Thin-wall structure laser powder bed fusion forming wall thickness prediction method based on deep learning
By using deep learning prediction models and spot compensation technology, the problem of controlling the wall thickness of thin-walled structures in laser powder bed fusion was solved, achieving efficient wall thickness accuracy control and reducing experimental costs and time.
Patent Information
- Application Number
- CN202510825687.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-19
- Publication Date
- 2025-10-17
AI Technical Summary
The wall thickness control of thin-walled structures during laser powder bed fusion processing is difficult to accurately predict, resulting in a large number of experimental iterations and high costs.
By employing a deep learning-based approach, a dataset is constructed and a deep learning prediction model is trained through designing a thin-walled structure model, setting a laser steering scanning strategy and process parameter combination, predicting the forming wall thickness of the thin-walled structure, and achieving wall thickness accuracy control through spot compensation adjustment.
It effectively reduced the cost of experimental iterations, enabled accurate prediction and control of the wall thickness of thin-walled structures under different process parameters, and improved the efficiency and precision of laser powder bed melting manufacturing.
Smart Images
Figure CN120805658A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of laser additive manufacturing, and particularly relates to a thin-wall structure laser powder bed fusion forming wall thickness prediction method based on deep learning. BACKGROUND
[0002] Laser powder bed fusion (LPBF) is a typical additive manufacturing technology based on the principle of dispersion / accumulation, which uses a high-energy laser beam to quickly melt metal powder along a set scanning path and quickly cools and solidifies to manufacture parts layer by layer. Compared with traditional manufacturing methods, LPBF has the advantages of high forming freedom, short processing cycle, low material consumption, etc., and can be used to manufacture high-precision, small-wall-thickness thin-wall structures.
[0003] However, the differences in laser scanning strategies and scanning path offsets of the thin-wall profile edges during the laser powder bed fusion process will lead to differences in the amount of deposited metal on the side walls of the parts, which will further affect the actual forming wall thickness. In addition, the changes in the size of the molten pool caused by the differences in heat input under different laser power and scanning speed combinations will also change the wall thickness size. Controlling the wall thickness through experimental methods requires multiple rounds of experiments under different process parameter combinations, which consumes a lot of time and cost.
[0004] In summary, there is an urgent need for a new method to solve the problem of controlling the actual forming wall thickness of laser powder fusion manufacturing thin walls. SUMMARY
[0005] The present application provides a thin-wall structure laser powder bed fusion forming wall thickness prediction method based on deep learning, which provides a means for predicting the forming wall thickness of laser powder bed fusion manufacturing thin walls and obtains the range of optimized spot compensation values within the wall thickness deviation threshold, overcoming the difficulty of thin-wall wall thickness prediction under multiple process parameter combination conditions.
[0006] To achieve the above purpose, the technical scheme adopted by the present application is as follows:
[0007] A thin-wall structure laser powder bed fusion forming wall thickness prediction method based on deep learning, comprising the following steps:
[0008] Step 1: Design different theoretical wall thickness thin-wall structure models, set the laser turning scanning strategy at the model profile, and perform layering processing;
[0009] Step 2: According to the thin-wall structure model designed in step 1, set different process parameter combinations, and perform laser powder bed fusion manufacturing of thin-wall structures;
[0010] Step three: According to the thin-walled structure manufactured in step two, measure the actual forming wall thickness data of the thin-walled structure under different process input conditions and construct a data set, and train an optimized thin-walled structure forming wall thickness deep learning prediction model based on the data set;
[0011] Step four: According to the optimal deep learning model optimized in step three, realize the prediction of the actual forming wall thickness of the thin-walled structure under the condition of the combination of the theoretical wall thickness and the process parameters.
[0012] Further, the step one specifically comprises:
[0013] Step 1.1: Design a thin-walled structure model as an overall model with varying wall thickness in different height ranges along the LPBF construction direction;
[0014] Step 1.2: Set the scanning strategy at the laser turning point of the thin-walled structure outer contour to be laser off jump or laser on continuous movement;
[0015] Step 1.3: Set the LPBF printing layer thickness and perform model slicing processing.
