Method for predicting temperature field in cold metal transition technology additive manufacturing process based on deep learning

Abstract

The invention provides a method for predicting a temperature field in a cold metal transition technology additive manufacturing process based on deep learning. According to the method, the problems of long acquisition period and insufficient quantity of the deep learning data set are solved. The method mainly comprises the following steps of: 1, establishing a finite element model in the CMT electric arc additive manufacturing process, and comparing a result obtained by simulation with experimental data to check the accuracy of the model; 2, creating an APDL command stream template file containing a basic simulation process on the basis of the checked finite element model, wherein key parameters are represented by placeholders; 3, Python and Ansys are subjected to joint simulation, and calculation, extraction and storage of CMT welding temperature cloud atlas data under different technological parameters are achieved; and 4, enhancing existing temperature cloud picture data by using a GAN (Generative Adversarial Network) model, forming a deep learning data set, inputting the deep learning data set into the ResNet model for training, verification and testing, and finally applying the trained model to online prediction of the electric arc additive manufacturing temperature field. The method is suitable for batch calculation of the welding temperature field under any welding method, simulation analysis efficiency is improved, a deep learning data set is expanded, deep learning is fully trained, and the generalization ability of deep learning is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of arc additive manufacturing, and in particular to a method for predicting the temperature field in the additive manufacturing process of cold metal transfer technology based on deep learning. Background Technique

[0002] Arc additive manufacturing is a technology that melts filamentous metal materials through an arc and stacks them layer by layer to form a shape. During the stacking process, as the number of stacked layers increases, heat accumulation gradually becomes severe, while the heat dissipation conditions gradually deteriorate. Severe heat accumulation will cause the molten pool to flow, resulting in local collapse of the additive part, leading to a decrease in the dimensional accuracy and surface quality of the additive part. Severe interlayer heat accumulation will also lead to the problem of grain coarsening, and even generate structures that are unfavorable to performance, thereby reducing the mechanical properties of the additive part. By predicting the temperature field, temperature anomaly regions that may cause defects can be detected in advance, realizing the early prediction and prevention of defects, and improving the quality and reliability of the manufactured parts.

[0003] Deep learning can automatically extract the most valuable features for temperature field prediction from a large amount of raw data, automatically learn and capture the complex non-linear relationships between various influencing factors in the additive manufacturing process, so as to realize the prediction of the temperature field. Compared with traditional numerical calculation methods for temperature field prediction, such as the finite element method, etc., the calculation time is greatly reduced, providing the possibility for real-time monitoring and feedback control.

[0004] With the help of technologies such as edge computing and cloud computing, deep learning models can be deployed on devices at the production site or cloud servers to realize real-time online prediction of the temperature field in arc additive manufacturing. By combining with the data collected in real time by sensors, the change of the temperature field can be monitored in time, providing timely decision-making support for process adjustment and quality control, and helping to improve production efficiency and product quality.

[0005] Through the above analysis, the problems and defects existing in the prior art are as follows:

[0006] 1. At present, there are still some deficiencies in the method of using deep learning for temperature field prediction in arc additive manufacturing. Among them, the problems in the dataset are particularly prominent. The methods for obtaining deep learning datasets mainly include experimental measurement, numerical simulation, and data enhancement of existing data using algorithms.

[0007] 2. Obtaining the temperature field data of arc additive manufacturing under different process parameters, materials, and component shapes through experimental means is often costly and time-consuming, thus limiting the richness and comprehensiveness of the data. This method of obtaining data is costly and time-consuming, and the obtained data is limited. Moreover, the temperature measurement equipment may also be affected by factors such as arc strong light, high temperature, and noise, resulting in inaccurate or incomplete measurement data.

[0008] 3. Numerical simulation methods are often used to study the distribution laws of thermal fields, stress fields, and flow fields during wire and arc additive manufacturing, so as to predict temperature distribution, deformation amount, surface topography, and dimensional accuracy, providing a theoretical basis for the optimization of wire and arc additive manufacturing processes. However, this method requires a certain simplification of the actual wire and arc additive manufacturing process, resulting in a deviation between the simulation results and the actual results. Moreover, numerical simulation often requires a large amount of time for calculation, data extraction, and analysis, reducing production efficiency. Summary of the Invention

[0009] Aiming at the problems existing in the prior art, the present invention provides a method for predicting the temperature field during the additive manufacturing process of cold metal transfer technology based on deep learning.

[0010] A method for predicting the temperature field during the additive manufacturing process of cold metal transfer technology based on deep learning includes the following steps:

[0011] Step 1: Conduct an arc additive experiment through a CMT arc welding platform, and collect and obtain the single-pass welding temperature field distribution under a certain process parameter through an infrared thermal imager.

