Method for rapidly screening scanning paths to reduce local heat accumulation in additive manufacturing

Abstract

The invention provides a method for rapidly screening scanning paths to reduce local heat accumulation in additive manufacturing, and relates to the technical field of additive manufacturing. Temperature field data of different deposition sequences are simulated through a finite element method, and then the deposition sequences and the temperature field data corresponding to the deposition sequences form a sample set; and detail features and change rules of the temperature field can be accurately captured. According to the method, the deep regression model is trained through the training set in the sample set, a large amount of temperature field data under different working conditions can be quickly generated through training and application of the deep regression model, and the calculation efficiency is greatly improved. According to the method, temperature field prediction results under different working conditions can be quickly obtained only by inputting the deposition sequence into the depth regression model, the complexity and difficulty of operation are greatly reduced, meanwhile, the requirement for computing resources can be reduced, temperature field data can be automatically generated and analyzed, and the labor cost is reduced.

Description

Technical Field

[0001] The invention relates to the technical field of additive manufacturing, in particular to a method for rapidly screening a scanning path to reduce local heat accumulation in additive manufacturing. Background Art

[0002] At present, the exploration of additive manufacturing processes still requires a lot of simulations and experiments. If some parameters change, the overall model and experimental specimens need to be re-prepared, which greatly limits the speed of optimization iteration of the additive manufacturing process. It is necessary to develop a fast prediction model, and machine learning models show great application potential in additive manufacturing. Machine learning methods do not require the creation and solution of complex regression equations for heat transfer models. Using data extracted from finite element simulation models to train models, complex engineering problems can be solved.

[0003] For the additive manufacturing process, the data collection required by the machine learning algorithm is the main problem currently faced. At present, the method of extracting experimental process data is generally used to establish a data set. The molten pool morphology detection based on the data collected by the coaxial infrared camera and the defect detection based on the online acoustic signal are the two current application directions. The currently available data types are divided into the following categories:

[0004] (1) Gas porosity and poor fusion are caused by the rapid solidification of materials during additive manufacturing. The main reason for the formation of porosity is the mismatch between powder supply and energy input. When the powder supply is high but the input energy is insufficient, a large amount of powder will be included in the molten pool without being completely melted. Such unmelted particles will greatly affect the mechanical properties.

[0005] (2) Dimensional errors in the additive manufacturing process, which are mainly caused by the shrinkage of the material after solidification. During the deposition process, the material shrinks when the molten pool area cools. This process is restricted by the deposited area and causes stress. These changes will cause obvious changes in the geometry.

[0006] (3) The melt pool size is an important parameter in the additive manufacturing process. It is closely related to process parameters such as laser power and scanning speed and is one of the main factors used to evaluate the microstructure. The regression model of the melt pool characteristics obtained using process parameters can better predict the microstructure and mechanical properties of parts.

[0007] (4) The thermal history of the deposited layer is the main means of analyzing the heat transfer mechanism of the additive process. Controlling heat accumulation can effectively reduce deformation and cracking. If the temperature near the molten pool is too low, the overlap between the molten pools will be insufficient, and pores will form nearby. Increasing the scanning speed will lead to a larger thermal gradient between the starting position and the end point, resulting in an increase in residual stress and an increase in the risk of cracks. Excessive heat accumulation near the molten pool will cause the microstructure to become coarser, resulting in a decrease in mechanical properties.

[0008] Machine learning has great advantages in data processing and can be used for regression and classification of large amounts of data collected during the experiment. According to the type and quantity of data in the machine learning model, the model can be divided into supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning. The difference between the first three methods lies mainly in whether the processed data is labeled in advance, while reinforcement learning is trained through a pre-designed reward and punishment mechanism. For the process optimization of additive manufacturing, data prediction and analysis are mainly carried out through offline examples, and the data are pre-labeled. Supervised learning is generally used for this type of problem. In supervised learning, it is considered whether the calculated model belongs to a classification problem or a regression problem. Classification methods are suitable for crack or micrometallographic identification, while regression methods are used for numerical prediction problems such as temperature field prediction.