[0016] Further, the step two specifically comprises:
[0017] Step 2.1: Set the laser spot compensation parameter, and the spot compensation value is defined as the distance between the laser spot outside and the thin-walled structure outer contour, which is used to adjust the distance between the laser scanning path and the thin-walled structure outer contour;
[0018] Step 2.2: Set the laser power and scanning speed, which are used to adjust the laser energy input in the manufacturing process of different thin-walled structures;
[0019] Step 2.3: Configure different process parameter combinations for the thin-walled structure model of the same design size, and perform laser powder layer melting manufacturing of the thin-walled structure, each set of process parameters including one scanning strategy, one laser power, one scanning speed and one spot compensation value.
[0020] Further, the step three specifically comprises:
[0021] Step 3.1: According to the thin-walled structure manufactured in step two, measure the actual forming wall thickness of the thin-walled structure under different process input conditions, construct a process parameter-design wall thickness-actual forming wall thickness correlation data set, and exclude abnormal data in the data set, the process input conditions including scanning strategy, spot compensation, laser power, scanning speed;
[0022] Step 3.2: Divide the data set obtained by measurement into training set, validation set and independent test set;
[0023] Step 3.3: Establish a laser powder bed fusion thin-walled forming wall thickness deep learning prediction model;
[0024] Step 3.4: The deep learning prediction model takes the design wall thickness, scanning strategy, laser power, scanning speed, and spot compensation of the training set and validation set as model input data, standardizes the design wall thickness, laser power, scanning speed, and spot compensation in the input data, and then the deep learning prediction model processes the standardized data to output the predicted forming wall thickness of the thin-walled structure,
[0025] The specific operation expression of the input data row standardization processing of the deep learning prediction model is formula (1):
[0026]
[0027] In the formula: is the new data after standardization, X i is the original data, E[x i ] is the mean value of the training set data, is the standard deviation of the training set data;
[0028] Step 3.5: During the training process of the deep learning prediction model, the K-fold cross-validation method is used to evaluate the influence of overfitting and selection bias, and the absolute percentage error APE and accuracy AR of the predicted forming wall thickness and the actual forming wall thickness of the deep learning prediction model under the same conditions are calculated as the indicators of the prediction accuracy of the deep learning prediction model. The specific operation expressions are formula (2) and formula (3) respectively:
[0029]
[0030] AR = (1-APE) (3)
[0031] In the formula: is the predicted forming wall thickness value, y i is the actual forming wall thickness value;
[0032] Step 3.6: Adjust the number of neurons and hidden layers of the deep learning prediction model, and then execute
[0033] Step 3.4-3.5, obtain multiple groups of prediction accuracy, compare the prediction accuracy of different deep learning prediction models, and determine the optimal deep learning prediction model with the highest accuracy;
[0034] Step 3.7: Input the design wall thickness and process parameter combination of the independent test set to obtain the predicted forming wall thickness under different input conditions, and verify the optimal deep learning prediction model.
[0035] Further, the step four specifically includes:
[0036] Adjust the spot compensation value under different design wall thickness and process parameter combination, obtain the corresponding thin-walled structure predicted forming wall thickness, so that the thin-walled structure predicted forming wall thickness is within the deviation threshold of the design wall thickness, obtain the corresponding optimization spot compensation range within the deviation threshold, establish the thin-walled structure forming wall thickness control process parameter table, realize the thin-walled structure forming wall thickness prediction meeting the accuracy requirement, and the thin-walled structure forming control process parameter table includes design wall thickness, scanning strategy, laser power, scanning speed, predicted forming wall thickness range and spot compensation range.
[0037] Further, the deep learning prediction model in step 3.3 is realized by a multilayer perceptron algorithm, adopts a rectified linear unit (ReLU) function as an activation function, and adopts an RMSprop algorithm as an optimizer.
[0038] Further, the deep learning prediction model in step 3.3 is realized by a multilayer perceptron algorithm, adopts a rectified linear unit (ReLU) function as an activation function, and adopts an RMSprop algorithm as an optimizer.