[0012] Step 2: Establish a finite element model for the CMT arc additive manufacturing process, and use the same parameters as in Step 1 for the welding process parameters. Complete the finite element simulation and save the result file.

[0013] Step 3: Compare the temperature field results collected in Step 1 with the calculation results in Step 2 to verify the accuracy of the finite element model.

[0014] Step 4: Create an APDL command stream template file containing the basic simulation process on the basis of the verified finite element model, where the key parameters are represented by placeholders, and save it as a template file.

[0015] Step 5: Use Python to read the APDL template file, replace the placeholders with specific parameter values, and generate multiple APDL command stream files; use the subprocess module to drive Ansys to execute these APDL files for batch calculation.

[0016] Step 6: Use Python software to read the simulation result files under each parameter, obtain the temperature nephograms under different process parameters during the arc additive manufacturing process, and perform preprocessing of image enhancement and normalization on the temperature nephograms.

[0017] Step 7: Build a GAN model, use a noise vector as the input, and let the generator learn the distribution of real welding temperature field data through training, so as to generate a large number of new welding temperature field data similar to the real data.

[0018] Step 8: Merge the preprocessed cloud map data and the data generated by GAN as the augmented data set, establish a deep learning database for the temperature field, and divide the data set into a training set, a validation set, and a test set;

[0019] Step 9: In the Python environment, use a deep learning framework to build a ResNet model, input the training data into the model batch by batch, and perform training, validation, and testing;

[0020] Step 10: Use the trained model to predict the temperature field of arc additive manufacturing.

[0021] Further, the key process parameters include welding current, welding voltage, wire feeding speed, welding speed, deposition path, and interlayer temperature.

[0022] Further, automate the extraction and processing of simulation result data through the Python language under the Pytorch framework; during the arc additive manufacturing process, for the same layer of weld bead, when the process parameters are fixed, the temperature distribution should be quasi-steady state. Therefore, extract the instantaneous temperature cloud map at the middle moment of each weld bead in each simulation model to establish deep learning data.

[0023] Further, since the ResNet model is based on labeled data for supervised learning, according to the simulation conditions, label and classify each extracted temperature distribution cloud map data; the labeling information includes welding current, welding voltage, wire feeding speed, welding speed, deposition path, and interlayer temperature.

[0024] Further, the method for image enhancement of the temperature cloud map includes a generative adversarial network (GAN); according to the existing welding temperature field data distribution, use GAN to generate a large number of new, similar but not exactly the same temperature field data, thereby augmenting the data set, providing richer training samples for the prediction model, reducing the risk of model overfitting, and improving the generalization ability of the model.

[0025] Further, merge the preprocessed cloud map data and the data generated by GAN as the augmented data set, establish a deep learning database for the temperature field, and divide the data set into a training set, a validation set, and a test set, where the training set accounts for 70%, used for model training; the validation set accounts for 15%, used to adjust the hyperparameters of the model, monitor the training process of the model, and prevent overfitting; the test set accounts for 15%, used to evaluate the final performance of the model to ensure that the model has good generalization ability.

[0026] Further, the ResNet model is a deep network that can learn complex features using a large amount of data, avoid overfitting, and fully explore the temperature change laws in the data.

[0027] Furthermore, combining GAN and ResNet and applying them to the prediction of the welding temperature field can give full play to the data generation ability of GAN and the feature extraction advantage of ResNet.

[0028] Another object of the present invention is to provide a hardware facility for an on-line temperature field prediction system based on cold metal transfer arc additive manufacturing process, including:

[0029] A cold metal transfer arc additive manufacturing work platform, a protective gas, and a temperature acquisition system;

[0030] The cold metal transfer arc additive manufacturing work platform includes a cold metal transfer digital welding system, an industrial robot, and a positioner;

[0031] Among them, the temperature sensor is a thermal imager temperature acquisition method.

[0032] Combined with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by the present invention are as follows:

[0033] 1. A method for on-line predicting the temperature in the additive manufacturing process of cold metal transfer welding technology based on deep learning. This system combines Python and Ansys co-simulation for batch calculation and data extraction, solves the problems that the traditional finite element analysis method is cumbersome in manually setting model parameters, cannot perform batch simulation, resulting in too long construction time and insufficient data for the deep learning data set, and effectively solves the problems of high computing resources and time costs.

