Intelligent 3D printing system
By using a dual-spectrum thermal imaging camera and a neural network model in a 3D printing system to construct material mapping relationships and automatically adjust printing parameters, the problems of differences in the fusion of various materials and blurred environmental boundaries are solved, achieving high-precision and rapid printing optimization design, and improving product quality and life prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HANGZHOU HANGLI ADDITIVE MFG TECH CO LTD
- Filing Date
- 2026-04-24
- Publication Date
- 2026-07-07
AI Technical Summary
Existing 3D printing technology lacks intelligent adjustment when dealing with differences in the fusion properties of various materials and blurred environmental boundaries, resulting in inaccurate printing parameters and an inability to correct deviations in real time, which affects product quality and lifespan.
A dual-spectrum thermal imaging camera is used to acquire printing data in real time. Combined with finite element analysis and neural network models, material mapping relationships are constructed, printing parameters are automatically adjusted, and real-time simulation and optimized design are achieved.
It enables product parameter adjustment based on different materials and working conditions, improves printing accuracy and product lifespan prediction, reduces the time of traditional trial and error methods, and enhances the robustness and accuracy of the system.
Smart Images
Figure CN122077935B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of additive manufacturing technology, and more particularly to an intelligent 3D printing system. Background Technology
[0002] Additive manufacturing, also known as 3D printing, is a manufacturing technology that builds three-dimensional entities by stacking materials layer by layer. Unlike traditional additive manufacturing, additive manufacturing is based on digital models, decomposing complex three-dimensional geometries into two-dimensional cross-sections, and then creating entities by stacking materials layer by layer. This technology can achieve the design of highly complex structures, while reducing material waste and shortening the production cycle.
[0003] While existing technologies can automatically identify the structure of drawings and perform additive manufacturing based on input drawing information, they still have shortcomings in terms of intelligence.
[0004] 1. The fusion properties of various materials can vary, which necessitates calculating or setting the printing parameters of the laser printhead in advance based on different product shapes and materials. This variation directly affects parameters such as the power, speed, spacing, and powder thickness of the printhead. Furthermore, 3D printing is a one-stop printing process that prints directly based on pre-set parameters and does not have the function of automatically correcting deviations based on the printed data while printing.
[0005] 2. The printing parameters are set only based on a wide range of usage environments, and the environmental boundary conditions are relatively vague. In actual application, products are divided into static and moving parts. The forces on static parts are relatively easy to predict, but the forces on moving parts are difficult to predict. This makes the 3D printing system not intelligent enough. Summary of the Invention
[0006] Based on the existing technical problems, this invention proposes an intelligent 3D printing system.
[0007] The present invention proposes an intelligent 3D printing system, including a frame and a controller. The frame is equipped with a laser print head and a powder bed mechanism. The powder bed mechanism lays up the powder, which is then melted by the laser print head to complete the cladding action.
[0008] The controller control method includes: S1, converting the product drawings to be additively manufactured and inputting them into the controller in a suitable format to construct a parametric model. .
[0009] S2, Model Perform finite element analysis and construct key parameter-model The mapping relationship Z(A) S →AF →A E Meanwhile, a manually / automatically adjustable fusion threshold R is set at the junctions of different materials to construct a mapping relationship Z between the different materials. R (A) S →A F →A E Finally, the optimal parameters for 3D printing are obtained.
[0010] S3. Set the theoretical additive profile based on the optimal parameters obtained in step S2 above. To carry out additive manufacturing.
[0011] S4. Use a dual-spectrum thermal imaging camera mounted on a 3D device to capture the actual additive outline of the bottom of the laser print head after additive processing. Data is collected.
[0012] S5. Theoretical additive contour line Compared with the actual additive profile The data is compared and the difference data is generated and input into S22 to correct the rapid proxy model.
[0013] S6. Automatically adjust the printing parameters of the laser printhead based on the data compared in S5.
[0014] Preferably, the model The parameters include product shape A. S Load A F and material property A E The load A F Including static load F 静 and dynamic load F 动 .
[0015] Preferably, a manually / automatically adjustable fusion threshold R is set at the junction of different materials to establish a mapping relationship Z between the different materials. R (A) S →A F →A E ).
[0016] The above technical solutions can ensure that the strength or load at the connection point remains consistent.