[0039] Further, each group of process parameters in step 2.3 includes three spot compensation values.
[0040] Further, the value of K in step 3.5 is 5.
[0041] Further, the deviation threshold in step four is 3%.
[0042] Compared with the prior art, the advantages of the present application include:
[0043] The thin-walled structure laser powder bed fusion forming wall thickness prediction method of the present application establishes the correlation between the process parameters and the forming wall thickness, realizes the laser powder bed fusion thin-walled structure forming wall thickness prediction under different working conditions, and controls the deviation between the thin-walled predicted forming wall thickness and the design wall thickness by adjusting the spot compensation parameter. Under the condition of giving different scanning strategy, laser power and scanning speed combination, the optimization spot compensation value range within the wall thickness deviation threshold can be directly output. Compared with the method of measuring the forming wall thickness by manufacturing each process parameter thin-walled structure through conventional experiment, the present application can effectively reduce the experimental parameter iteration trial and error cost, and support the laser powder bed fusion thin-walled wall thickness precision optimization.
[0044] In addition to the features and advantages described above, the principles of the present application and other features and advantages. The present application will be further described in detail below with reference to the drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS
[0045] Figure 1 The model diagram of the thin-walled structure for different design wall thicknesses;
[0046] Figure 2 The schematic diagram of different laser steering scanning strategies for the laser powder bed fusion manufacturing thin-walled profile;
[0047] Figure 3 A schematic diagram of a thin-walled structure forming wall thickness prediction process;
[0048] Figure 4 A deep learning prediction model for predicting the correlation between the forming wall thickness and the actual forming wall thickness;
[0049] Figure 5 A deep learning-based laser powder bed fusion thin-walled structure forming wall thickness prediction process diagram. DETAILED DESCRIPTION
[0050] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application. The implementation of the present application is described in detail below in combination with specific examples.
[0051] In combination Figure 5 A deep learning-based thin-walled structure laser powder bed fusion forming wall thickness prediction method, comprising the following steps:
[0052] Step one: design different wall thickness thin-walled structure model, set the laser steering scanning strategy at the model contour, and use Magics software for slicing processing;
[0053] Step two: according to the thin-walled structure model designed in step one, set different process parameter inputs, and manufacture thin-walled structure laser powder bed fusion;
[0054] Step three: according to the thin-walled structure manufactured in step two, measure the actual forming wall thickness data of the thin-walled structure using a vernier caliper and build a data set, and train and optimize the thin-walled forming wall thickness deep learning prediction model based on the data set;
[0055] Step four: according to the optimal deep learning prediction model optimized in step three, adjust the spot compensation value under different design wall thickness and process parameter combinations to realize the forming wall thickness prediction of thin-walled structure meeting the accuracy requirement.
[0056] Further, the step one specifically comprises:
[0057] Step 1.1: design the thin-walled structure model as an overall model with varying wall thickness in different height ranges along the LPBF construction direction, such as Figure 1 The total height of the thin-walled design model is 20mm, the length is 10mm, and the height direction is divided into 4 segments of 5mm area, with wall thicknesses of 2mm, 1.0mm, 0.5mm and 0.3mm respectively;
[0058] Step 1.2: set the scanning strategy at the laser steering position of the thin-walled structure outer contour to laser off jump or laser on continuous movement, such as Figure 2Fig. 1(a) shows the laser turning off at the end of the scanning path and directly jumping to the next path under the laser jump scanning strategy, and there is no scanning path between adjacent deposition paths. Figure 2 Fig. 1(b) shows the laser turning on and turning between adjacent scanning paths under the laser continuous scanning strategy, and there is no laser turning off and restarting phase.
[0059] Step 1.3: Set the LPBF printing layer thickness using the Magics software, perform model slicing processing, and input the sliced data into the LPBF manufacturing equipment.
