt growth at the top and bottom of the weld joint is lowest.

[0009] S3. Input the one-dimensional gray-scale time-series signal into the time-domain convolutional neural network model to obtain the induced current modulation depth required to maintain the turbulent fine-grained state in the center region of the molten pool.

[0010] S4. When the modulation depth of the induced current changes, the carrier reference amplitude is calculated in real time using the constant effective value compensation algorithm. By performing reverse dynamic compensation on the carrier reference amplitude, the final output effective value of the total current is ensured to be equal to the target effective value of the induced current.

[0011] S5. Based on the calculated carrier reference amplitude and induced current modulation depth, output composite alternating current. While applying varying pulsating stirring force to the molten pool using low-frequency pulsating magnetic pressure, lock the total heat input power to achieve constant temperature induction heating.

[0012] Preferred temporal convolutional neural network models include: The input layer is used to receive one-dimensional grayscale timing signals.

[0013] The feature extraction layer consists of multiple stacked one-dimensional convolutional layers. It extracts the inherent frequency oscillation features and waveform envelope features of the molten pool in the one-dimensional gray-scale time-series signal by sliding convolution kernels. Residual connections are introduced between each convolutional layer to superimpose the input and output of the current layer and pass them to the next layer to preserve the gradient information of the original signal and output the molten pool oscillation feature map.

[0014] The attention enhancement layer receives the molten pool oscillation feature map output by the feature extraction layer and has a weighting module with a self-attention mechanism. The weighting module calculates the weight distribution of the molten pool oscillation feature map, reduces the feature weights corresponding to transient high-frequency noise, and enhances the feature weights corresponding to the periodic oscillation features of the molten pool, and outputs the weighted enhanced molten pool oscillation feature map.

[0015] The output layer receives the enhanced molten pool oscillation feature map output by the attention enhancement layer, and maps it to the output induced current modulation depth through the fully connected layer.

[0016] The preferred constant effective value compensation algorithm calculation formula is as follows:

[0017] in, The reference amplitude of the carrier current; The modulation depth of the induced current, and satisfying , This represents the target effective value of the induced current.

[0018] The preferred formula for calculating composite alternating current is as follows:

[0019] in, It is a high-frequency carrier. It is a low-frequency pulsating wave. t For time.

[0020] Preferably, S3 also includes a process of training the temporal convolutional neural network model, specifically including: Constructing a training sample set: In the early trial welding experiments, the modulation depth of the induced current under different working conditions was recorded, and the corresponding one-dimensional grayscale time sequence signal was collected simultaneously; after welding, metallographic analysis was performed to screen out the working conditions that minimized the average grain size and stabilized the molten pool oscillation, and the corresponding modulation depth of the induced current was output as the optimal modulation depth label.

[0021] Model parameter optimization: Using a one-dimensional gray-scale time series signal as the model input and the optimal modulation depth label as the desired output, a loss function is constructed. The error between the predicted value and the label value is calculated through the backpropagation algorithm, and the model parameters are optimized.

[0022] Establish mapping relationship: Establish a time-domain convolutional neural network model to learn the nonlinear mapping relationship between the input one-dimensional gray-scale time-series signal and the output optimal induced current modulation depth.

[0023] Preferably, the composite alternating current in S5 consists of a high-frequency component. and low frequency components The high-frequency carrier wave corresponds to the high-frequency component of the composite alternating current. By utilizing the constant power heat input provided by this high-frequency component, the weld is kept at the target temperature for heat treatment. The low-frequency pulsating wave corresponds to the low-frequency component of the composite alternating current. At the same time, the low-frequency pulsating magnetic pressure generated by this low-frequency component is used as a pulsating Lorentz force to stir the molten pool and refine the grains.

[0024] In a second aspect, the present invention also relates to a control device for improving the performance of electrothermal alloy laser welding joints, comprising: a laser welding head, a single multi-functional induction coil, a programmable digital induction power supply, an infrared vision sensing system, and a computer control system. A single multi-functional induction coil is coaxially mounted below the laser welding head; The programmable digital induction power supply connects to a single multi-functional induction coil and outputs a composite alternating current. An infrared vision sensing system acquires image sequences of the molten pool area during the laser welding process, extracts data on the change of the average gray value of the central region of the molten pool over time, and constructs a one-dimensional gray-scale time-series signal. Steps S2-S5 in the execution method of the computer control system.

