Electric arc additive manufacturing temperature field prediction method, device and system and storage medium

Abstract

The invention discloses an electric arc additive manufacturing temperature field prediction method, device and system and a storage medium. The method comprises the steps that S1, an electric arc additive manufacturing physical information temperature field data set is obtained; s2, obtaining an electric arc additive manufacturing temperature field prediction model based on physical information machine learning according to the electric arc additive manufacturing physical information temperature field data set; and S3, according to the electric arc additive manufacturing temperature field prediction model, transfer learning is carried out on electric arc additive manufacturing experiment data, and real-time prediction of the temperature field is achieved. By adopting the technical scheme of the invention, the problems of real-time performance and accuracy of temperature field prediction in the prior art are solved, and real-time monitoring and prediction of the temperature field in the rapid manufacturing process can be realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of arc additive manufacturing, and particularly relates to a method and device, system, and storage medium for predicting the temperature field in arc additive manufacturing. Background Art

[0002] Arc additive manufacturing, also known as arc directed energy deposition, is an advanced metal additive manufacturing technology that produces near-net-shaped parts by using an arc to melt wire materials. This process has high deposition efficiency and material utilization rate, low equipment cost, and can achieve the processing of high-strength materials. Therefore, it is widely used in fields such as aerospace, maritime, energy, and heavy industry, providing more feasible solutions for the production of complex parts.

[0003] Although arc has some advantages compared with other AM technologies, reducing the influence of the temperature field on the final properties of parts during the WAAM deposition process remains a challenge. For example, large thermal gradients can bring unacceptable residual stresses and thermal deformations, resulting in part failure, while small thermal gradients and cooling rates will reduce the hardness of parts. Therefore, studying the thermal behavior during the printing process, such as temperature distribution, gradient, heat accumulation, heat transfer mechanism, and its relationship with process parameters, is crucial for reducing defects in WAAM printed parts. The spatio-temporal distribution of temperature during the deposition process affects the formation of defects and the evolution of the microstructure, thus controlling the quality and performance obtained in the printed state. Therefore, rapid and accurate temperature field prediction is the key to ensuring high-quality production of WAAM. During the metal additive manufacturing process, accurate prediction of the temperature field is crucial for preventing overheating, adjusting process parameters, and ensuring process stability.

[0004] Traditional physical methods for predicting the temperature field in arc additive manufacturing include the finite element method, finite volume method, and finite difference method. Although the physical-based computational models are accurate, they are usually time-consuming and not suitable for real-time prediction and online control. In addition, numerical methods are also used for temperature field prediction, and some researchers use semi-analytical methods to solve the problems of low computational efficiency and complex boundary solving.

[0005] Although traditional physical and numerical-based computational models have high accuracy and interpretability, they are usually time-consuming and not suitable for real-time prediction and online control. Machine learning models, on the other hand, rely on high-quality datasets, which are costly and challenging to obtain in the field of arc additive manufacturing. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and device, system, and storage medium for predicting the temperature field in arc additive manufacturing, to solve the problems of real-time performance and accuracy of temperature field prediction in the prior art, and to enable real-time monitoring and prediction of the temperature field during the rapid manufacturing process.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] An arc additive manufacturing temperature field prediction method, comprising:

[0009] Step S1, obtaining an arc additive manufacturing physical information temperature field data set;

[0010] Step S2, based on the arc additive manufacturing physical information temperature field data set and physical information machine learning, obtaining an arc additive manufacturing temperature field prediction model;

[0011] Step S3, according to the arc additive manufacturing temperature field prediction model, performing transfer learning on the arc additive manufacturing experimental data to achieve real-time temperature field prediction.

[0012] Preferably, in step S1, according to the temperature field simulation model of the arc additive manufacturing process, finite element temperature field simulation analysis is performed, and the temperature field data, geometric features, and physical information at each time step are extracted to obtain the arc additive manufacturing physical information temperature field data set.

[0013] Preferably, the arc additive manufacturing temperature field prediction model is a discretized representation of the physical information, geometric features, and corresponding temperature field of the arc additive manufacturing process, and is used for spatio-temporal sequence prediction by a physical information deep learning model; a decoder is used to extract features and encode data for the input three matrices, multiple ConvLSTM units are used to extract spatio-temporal features, and the output result of the ConvLSTM module is decoded by the decoder into a temperature field matrix data format; the model parameters are iteratively updated through data loss and physical information loss to fit the mapping relationship between the input current physical information temperature field data time series matrix and the future time series temperature field matrix.