[0009] The challenge of applying machine learning in the field of additive manufacturing is how to establish a usable data set. Current research relies on a large amount of real experimental data, which requires a lot of time and cost. As a mature computational method, numerical modeling can serve as a good source of data containing physical information about the heat transfer process of laser energy in the deposited layer. Hosseini et al. adopted an unsupervised learning strategy to solve the heat transfer equation based on the one-way deposition model and obtained the temperature curves under different combinations of process conditions. Li et al. established a physical information neutral network (PINN) model for one-way deposition, which includes a basic training part and a transfer learning part, and achieved accurate and efficient prediction of the temperature field. Zhao et al. constructed a data set based on the data of the numerical prediction model and established a multi-layer perceptron (MLP) deep learning model to achieve bidirectional prediction of the molten pool size and process parameters.

[0010] In the deposition of larger parts, the temperature field of different scanning modes is also an issue that needs attention in the manufacturing process. Roy et al. established a multi-layer deposition agent model based on machine learning, which can predict the temperature field distribution of GCode, deposition area and heat-affected zone with high accuracy at a low computational cost. Demir et al. and Ren et al. used finite element models to establish arbitrary path temperature field data sets, which can realize the prediction of in-plane temperature fields under different scanning modes. Ogoke et al. proposed a deep reinforcement learning framework for obtaining optimized scanning speed and power parameters to reduce the molten pool changes during scanning. Ren et al. developed a temperature pattern recursive neural network model and used different scanning pattern pairings between layers to establish a data set to select the best path planning to reduce local heat. Mozaffar et al. proposed a graph-based neural network model to predict the temperature field distribution of different part shapes during deposition.

[0011] There are some defects and problems in the research on using deep learning methods to predict the temperature field of additive manufacturing at home and abroad, as well as the reasons for these problems. The following are some key points:

[0012] (1) Challenges in data acquisition: Deep regression models rely heavily on high-quality datasets, which are expensive and challenging to obtain in the field of metal additive manufacturing. This is because data collection in the additive manufacturing process requires high-precision sensors and monitoring tools, which are often expensive and difficult to deploy and maintain in high-temperature, dynamic manufacturing environments.

[0013] (2) Accuracy of real-time prediction: Although deep regression models have made some progress in predicting temperature fields, their prediction effect is limited under real-time changing conditions. This is because the temperature field in the additive manufacturing process changes very quickly and complexly, and the model may find it difficult to capture all the subtle changes, resulting in a decrease in the accuracy of the prediction.

[0014] (3) Model generalization ability: A deep regression model may perform well under certain conditions, but may not generalize well under different process parameters, geometries, and deposition modes. This is because model training is often based on a specific data set, and when faced with new or unseen situations, the model’s predictive performance may decline.

[0015] (4) Computational resources and time: Although physics-based computational models provide accuracy, they are usually very time-consuming and not suitable for real-time prediction and online control in iterative design scenarios. Although deep regression models have advantages in speed, training these models requires a lot of computing resources and time.

[0016] (5) Model interpretability: Deep regression models are often considered “black box” models, and their decision-making process lacks transparency and interpretability. This is particularly problematic in additive manufacturing, as understanding and interpreting temperature field predictions is critical to optimizing processes and improving product quality.

[0017] (6) Challenges of physical information fusion: Although the physical information neural network (PINN) framework combines physical principles and deep learning to accurately predict the temperature and dynamics of the molten pool, this method of fusing physical information may face challenges in practical applications because it requires accurate physical models and a large amount of labeled data to train the network.

[0018] (7) Accuracy of defect detection: Although deep learning methods have shown potential in defect detection, they have a high recognition ability for anomalies caused by porosity defects, with a recognition accuracy of over 90%. However, in practical applications, the accuracy of defect detection may be affected by many factors, including the complexity of the melt pool characteristics, environmental noise, and the training data of the model. In summary, although deep learning has made some progress in predicting the temperature field of additive manufacturing, it still faces challenges such as data acquisition, real-time prediction accuracy, model generalization ability, computing resources, model interpretability, and physical information fusion. Solving these problems requires interdisciplinary research efforts, including collaboration in sensor technology, data science, materials science, and computer science.