[0034] Utilizing the powerful programming function of Python and the advantage of the APDL command flow file of Ansys, more influencing factors can be considered in the finite element simulation of arc additive manufacturing. Represent the values of different process parameters with placeholders, and then use Python to form different combinations of process parameters and perform calculations. Through the comprehensive analysis of multiple influencing factors, the deep learning model can better establish the non-linear relationship between various influencing factors and improve the accuracy of temperature field prediction.

[0035] 2. Combining the GAN and ResNet models and applying them to temperature field prediction can give full play to the data generation ability of GAN and the feature extraction advantage of ResNet, and improve the prediction accuracy and reliability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It shows a schematic diagram of the basic process of an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0037] To make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0038] For a specific embodiment of the method for online predicting temperature during the additive manufacturing process of cold metal transfer welding technology through deep learning, it can be described based on the above features and working principles. The following are the embodiments:

[0039] As Figure 1 shown, a method for online predicting temperature during the additive manufacturing process of cold metal transfer welding technology based on deep learning provided by an embodiment of the present invention is as follows:

[0040] S101: Conduct an arc additive experiment through a CMT arc welding platform, and collect and obtain the single-pass welding temperature field distribution under a certain process parameter through an infrared thermal imager;

[0041] S102: Establish a finite element model for the CMT arc additive manufacturing process, use the same parameters as in S101 for the welding process parameters, complete the finite element simulation and save the result file;

[0042] S103: Compare the temperature field results collected in S101 with the calculation results in S102 to verify the accuracy of the finite element model;

[0043] S104: Create an APDL command flow template file containing the basic simulation process on the basis of the calibrated finite element model, use placeholder symbols for the key parameters therein, and save it as a template file;

[0044] S105: Use Python to read the APDL template file, replace the placeholder symbols with specific parameter values to generate multiple APDL command flow files; use the subprocess module to drive Ansys to execute these APDL files for batch calculation;

[0045] S106: Use Python software to read the simulation result files under each parameter, obtain the temperature nephograms under different process parameters during the arc additive manufacturing process, and perform preprocessing of image enhancement and normalization on the temperature nephograms;

[0046] S107: Construct a GAN model, use the noise vector as the input, and let the generator learn the distribution of real welding temperature field data through training, so as to generate a large number of new welding temperature field data similar to the real data;

[0047] S108: Combine the preprocessed nephogram data and the data generated by GAN as the expanded data set, establish a deep learning database for the temperature field, and divide the data set into a training set, a validation set and a test set;

[0048] S109: In a Python environment, use a deep learning framework to build a ResNet model, input the training data into the model in batches, and perform training, validation, and testing.

[0049] S110: Use the trained model to predict the temperature field of arc additive manufacturing.

[0050] The key process parameters provided in the embodiments of the present invention include welding current, welding voltage, wire feeding speed, welding speed, deposition path, and interlayer temperature.

[0051] The embodiments of the present invention provide the automatic extraction and processing of simulation result data through the Python language under the Pytorch framework; during the arc additive manufacturing process, for the same layer of weld bead, when the process parameters are fixed, the temperature distribution should be quasi-steady state. Therefore, extract the instantaneous temperature contour maps at the intermediate moment of each weld seam in each simulation model to establish deep learning data.

[0052] Since the ResNet model in the embodiments of the present invention performs supervised learning based on labeled data, according to the simulation conditions, label and classify the temperature distribution contour map data extracted for each; the labeling information includes welding current, welding voltage, wire feeding speed, welding speed, deposition path, and interlayer temperature.

[0053] The method for image enhancement of the temperature contour map provided in the embodiments of the present invention includes a generative adversarial network (GAN); according to the existing welding temperature field data distribution, use GAN to generate a large number of new, similar but not exactly the same temperature field data, thereby expanding the data set, providing richer training samples for the prediction model, reducing the risk of model overfitting, and improving the generalization ability of the model.

[0054] The embodiments of the present invention provide that the preprocessed contour map data and the data generated by GAN are combined as the expanded data set, and a temperature field deep learning database is established. The data set is divided into a training set, a validation set, and a test set, where the training set accounts for 70% and is used for model training; the validation set accounts for 15% and is used to adjust the hyperparameters of the model, monitor the training process of the model, and prevent overfitting; the test set accounts for 15% and is used to evaluate the final performance of the model to ensure that the model has good generalization ability.

[0055] The ResNet model provided in the embodiments of the present invention is a deep network that can learn complex features using a large amount of data, avoid overfitting, and fully explore the temperature change laws in the data.

[0056] The combination of GAN and ResNet applied to the prediction of welding temperature field provided by the embodiments of the present invention can give full play to the data generation ability of GAN and the feature extraction advantage of ResNet.