[0014] The present invention also provides an arc additive manufacturing temperature field prediction device, comprising:

[0015] An acquisition module, configured to acquire an arc additive manufacturing physical information temperature field data set;

[0016] A training module, configured to obtain an arc additive manufacturing temperature field prediction model based on the arc additive manufacturing physical information temperature field data set and physical information machine learning;

[0017] A prediction module, configured to perform transfer learning on the arc additive manufacturing experimental data according to the arc additive manufacturing temperature field prediction model to achieve real-time temperature field prediction.

[0018] Preferably, the acquisition module performs finite element temperature field simulation analysis according to the temperature field simulation model of the arc additive manufacturing process, and extracts the temperature field data, geometric features, and physical information at each time step to obtain the arc additive manufacturing physical information temperature field data set.

[0019] Preferably, the arc additive manufacturing temperature field prediction model discretizes the physical information, geometric features and corresponding temperature field of the arc additive manufacturing process, and uses the physical information deep learning model to perform spatiotemporal series prediction; a decoder is used to extract features and encode data of the three input matrices, and multiple ConvLSTM units are used to extract spatiotemporal features. The output results of the ConvLSTM module are decoded into a temperature field matrix data format through a decoder; the model parameters are iteratively updated through data loss and physical information loss to fit the mapping relationship between the input current physical information temperature field data time series matrix and the future time series temperature field matrix.

[0020] The present invention also provides an arc additive manufacturing temperature field prediction system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes an arc additive manufacturing temperature field prediction method when executed by the processor.

[0021] An embodiment of the present invention further provides a storage medium, on which a computer program is stored, and when the computer program is run, the method for predicting the temperature field of arc additive manufacturing is executed.

[0022] The present invention designs a suitable data structure to represent the physical information of the arc additive manufacturing process and its corresponding temperature field during the arc additive manufacturing process, and designs a model with physical significance to mine the mapping relationship between manufacturing status data and temperature field data. The present invention uses finite element analysis software to establish arc additive manufacturing temperature field data sets of various geometric shapes, establishes and trains an arc additive manufacturing temperature field prediction model based on deep learning of physical information, and can perform transfer learning on arc additive manufacturing to achieve real-time prediction of the temperature field. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0024] Figure 1 Flow chart of the method of the present invention

[0025] Figure 2 Flowchart for establishing a simulation-based temperature field dataset for arc additive manufacturing process

[0026] Figure 3 Generate arc additive manufacturing process simulation model and mesh division for the example

[0027] Figure 4 Structural diagram of the physical information deep learning model in the embodiment

[0028] Figure 5 Visualization of the results predicted by the model after training based on simulation data in the embodiment

[0029] Figure 6 Visualization of the result error of the model predicted after training based on simulation data in the embodiment

[0030] Figure 7 Visualization of the results predicted by the model after training with transfer learning based on experimental data in the embodiment

[0031] Figure 8 Visualization of the result error of the model predicted after training with transfer learning based on experimental data in the embodiment Detailed implementation manners

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0033] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0034] Embodiment 1:

[0035] As Figure 1 shown, a method for predicting the temperature field of arc additive manufacturing in an embodiment of the present invention includes:

[0036] Step S1, establish a temperature field simulation model for the arc additive manufacturing process, perform finite element temperature field simulation analysis using finite element analysis software, extract the temperature field data, geometric features, and physical information at each time step, and establish an arc additive manufacturing physical information temperature field dataset

[0037] Taking a single-pass multi-layer thin-wall structure as an example, establish a temperature field simulation model for the arc additive manufacturing process in ANSYS, perform finite element simulation analysis, extract the deposition state data, temperature field data, and physical information data at each simulation step time, and establish a temperature field dataset for the arc additive manufacturing process. The process is as Figure 2 shown, including the following steps:

[0038] 1.1), establish a temperature field simulation model for arc additive manufacturing: establish a temperature field simulation model for the single-pass multi-layer thin-wall structure in the arc additive manufacturing space. The specific steps are as follows:

[0039] 1.1.1). Determine the process parameters of arc additive manufacturing, including process parameters, materials, geometries, etc. The specific arc current is 109 A, the arc voltage is 10.9 V, the arc moving speed is 8 mm / s, and the arc thermal efficiency is 80%; the wire diameter is 1.2 mm, the wire feeding speed is 4.9 m / min, and the wire material is ER70s-6; the length of the single-pass multi-layer thin-walled structure is 100 mm, the width is 5 mm, the height is 40 mm, and the number of layers is 20; the substrate geometry is square, the length is 300 mm, the thickness is 10 mm, and the substrate material is Q235b.