[0019] The method currently used is mainly based on the finite element simulation method to simulate the temperature field of the sedimentary layer. After a long period of verification, the traditional finite element analysis method has been proven to be able to make a relatively accurate simulation of the temperature field, but once the conditions change, the entire model needs to be recalculated. Each model calculation takes several hours to several days, and there is a bottleneck problem in computational efficiency, which makes it impossible to apply it to the prediction of a large number of deposition paths. The traditional temperature field simulation method requires complex steps such as model establishment, meshing, and boundary condition setting. The operation is cumbersome and prone to errors. In addition, the traditional temperature field simulation method requires a large amount of computing resources and the participation of professionals, which is costly. Summary of the invention

[0020] In view of the shortcomings of the prior art, the object of the present invention is to propose a method for quickly screening scanning paths to reduce local heat accumulation in additive manufacturing, comprising:

[0021] Step 1: construct a finite element model, obtain multiple random deposition sequences, and simulate the additive manufacturing process for each deposition sequence based on the finite element model to obtain temperature field data corresponding to each deposition sequence, wherein the deposition sequence includes multiple deposition positions arranged in time;

[0022] Step 2: taking the deposition sequence as an input sample, and taking the temperature field data corresponding to the deposition sequence as an output sample, the input sample and the output sample constitute a training sample, a plurality of training samples constitute a sample set, and the sample set is divided into a training set and a validation set according to a preset ratio;

[0023] Step 3: Train the initial deep regression model according to the training set, and verify the trained deep regression model through the verification set to obtain the trained deep regression model;

[0024] Step 4: Obtain multiple random deposition sequences, input each deposition sequence into the trained deep regression model, obtain the temperature field data corresponding to each deposition sequence, obtain the temperature uniformity of each deposition sequence based on the temperature field data, obtain the minimum value of all temperature uniformities, and take the deposition sequence corresponding to the minimum value as the optimal deposition sequence.

[0025] Optionally, step 1 specifically includes:

[0026] The finite element model irradiates the laser beam to the deposition position and sprays the powder to the deposition position at the same time, and then extracts the temperature field data of the deposition position at the moment of powder solidification. Then, starting from the first deposition position in the deposition sequence, the above operation is repeated to obtain the temperature field data corresponding to each deposition position, and then the temperature field data corresponding to each deposition sequence is obtained.

[0027] Optionally, the deep regression model in step 3 includes a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, an activation function, upsampling and a skip connection.

[0028] Optionally, step 3 specifically includes:

[0029] The input samples are input into the initial deep regression model, and the input samples are encoded, decoded and jump-connected to obtain the predicted temperature field data. According to the predicted temperature field data and the output samples, the parameters in the initial deep regression model are updated, and the trained deep regression model is verified through the validation set until the model converges to obtain the trained deep regression model.

[0030] Optionally, the input samples are encoded, decoded and skipped to obtain predicted temperature field data, including:

[0031] The deposition sequence is sequentially subjected to a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, and an activation function to obtain first data; the first data is sequentially subjected to a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, and an activation function to obtain second data; the second data is subjected to a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, an activation function, a three-dimensional convolution layer, a batch normalization layer, an activation function, and upsampling to obtain third data; the third data and the second data are jump-connected to obtain fourth data; the fourth data is upsampled to obtain the upsampled fourth data; the upsampled fourth data and the first data are jump-connected to obtain fifth data; the fifth data is upsampled to obtain the upsampled fifth data; the upsampled fifth data is jump-connected to the deposition sequence to obtain predicted temperature field data.

[0032] Optionally, the third data has the same dimension as the second data, the upsampled fourth data has the same dimension as the first data, and the upsampled fifth data has the same dimension as the deposition order.

[0033] Optionally, in step 4, the temperature uniformity of each deposition sequence is obtained according to the temperature field data, including:

[0034] For the temperature field data of each deposition position in the deposition sequence, the average temperature variance is calculated and the average temperature variance is used as the temperature uniformity.

[0035] The beneficial effects of adopting the above technical solution are:

[0036] The present invention simulates the temperature field data of different deposition orders by the finite element method, and then the deposition order and the temperature field data corresponding to the deposition order constitute a sample set, which can accurately capture the detailed characteristics and change rules of the temperature field, and provide high-quality input data for the subsequent training of the deep regression model. During the training process, the deep regression model can learn the complex mapping relationship between a large number of temperature field data under different working conditions, further improving the accuracy and quality of the generated temperature field data. The present invention trains the deep regression model through the training set in the sample set to obtain the trained deep regression model. Through the training and application of the deep regression model, a large number of temperature field data under different working conditions can be quickly generated, greatly improving the calculation efficiency. This makes it possible to quickly predict and analyze the temperature field under different working conditions in actual engineering applications, providing strong support for engineering design and optimization. Compared with the traditional temperature field simulation method, which requires complex model establishment, grid division, boundary condition setting and other steps, the present invention only needs to input the deposition order into the trained deep regression model to quickly obtain the predicted temperature field data under different working conditions. This greatly reduces the complexity and difficulty of the operation, allowing non-professional technicians to easily predict and analyze the temperature field. At the same time, it can reduce the demand for computing resources and automatically generate and analyze temperature field data, reducing labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic flow chart of a method for rapidly screening scanning paths to reduce local heat accumulation in additive manufacturing according to an embodiment of the present invention;