[0057] An on-line temperature field prediction system for CMT arc additive manufacturing process provided by the embodiments of the present invention, the hardware facilities of the on-line temperature field prediction system for CMT arc additive manufacturing process include:

[0058] CMT arc additive manufacturing work platform, protective gas, temperature acquisition system;

[0059] The CMT arc additive manufacturing work platform includes a CMT digital welding system, an industrial robot, and a positioner;

[0060] Among them, the temperature sensor is a thermal imager temperature acquisition method.

[0061] The above are the specific implementation manners of the present invention, but the protection scope of the present invention should not be limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope defined by the claims.

Claims

1. A method for predicting the temperature field of the additive manufacturing process of cold metal transition technology based on deep learning, characterized in that: The following steps are involved: Step 1: Perform arc additive experiments on the CMT arc welding platform, and use a thermal imager to collect and obtain the single-pass welding temperature field distribution under certain process parameters; Step 2: Establish a finite element model of the CMT arc additive manufacturing process. The welding process parameters use the same parameters as in step 1. Complete the finite element simulation and save the result file. Step 3: Compare the temperature field results collected in step 1 with the calculation results in step 2 to verify the accuracy of the finite element model; Step 4: Create an APDL command flow template file containing the basic simulation process based on the verified finite element model. The key parameters are represented by placeholders and saved as a template file. Step 5: Use Python to read the APDL template file, replace the placeholders with specific parameter values, and generate multiple APDL command flow files; Use the subprocess module to drive Ansys to execute these APDL files for batch calculations; Step 6: Use Python software to read the simulation result file under each parameter, obtain the temperature cloud map under different process parameters in the arc additive manufacturing process, and perform image enhancement and normalization preprocessing on the temperature cloud map; Step 7: Build a GAN model, take the noise vector as input, and train the generator to learn the distribution of the real welding temperature field data, thereby generating a large amount of new welding temperature field data similar to the real data; Step 8: Combine the preprocessed cloud image data and the data generated by GAN as the expanded data set, establish a temperature field deep learning database, and divide the data set into training set, validation set, and test set; Step 9: In the Python environment, use the deep learning framework to build the ResNet model, input the training data into the model in batches, and perform training, verification, and testing; Step 10: Use the trained model to predict the temperature field of arc additive manufacturing.

2. The method according to claim 1, characterized in that The key process parameters include welding current, welding voltage, wire feeding speed, welding speed, deposition path, and interlayer temperature.

3. The method according to claim 1, characterized in that: The simulation result data is automatically extracted and processed using Python language under the Pytorch framework. During the arc additive manufacturing process, for the same layer of welds, the temperature distribution should be quasi-steady-state when the process parameters are fixed. Therefore, the instantaneous temperature cloud map of each weld in the middle of each simulation model is extracted to establish deep learning data.

4. The method according to claim 1, characterized in that: Since the ResNet model is supervised learning based on labeled data, each extracted temperature distribution cloud map data is labeled and classified according to the simulation conditions; the labeled information includes welding current, welding voltage, wire feeding speed, welding speed, deposition path, and interlayer temperature.

5. The method according to claim 1, characterized in that The method of image enhancement for temperature cloud map includes generative adversarial network (GAN); according to the existing distribution of welding temperature field data, GAN is used to generate a large amount of new, similar but not completely the same temperature field data, thereby expanding the data set, providing more abundant training samples for the prediction model, reducing the risk of model overfitting, and improving the generalization ability of the model.

6. The method according to claim 1, characterized in that The preprocessed cloud map data and the data generated by GAN are combined as an expanded data set to establish a temperature field deep learning database. The data set is divided into a training set, a validation set and a test set, of which the training set accounts for 70% and is used for model training; the validation set accounts for 15% and is used to adjust the hyperparameters of the model, monitor the training process of the model, and prevent overfitting; the test set accounts for 15% and is used to evaluate the final performance of the model to ensure that the model has good generalization ability.

7. The method according to claim 1, characterized in that The ResNet model is a deep network that can use a large amount of data to learn complex features, avoid overfitting, and fully explore the temperature change patterns in the data.

8. The method according to claim 1, characterized in that: Combining GAN and ResNet for welding temperature field prediction can give full play to the data generation capability of GAN and the feature extraction advantages of ResNet.

9. An online temperature field prediction system based on CMT arc additive manufacturing process according to any one of claims 1 to 8, characterized in that: The hardware facilities of the online temperature field prediction system based on the CMT arc additive manufacturing process include: CMT arc additive manufacturing work platform, shielding gas, and temperature acquisition system; The CMT arc additive manufacturing work platform includes the CMT digital welding system, industrial robot, and positioner; The temperature sensor is a thermal imager temperature acquisition method.

Images