[0017] Preferably, S2 further includes S21, using Ansys or Motion software to apply load A to all locations of the product. F The forces are labeled and displayed, and the material properties A are analyzed in conjunction with dynamics. E The SN curve is obtained, and the service life of the structure is calculated based on Miner's linear cumulative damage theory. .
[0018] The above technical solutions allow for parameter adjustments to products made from different materials and under different working conditions.
[0019] Preferably, S2 further includes S22, mapping relationship Z (A S →A F →A E Forward mapping is used to build a fast proxy model.
[0020] S22 includes S221, generating parameter combinations.
[0021] S222. Data is automatically generated using finite element method.
[0022] S223, Dataset Construction and Preprocessing.
[0023] S224, Neural Network Model Training.
[0024] S225, Verification and Application.
[0025] The above technical solution offers the advantages of real-time simulation and optimized design.
[0026] Preferably, in step S222, the Abaqus script interface is used for automated modeling, then the logic is executed in a loop, and finally the data is cleaned. During the cleaning process, the simulation is checked for convergence, and abnormal results are removed.
[0027] In step S223, the format of the data samples is defined to form a matrix, the data samples are then normalized, and finally a dataset is generated.
[0028] Preferably, the dataset includes a training set, a validation set, and a test set.
[0029] In S225, the evaluation criteria include calculating the coefficient of determination R², root mean square error RMSE, and maximum relative error on the test set.
[0030] Preferably, in step S5, reverse mapping is used for correction.
[0031] The above technical solution can quickly and accurately identify complex material model parameters and, based on the dynamic response of the structure, infer the input or physical properties of the system.
[0032] Preferably, in the reverse mapping, the objective function is minimized using the iterative optimization method:
[0033] .
[0034] in, For shape A S Load AF and material property A E Any one or more parameter vectors to be identified in the above data are the difference data generated after the comparison.
[0035] Represented by the summation symbol, for summation from the 1st to the 2nd... The errors of each measurement point are accumulated.
[0036] This represents the total number of measurement points; Represented as weighting coefficients; The square of the error.
[0037] Represented as simulation results; The results are presented as experimental findings.
[0038] The above technical solution can be used to repeatedly adjust parameters through optimization algorithms to make the simulation results approximate the experiment; it has high accuracy and good robustness.
[0039] The beneficial effects of this invention are as follows:
[0040] 1. By setting the model Finite element analysis offers several advantages. First, it leverages forward mapping to enable trained networks to output results in seconds or even milliseconds, significantly accelerating real-time simulation of design iterations. Second, it allows neural networks to replace time-consuming finite element calculations in optimization processes like genetic algorithms, quickly finding optimal design parameters and effectively achieving optimized design. Third, it utilizes backward mapping to rapidly and accurately identify complex material model parameters, several times faster than traditional trial-and-error methods, and can infer system inputs or physical properties based on the structure's dynamic response.
[0041] 2. Set the theoretical additive profile based on optimal parameters. Based on the static load F 静 and dynamic load F 动 Additive manufacturing is used to obtain fatigue life cloud maps, critical node locations, and corresponding fatigue life of the overall structure based on the load magnitude and frequency under different static or dynamic load conditions. . Attached Figure Description
[0042] Figure 1 This is a schematic diagram of an intelligent 3D printing system proposed in this invention;
[0043] Figure 2 This is a front view of the laser print head of an intelligent 3D printing system proposed in this invention;
[0044] Figure 3 This invention proposes an intelligent 3D printing system. Figure 2 Figure AA in the diagram;
[0045] Figure 4 This invention proposes an intelligent 3D printing system. Figure 3 Image at point BB in the middle;
[0046] Figure 5 This invention proposes an intelligent 3D printing system. Figure 4 Enlarged view of point A in the image.