[0038] Figure 2 It is a schematic diagram of the data processing flow of the deep regression model in an embodiment of the present invention;

[0039] Figure 3 Graphs showing the training accuracy of the SC-3DCAE model in an embodiment of the present invention, wherein (a) is a graph showing the loss function of the training model, and (b) is a graph showing the model accuracy rate;

[0040] Figure 4 Schematic diagrams comparing the finite element simulation results and the model prediction results of several deposition sequences with the smallest variance in the embodiments of the present invention, wherein (a) is a schematic diagram comparing the finite element simulation results and the model prediction results of the first deposition sequence, (b) is a schematic diagram comparing the finite element simulation results and the model prediction results of the second deposition sequence, and (c) is a schematic diagram comparing the finite element simulation results and the model prediction results of the third deposition sequence. DETAILED DESCRIPTION

[0041] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0042] In the additive manufacturing process, local heat accumulation is a common technical problem, especially in technologies such as laser directed energy deposition. Heat accumulation can cause overheating and deformation of the material, thus affecting the geometric accuracy and mechanical properties of the formed part. This heat accumulation not only causes dimensional deviations of the parts, but may also cause defects such as cracks and stress concentration, seriously affecting the performance and reliability of the parts.

[0043] In order to solve this problem, this patent proposes a fast screening method for optimizing the scanning path, which uses a deep regression model for rapid prediction and screening. Traditional scanning path planning methods usually rely on empirical rules and trial and error methods, which are inefficient and difficult to adapt to complex manufacturing environments. The deep regression model is trained with a large amount of historical data and can accurately predict the heat accumulation under different scanning paths, thereby quickly screening out the optimal scanning path. This method not only improves the efficiency of scanning path planning, but also significantly reduces heat accumulation and reduces residual stress and deformation during the forming process. In addition, the scalability and flexibility of the deep regression model enable it to adapt to different materials and process parameters, and has broad application prospects. Through this optimization method, the quality and performance of additively manufactured parts can be significantly improved, promoting the application of additive manufacturing technology in aviation, aerospace, medical and other fields.

[0044] For the prediction of temperature fields under different deposition paths of additive processes, the use of deep regression models can significantly reduce the calculation time, so that tasks such as temperature field prediction can be completed more efficiently. The deep regression model relies on a large amount of data support and has better performance in data classification and regression. The deep regression model has the following advantages. First, the deep regression model is very powerful and can realize the automatic extraction of data features, avoiding tedious manual intervention and adjustment. Secondly, the deep regression model has a large number of network layers, and the function mapping of complex models can be realized under the existing computing power. Then, there are many ready-made frameworks for deep regression models that can be used, and these frameworks can implement computing functions on multiple platforms. Finally, the deep regression model can achieve fast prediction. Although a lot of resources are required during the training process, it can be predicted quickly after the training is completed, which is very suitable for the deposition path screening required by this patent. Therefore, the present invention provides a method for quickly screening scanning paths to reduce local heat accumulation in additive manufacturing, combined with Figure 1 , which may include the following steps:

[0045] Step 1: Construct a finite element model to obtain multiple random deposition sequences. For each deposition sequence, simulate the additive manufacturing process based on the finite element model to obtain the temperature field data corresponding to each deposition sequence, that is, Figure 1 The temperature field simulation result in the deposition sequence includes a plurality of deposition positions arranged in time;

[0046] The finite element model irradiates the laser beam to the deposition position and sprays the powder to the deposition position at the same time, and then extracts the temperature field data of the deposition position when the powder solidifies, and then starts from the first deposition position in the deposition sequence, repeats the above operation to obtain the temperature field data corresponding to each deposition position, and then obtains the temperature field data corresponding to each deposition sequence. That is to say, after performing the above operation at the first deposition position, the laser beam is irradiated to the second deposition position, and the powder is also sprayed to the second deposition position, and then the temperature field data of the second deposition position is extracted when the powder solidifies, and then the above operation is performed in sequence according to the deposition sequence.