[0040] 1.1.2). Establish a finite element model for single-pass multi-layer thin-walled arc additive manufacturing, and determine the deposition path of arc additive manufacturing. The finite element model and deposition path of arc additive manufacturing are as Figure 3 shown. Mesh the finite element model of arc additive manufacturing. The formed part is meshed with hexahedron meshes, and the mesh size is 2×2×2.5 mm 3 , and the substrate is meshed with tetrahedron free meshes.

[0041] 1.1.3). Determine the heat source of the temperature field simulation model: Considering the influence of heat source movement on the temperature field distribution, select the double ellipsoidal heat source as the heat source of the temperature field simulation model. The heat flux density distribution functions q 1 and q 2 of its front and rear semi-ellipsoids are calculated as follows:

[0042]

[0043] f 1 +f 2 =2

[0044] Q = ηUI

[0045] where a, b 1 , b 2 , c are the shape parameters of the double ellipsoidal heat source; f 1 and f 2 are the energy ratios of the front and rear semi-ellipsoids respectively, f 1 =0.6, f 2 =1.4. Q is the arc power, which is calculated through the relational formula. η, U, and I are the arc thermal efficiency, voltage, and current respectively.

[0046] 1.1.4). Calibrate the shape parameters of the double ellipsoidal heat source through monitoring a single-pass experiment with a thermal imager. Install the thermal imager off-axis and take pictures at the position where the temperature field data is the richest to monitor the area where the temperature is higher than the melting point, which is the molten pool. Compare and determine the shape parameters of the molten pool, a = 2 mm, b 1 =2 mm, b 2a = 4 mm, c = 2 mm.

[0047] 1.1.5), Determine the time step sequence of the temperature field simulation model: Based on the wire arc additive manufacturing process, determine the time step sequence, and determine the entity elements that need to be activated for each time step in the time step sequence. To meet the requirements of the activation area of the double ellipsoidal heat source, more elements need to be activated at the beginning of each layer time step.

[0048] 1.2), Finite element model calculation: Set the initial conditions and boundary conditions of the substrate and the formed part. The heat exchange between the formed part and the surrounding environment includes heat convection and heat radiation; Set the time step as the time required to deposit one grid, and set the interlayer cooling after each layer is deposited.

[0049] 1.3), Design the data structure: Extract the geometric feature data, physical information data, and temperature field data at each time step of the temperature field simulation model. The geometric feature matrix and the temperature field matrix are represented by the activation order data of the unit set of the temperature field simulation model and the point set temperature data. One unit corresponds to four nodes, that is, these 4 points are in the activated state.

[0050] 1.3.1), The length of the square matrix is determined in the y-z direction of the formed part. The size of the matrix is 51×51, the spatial resolution is 2 mm, and the time resolution is 0.25 s. Three matrices can be collected in the x direction, and a total of 9 matrices can be collected in one time step.

[0051] 1.3.2), Extract the calculation result file of the finite element model to obtain the temperature value of each node of the formed part at each time step, and fill the temperature field matrix of each time step. For the nodes where the temperature data is not activated, set the temperature data as the ambient temperature.

[0052] 1.3.3), The geometric feature data is based on the activation state of the unit. If the unit is activated, set the value of the corresponding node to 1, otherwise set its value to 0. Based on the physical information data represented in step 1.2.3), determine the position of the heat source and the applied load node sets of the front and rear ellipsoids according to the activation order, and calculate the magnitude and position of the applied load for each node.

[0053] 1.3.3), The temperature field matrix data set Tt, geometric feature matrix data set Gt, and physical information matrix data set St for each time step can be obtained through sufficient extraction.

[0054] 1.4), Establish the physical information temperature field data set of wire arc additive manufacturing to obtain three different forms of matrix data sets. Based on the data structure designed in step 1.5), extract the activation order of the node set and the node set temperature data, geometric feature data, and physical information data of the formed part at each time step, and construct the data set {(Tt, Gt, St)} of the temperature field simulation model.