[0047] In the image: 1. Product; 2. Dual-spectrum thermal imaging camera; 3. Laser printhead. Detailed Implementation
[0048] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0049] Reference Figures 1-5 An intelligent 3D printing system, such as Figures 1-2 As shown, in the field of additive manufacturing, additive powder is still granular at the microscopic level. Therefore, when laying the powder onto a powder bed, the "particle size" and the laying thickness of the powder need to be considered to avoid problems such as over-melting or incomplete fusion caused by mismatched laser melting power of the laser printhead 3 during additive manufacturing. Specifically, the laser melting power of the laser printhead 3 is intelligently matched as follows:
[0050] like Figures 1-2 As shown, a dual-spectrum thermal imaging camera 2 is installed on the surface of the laser printhead 3 to irradiate the additive product 1 on the powder bed. This camera can simultaneously output visible light video and infrared thermal images, allowing for direct location of areas with abnormal temperatures (such as incomplete fusion or pores) and correlation with morphological features. The dual-spectrum thermal imaging camera 2 collects data on the temperature of the powder laser-melted area at the bottom of the laser printhead 3 and the morphological data of the molten additive layer. Finally, the morphological data is input into the controller for comparison and analysis. Specific steps include:
[0051] S1. Convert the product drawings that require additive manufacturing and input them into the controller in a suitable format. The product drawings include drawings created by 3D software such as CAD and SolidWorks. Then, simplify unnecessary or non-additive manufacturing parts in the electronic drawings. Finally, convert the electronic drawing format into a format that the additive manufacturing equipment can recognize, define the key parameters of product 1, and construct a parametric model. .
[0052] The model The parameters include the shape A of product 1. S Load A F and material property A E Among them, load A F Including static load F 静 and dynamic load F 动 Material property A E Including but not limited to any one or more components or any combination of components from titanium alloys, aluminum alloys, stainless steel or high-temperature alloys.
[0053] The reason why shape, load, and material properties are used as the main parameters of model A is that product 1, once manufactured, will be used on actual equipment and will be subject to forces and loads. Regardless of the static load F... 静 or dynamic load F 动 The load capacity is determined by the shape of product 1 and the type of 3D printing material. Meanwhile, shape A... S Load A F and material property A E The parameters between them can also affect each other, such as shape A. S The load A was determined F Size, material property A E It will also be due to shape A S Differences and load A F The size requires matching with the corresponding materials, and the three factors influence each other and jointly determine the performance of the final product 1. Therefore, the following steps are needed to express the correspondence between these three factors using parameters.
[0054] S2, such as Figure 5 As shown, for the model Perform finite element analysis and construct key parameter-model The mapping relationship Z(A) S →A F →A E Meanwhile, a manually / automatically adjustable fusion threshold R is set at the junctions of different materials to construct a mapping relationship Z between the different materials. R (A) S →A F →A E ).
[0055] The reason for establishing a mapping relationship between different materials is that there are differences in the connection strength between different materials. Compared to product 1 printed from the same material, the joints between multiple materials are most prone to breakage and cracking, which directly affects the load-bearing capacity of product 1. Therefore, a separate mapping relationship Z needs to be established. R (A) S →AF →A E Furthermore, for higher precision, the manual / automatic setting of the fusion threshold R can adjust the fusion properties between different materials, ensuring that the strength or load at the connection point remains consistent.
[0056] In actual use, the smaller the fusion threshold R, the better the fusion between the multiple materials. Therefore, the judgment criterion in the later actual calculation can be set to take the one with the smallest fusion threshold R.
[0057] S21, In order to be able to handle load A F To achieve parameterized continuous representation, Ansys or Motion software can be used to apply load A at all locations on product 1. F The forces applied are labeled and displayed.
[0058] Due to load A F Divided into static load F 静 and dynamic load F 动 Therefore, it is also necessary to further consider the stress load situation of product 1 under these two working conditions. In this embodiment, the static load F 静 It can be obtained through statics; while the dynamic load F 动 The load A can be obtained through transient dynamic analysis. F Response in static or dynamic processes.
[0059] Load A obtained from static and transient dynamic analysis F Load distribution diagrams can also be used in conjunction with dynamic analysis of material properties A E The SN curve is obtained, and the service life of the structure is calculated based on Miner's linear cumulative damage theory. That is, it can be based on material property A. E The fatigue life cloud map, critical node locations, and corresponding fatigue life of the overall structure are obtained by measuring the load magnitude and frequency under different static or dynamic load conditions. This approach closely matches the operational conditions of 3D printing, which is well-suited for producing small batches of products under varying conditions, such as aerospace and military components. These components typically require small batch sizes and fine-tuning of printing parameters based on the specific application conditions. Therefore, this embodiment can precisely adjust parameters for products using different materials and under different conditions, and then, based on the simulated load A... F The lifespan of the 3D-printed product 1 is predicted by measuring the size, static and dynamic load conditions and frequencies.