[0025] Preferably, the computer control system includes: Signal acquisition module: Acquires image sequences of the molten pool area during the laser welding process, extracts data on the average gray value of the central region of the molten pool as a function of time, and constructs a one-dimensional gray-scale time-series signal; Intelligent decision-making module of temporal convolutional neural network model: Input the one-dimensional gray-scale time-series signal into the temporal convolutional neural network model to obtain the induced current modulation depth required to maintain the turbulent fine-grained state in the central region of the molten pool; Constant RMS Compensation Module: When the modulation depth of the induced current changes, the constant RMS compensation algorithm is used to calculate the carrier reference amplitude in real time. By performing reverse dynamic compensation on the carrier reference amplitude, it is ensured that the final output total current RMS value is equal to the target RMS value of the induced current. Composite field control module: Based on the calculated carrier reference amplitude and induced current modulation depth, it outputs a composite alternating current. While applying a changing pulsating stirring force to the molten pool using low-frequency pulsating magnetic pressure, it locks the total heat input power to achieve constant temperature induction heating.

[0026] Thirdly, the present invention also relates to an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described method for regulating the performance of an electrothermal alloy laser welding joint.

[0027] Fourthly, the present invention also relates to a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the above-described method for regulating the performance of an electrothermal alloy laser welding joint are implemented.

[0028] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method and apparatus for controlling the performance of electrothermal alloy laser welding joints. The beneficial effects of the above technical solutions provided by the embodiments of the present invention include at least the following: ① By simultaneously achieving induction heating heat treatment and electromagnetic stirring through a single coil, the problem of arranging welding equipment in a narrow space is solved.

[0029] ② The decoupled control of thermal and mechanical effects was achieved through a constant effective value compensation algorithm.

[0030] ③ By using a time-domain convolutional neural network model, the dynamic characteristics of the molten pool can be identified and adaptively controlled in real time, thereby refining the grains, reducing the degree of grain tilt, and reducing or breaking continuous grain boundaries, thus significantly improving the mechanical properties of the welded joint. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0032] Figure 1 This is a flowchart of a method for improving the performance of laser-welded electrothermal alloy joints provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a laser welding device for improving the performance of laser-welded joints of electrothermal alloys provided in an embodiment of the present invention; Figure 3 The tensile properties diagram of the electrothermal alloy laser welded joint provided in the embodiments of the present invention; Figure 4 The grain morphology provided in this embodiment of the invention leads to poor tensile properties of the laser-welded joint of the electrothermal alloy; Figure 5 The figures show the results of slow strain tensile tests on the original state and the welded joint after heat treatment, as provided in the embodiments of the present invention.

[0033] In the diagram, 1-laser welding head, 2-single multi-functional induction coil, 3-programmable digital induction power supply, 4-infrared vision sensing system, and 5-computer control system. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] like Figure 1 As shown in the figure, this invention discloses a method for controlling the performance of laser-welded electrothermal alloy joints, specifically including the following steps: S1. Collect image sequences of the molten pool area during the laser welding process, extract the data of the average gray value of the central area of ​​the molten pool changing over time, and construct a one-dimensional gray-scale time sequence signal.

[0036] S2. Set the target effective value of the induced current according to the target temperature at which the grain tilt growth at the top and bottom of the weld joint is lowest.

[0037] S3. Input the one-dimensional gray-scale time-series signal into the time-domain convolutional neural network model to obtain the induced current modulation depth required to maintain the turbulent fine-grained state in the center region of the molten pool.

[0038] Temporal convolutional neural network models include: The input layer is used to receive one-dimensional grayscale timing signals.

[0039] The feature extraction layer consists of multiple stacked one-dimensional convolutional layers. It extracts the inherent frequency oscillation features and waveform envelope features of the molten pool in the one-dimensional gray-scale time-series signal by sliding convolution kernels. Residual connections are introduced between each convolutional layer to superimpose the input and output of the current layer and pass them to the next layer to preserve the gradient information of the original signal and output the molten pool oscillation feature map.