[0047] In the present invention, each deposition sequence is written into the finite element model by Pycharm software to perform the above operations according to the deposition position in the deposition sequence. The written model is Figure 1 The scanning strategy for the input in .

[0048] In the applied laser directed energy deposition additive manufacturing process, the laser and the input powder are focused on the corresponding position of the substrate, accompanied by the melting and solidification process. The heat source movement and material addition are realized according to the predetermined deposition path program. In order to be consistent with the actual thermal physical process, the material parameters used change with temperature. In order to reduce the calculation time, a denser grid arrangement is used near the deposited layer. The material parameters and boundary conditions have been described in previous work. The developed model can collect temperature field data at a predetermined time and location.

[0049] Step 2: Take the deposition sequence as the input sample, and take the temperature field data corresponding to the deposition sequence as the output sample. The input sample and the output sample constitute a training sample, and multiple training samples constitute a sample set, that is, Figure 1 The temperature data set in the example is divided into a training set and a validation set according to a preset ratio;

[0050] Step 3: According to the scanning strategy, that is, the scanning order, the initial deep regression model is trained according to the training set, and the trained deep regression model is verified by the verification set to obtain the trained deep regression model;

[0051] The deep regression model includes a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, an activation function, upsampling and a skip connection.

[0052] Among them, the model contains maximum pooling downsampling and transposed convolution upsampling, which can be used to extract important features between temperature field data and scanning patterns. The developed model can replace the finite element model to predict temperature field data in additive manufacturing. The input scanning pattern is represented in the form of three-dimensional data to keep consistent with the size of the temperature field data. The encoding process included in the model is mainly to gradually extract the features of the input temperature field through a series of convolutional layers and pooling layers. The convolutional layer is responsible for extracting local features, while the pooling layer is used to reduce the spatial resolution of the feature map, thereby reducing the amount of calculation and extracting higher-level abstract features. This process enables the network to capture important information in the data. Through multiple downsampling operations, the encoder gradually reduces the spatial size of the input temperature field data. This not only reduces the amount of calculation, but also enables the network to better process the overall structure and contextual information of the image. The decoding process of the model is mainly to gradually restore the spatial resolution of the feature map through upsampling operations, so that the size of the temperature field is gradually restored to the same as the original input data. In the decoding process, the features extracted in the encoder are fused with the features in the decoder through jump connections. This design enables the decoder to combine low-level details and high-level semantic information, thereby maintaining an understanding of the overall structure while restoring image details. The model consists of 3D convolutional layers, downsampling layers, and upsampling layers. Each convolutional layer is followed by a batch normalization layer and an activation function, namely the rectified linear unit (ReLU). The input data goes through the encoding and decoding stages. In the encoding stage, the convolutional layers gradually extract high-dimensional features, while the downsampling layers can increase the receptive field and capture more feature representations with fewer parameters. After convolution, the model enhances the ability to represent the temperature field features within the sedimentary layer. The decoding stage consists of convolutional layers and upsampling layers. In the first two upsampling and convolutional layers, the number of feature channels is halved. After the last upsampling layer, the feature channels are further reduced to match the size of the temperature field data. The upsampled features obtained from each layer are merged with the corresponding downsampled features. The fused features continue to be input into the sampling process, preserving data details for higher accuracy.

[0053] Based on this, the present invention is specifically implemented in combination with Figure 2 , specifically including: inputting the input sample into the initial deep regression model, encoding, decoding and skipping the input sample to obtain the predicted temperature field data, updating the parameters in the initial deep regression model according to the predicted temperature field data and output samples, and verifying the trained deep regression model through the validation set until the model converges, that is, Figure 1 By comparative verification in , a trained deep regression model is obtained, wherein the initial deep regression model can be a SC-3DCAE model.

[0054] Among them, the input samples are encoded, decoded and skipped to obtain the predicted temperature field data, including:

[0055] The deposition sequence is sequentially subjected to a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, and an activation function to obtain first data; the first data is sequentially subjected to a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, and an activation function to obtain second data; the second data is subjected to a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, an activation function, a three-dimensional convolution layer, a batch normalization layer, an activation function, and upsampling to obtain third data; the third data and the second data are jump-connected to obtain fourth data; the fourth data is upsampled to obtain the upsampled fourth data; the upsampled fourth data and the first data are jump-connected to obtain fifth data; the fifth data is upsampled to obtain the upsampled fifth data; the upsampled fifth data is jump-connected to the deposition sequence to obtain predicted temperature field data.