[0060] S22. In order to quickly find the optimal shape A S Load A Fand material property A E Design parameters for mapping relationship Z(A) S →A F →A E Forward mapping is used to build a fast proxy model.
[0061] Specifically: Use the finite element method to represent shape A S Load A F and material property A E A large number of simulation results with different parameter combinations are generated as a dataset, and a neural network is trained to learn the mapping relationship from parameters to results.
[0062] Including S221, generating parameter combinations; for shape A S Load A F and material property A E Define the parameters:
[0063] Let shape A S Variable range: length L ∈ [50, 200] mm, thickness t ∈ [2, 10] mm, aperture d ∈ [5, 30] mm, etc.;
[0064] Let load A F Variable range: Concentrated force F ∈ [100, 5000] N, pressure P ∈ [0.1, 10] MPa, torque M ∈ [10, 1000] Nm;
[0065] Material Property A E Variable range: titanium alloy, aluminum alloy, stainless steel or high-temperature alloy, etc.
[0066] The parameters generated above can be sampled using the Latin hypercube method, which, compared to random sampling, can more uniformly cover the multidimensional space. Finally, the output sample is determined.
[0067] S222. Data is automatically generated using finite element analysis; specifically, from automated modeling, the shape of the component can be parametrically created using the Abaqus script interface (abaqus python). S Set material properties A E Define load A F The process then proceeds to the loop execution logic, and finally, the data is cleaned. During cleaning, the simulation convergence is checked, and abnormal results are removed. Because some combinations may cause mesh distortion or non-convergence of calculations, it is necessary to record and supplement samples.
[0068] S223, Dataset Construction and Preprocessing; This step mainly defines the format of the data samples. In this embodiment, the data samples ultimately form a matrix. Due to shape A S Load AF and material property A E Because the data samples are different, they need to be normalized after forming the matrix. Finally, a dataset is generated, which includes a training set, a validation set, and a test set.
[0069] S224. Training the neural network model; the network architecture of the model can be any one of a multilayer perceptron, graph neural network, or convolutional neural network. The specific selection criteria can be based on the specific environment to choose an appropriate network architecture.
[0070] S225, Validation and Application; this includes accurate evaluation, with evaluation criteria including calculating the coefficient of determination R², root mean square error RMSE, and maximum relative error on the test set.
[0071] By analyzing the mapping relationship Z(A) S →A F →A E Forward mapping enables real-time simulation: a trained network can output results in seconds or even milliseconds, greatly accelerating design iterations. It also excels at optimizing designs: in optimization processes such as genetic algorithms, neural networks replace time-consuming finite element calculations, quickly finding the optimal design parameters.
[0072] S3. Set the theoretical additive profile based on the optimal parameters obtained in step S2 above. And set the laser power required for the corresponding 3D printer bed according to the optimal parameters. Scanning speed Scanning Spacing and powder layer thickness The reason for setting the theoretical additive manufacturing profile is... This is because the optimal design parameters mentioned above are affected by the 3D printer bed model, the "particle size" of the powder, and the layup thickness. In particular, the "particle size" of the powder and the layup thickness directly affect the laser melting power of the laser print head 3, which can easily cause the laser print head 3 to be mismatched, resulting in problems such as over-melting or incomplete fusion.
[0073] S4. After additive manufacturing begins, as follows: Figure 1 , Figure 4 as well as Figure 5 As shown, the dual-spectrum thermal imaging camera 2 captures the actual additive outline of the bottom of the laser print head 3 after additive processing. Data is collected;
[0074] S5. Theoretical additive contour line Compared with the actual additive profile The data is compared and the resulting difference data is input into S22 to correct or compensate for the fast proxy model. In this embodiment, the fast proxy model uses inverse mapping, which can use the minimization objective function in iterative optimization:
[0075] ;
[0076] in, For shape A S Load A F and material property A E Any one or more parameter vectors to be identified in the above comparison are the difference data generated after the comparison.
[0077] Represented by the summation symbol, for summation from the 1st to the 2nd... The errors of each measurement point are accumulated;
[0078] This represents the total number of measurement points; Represented as weighting coefficients; The square of the error;
[0079] Represented as simulation results; The results are presented as experimental findings.
[0080] The iterative optimization method can be used to repeatedly adjust parameters through optimization algorithms to make the simulation results approximate the experiment; it has high accuracy and good robustness.