[0040] The attention enhancement layer receives the molten pool oscillation feature map output by the feature extraction layer and has a weighting module with a self-attention mechanism. The weighting module calculates the weight distribution of the molten pool oscillation feature map, reduces the feature weights corresponding to transient high-frequency noise, and enhances the feature weights corresponding to the periodic oscillation features of the molten pool, and outputs the weighted enhanced molten pool oscillation feature map.

[0041] The output layer receives the enhanced molten pool oscillation feature map output by the attention enhancement layer, and maps it to the output induced current modulation depth through the fully connected layer.

[0042] S3 also includes the process of training a temporal convolutional neural network model, specifically including: Constructing a training sample set: In the early trial welding experiments, the modulation depth of the induced current under different working conditions was recorded, and the corresponding one-dimensional grayscale time sequence signal was collected simultaneously; after welding, metallographic analysis was performed to screen out the working conditions that minimized the average grain size and stabilized the molten pool oscillation, and the corresponding modulation depth of the induced current was output as the optimal modulation depth label.

[0043] Model parameter optimization: Using a one-dimensional gray-scale time series signal as the model input and the optimal modulation depth label as the desired output, a loss function is constructed. The error between the predicted value and the label value is calculated through the backpropagation algorithm, and the model parameters are optimized.

[0044] Establish mapping relationship: Establish a time-domain convolutional neural network model to learn the nonlinear mapping relationship between the input one-dimensional gray-scale time-series signal and the output optimal induced current modulation depth.

[0045] S4. When the modulation depth of the induced current changes, the carrier reference amplitude is calculated in real time using the constant effective value compensation algorithm. By performing reverse dynamic compensation on the carrier reference amplitude, the final output effective value of the total current is ensured to be equal to the target effective value of the induced current.

[0046] The calculation formula for the constant effective value compensation algorithm is as follows:

[0047] in, The reference amplitude of the carrier current; The modulation depth of the induced current, and satisfying , This represents the target effective value of the induced current.

[0048] Reverse dynamic compensation is used to offset the additional power fluctuations caused by changes in modulation depth, thereby providing accurate reference parameters for the composite alternating current synthesized in subsequent steps, ensuring that the effective value of the total output current of the coil is equal to the target effective value of the induced current.

[0049] S5. Based on the calculated carrier reference amplitude and induced current modulation depth, output composite alternating current. While applying varying pulsating stirring force to the molten pool using low-frequency pulsating magnetic pressure, lock the total heat input power to achieve constant temperature induction heating.

[0050] The formula for calculating composite alternating current is as follows:

[0051] in, It is a high-frequency carrier. It is a low-frequency pulsating wave. t For time.

[0052] The composite alternating current in S5 consists of high-frequency components. and low frequency components The high-frequency carrier wave corresponds to the high-frequency component of the composite alternating current. By utilizing the constant power heat input provided by this high-frequency component, the weld is kept at the target temperature for heat treatment. The low-frequency pulsating wave corresponds to the low-frequency component of the composite alternating current. At the same time, the low-frequency pulsating magnetic pressure generated by this low-frequency component is used as a pulsating Lorentz force to stir the molten pool and refine the grains.

[0053] The present invention also discloses a control device for improving the performance of laser welding joints of electrothermal alloys, comprising: a laser welding head 1, a single multi-functional induction coil 2, a programmable digital induction power supply 3, an infrared vision sensing system 4, and a computer control system 5.

[0054] The single-unit multi-functional induction coil 2 is coaxially mounted below the laser welding head 1.

[0055] The programmable digital induction power supply 3 is connected to the single multi-functional induction coil 2 and outputs a composite alternating current.

[0056] The infrared vision sensing system 4 acquires image sequences of the molten pool area during the welding process of the laser welding head 1, extracts data on the average gray value of the central region of the molten pool as a function of time, and constructs a one-dimensional gray-scale time sequence signal.

[0057] The computer control system 5 includes: Signal acquisition module: Acquires image sequences of the molten pool area during the laser welding process, extracts data on the average gray value of the central region of the molten pool as a function of time, and constructs a one-dimensional gray-scale time-series signal.

[0058] Intelligent decision-making module of temporal convolutional neural network model: Input the one-dimensional gray-scale time-series signal into the temporal convolutional neural network model to obtain the induced current modulation depth required to maintain the turbulent fine-grained state in the center region of the molten pool.