[0056] The third data has the same dimension as the second data, the upsampled fourth data has the same dimension as the first data, and the upsampled fifth data has the same dimension as the deposition sequence.

[0057] Step 4: Obtain multiple random deposition orders, that is, randomly generate a large number of deposition orders, namely Figure 1 A large amount of random scanning path data is generated, and each deposition sequence is input into the trained deep regression model to obtain the temperature field data corresponding to each deposition sequence. According to the temperature field data, the temperature uniformity of each deposition sequence is obtained, and the minimum value of all temperature uniformities is obtained. The deposition sequence corresponding to the minimum value is taken as the optimal deposition sequence.

[0058] For the temperature field data of each deposition position in the deposition sequence, the average temperature variance is calculated and the average temperature variance is used as the temperature uniformity.

[0059] That is to say, based on the minimum variance of the temperature distribution at each position in the deposition layer as the judgment criterion, the optimal scanning path screens and optimizes the predicted value of the path. In order to obtain lower heat accumulation and more uniform temperature distribution, a large number of random paths need to be generated and input into the model. The corresponding thermal data can be quickly obtained using the existing model. The predicted temperature field is output, the average temperature variance is calculated, and the average temperature variance is used as the temperature uniformity to sort all the temperature uniformities, that is, Figure 1 The temperature distribution variance is sorted, where the smaller the variance, the better the uniformity, which can greatly reduce the local heat accumulation of the deposition layer. The deposition sequence corresponding to the minimum temperature uniformity is taken as the optimal deposition sequence, that is, Figure 1 The optimized scan path is obtained in .

[0060] Based on the above scheme, the present invention carried out the following experiments:

[0061] The temperature field data is divided into a training set and a test set in a ratio of 3:1. The training set is used to train the model and fit the data distribution pattern. The test set is used to evaluate the generalization ability of the model and the accuracy of the model. The training loss curve and test accuracy curve on the test set during the training phase are shown in Figure 2. Figure 3 As shown, (a) is the training model loss function curve, and (b) is the model accuracy curve. It can be seen that when the model training is completed and reaches a stable state, compared with the test set that did not participate in the training process, the final accuracy calculated reaches a high level, which shows that the prediction model can match the temperature field data well and obtain a relatively stable training process.

[0062] The present invention selects several scanning paths with the smallest variance, i.e., deposition order, as the scanning paths recommended by the model, and compares the predicted results with the simulation results of the finite element model, such as Figure 4 As shown, (a) is a schematic diagram comparing the finite element simulation results of the first deposition sequence with the model prediction results, (b) is a schematic diagram comparing the finite element simulation results of the second deposition sequence with the model prediction results, and (c) is a schematic diagram comparing the finite element simulation results of the third deposition sequence with the model prediction results. By using cloud charts to compare the data of the prediction model and the data of the finite element simulation, the accuracy of the prediction can be intuitively understood. It can be seen that the temperature distribution shown by the model prediction and the finite element simulation is relatively close. Uniform temperature distribution can not only reduce part deformation, but also obtain a finer and more uniform microstructure distribution. The developed model can perform a simulation prediction for a specified deposition sequence within one millisecond. Compared with several hours for a single finite element model, this greatly improves efficiency and makes scanning pattern optimization possible.

[0063] It can be seen from this that the present invention simulates the temperature field data of different working conditions by the finite element method, which can accurately capture the detailed characteristics and change laws of the temperature field, and provide high-quality input data for the subsequent training of the deep regression model. During the training process, the deep regression model can learn the complex mapping relationship between a large number of temperature field data under different working conditions, further improving the accuracy and quality of the generated temperature field data. Faster calculation speed in terms of efficiency: Traditional temperature field simulation methods, such as finite element analysis, usually require a lot of computing resources and time, especially for the simulation of large-scale complex working conditions, the calculation process may take hours or even days. However, this patent can quickly generate a large amount of temperature field data under different working conditions through the training and application of the deep regression model, greatly improving the calculation efficiency. This makes it possible to quickly predict and analyze the temperature field under different working conditions in actual engineering applications, providing strong support for engineering design and optimization. This patent simplifies the operational process of temperature field prediction through the training and application of the deep regression model. Traditional temperature field simulation methods require complex model establishment, meshing, boundary condition setting and other steps, which are cumbersome and error-prone. The method of this technology only needs to input the deposition sequence into the trained deep regression model to quickly obtain the predicted temperature field data under different working conditions. This greatly reduces the complexity and difficulty of the operation, allowing non-professional technicians to easily predict and analyze the temperature field. The method of this patent can effectively reduce the cost of temperature field prediction. Traditional temperature field simulation methods require a lot of computing resources and the participation of professionals, which is costly. The method of this technology can reduce the demand for computing resources, and can automatically generate and analyze temperature field data, reducing labor costs.

