Welding temperature field real-time prediction method based on data-mechanism combined driving

Abstract

The invention discloses a data-mechanism combined drive-based welding temperature field real-time prediction method, which comprises the following steps of: solving temperature fields in a welding process under different welding parameters by using a finite element method, and constructing a welding temperature field data set and a heat source parameter data set; utilizing the heat source parameter data set to train a heat source parameter prediction model based on the long-short memory network; training a welding temperature field real-time prediction model based on a DeepONet neural operator by using the temperature field data set; real-time heat source parameters are obtained from the preprocessed welding real-time monitoring data through a heat source parameter model; the real-time heat flux density of each unit node of the workpiece is obtained in combination with a double-ellipsoid moving heat source function; and inputting the real-time heat flux density into the temperature field real-time prediction model to obtain a predicted temperature value of each unit node. According to the method, real-time prediction of the temperature field in the welding process is achieved in a data driving and physical mechanism combined driving mode, and a generalized solution is provided for sensing and prediction of the welding process.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a real-time prediction method for welding temperature field based on data-mechanism joint drive. Background Art

[0002] The welding process temperature field represents the distribution of temperature on the material during the welding process. Only by determining the temperature field changes throughout the welding process can we further analyze and control various factors such as thermal stress, deformation and residual stress, which is essential to ensure the structural integrity and mechanical properties of the weld. By optimizing welding parameters and ensuring that the welding process temperature field remains within the optimal range, the welding quality and efficiency of the manufacturing process can be improved, and high-precision control of the welding process can be achieved.

[0003] At present, the monitoring methods for welding temperature field mainly include thermocouple measurement, infrared thermal imaging and other methods. However, these methods can only measure the surface temperature field and are powerless for the internal temperature field. Numerical simulation methods, such as finite element method (FEM) and finite difference method (FDM), can make detailed predictions and analyses of the three-dimensional temperature distribution under different welding parameters by establishing a transient heat conduction model of the welding process. However, these numerical simulation methods have extremely high computing resources and time costs when facing complex geometric shapes and high-dimensional problems, and cannot meet the real-time requirements of high-precision control. Faced with dynamic changes in the actual welding process, digital simulation methods cannot integrate real-time monitoring data with partial differential equations (PDEs) in real time. Each parameter change needs to be re-simulated, which limits the flexibility of temperature field prediction in the welding process.

[0004] Therefore, how to achieve high-precision real-time prediction of the temperature field of the welding process through the application of artificial intelligence technology, so as to more effectively optimize the welding process parameters, achieve high-precision control of the welding process, and improve welding quality and production efficiency has become a problem that needs to be solved urgently. Summary of the invention

[0005] 1. Technical problems to be solved:

[0006] In view of the above technical problems, the present invention provides a real-time prediction method for welding temperature field based on data-mechanism joint drive, and establishes a temperature field data set and a heat source parameter data set according to the heat transfer mechanism in the welding process. Create a data-driven real-time prediction model of heat source parameters and a real-time prediction model of welding temperature field, and then merge the actual process data with the theoretical data. Through the data-driven and physical mechanism-jointly driven method, the real-time prediction of the temperature field of the welding process is realized, providing a generalized solution for the perception and prediction of the welding process, which can be applied to manufacturing scenarios with high quality standards.

[0007] 2. Technical solution:

[0008] A real-time prediction method for welding temperature field based on data-mechanism joint drive, characterized in that it includes the following steps:

[0009] Step 1: Using the finite element method to solve the temperature field in the welding process under different welding parameters, constructing a welding temperature field data set and a heat source parameter data set; each data in the welding temperature field data set includes the coordinate value, heat flux density and simulated temperature value of each unit node on the welding workpiece geometric model at the same time; each data in the heat source parameter data set includes the welding current, arc voltage, simulated molten pool surface width and heat source parameters at the same time;

[0010] Step 2: Construct a real-time prediction model for heat source parameters based on long short-term memory network, take minimizing the model loss function as the objective function, and use the heat source parameter data set to train the real-time prediction model for heat source parameters to obtain a trained real-time prediction model for heat source parameters;