[0059] Constant RMS Compensation Module: When the modulation depth of the induced current changes, the constant RMS compensation algorithm is used to calculate the carrier reference amplitude in real time. By performing reverse dynamic compensation on the carrier reference amplitude, it is ensured that the final output total current RMS value is equal to the target RMS value of the induced current.

[0060] Composite field control module: Based on the calculated carrier reference amplitude and induced current modulation depth, it outputs a composite alternating current. While applying a changing pulsating stirring force to the molten pool using low-frequency pulsating magnetic pressure, it locks the total heat input power to achieve constant temperature induction heating.

[0061] The present invention will be described in detail through specific embodiments.

[0062] Example 1: System Hardware Construction like Figure 2 As shown in the figure, the FeCrAl electrothermal alloy laser welding control device based on time-domain convolutional neural network model feedback and constant effective power compensation provided in this embodiment mainly consists of a laser welding system composed of a fiber laser, a six-axis welding robot, and a laser welding head 1, an infrared vision sensing system 4, a computer control system 5, and an induction heating and stirring assembly.

[0063] The core of the induction heating and stirring assembly is a single multi-functional induction coil 2, coaxially mounted below the laser welding head 1. It is made of copper tubing and internally cooled by water. The coil is connected to a 20kW digital programmable induction power supply 3, which has amplitude modulation capabilities, and its output current waveform can be defined in real time by a computer via an analog interface or communication protocol. The infrared visual sensing system 4 uses a high-speed infrared camera with a sampling rate of 1000Hz, and a narrow-band filter for a 1064nm wavelength is installed in front of the lens to filter out interference from laser reflection light.

[0064] Example 2: Construction and Training of Temporal Convolutional Neural Network Model The computer control system 5 is equipped with a time-domain convolutional neural network model.

[0065] Model Architecture: The input layer receives a 50ms time-series signal of the center grayscale of the melt pool. The feature extraction layer consists of three stacked one-dimensional convolutional layers with a kernel size of 3 and a ReLU activation function. Residual connections are provided between layers to prevent gradient vanishing. A self-attention module is connected after the convolutional layers to calculate the weight distribution at each time step and automatically reduce the weights of sudden splash noise. The output layer is a fully connected layer, outputting a single control variable: modulation depth. (Value range: 0~1).

[0066] Model training: In the initial experiments, the modulation depth was changed. A test weld was performed, and infrared signals were collected simultaneously. Metallographic analysis was then conducted on the sample after welding, such as... Figure 4 As shown, the average grain size and aspect ratio are statistically analyzed. The model is trained with "grain size minimization" and "molten pool oscillation stability" as labels, enabling the model to learn to determine whether to increase or decrease the stirring intensity based on the current molten pool oscillation waveform, that is, to accurately predict and output the required optimal modulation depth.

[0067] Example 3: Specific Operation of the Control Method This embodiment describes laser butt welding of a 5mm thick FeCrAl electrothermal alloy sheet. The specific steps are as follows: Step 1: Preset process parameters Preset the parameters before actual welding. Configure the induction power supply to output only an unmodulated high-frequency carrier wave. =200kHz, =0), adjust the current amplitude until the infrared thermometer displays a stable temperature in the cooling section of the welded joint. =800℃. Record the effective value of the current at this point. Repeat this experiment multiple times, and take the average value as the target effective value of the induced current in subsequent steps. Simultaneously, based on existing process experiments, the optimal turbulent fine-grained state during molten pool stirring was determined, along with the spectrum and amplitude range of the gray-scale time-series signal collected at the center of the molten pool under this state, serving as the standard range for subsequent control. The standard range of the gray-scale time-series signal spectrum and amplitude is stored in the computer control system (5).

[0068] Step 2: Welding and Signal Acquisition Laser welding is initiated, and individual induction coils are synchronously energized. An infrared camera captures real-time images of the molten pool, and the algorithm automatically locks onto a 1mm × 1mm area at the geometric center of the molten pool, extracts the curve of the average grayscale change of this area over time, and generates a one-dimensional time-series signal.

[0069] Step 3: Intelligent Decision-Making and Compensation The computer inputs real-time acquired time-series signals into a time-domain convolutional neural network model. If the model detects a single signal spectrum and low amplitude, it indicates that the molten pool is in a laminar state, which easily leads to the growth of columnar crystals, and the model outputs a larger modulation depth. The instruction was given. Simultaneously, the computer immediately invoked the constant effective value compensation formula:

[0070] The system command power supply will transfer the carrier reference from Downgraded to At the same time, the superposition amplitude is low-frequency pulsations ( =20Hz).