[0055] Step S2: Based on the physical information temperature field dataset of the arc additive manufacturing process obtained from simulation, establish and train a deep learning model with physical information, and use the PyTorch library to establish and train the deep learning model with physical information

[0056] The specific steps are as follows:

[0057] 2.1) Define the input data and label data of the model: The dataset of the deep learning model with physical information is generated based on the temperature field dataset of the arc additive manufacturing process obtained from simulation, and is further divided into a training set and a test set. The specific steps are as follows:

[0058] 2.1.1) Define the input data and label data of the dataset of the deep learning model with physical information: The model predicts the temperature field matrix data of the next 5 time steps based on the temperature field matrix data, geometric feature matrix data, and physical information matrix data of the first 5 time steps before arc additive manufacturing.

[0059] 2.1.2) Divide the dataset of the deep learning model with physical information to obtain a training set and a test set: Matrix data is extracted with a window of 10 time steps to establish the model dataset. The data of the first 15 layers is divided into the training set, and the data of the last 5 layers is divided into the prediction set.

[0060] 2.2) Design a machine learning model with physical information, which specifically includes the following steps:

[0061] 2.2.1) The machine learning model with physical information consists of a decoder, an encoder, a ConvLSTM layer, and a physical information loss function. Each ConvLSTM layer is composed of two ConvLSTM units. The structure of the model is as Figure 4 shown

[0062] 2.2.2) Normalize the input matrix data: Select the maximum and minimum values of the entire sequence for normalization so that the temperature field matrix Tt and the physical information matrix St in the arc additive manufacturing process are mapped to the interval [0, 1]. As shown in the following formula:

[0063]

[0064] where, T max and T min respectively represent the highest temperature value and the lowest temperature value in the temperature field of all time steps; S max and S min respectively represent the maximum and minimum load values in the physical information matrix of all time steps, and represent the normalized matrix data.

[0065] 2.2.3) The geometric feature matrix sequence is averaged as shown in the following formula:

[0066]

[0067] where G 1 is the geometric feature matrix at the second time step, and G m is the geometric feature matrix at the penultimate time step. is used to represent the averaged processing matrix of the input geometry from the first to the penultimate sequence. Therefore, the original geometric matrix sequence from G 1 to G m is averaged to

[0068] 2.2.4) The and sequence matrices are used as the input matrices of the model. The number of matrices is 15, which is used to predict the temperature field in the next 5 time steps.

[0069] 2.2.5) The input matrix sequence undergoes feature fusion through the encoding layer to extract important features.

[0070] 2.2.6) Two ConvLSTM units are used to capture spatial features and temporal dependencies of the processed matrix sequence.

[0071] 2.2.7) The decoder restores the output result of the ConvLSTM layer to the original data format, which is to convert it into the temperature field matrix format here.

[0072] 2.3) After completing the establishment of the physics-informed deep learning model and its dataset, the model is trained. The specific steps are as follows:

[0073] 2.3.1) Select the mini-batch method for training, and set the batch size to 8.

[0074] 2.3.2) Establish a loss function: The physics-informed loss function usually includes two core elements: data-based loss and physics-based loss. Due to the hard enforcement of the I / BCs conditions, for the physics-based loss, we only need to control the partial differential equation to construct the loss function. The deviation between the prediction and the partial differential equation is quantified through the residual of the partial differential equation. The partial differential equation loss can be represented by a difference equation. The difference form of the partial differential equation is:

[0075]

[0076] where T(x i , y j ) is the temperature at the node in the predicted temperature field at (x i , yj ) The temperature value at; T t+1 (x i , y j ) and T t-1 (x i , y j ) are the temperature values of the node at (x i , y j ) in the previous and the next time steps respectively. In the ideal case, the output temperature of the model should make R(x i , y j ) close to zero. The physical constraints can be imposed on the model by minimizing the loss function L pde . L pde is equivalent to the mean square of the residual of the partial differential equation in the spatio-temporal discretization, and can be expressed as:

[0077]

[0078] where n and m represent the height and width of the output matrix respectively, and t is the total number of time steps for prediction.

[0079] The data loss L data can be expressed as:

[0080]

[0081] where, is the predicted temperature field matrix, and T(x i , y j ) is the true temperature field matrix. Therefore, the total loss L total of the deep learning of physical information can be expressed as:

[0082] L total = w p L pde + w d L data

[0083] 2.3.3), Calculate the total loss function value of the batch data, set the learning rate to 0.001, and train through the Stochastic Gradient Descent with Adaptive Moment Estimation optimizer.

[0084] 2.3.4), The training process stops after 3000 epochs, save all the parameters after the model is trained, and the entire training process only takes 3 minutes. To further evaluate the performance of the deep learning model of physical information, the mean squared error is tested and recorded on the test set every 500 epochs, and the mean squared error of each time step is tested.