[0011] Step 3: Construct a real-time prediction model for welding temperature field based on DeepONet neural operator, take minimizing the model loss function as the objective function, and use the welding temperature field data set to train the real-time prediction model for welding temperature field to obtain a trained real-time prediction model for welding temperature field;

[0012] Step 4: collecting real-time monitoring data during the welding process; the real-time monitoring data includes welding current, arc voltage and molten pool surface image; extracting the actual molten pool surface width from the real-time molten pool surface image by image processing method;

[0013] Step 5: Taking the real-time collected monitoring data as the object, after the data processing in step 4, it is input into the real-time prediction model of heat source parameters trained in step 2 to obtain the predicted real-time heat source parameters; the real-time heat flux density of each unit node on the welding workpiece geometric model is calculated by using the double ellipsoid moving heat source function combined with the predicted real-time heat source parameters; the real-time heat flux density is input into the real-time prediction model of welding temperature field trained in step 3, that is, the predicted temperature value of each unit node on the welding workpiece geometric model at the current moment is obtained.

[0014] Furthermore, the welding parameters include welding current, arc voltage, material properties of the workpiece, and heat source parameters; the heat source parameters include the front semi-axis length of the heat source in the welding direction, the rear semi-axis length in the welding direction, the semi-axis length in the weld width direction, and the semi-axis length in the weld depth direction.

[0015] Furthermore, in step 1, a finite element analysis method is used to solve the temperature field during the welding process, which specifically includes the following steps:

[0016] S11: Create a geometric model of the welding workpiece and define the geometric dimensions of the welding workpiece;

[0017] S12: Setting material property data of the welding workpiece; the material property data includes thermal conductivity, density, latent heat and specific heat; wherein thermal conductivity and latent heat vary with temperature;

[0018] S13: define the parameters of the analysis steps in the finite element solution to ensure the convergence of the algorithm;

[0019] S14: define the initial temperature conditions and thermal convection boundary conditions of the welding workpiece;

[0020] S15: Based on the welding parameters and heat source parameters, a subroutine is created to create a double ellipsoid moving heat source function to obtain the heat flux density of each unit node on the welding workpiece geometric model;

[0021] S16: The heat flux density of each unit node on the welding workpiece geometric model is used to solve the temperature value of the corresponding unit node.

[0022] Furthermore, in the temperature field data set, the heat flux density of each unit node on the welding workpiece geometric model at the same moment is the accumulation of the heat flux density of the unit node corresponding to all historical moments before that moment; in the heat source parameter data set, the simulated molten pool surface width is obtained by calculating the distance between the unit nodes at the left and right edges of the welding workpiece geometric model whose temperature value is the metal melting point temperature.

[0023] Furthermore, step 2 specifically includes:

[0024] S21: Construct a real-time prediction model for heat source parameters based on long short-term memory network; the current moment is t, the welding current, arc voltage and simulated molten pool surface width at the moment [tn, t] are input into the model, where tn represents the first n sampling moments before the current moment, and n is an integer greater than or equal to 1; the predicted value of the heat source parameter at the current moment t is obtained based on regression;

[0025] S22: Use the heat source parameter data set to train a real-time prediction model for heat source parameters, minimize the model loss function through a gradient descent algorithm, and then iteratively update the model weights to obtain the trained model weights;

[0026] S23: Loading the trained model weights into the real-time prediction model of the heat source parameters, that is, the trained real-time prediction model of the heat source parameters.

[0027] Furthermore, step three specifically includes:

[0028] S31: construct a real-time prediction model for welding temperature field based on DeepONet neural operator; the real-time prediction model for welding temperature field includes a main network and a branch network; the main network inputs the heat flux density of each unit node on the welding workpiece geometric model in the temperature field data set, and the branch network inputs the coordinate value corresponding to the unit node; the outputs of the branch network and the main network are combined and output after performing a dot multiplication operation, and the output is the predicted temperature value of each unit node on the welding workpiece geometric model at the moment corresponding to the input data set;

[0029] S32: Use the temperature field data set to train the real-time prediction model of the welding temperature field, optimize the network parameters by minimizing the mean square error between the predicted temperature value and the simulated temperature value through the gradient descent algorithm, and obtain the trained model weights; load the trained model weights to obtain the trained real-time prediction model of the welding temperature field.