[0071] Step 4: Physical Process - Power Supply Outputs Composite Alternating Current. The specific current magnitude is...

[0072] in, The high-frequency carrier wave, with a value of 200kHz, is used to generate induced electromotive force and induced current at the weld, providing Joule heating to maintain the weld at the target temperature. ; The low-frequency modulation frequency is set to 20Hz. The induced current of low-frequency amplitude modulation excites a pulsating Lorentz force synchronized with the modulation frequency in the molten pool, which induces strong shear convection in the melt, thus stirring the molten pool. Based on the temporal convolutional neural network model in step three... It adjusts dynamically in the opposite direction to the changes.

[0073] A thermal effect is generated through composite alternating current: although the current waveform fluctuates drastically, its effective value remains constant. Therefore, the Joule heating effect is constant, and the weld joint and heat-affected zone are always maintained at 800℃, effectively reducing the temperature gradient at the weld during welding, reducing the tilt of the grains at the top and bottom of the joint, and further reducing the brittle transgranular fracture zone of the joint. At the same time, the composite alternating current generates a force effect: the superimposed 20Hz low-frequency component excites a strong pulsating Lorentz force inside the molten pool, causing the melt to generate volume wave oscillations, strongly scouring the solid-liquid interface, mechanically breaking the growing coarse columnar grain tips. The broken grain arms are free in the molten pool, becoming new nucleation points, ultimately inducing the formation of fine equiaxed grain structures in the weld, resulting in grain refinement. At the same time, it breaks up continuous grain boundaries that penetrate the plate thickness, further reducing the brittle intergranular fracture zone of the joint. Locking the total heat input power, it achieves constant temperature induction heating, thereby making... Figure 3 The mechanical properties of the brittle welded joints are improved as shown in the figure. Figure 5 As shown.

[0074] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0075] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for controlling the performance of laser-welded electrothermal alloy joints, characterized in that, Specifically, the following steps are included: S1. Collect image sequences of the molten pool area during the laser welding process, extract the data of the average gray value of the central region of the molten pool changing with time, and construct a one-dimensional gray-scale time sequence signal. S2. Set the target effective value of the induced current according to the target temperature at which the grain tilt growth at the top and bottom of the welded joint is lowest. S3. Input the one-dimensional gray-scale time-series signal into the time-domain convolutional neural network model to obtain the induced current modulation depth required to maintain the turbulent fine-grained state in the central region of the molten pool. S4. When the modulation depth of the induced current changes, the carrier reference amplitude is calculated in real time using the constant effective value compensation algorithm. By performing reverse dynamic compensation on the carrier reference amplitude, the final output effective value of the total current is ensured to be equal to the target effective value of the induced current. S5. Based on the calculated carrier reference amplitude and the induced current modulation depth, output a composite alternating current. While applying a varying pulsating stirring force to the molten pool using the low-frequency pulsating magnetic pressure, lock the total heat input power to achieve constant temperature induction heating.

2. The method for controlling the performance of electrothermal alloy laser welding joints according to claim 1, characterized in that, The temporal convolutional neural network model includes: The input layer is used to receive the one-dimensional grayscale timing signal; The feature extraction layer consists of multiple stacked one-dimensional convolutional layers. It extracts the inherent frequency oscillation features and waveform envelope features of the molten pool in the one-dimensional gray-scale time-series signal by sliding convolution kernels. Residual connections are introduced between each convolutional layer to superimpose the input and output of the current layer and pass them to the next layer to preserve the gradient information of the original signal and output the molten pool oscillation feature map. An attention enhancement layer receives the molten pool oscillation feature map output by the feature extraction layer and is equipped with a weighting module with a self-attention mechanism. The weighting module calculates the weight distribution of the molten pool oscillation feature map, reduces the feature weights corresponding to transient high-frequency noise, and enhances the feature weights corresponding to the periodic oscillation features of the molten pool, and outputs a weighted enhanced molten pool oscillation feature map. The output layer receives the enhanced molten pool oscillation feature map output by the attention enhancement layer, and maps and outputs the induced current modulation depth through the fully connected layer.