[0060] Further, the step two specifically includes:
[0061] Step 2.1: Set the laser spot compensation parameter, and the spot compensation value is defined as the distance between the outer side of the laser spot and the outer contour of the thin-walled structure, which is used to adjust the distance between the laser scanning path and the outer contour of the thin-walled structure. When the spot compensation value is larger, the laser printing path is away from the outer contour of the part. When the spot compensation value is negative, the laser spot can heat the metal powder outside the outer contour of the designed thin-walled model.
[0062] Step 2.2: Set the laser power and scanning speed, which are used to adjust the laser energy input during the manufacturing of different thin-walled structures. The laser power and scanning speed directly affect the laser energy density. Under the condition of high energy density, the laser heats and melts the metal powder to form a molten pool size, which is larger than that under the condition of low energy density. Under the premise of the same scanning path, the thin-walled structure formed under the condition of high energy density has a larger forming wall thickness.
[0063] Step 2.3: Configure different process parameter combinations for the same designed size of thin-walled structure model, each process parameter combination including one scanning strategy, one laser power, one scanning speed, and one spot compensation value. Perform thin-walled structure laser powder layer melting manufacturing. The parameters are shown in Table 1. Under the premise of ensuring the internal quality of thin-walled manufacturing, two strategies of jump scanning and continuous scanning are used respectively, six groups of laser power and scanning speed are set, and three spot compensation values are designed for each group, totaling 36 parameter combinations for manufacturing variable-thickness thin-walled structures.
[0064] Table 1: Thin-walled structure manufacturing process parameters
[0065]
[0066] Further, the step three specifically includes:
[0067] Step 3.1: According to the thin-walled structure manufactured in step two, the actual forming wall thickness of the thin-walled structure under different process input conditions is measured using a vernier caliper, and a process parameter-design wall thickness-actual forming wall thickness correlation data set is constructed. A single thin-walled structure can obtain 4 sets of measured values under different design wall thickness conditions (corresponding to 4 5mm regions in step one), and a total of 144 groups of data are measured to construct the data set. By analyzing the correlation between process input (including scanning strategy, spot compensation, laser power, scanning speed) and actual forming wall thickness, abnormal data in the data set is excluded (no abnormal data in this embodiment);
[0068] Step 3.2: The data set obtained by measurement is divided into a training set, a validation set and an independent test set. The training set is used to adjust the internal parameter weight of the deep learning prediction model, and helps the model to fit the data distribution by minimizing the training error; the validation set is used to monitor the training process to optimize and evaluate the ability of the deep learning prediction model, and prevent overtraining; the independent test set is used to evaluate the generalization ability of the deep learning prediction model, and check the prediction accuracy of the MLP model. The training set and the validation set are used multiple times in the optimization process, and the independent test set is reserved in advance;
[0069] Step 3.3: The laser powder bed fusion thin-walled forming wall thickness deep learning prediction model established by the present application adopts TensorFlow and Keras modules to write a multilayer perceptron algorithm in a Python 3.6 environment, and adopts a rectified linear unit (ReLU) function as an activation function and an RMSprop algorithm as an optimizer;
[0070] Step 3.4: As Figure 3 , the deep learning prediction model of the present application takes the design wall thickness and scanning strategy, laser power, scanning speed, and spot compensation manufacturing process parameters of the training set and the validation set as model input data, considers the differences in dimensions, process parameters and forming wall thickness, and the deep learning prediction model standardizes part of the input data (including design wall thickness, laser power, scanning speed, and spot compensation). Then the deep learning prediction model processes the standardized data, and outputs the predicted forming wall thickness of the thin-walled structure.
[0071] The specific operation expression of the input data row standardization processing of the deep learning prediction model is as formula (1):
[0072]
[0073] In the formula, is the new data after standardization, X i is the original data, E[x i ] is the average value of the training set data,
[0074] is the standard deviation of the training set data;
[0075] Step 3.5: During the training process of the MLP model, the K-fold cross-validation method is used to evaluate the influence of overfitting and selection bias, where K is set to 5. The absolute percentage error (APE) and accuracy (AR) of the predicted forming wall thickness and the actual forming wall thickness of the deep learning prediction model under the same conditions are calculated as the indicators of the prediction accuracy of the deep learning prediction model, and the specific operation expressions are formula (2) and formula (3) respectively:
[0076]
[0077] AR = (1-APE) (3)
[0078] In the formula: is the predicted value, y i is the actual value.