[0085]

[0086] The comparison after visualizing the two sequences of real data and model prediction data in the test set is as Figure 5 shown. The first row is the real data, and the second row is the predicted data. It can be seen that the simulation results and the predicted results have good consistency. In addition, the error visualization analysis of the real data and model prediction data of these two sequences is as Figure 6 shown. The first row is the error of sequence 1, and the second row is the error of sequence 2. It can be seen that the maximum error of the predicted results is concentrated in the heat source area, and the maximum error is 8.7%. There are smaller errors in other areas. The prediction of the model is in milliseconds, meeting the accuracy and real-time requirements of real-time prediction.

[0087] Step S3: After the training and saving of the physics-informed machine learning model, predict the temperature field during the arc additive manufacturing experiment

[0088] Specifically, it includes the following:

[0089] 3.1) Conduct an arc additive manufacturing deposition experiment with the same process parameters, and use an infrared thermal imager to monitor the temperature field during the process.

[0090] 3.2) The infrared thermal imager is installed on the side that can capture the most temperature field data, which has the same effect as the data collected in the simulation. The temperature field data of the node set is extracted through post-processing software to obtain the required physical information temperature field dataset of the arc additive manufacturing experiment.

[0091] Since there will be a deviation between the spatio-temporal resolution of the infrared thermal imager and the experiment, a square area needs to be extracted when processing the temperature field matrix sequence data. This area will present the temperature field of the formed part as much as possible, and Resize processing is required to make the size of the square matrix 51×51, so that the spatial resolution is 2mm. The sampling frequency of the infrared thermal imager is set to 6.25Hz, and the time resolution is 0.16s.

[0092] 3.3) Use transfer learning with the model trained by simulation data. Input the collected physical information temperature field matrix data of the first 15 layers into the physics-informed deep learning model for training, and input the data collected by the subsequent layers and preprocess them into the model to verify the prediction effect of the model.

[0093] The comparison after visualizing the real data and model prediction data of a sequence is as Figure 7 shown. The first row is the real data, and the second row is the predicted data. It can be seen that the predicted results and the experimental results have good consistency. In addition, the error visualization analysis of the real data and model prediction data of this sequence is as Figure 8As shown, it can be seen that the maximum errors of the prediction results are concentrated in the heat source area, with a maximum error of 13.2%, and smaller errors in other areas. The prediction of the model is at the millisecond level, meeting the accuracy and real-time requirements of real-time prediction.

[0094] The present invention realizes the discretized representation of the physical information, geometric features, and corresponding temperature field of the arc additive manufacturing process, enabling it to be used for spatio-temporal sequence prediction by a physical information deep learning model. It fully considers the spatio-temporal correlation between discrete input data, uses a decoder to extract features and encode data for three input matrices, utilizes multiple ConvLSTM units to extract spatio-temporal features, and finally decodes the output result of the ConvLSTM module into the temperature field matrix data format by the decoder. The model parameters are iteratively updated through data loss and physical information loss to fit the mapping relationship between the input current physical information temperature field data time series matrix and the future time series temperature field matrix. The predicted temperature field matrix has a high degree of coincidence with the actual temperature field matrix. Also, since the model is pre-trained with simulation data before actual experiments, the present invention has fast-converging data and high prediction accuracy during the experiment process, and can meet the needs of real-time prediction at the same time. When predicting the temperature field during the arc additive manufacturing experiment process, the present invention only needs to collect the temperature field time series matrix data set corresponding to this process, input this data set and other feature matrices into the model trained with simulation data for training, and predict the temperature field at future time steps through the data collected next.

[0095] Embodiment 2:

[0096] The embodiment of the present invention further provides an arc additive manufacturing temperature field prediction device, including:

[0097] An acquisition module, configured to acquire an arc additive manufacturing physical information temperature field data set;

[0098] A training module, configured to obtain an arc additive manufacturing temperature field prediction model based on physical information machine learning according to the arc additive manufacturing physical information temperature field data set;

[0099] A prediction module, configured to perform transfer learning on the arc additive manufacturing experimental data according to the arc additive manufacturing temperature field prediction model to achieve real-time temperature field prediction.

[0100] As an implementation manner of the embodiment of the present invention, the acquisition module performs finite element temperature field simulation analysis according to the temperature field simulation model of the arc additive manufacturing process, extracts the temperature field data, geometric features, and physical information at each time step, and obtains the arc additive manufacturing physical information temperature field data set.