[0030] Furthermore, in step four, the welding current and arc voltage of the welding machine during the welding process are monitored in real time by an arc sensor; the molten pool surface image is captured in real time by a high dynamic range HDR industrial camera; and a bandpass filter for suppressing arc light is installed on the lens of the industrial camera.

[0031] Furthermore, in step 4, the actual molten pool surface width is extracted from the real-time molten pool surface image by an image processing method, comprising the following steps:

[0032] S41: Correct the melt pool surface image using the checkerboard method so that the pixel position in the melt pool surface image accurately matches the point position in the actual space; extract the region of interest DOI in the image;

[0033] S42: using the Retinex algorithm to perform image enhancement processing on the extracted image, so as to obtain an image with a clearer and more obvious molten pool boundary;

[0034] S43: Use Canny edge detection to detect edges in the image, and use morphological operations to make the image have continuous and complete edges;

[0035] S44: Use the least squares method to fit the straight line to extract the left and right boundaries; calculate the actual molten pool surface width based on the pixel distance between the left and right boundaries.

[0036] Furthermore, in step 5, the monitoring data collected in real time is taken as the object, and after the data processing in step 4, it is input into the real-time prediction model of heat source parameters trained in step 2, specifically: the current time is t, the welding current, arc voltage collected at the time [tn, t] and the actual molten pool surface width obtained after image processing are input into the real-time prediction model of heat source parameters trained in step 3 to predict the real-time heat source parameters;

[0037] The predicted real-time heat source parameters are brought into the double ellipsoid moving heat source function to obtain the heat flux density on each unit node on the welding workpiece geometric model at the current moment; it is accumulated with the historical heat flux density at all historical moments and sent into the main network of the welding temperature field real-time prediction model; the coordinate values ​​of all unit nodes on the welding workpiece geometric model in three-dimensional space are sent to the branch network of the welding temperature field real-time prediction model, and the predicted temperature value of each unit node on the welding workpiece geometric model at the current moment is obtained.

[0038] 3. Beneficial effects:

[0039] (1) In this method, a real-time prediction method of welding temperature field based on data-mechanism joint drive is firstly based on the heat transfer mechanism in the welding process, and the finite element method is used to construct the data of welding process temperature field under different welding parameters. This method can quickly and safely generate a large amount of data, which covers a wide range of experimental conditions and provides a rich training set for subsequent model training.

[0040] (2) In a real-time prediction method of welding temperature field based on data-mechanism joint drive, a real-time prediction model of heat source parameters based on long short-term memory network and a real-time prediction model of welding temperature field based on DeepONet neural operator are constructed to realize real-time prediction of heat source parameters and temperature field in the welding process in a data-driven way. The data-driven way can not only achieve real-time fusion of real-time monitoring data and partial differential equations of heat transfer mechanism, but also greatly reduce computing resources and time costs to achieve real-time prediction. Furthermore, the heat source parameters are continuously adjusted according to the actual weld shape to make the simulation results consistent with the actual molten pool boundary, thereby ensuring the accuracy of the predicted temperature field.

[0041] (3) In this method, the heat source parameters output by the real-time prediction model of heat source parameters are calculated through real-time heat flux density, and the calculated real-time heat flux density is used as the input of the real-time prediction model of welding temperature field. This process is driven by data and physical mechanisms, combined with real-time monitoring data during the welding process, and uses artificial intelligence technology to realize real-time prediction of the temperature field of the welding process. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is the overall flow chart of this method;

[0043] Figure 2 It is a schematic diagram of the temperature field in the welding process under different welding parameters solved by the finite element method in the specific implementation;

[0044] Figure 3 is a schematic diagram of a hardware acquisition device involved in this method;

[0045] Figure 4Schematic diagram of a double ellipsoid moving heat source involved in the method;

[0046] Figure 5 Comparison of the predicted results with the simulation results obtained by finite element analysis for a typical case. DETAILED DESCRIPTION

[0047] The present invention will be described in detail below with reference to the accompanying drawings.