[0079] Step 3.6: Adjust the number of neurons and hidden layers of the deep learning prediction model (the hidden layer and neuron number combination parameters are shown in Table 2), and then perform steps 3.4-3.5 again to obtain multiple sets of prediction accuracy (6 sets of accuracy are obtained in this embodiment, as shown in Table 3). Compare the prediction accuracy of different deep learning prediction models to determine the optimal deep learning prediction model (in this embodiment, model 5 in Table 2 is determined as the optimal deep learning prediction model);
[0080] Table 2 Deep learning prediction model hidden layer and neuron number combination
[0081]
[0082] Table 3 Deep learning prediction model thin-walled structure forming wall thickness prediction accuracy of different hidden layers and neuron numbers
[0083]
[0084] Step 3.7: According to the optimal deep learning prediction model in step 3.6, input the design wall thickness and process parameter combination of the independent test set to obtain the predicted forming wall thickness under different input conditions, and combine Figure 4 , the predicted forming wall thickness of the optimized deep learning prediction model is in good consistency with the actual value;
[0085] Further, the step four specifically comprises:
[0086] Adjusting the light spot compensation value under different design wall thickness and process parameter combination, obtaining the corresponding thin wall predicted forming wall thickness, so that the thin wall predicted forming wall thickness and the design wall thickness are within the deviation threshold (such as 3%), obtaining the corresponding optimization light spot compensation range within the deviation threshold (such as 3%). Establishing the thin wall forming wall thickness control process parameter table (part of the reference data is shown in Table 4), and through the optimization of the light spot compensation value, the thin wall structure forming wall thickness prediction meeting the accuracy requirement is realized.
[0087] Table 4 Thin wall forming wall thickness control process parameter table
[0088]
[0089] The basic principles, main features and advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited by the above examples, and the above examples and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.
Claims
1. A method for predicting wall thickness of thin-walled structure laser powder bed fusion forming based on deep learning, characterized in that: The following steps are involved: Step 1: Design thin-walled structure models with different theoretical wall thicknesses, set the laser steering scanning strategy at the model contour and perform layer cutting; Step 2: According to the thin-walled structure model designed in step 1, different process parameter combinations are set to perform laser powder bed fusion manufacturing of the thin-walled structure; Step 3: Based on the thin-walled structure manufactured in step 2, measure the actual wall thickness data of the thin-walled structure under different process input conditions and construct a data set. Based on the data set, train and optimize the deep learning prediction model for the wall thickness of the thin-walled structure; Step 4: Based on the optimal deep learning model optimized in step 3, the actual formed wall thickness of the thin-walled structure is predicted under the conditions of theoretical wall thickness and process parameter combination.
2. The method for predicting wall thickness of thin-walled structure laser powder bed fusion forming based on deep learning according to claim 1 is characterized in that: The step 1 specifically includes: Step 1.1: Design a thin-walled structure model as an overall model with varying wall thicknesses at different heights along the LPBF construction direction; Step 1.2: Set the scanning strategy for the laser turning point of the thin-walled structure outer contour to laser light jump or laser light continuous movement; Step 1.3: Set the LPBF printing layer thickness and perform model slicing.
3. The method for predicting wall thickness of thin-walled structure laser powder bed fusion forming based on deep learning according to claim 1 is characterized in that: The second step specifically includes: Step 2.1: Set the laser spot compensation parameters. The spot compensation value is defined as the distance between the outer side of the laser spot and the outer contour of the thin-walled structure. It is used to adjust the distance between the laser scanning path and the outer contour of the thin-walled structure. Step 2.2: Set the laser power and scanning speed to adjust the laser energy input during the manufacturing process of different thin-walled structures; Step 2.3: Configure different process parameter combinations for thin-walled structure models of the same design size and perform thin-walled structure laser powder layer fusion manufacturing. Each set of process parameters includes a scanning strategy, a laser power, a scanning speed, and a spot compensation value.