[0064] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) to form a technical solution.

Claims

1. A method for rapidly screening scanning paths to reduce local heat accumulation in additive manufacturing, characterized in that: include: Step 1: construct a finite element model, obtain multiple random deposition sequences, and simulate the additive manufacturing process for each deposition sequence based on the finite element model to obtain temperature field data corresponding to each deposition sequence, wherein the deposition sequence includes multiple deposition positions arranged in time; Step 2: taking the deposition sequence as an input sample, and taking the temperature field data corresponding to the deposition sequence as an output sample, the input sample and the output sample constitute a training sample, a plurality of training samples constitute a sample set, and the sample set is divided into a training set and a validation set according to a preset ratio; Step 3: Train the initial deep regression model according to the training set, and verify the trained deep regression model through the verification set to obtain the trained deep regression model; Step 4: Obtain multiple random deposition sequences, input each deposition sequence into the trained deep regression model, obtain the temperature field data corresponding to each deposition sequence, obtain the temperature uniformity of each deposition sequence based on the temperature field data, obtain the minimum value of all temperature uniformities, and take the deposition sequence corresponding to the minimum value as the optimal deposition sequence.

2. A method for rapidly screening scanning paths to reduce local heat accumulation in additive manufacturing according to claim 1, characterized in that: Step 1 specifically includes: The finite element model irradiates the laser beam to the deposition position and sprays the powder to the deposition position at the same time, and then extracts the temperature field data of the deposition position at the moment of powder solidification. Then, starting from the first deposition position in the deposition sequence, the above operation is repeated to obtain the temperature field data corresponding to each deposition position, and then the temperature field data corresponding to each deposition sequence is obtained.

3. The method of quickly screening scanning paths to reduce local heat accumulation in additive manufacturing according to claim 1, characterized in that: The deep regression model described in step 3 includes a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, an activation function, upsampling, and a skip connection.

4. The method of quickly screening scanning paths to reduce local heat accumulation in additive manufacturing according to claim 1, characterized in that: Step 3 specifically includes: The input samples are input into the initial deep regression model, and the input samples are encoded, decoded and jump-connected to obtain the predicted temperature field data. According to the predicted temperature field data and the output samples, the parameters in the initial deep regression model are updated, and the trained deep regression model is verified through the validation set until the model converges to obtain the trained deep regression model.

5. The method for rapidly screening scanning paths to reduce local heat accumulation in additive manufacturing according to claim 4, characterized in that: Encode, decode and skip the input samples to obtain the predicted temperature field data, including: The deposition sequence is sequentially subjected to a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, and an activation function to obtain first data; the first data is sequentially subjected to a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, and an activation function to obtain second data; the second data is subjected to a maximum pooling layer, a three-dimensional convolution layer, a batch normalization layer, an activation function, a three-dimensional convolution layer, a batch normalization layer, an activation function, and upsampling to obtain third data; the third data and the second data are jump-connected to obtain fourth data; the fourth data is upsampled to obtain the upsampled fourth data; the upsampled fourth data and the first data are jump-connected to obtain fifth data; the fifth data is upsampled to obtain the upsampled fifth data; the upsampled fifth data is jump-connected to the deposition sequence to obtain predicted temperature field data.

6. The method of quickly screening scanning paths to reduce local heat accumulation in additive manufacturing according to claim 5, characterized in that: The third data has the same dimension as the second data, the upsampled fourth data has the same dimension as the first data, and the upsampled fifth data has the same dimension as the deposition sequence.

7. The method of quickly screening scanning paths to reduce local heat accumulation in additive manufacturing according to claim 1, characterized in that: In step 4, the temperature uniformity of each deposition sequence is obtained based on the temperature field data, including: For the temperature field data of each deposition position in the deposition sequence, the average temperature variance is calculated and the average temperature variance is used as the temperature uniformity.

Images