[0048] As attached Figure 1 , a welding temperature field prediction method based on data-mechanism joint drive, characterized in that it includes the following steps:

[0049] Step 1: Using the finite element method to solve the temperature field in the welding process under different welding parameters, constructing a welding temperature field data set and a heat source parameter data set; each data in the welding temperature field data set includes the coordinate value, heat flux density and simulated temperature value of each unit node on the welding workpiece geometric model at the same time; each data in the heat source parameter data set includes the welding current, arc voltage, simulated molten pool surface width and heat source parameters at the same time;

[0050] Step 2: Construct a real-time prediction model for heat source parameters based on long short-term memory network, take minimizing the model loss function as the objective function, and use the heat source parameter data set to train the real-time prediction model for heat source parameters to obtain a trained real-time prediction model for heat source parameters;

[0051] Step 3: Construct a real-time prediction model for welding temperature field based on DeepONet neural operator, take minimizing the model loss function as the objective function, and use the welding temperature field data set to train the real-time prediction model for welding temperature field to obtain a trained real-time prediction model for welding temperature field;

[0052] Step 4: collecting real-time monitoring data during the welding process; the real-time monitoring data includes welding current, arc voltage and molten pool surface image; extracting the actual molten pool surface width from the real-time molten pool surface image by image processing method;

[0053] Step 5: Taking the real-time collected monitoring data as the object, after the data processing in step 4, it is input into the real-time prediction model of heat source parameters trained in step 2 to obtain the predicted real-time heat source parameters; the real-time heat flux density of each unit node on the welding workpiece geometric model is calculated by using the double ellipsoid moving heat source function combined with the predicted real-time heat source parameters; the real-time heat flux density is input into the real-time prediction model of welding temperature field trained in step 3, that is, the predicted temperature value of each unit node on the welding workpiece geometric model at the current moment is obtained.

[0054] The power density distribution of the first half of the double ellipsoid moving heat source function used in this method is as follows:

[0055]

[0056] The power density distribution of the second half of the double ellipsoid mobile heat source is as follows:

[0057]

[0058] In the above two formulas, η is the welding thermal efficiency, V is the arc voltage, I is the welding current, and a f is the front half axis length in the welding direction, a r is the rear semi-axis length in the welding direction, b is the semi-axis length in the direction of weld width, c is the semi-axis length in the direction of weld depth, x is the distance from the center of the heat source in the welding direction, y is the distance from the center of the heat source in the direction of weld width, z is the distance from the center of the heat source in the direction of weld depth, f1 and f2 are the energy ratios of the front and rear parts, f1+f2=2.

[0059] According to the double ellipsoid moving heat source function of the above two equations, the real-time monitoring data V, I and the predicted real-time heat source parameter a are f , a r , b, c are substituted into the function to obtain the heat flux density at each unit node on the welding workpiece geometric model.

[0060] Furthermore, the welding parameters include welding current, arc voltage, material properties of the workpiece, and heat source parameters; the heat source parameters include the front semi-axis length of the heat source in the welding direction, the rear semi-axis length in the welding direction, the semi-axis length in the weld width direction, and the semi-axis length in the weld depth direction.

[0061] Furthermore, in step 1, a finite element analysis method is used to solve the temperature field during the welding process, which specifically includes the following steps:

[0062] S11: Create a geometric model of the welding workpiece and define the geometric dimensions of the welding workpiece;

[0063] S12: Setting material property data of the welding workpiece; the material property data includes thermal conductivity, density, latent heat and specific heat; wherein thermal conductivity and latent heat vary with temperature;

[0064] S13: define the parameters of the analysis steps in the finite element solution to ensure the convergence of the algorithm;

[0065] S14: define the initial temperature conditions and thermal convection boundary conditions of the welding workpiece;

[0066] S15: Based on the welding parameters and heat source parameters, a subroutine is created to create a double ellipsoid moving heat source function to obtain the heat flux density of each unit node on the welding workpiece geometric model;

[0067] S16: The heat flux density of each unit node on the welding workpiece geometric model is used to solve the temperature value of the corresponding unit node.