4. The method for predicting wall thickness of thin-walled structure laser powder bed fusion forming based on deep learning according to claim 3 is characterized in that: The step three specifically includes: Step 3.1: Based on the thin-walled structure manufactured in Step 2, measure the actual formed wall thickness of the thin-walled structure under different process input conditions, construct a data set of process parameters-design wall thickness-actual formed wall thickness, and eliminate abnormal data in the data set. The process input conditions include scanning strategy, spot compensation, laser power, and scanning speed; Step 3.2: Divide the measured data set into a training set, a validation set, and an independent test set; Step 3.3: Establish a deep learning prediction model for wall thickness of laser powder bed fusion thin-wall forming; Step 3.4: The deep learning prediction model is based on the design wall thickness, scanning strategy, Laser power, scanning speed, and spot compensation are used as model input data. The designed wall thickness, laser power, scanning speed, and spot compensation in the input data are standardized. Then, the deep learning prediction model processes the standardized data and outputs the predicted forming wall thickness of the thin-walled structure. The specific operation expression for the normalization of the input data of the deep learning prediction model is as follows: Where: is the new data after standardization, X i is the original data, E[x i ] is the average value of the training set data, is the standard deviation of the training set data; Step 3.5: During the training process of the deep learning prediction model, the K-fold cross-validation method is used to evaluate the influence of overfitting and selection bias. The absolute percentage error (APE) and accuracy (AR) between the predicted formed wall thickness and the actual formed wall thickness of the deep learning prediction model under the same conditions are calculated as indicators of the prediction accuracy of the deep learning prediction model. The specific operation expressions are formula (2) and formula (3), respectively: AR=(1-APE)(3) Where: To predict the formed wall thickness value, y i is the actual formed wall thickness value; Step 3.6: Adjust the number of neurons and hidden layers in the deep learning prediction model, then perform steps 3.4-3.5 again to obtain multiple sets of prediction accuracy. Compare the prediction accuracy of different deep learning prediction models, and determine the optimal deep learning prediction model by taking the deep learning prediction model with the highest accuracy. Step 3.7: Input the design wall thickness and process parameter combinations of the independent test set to obtain the predicted formed wall thickness under different input conditions and verify the optimal deep learning prediction model.
5. The method for predicting wall thickness of thin-walled structure laser powder bed fusion forming based on deep learning according to claim 4 is characterized in that: The step 4 specifically includes: Adjust the spot compensation value under different design wall thickness and process parameter combinations to obtain the corresponding predicted forming wall thickness of the thin-walled structure, so that the predicted forming wall thickness of the thin-walled structure and the design wall thickness are within the deviation threshold, obtain the optimized spot compensation range corresponding to the deviation threshold, establish a thin-walled structure forming wall thickness control process parameter table, and realize the thin-walled structure forming wall thickness prediction that meets the accuracy requirements. The thin-walled structure forming control process parameter table includes the design wall thickness, scanning strategy, laser power, scanning speed, predicted forming wall thickness range, and spot compensation range.
6. The method for predicting wall thickness of thin-walled structure laser powder bed fusion forming based on deep learning according to claim 5 is characterized in that: The deep learning prediction model in step 3.3 is implemented by the multi-layer perceptron algorithm, using the rectified linear unit (ReLU) function as the activation function and the RMSprop algorithm as the optimizer.
7. The method for predicting wall thickness of thin-walled structure laser powder bed fusion forming based on deep learning according to claim 5 is characterized in that: Each set of process parameters in step 2.3 includes three spot compensation values.
8. The method for predicting wall thickness of thin-walled structure laser powder bed fusion forming based on deep learning according to claim 5, characterized in that: The value of K in step 3.5 is 5.
9. The method for predicting wall thickness of thin-walled structure laser powder bed fusion forming based on deep learning according to claim 5, characterized in that: The deviation threshold in step 4 is 3%.