3. The method for controlling the performance of electrothermal alloy laser welding joints according to claim 1, characterized in that, The calculation formula for the constant effective value compensation algorithm is as follows: in, The reference amplitude of the carrier current; The modulation depth of the induced current, and satisfying , This represents the target effective value of the induced current.

4. The method for controlling the performance of electrothermal alloy laser welding joints according to claim 1, characterized in that, The formula for calculating the composite alternating current is as follows: in, It is a high-frequency carrier. It is a low-frequency pulsating wave. t For time.

5. The method for controlling the performance of electrothermal alloy laser welding joints according to claim 1, characterized in that, S3 also includes the process of training the temporal convolutional neural network model, specifically including: Constructing a training sample set: In the early trial welding experiments, the modulation depth of the induced current under different working conditions was recorded, and the corresponding one-dimensional gray-scale time-series signal was collected simultaneously; after welding, metallographic analysis was performed to screen out the working conditions that minimized the average grain size and stabilized the molten pool oscillation, and the corresponding modulation depth of the induced current was output as the optimal modulation depth label. Model parameter optimization: Using the one-dimensional gray-scale time-series signal as the model input and the optimal modulation depth label as the desired output, a loss function is constructed. The error between the predicted value and the label value is calculated using the backpropagation algorithm, and the model parameters are optimized. Establish mapping relationship: Establish the nonlinear mapping relationship between the input one-dimensional gray-scale time-series signal and the output optimal induced current modulation depth of the time-domain convolutional neural network model.

6. The method for controlling the performance of laser-welded electrothermal alloy joints according to claim 4, characterized in that, The composite alternating current described in S5 consists of high-frequency components. and low frequency components The high-frequency carrier wave is superimposed and modulated, and corresponds to the high-frequency component of the composite alternating current. By utilizing the constant power heat input provided by the high-frequency component, the weld is kept at the target temperature for heat treatment. The low-frequency pulsating wave corresponds to the low-frequency component of the composite alternating current, and the low-frequency pulsating magnetic pressure generated by the low-frequency component is used as a pulsating Lorentz force to stir the molten pool and refine the grains.

7. A control device for improving the performance of laser-welded joints of electrothermal alloys, characterized in that, include: Laser welding head (1), single multi-functional induction coil (2), programmable digital induction power supply (3), infrared vision sensing system (4) and computer control system (5). A single multi-functional induction coil (2) is coaxially mounted below the laser welding head (1); The programmable digital induction power supply (3) is connected to the single multi-functional induction coil (2) and outputs a composite alternating current; The infrared vision sensing system (4) collects the image sequence of the molten pool area during the welding process of the laser welding head (1), extracts the data of the average gray value of the central area of ​​the molten pool changing with time, and constructs a one-dimensional gray-scale time sequence signal. The computer control system (5) performs steps S2-S5 of the method as described in any one of claims 1-6.

8. The control device for improving the performance of laser-welded electrothermal alloy joints according to claim 7, characterized in that, The computer control system (5) includes: Signal acquisition module: Acquires image sequences of the molten pool area during the laser welding process, extracts data on the average gray value of the central region of the molten pool as a function of time, and constructs a one-dimensional gray-scale time-series signal; Intelligent decision-making module of temporal convolutional neural network model: The one-dimensional gray-scale time-series signal is input into the temporal convolutional neural network model to obtain the induced current modulation depth required to maintain the turbulent fine-grained state in the central region of the molten pool; Constant RMS Compensation Module: When the modulation depth of the induced current changes, the constant RMS compensation algorithm is used to calculate the carrier reference amplitude in real time. By performing reverse dynamic compensation on the carrier reference amplitude, the final output total current RMS value is ensured to be equal to the target RMS value of the induced current. Composite field control module: Based on the calculated carrier reference amplitude and the induced current modulation depth, output composite alternating current, while applying a changing pulsating stirring force to the molten pool using the low-frequency pulsating magnetic pressure, lock the total heat input power to achieve constant temperature induction heating.

9. An electronic 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 steps of a method for controlling the performance of an electrothermal alloy laser welding joint as described in any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the control method for improving the performance of electrothermal alloy laser welding joints as described in any one of claims 1-6.

Images