[0068] Furthermore, in the temperature field data set, the heat flux density of each unit node on the welding workpiece geometric model at the same moment is the accumulation of the heat flux density of the unit node corresponding to all historical moments before that moment; in the heat source parameter data set, the simulated molten pool surface width is obtained by calculating the distance between the unit nodes at the left and right edges of the welding workpiece geometric model whose temperature value is the metal melting point temperature.

[0069] Furthermore, step 2 specifically includes:

[0070] S21: Construct a real-time prediction model for heat source parameters based on long short-term memory network; the current time is t, the welding current, arc voltage and simulated molten pool surface width at the time [t-10, t] are input into the model; the predicted value of the heat source parameter at the current time t is obtained based on regression;

[0071] S22: Use the heat source parameter data set to train a real-time prediction model for heat source parameters, minimize the model loss function through a gradient descent algorithm, and then iteratively update the model weights to obtain the trained model weights;

[0072] S23: Loading the trained model weights into the real-time prediction model of the heat source parameters, that is, the trained real-time prediction model of the heat source parameters.

[0073] Furthermore, step three specifically includes:

[0074] S31: construct a real-time prediction model for welding temperature field based on DeepONet neural operator; the real-time prediction model for welding temperature field includes a main network and a branch network; the main network inputs the heat flux density of each unit node on the welding workpiece geometric model in the temperature field data set, and the branch network inputs the coordinate value corresponding to the unit node; the outputs of the branch network and the main network are combined and output after performing a dot multiplication operation, and the output is the predicted temperature value of each unit node on the welding workpiece geometric model at the moment corresponding to the input data set;

[0075] S32: Use the temperature field data set to train the real-time prediction model of the welding temperature field, optimize the network parameters by minimizing the mean square error between the predicted temperature value and the simulated temperature value through the gradient descent algorithm, and obtain the trained model weights; load the trained model weights to obtain the trained real-time prediction model of the welding temperature field.

[0076] Furthermore, in step four, the welding current and arc voltage of the welding machine during the welding process are monitored in real time by an arc sensor; the molten pool surface image is captured in real time by a high dynamic range HDR industrial camera; and a bandpass filter for suppressing arc light is installed on the lens of the industrial camera.

[0077] Furthermore, in step 4, the actual molten pool surface width is extracted from the real-time molten pool surface image by an image processing method, comprising the following steps:

[0078] S41: Correct the melt pool surface image using the checkerboard method so that the pixel position in the melt pool surface image accurately matches the point position in the actual space; extract the region of interest DOI in the image;

[0079] S42: using the Retinex algorithm to perform image enhancement processing on the extracted image, so as to obtain an image with a clearer and more obvious molten pool boundary;

[0080] S43: Use Canny edge detection to detect edges in the image, and use morphological operations to make the image have continuous and complete edges;

[0081] S44: Use the least squares method to fit the straight line to extract the left and right boundaries; calculate the actual molten pool surface width based on the pixel distance between the left and right boundaries.

[0082] Further, in step 5, the monitoring data collected in real time is taken as the object, and after the data processing in step 4, it is input into the real-time prediction model of heat source parameters trained in step 2, specifically: the current time is t, the welding current, arc voltage collected at the time [tn, t] and the actual molten pool surface width obtained after image processing are input into the real-time prediction model of heat source parameters trained in step 3 to predict the real-time heat source parameters;

[0083] The predicted real-time heat source parameters are brought into the double ellipsoid moving heat source function to obtain the heat flux density on each unit node on the welding workpiece geometric model at the current moment; it is accumulated with the historical heat flux density at all historical moments and sent into the main network of the welding temperature field real-time prediction model; the coordinate values ​​of all unit nodes on the welding workpiece geometric model in three-dimensional space are sent to the branch network of the welding temperature field real-time prediction model, and the predicted temperature value of each unit node on the welding workpiece geometric model at the current moment is obtained. Specific embodiment:

[0085] The schematic diagram of the device structure of the hardware operating environment involved in this embodiment is shown in the attached figure. Figure 3 The schematic diagram of the double ellipsoid moving heat source is shown in the attached Figure 4 Show.