[0101] As an implementation mode of the embodiment of the present invention, the temperature field prediction model for arc additive manufacturing is a discretized representation of the physical information, geometric features, and corresponding temperature field of the arc additive manufacturing process, and is used for spatio-temporal sequence prediction by the physical information deep learning model; a decoder is used to extract features and encode data for the three input matrices, multiple ConvLSTM units are used to extract spatio-temporal features, and the decoder decodes the output result of the ConvLSTM module into the temperature field matrix data format; the model parameters are iteratively updated through data loss and physical information loss to fit the mapping relationship between the input current physical information temperature field data time series matrix and the future time series temperature field matrix.

[0102] Embodiment 3:

[0103] The embodiment of the present invention further provides an arc additive manufacturing temperature field prediction system, including: a memory and a processor, where a computer program run by the processor is stored on the memory, and the computer program executes the arc additive manufacturing temperature field prediction method when run by the processor.

[0104] Embodiment 4:

[0105] The embodiment of the present invention further provides a storage medium, where a computer program is stored on the storage medium, and the computer program executes the arc additive manufacturing temperature field prediction method when running.

[0106] The above-described embodiments are only descriptions of the preferred modes of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for predicting temperature field in arc additive manufacturing, characterized in that it includes: Step S1, obtaining a temperature field data set of arc additive manufacturing physical information; Step S2, obtaining an arc additive manufacturing temperature field prediction model based on physical information machine learning according to the arc additive manufacturing physical information temperature field data set; Step S3: Based on the arc additive manufacturing temperature field prediction model, transfer learning is performed on the arc additive manufacturing experimental data to achieve real-time prediction of the temperature field.

2. The arc additive manufacturing temperature field prediction method as described in claim 1 is characterized in that, in step S1, a finite element temperature field simulation analysis is performed based on the arc additive manufacturing process temperature field simulation model, and the temperature field data, geometric features and physical information of each time step are extracted to obtain an arc additive manufacturing physical information temperature field data set.

3. The arc additive manufacturing temperature field prediction method as described in claim 2 is characterized in that the arc additive manufacturing temperature field prediction model discretizes the physical information, geometric characteristics and corresponding temperature field of the arc additive manufacturing process, and is used for the physical information deep learning model to perform spatiotemporal series prediction; a decoder is used to extract features and encode data of the three input matrices, and multiple ConvLSTM units are used to extract spatiotemporal features. The output results of the ConvLSTM module are decoded into a temperature field matrix data format through the decoder; the model parameters are iteratively updated through data loss and physical information loss to fit the mapping relationship between the input current physical information temperature field data time series matrix and the future time series temperature field matrix.

4. An arc additive manufacturing temperature field prediction device, characterized in that it includes: An acquisition module is used to acquire a temperature field data set of arc additive manufacturing physical information; A training module is used to obtain an arc additive manufacturing temperature field prediction model based on the arc additive manufacturing physical information temperature field data set and physical information machine learning; The prediction module is used to perform transfer learning on arc additive manufacturing experimental data according to the arc additive manufacturing temperature field prediction model to achieve real-time prediction of the temperature field.

5. The arc additive manufacturing temperature field prediction device as described in claim 4 is characterized in that the acquisition module performs finite element temperature field simulation analysis based on the arc additive manufacturing process temperature field simulation model, extracts the temperature field data, geometric features and physical information of each time step, and obtains the arc additive manufacturing physical information temperature field data set.

6. The arc additive manufacturing temperature field prediction device as described in claim 5 is characterized in that the arc additive manufacturing temperature field prediction model discretizes the physical information, geometric characteristics and corresponding temperature field of the arc additive manufacturing process, and is used for the physical information deep learning model to perform spatiotemporal series prediction; a decoder is used to extract features and encode data of the three input matrices, and multiple ConvLSTM units are used to extract spatiotemporal features. The output results of the ConvLSTM module are decoded into a temperature field matrix data format through the decoder; the model parameters are iteratively updated through data loss and physical information loss to fit the mapping relationship between the input current physical information temperature field data time series matrix and the future time series temperature field matrix.

7. An arc additive manufacturing temperature field prediction system, characterized in that: include: A memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the arc additive manufacturing temperature field prediction method according to any one of claims 1 to 3 is executed.

8. A storage medium, characterized in that: The storage medium stores a computer program, which, when running, executes the arc additive manufacturing temperature field prediction method according to any one of claims 1 to 3.

Images