[0081] Inverse mapping has the following advantages: 1. Based on experimental data, it can quickly and accurately identify complex material model parameters, which is several times faster than the traditional trial and error method.
[0082] 2. System identification: Based on the dynamic response of the structure, infer the effect of the system's input or physical characteristics.
[0083] S6. Automatically adjust the parameters of laser printhead 3 based on the data compared in S5; that is, the laser power. Scanning speed Scanning Spacing and powder layer thickness .
[0084] This application, on the one hand, fully utilizes forward mapping to enable trained networks to output results in seconds or even milliseconds, greatly accelerating the real-time simulation effect of design iteration. Simultaneously, it can replace time-consuming finite element calculations with neural networks in optimization processes such as genetic algorithms, quickly finding optimal design parameters and effectively achieving optimized design results. On the other hand, it utilizes backward mapping to quickly and accurately identify complex material model parameters, several times faster than traditional trial-and-error methods, and can infer the system's input or physical properties based on the structure's dynamic response.
[0085] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. An intelligent 3D printing system, comprising a frame and a controller, wherein the frame is equipped with a laser print head and a powder bed mechanism, wherein the powder bed mechanism lays up powder and then the laser print head melts it with a laser to complete the cladding action; Its features are: The controller control method includes: S1. Convert the product drawings that require additive manufacturing and input them into the controller in a suitable format to build a parametric model. ; The model The parameters include product shape A S Load A F and material property A E The load A F Including static load F 静 and dynamic load F 动 ; S2, Model Perform finite element analysis and construct key parameter-model The mapping relationship Z is established, and a fusion threshold R that can be manually / automatically adjusted is set at the junction of different materials to construct the mapping relationship Z between different materials. R Ultimately, the optimal parameters for 3D printing are obtained; S2 also includes; S21. Use Ansys or Motion software to apply load A to all locations of the product. F The forces are labeled and displayed, and the material properties A are analyzed in conjunction with dynamics. E The SN curve is obtained, and the service life of the structure is calculated based on Miner's linear cumulative damage theory. ; S2 also includes S22, and the mapping relationship Z adopts a forward mapping to construct a fast proxy model; S22 includes S221, generating parameter combinations; S222. Data is automatically generated using finite element analysis. S223, Dataset Construction and Preprocessing; S224, Neural network model training; S225, Verification and Application; S3. Set the theoretical additive profile based on the optimal parameters obtained in step S2 above. Based on the static load F 静 and dynamic load F 动 To perform additive manufacturing; S4. Use a dual-spectrum thermal imaging camera mounted on a 3D device to capture the actual additive outline of the bottom of the laser print head after additive processing. Data is collected; S5. Theoretical additive contour line Compared with the actual additive profile The data is compared and the difference data is generated and input into S22 to correct the rapid proxy model; S6. Automatically adjust the printing parameters of the laser printhead based on the data compared in S5.
2. The intelligent 3D printing system according to claim 1, characterized in that: A manually / automatically adjustable fusion threshold R is set at the junction of different materials to construct the mapping relationship Z between the different materials. R .
3. The intelligent 3D printing system according to claim 2, characterized in that: In S222, the Abaqus script interface is used for automated modeling, then the logic is executed in a loop, and finally the data is cleaned. During the cleaning process, the simulation is checked for convergence and abnormal results are removed. In step S223, the format of the data samples is defined to form a matrix, the data samples are then normalized, and finally a dataset is generated.
4. The intelligent 3D printing system according to claim 3, characterized in that: The dataset includes a training set, a validation set, and a test set; In S225, the evaluation criteria include calculating the coefficient of determination R², root mean square error RMSE, and maximum relative error on the test set.
5. The intelligent 3D printing system according to claim 4, characterized in that: In step S5, reverse mapping is used for correction.
6. The intelligent 3D printing system according to claim 5, characterized in that: In the reverse mapping, the objective function is minimized using the iterative optimization method: ； in, For shape A S Load A F and material property A E Any one or more parameter vectors to be identified in the above comparison are the difference data generated after the comparison. Represented by the summation symbol, for summation from the 1st to the 2nd... The errors of each measurement point are accumulated; This represents the total number of measurement points; Represented as weighting coefficients; Expressed as the square of the error; Represented as simulation results; The results are presented as experimental findings.