[0086] As attached Figure 3The equipment structure is configured with corresponding sensors to collect real-time monitoring data during the welding process; the real-time monitoring data includes welding current, arc voltage and molten pool image; the welding process is pulsed gas shielded tungsten arc welding, the tungsten diameter is 2.4mm, the shielding gas is argon, the gas flow is 7L / min, the frequency is 30HZ, and the duty cycle is 50%; the filter center installed on the industrial camera is 685±2 (nm), the filter half-peak full width is 10±2 (nm), the format is Mono8, the image size is 300×300, the sharpness is 3000, the gamma is 2.5, and the gain is 0 (db); the image acquisition frequency is adjusted to be consistent with the welding pulse frequency, and the camera can collect a molten pool surface image at each base current and peak current stage respectively.

[0087] The real-time monitoring data during the welding process is analyzed and processed, and the molten pool surface width is extracted from the real-time molten pool image by image processing method.

[0088] Based on the heat transfer mechanism during the welding process, the finite element method is used to construct the data of the temperature field of the welding process under different welding parameters, and the welding temperature field data set and the heat source parameter data set are constructed. In this embodiment, the finite element simulation software used is Abaqus, and the material properties of the welding workpiece are set to 304L; the dimensions of the workpiece in the X, Y, and Z directions are 40mm, 30mm, and 2mm, respectively, the X direction is along the weld direction, the Y direction is perpendicular to the weld, and the Z direction is along the thickness direction; the unit side length is 1mm, the total number of grids is 2400, and the total number of nodes is 3813; the ambient temperature is 298.15K; the welding heat source moves along the X-axis direction, the moving speed is 4mm / s, the welding time is 10s, and the time increment step is 0.16s, so that the data of 10 / 0.16 sampling moments can be obtained. The schematic diagram of the simulated temperature field constructed is shown in the attached figure. Figure 2 As shown, in this embodiment, 35 experiments were conducted using different welding parameters, and the temperature field data of the welding process under different welding parameters were calculated.

[0089] A real-time prediction model for heat source parameters based on long short-term memory network is constructed. In this embodiment, the number of hidden layers of the model is 1, the number of neurons in each layer is 15, and the length of the input data sequence is 10. The hidden state of the last time step used is passed through the fully connected layer to obtain the final output heat source parameters.

[0090] A real-time prediction model of welding temperature field based on DeepONet neural operator is constructed. The main network and trunk network of the DeepONet neural operator model are 5-layer and 3-layer fully connected layer structures respectively, with 3813 neurons in the input layer, 2000 neurons in the hidden layer, and 500 neurons in the output layer.

[0091] The operating environment of this embodiment is shown in the following table.

[0092]

[0093]

[0094] After preprocessing, the collected real-time monitoring data is input into the real-time prediction model of heat source parameters at this moment and the previous 10 sampling moments to obtain the predicted real-time heat source parameters at this moment; the real-time heat flux density of each unit node on the welding workpiece geometric model is calculated using the double ellipsoid moving heat source function combined with the predicted real-time heat source parameters; the real-time heat flux density is input into the real-time prediction model of the welding temperature field to obtain the predicted temperature value of each unit node on the welding workpiece geometric model at the current moment. The comparison between the obtained predicted data and the real data is shown in the attached figure. Figure 5 As shown, the dynamic characteristics of the temperature field changing with time can be clearly captured from the image. The predicted temperature field of the welding process is highly consistent with the actual temperature field, which proves the accuracy of the temperature field prediction.

[0095] In this embodiment, the temperature field data of the welding process is constructed based on the heat transfer mechanism of the welding process. By jointly driving the data and physical mechanism, combined with the real-time monitoring data of the welding process, the real-time prediction of the temperature field of the welding process is realized by using artificial intelligence technology. It has a clear structure, a specific implementation method and a certain degree of portability, and provides a new method with high efficiency and practical application value for realizing real-time monitoring of the welding process, and provides a basis for high-precision control of the welding process.

[0096] Although the present invention has been disclosed as above in terms of preferred embodiments, they are not intended to limit the present invention. Anyone skilled in the art can make various changes or modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection defined by the claims of this application.

Claims

1. A real-time prediction method for welding temperature field based on data-mechanism joint drive, characterized in that: The following steps are involved: Step 1: Use the finite element method to solve the temperature field during the welding process under different welding parameters, and construct the welding temperature field data set and heat source parameter data set; Each data in the welding temperature field data set includes the coordinate value, heat flux density and simulated temperature value of each unit node on the welding workpiece geometric model at the same time; each data in the heat source parameter data set includes the welding current, arc voltage, simulated molten pool surface width and heat source parameters at the same time; Step 2: Construct a real-time prediction model for heat source parameters based on long short-term memory network, take minimizing the model loss function as the objective function, and use the heat source parameter data set to train the real-time prediction model for heat source parameters to obtain a trained real-time prediction model for heat source parameters; Step 3: Construct a real-time prediction model for welding temperature field based on DeepONet neural operator, take minimizing the model loss function as the objective function, and use the welding temperature field data set to train the real-time prediction model for welding temperature field to obtain a trained real-time prediction model for welding temperature field; Step 4: collecting real-time monitoring data during the welding process; the real-time monitoring data includes welding current, arc voltage and molten pool surface image; extracting the actual molten pool surface width from the real-time molten pool surface image by image processing method; Step 5: Taking the real-time collected monitoring data as the object, after the data processing in step 4, it is input into the real-time prediction model of heat source parameters trained in step 2 to obtain the predicted real-time heat source parameters; using the double ellipsoid moving heat source function combined with the predicted real-time heat source parameters to calculate the real-time heat flux density of each unit node on the welding workpiece geometric model; The real-time heat flux density is input into the real-time prediction model of welding temperature field trained in step 3, and the predicted temperature value of each unit node on the welding workpiece geometric model at the current moment is obtained.

2. The method for real-time prediction of welding temperature field based on data-mechanism joint drive according to claim 1 is characterized in that: The welding parameters include welding current, arc voltage, material properties of the workpiece, and heat source parameters; the heat source parameters include the front semi-axis length of the heat source in the welding direction, the rear semi-axis length in the welding direction, the semi-axis length in the weld width direction, and the semi-axis length in the weld depth direction.

3. The method for real-time prediction of welding temperature field based on data-mechanism joint drive according to claim 2 is characterized in that: In step 1, the finite element analysis method is used to solve the temperature field during the welding process, which specifically includes the following steps: S11: Create a geometric model of the welding workpiece and define the geometric dimensions of the welding workpiece; S12: Setting material property data of the welding workpiece; the material property data includes thermal conductivity, density, latent heat and specific heat; wherein thermal conductivity and latent heat vary with temperature; S13: define the parameters of the analysis steps in the finite element solution to ensure the convergence of the algorithm; S14: define the initial temperature conditions and thermal convection boundary conditions of the welding workpiece; S15: Based on the welding parameters and heat source parameters, a subroutine is created to create a double ellipsoid moving heat source function to obtain the heat flux density of each unit node on the welding workpiece geometric model; S16: The heat flux density of each unit node on the welding workpiece geometric model is used to solve the temperature value of the corresponding unit node.

4. The method for real-time prediction of welding temperature field based on data-mechanism joint drive according to claim 3 is characterized in that: In the temperature field data set, the heat flux density of each unit node on the welding workpiece geometric model at the same moment is the accumulation of the heat flux density of the unit node corresponding to all historical moments before that moment; in the heat source parameter data set, the simulated molten pool surface width is obtained by calculating the distance between the unit nodes at the left and right edges of the welding workpiece geometric model whose temperature value is the metal melting point temperature.

5. The method for real-time prediction of welding temperature field based on data-mechanism joint drive according to claim 4 is characterized in that: Step 2 specifically includes: S21: Construct a real-time prediction model for heat source parameters based on long short-term memory network; the current moment is t, the welding current, arc voltage and simulated molten pool surface width at the moment [tn, t] are input into the model, where tn represents the first n sampling moments before the current moment, and n is an integer greater than or equal to 1; the predicted value of the heat source parameter at the current moment t is obtained based on regression; S22: Use the heat source parameter data set to train a real-time prediction model for heat source parameters, minimize the model loss function through a gradient descent algorithm, and then iteratively update the model weights to obtain the trained model weights; S23: Loading the trained model weights into the real-time prediction model of the heat source parameters, that is, the trained real-time prediction model of the heat source parameters.

6. A method for real-time prediction of welding temperature field based on data-mechanism joint drive according to claim 5, characterized in that Step three specifically includes: S31: construct a real-time prediction model for welding temperature field based on DeepONet neural operator; the real-time prediction model for welding temperature field includes a main network and a branch network; the main network inputs the heat flux density of each unit node on the welding workpiece geometric model in the temperature field data set, and the branch network inputs the coordinate value corresponding to the unit node; the outputs of the branch network and the main network are combined and output after performing a dot multiplication operation, and the output is the predicted temperature value of each unit node on the welding workpiece geometric model at the moment corresponding to the input data set; S32: Use the temperature field data set to train the real-time prediction model of the welding temperature field, optimize the network parameters by minimizing the mean square error between the predicted temperature value and the simulated temperature value through the gradient descent algorithm, and obtain the trained model weights; load the trained model weights to obtain the trained real-time prediction model of the welding temperature field.

7. The method for real-time prediction of welding temperature field based on data-mechanism joint drive according to claim 6 is characterized in that: In step 4, the welding current and arc voltage of the welding machine during the welding process are monitored in real time by an arc sensor; a high dynamic range HDR industrial camera is used to capture the molten pool surface image in real time; and a bandpass filter for suppressing arc light is installed on the lens of the industrial camera.

8. The method for real-time prediction of welding temperature field based on data-mechanism joint drive according to claim 7 is characterized in that: In the fourth step, the actual molten pool surface width is extracted from the real-time molten pool surface image by an image processing method, comprising the following steps: S41: Correct the melt pool surface image using the checkerboard method so that the pixel position in the melt pool surface image accurately matches the point position in the actual space; extract the region of interest DOI in the image; S42: Use the Retinex algorithm to perform image enhancement processing on the extracted image to obtain an image with a clearer and more obvious molten pool boundary; S43: Use Canny edge detection to detect edges in the image, and use morphological operations to make the image have continuous and complete edges; S44: Use the least squares method to fit the straight line to extract the left and right boundaries; calculate the actual molten pool surface width based on the pixel distance between the left and right boundaries.

9. The method for real-time prediction of welding temperature field based on data-mechanism joint drive according to claim 8 is characterized in that: In step 5, the monitoring data collected in real time is taken as the object, and after the data processing in step 4, it is input into the real-time prediction model of heat source parameters trained in step 2, specifically: the current time is t, the welding current, arc voltage collected at the time [tn, t] and the actual molten pool surface width obtained after image processing are input into the real-time prediction model of heat source parameters trained in step 3 to predict the real-time heat source parameters; The predicted real-time heat source parameters are brought into the double ellipsoid moving heat source function to obtain the heat flux density at each unit node on the welding workpiece geometric model at the current moment; The accumulated value is added to the historical heat flux density at all historical moments and sent to the main network of the real-time prediction model of the welding temperature field. The coordinate values ​​of all unit nodes on the welding workpiece geometric model in three-dimensional space are sent to the branch network of the real-time prediction model of the welding temperature field, so as to obtain the predicted temperature value of each unit node on the welding workpiece geometric model at the current moment.

Images