A method for compensating deformation of 3D printing aerospace engines

Abstract

The present invention discloses a method for compensating for deformation in 3D printing of aerospace engines, comprising the following steps: S1, constructing a three-dimensional CAD model and generating a set of printed layers; S2, obtaining the residual stress tensor field of the current printed layer and the historical printed layers; S3, constructing a stress shadow mapping tensor; S4, inputting the stress shadow mapping tensor and the current printed layer parameters into an inverse stress propagation neural network model, and outputting a deformation prediction vector field; S5, performing geometric compensation processing on the geometric area of ​​the printed layer based on the deformation prediction vector field; S6, generating a printing code based on the compensated geometric model; S7, comparing the actual deformation vector field with the overall predicted deformation vector field; if the error is greater than a preset tolerance threshold, feeding the error information back to step S5, and repeating steps S5 to S7 until the error vector field does not exceed the tolerance threshold. The present invention integrates residual stress modeling and inverse stress propagation neural network to achieve deformation compensation in 3D printing of aerospace engines.

Description

Technical Field

[0001] The present invention relates to the technical field of additive manufacturing and geometric modeling compensation, and in particular to a method for compensating for deformation in 3D printing of aerospace engines. Background Art

[0002] With the growing demand for high-performance, high-precision parts in the aerospace sector, additive manufacturing (AM), particularly metal 3D printing, is playing an increasingly important role in the complex fabrication of key aerospace engine components. This technology, which manufactures integral structural parts by stacking molten metal powder layer by layer, offers significant advantages such as high design freedom, high material utilization, and short manufacturing cycles. Typical processes, such as selective laser melting (SLM) and electron beam melting (EBM), have been widely used in complex geometries such as combustion chambers, nozzles, and turbine components.

[0003] However, because the metal 3D printing process is essentially a highly nonlinear thermal-mechanical coupling process, it is affected by factors such as laser energy input, scanning path, and material thermal properties. This can generate severe thermal gradients and non-uniform cooling during printing, which in turn induces complex residual stress field distributions. These residual stress fields will cause varying degrees of warping, shrinkage, stress concentration, and microcracks in structural components, significantly affecting the geometric accuracy and mechanical properties of the components, and in severe cases, even causing the parts to be scrapped. To suppress this printing deformation, existing technologies mainly adopt two types of solutions: one is to reduce stress concentration through process parameter optimization (such as adjusting the scanning speed and preheating temperature); the other is to perform geometric compensation on the CAD model before printing, that is, "reverse design" the pre-printing model based on the known deformation trend, so that it naturally rebounds to the target shape after printing.

[0004] Current research has attempted to use finite element analysis (FEA) to simulate printing residual stresses to predict deformation behavior, and then perform geometric compensation through linear offsets of model points. While these methods can improve forming accuracy to a certain extent, they generally suffer from three core issues: First, the finite element simulation process consumes large amounts of computational resources, making it difficult to achieve rapid response. Furthermore, they typically only model the stress field of the current layer, ignoring the cumulative impact of previous printed layers on the current layer, making it difficult to accurately capture cross-layer stress propagation and nonlinear response behavior. Second, geometric compensation methods often involve simple linear displacements of control points, lacking a nonlinear geometric reconstruction mechanism tightly coupled to stress prediction accuracy. This makes it impossible to effectively address multi-scale deformations on complex boundary surfaces or in regions with high-order curvature. Third, compensation strategies are typically one-way processes lacking a closed-loop feedback mechanism. Once the printing error deviates from the initial assumption, it is impossible to implement adaptive iterative corrections based on physical feedback.

[0005] In addition, with the rise of deep learning applications in engineering problems, some studies have attempted to build data-driven deformation prediction models, such as using convolutional neural networks or recurrent neural networks to predict deformation trends. However, most network structures are still relatively simple and fail to fully model the propagation path of stress tensors in spatial and temporal dimensions. They also lack the interpretability and physical consistency that matches the actual physical properties of residual stress, resulting in unstable prediction accuracy and weak generalization ability.

[0006] Therefore, how to provide a method for compensating for deformation in 3D printing of aerospace engines is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] One purpose of the present invention is to propose a method for compensating for deformation in 3D printing of aerospace engines. The present invention integrates residual stress tensor modeling, stress propagation modeling, and inverse stress propagation neural network prediction methods to construct a complete aerospace engine 3D printing deformation compensation process. The stress shadow mapping tensor and control point compensation function are introduced to realize multi-layer residual stress perception, prediction, and feedback compensation during the printing process. The method has high-precision, adaptive, and closed-loop optimization capabilities, significantly improving the forming quality and printing stability of complex structural parts.

[0008] A method for compensating for deformation in 3D printing of an aerospace engine according to an embodiment of the present invention includes the following steps:

[0009] S1. Constructing a three-dimensional CAD model of an aerospace engine component, and slicing the three-dimensional CAD model to generate a set of printing layers;

[0010] S2. Obtain the printing parameters of each printing layer, and obtain the residual stress tensor field of the current printing layer and several previous historical printing layers based on finite element simulation, and construct a residual stress input data set;

[0011] S3. The residual stress is input into the data set to perform propagation modeling in the time and space directions, and a stress shadow mapping tensor is constructed. The stress shadow mapping tensor is used to describe the residual stress influence area formed by the projection of the historical printing layer on the current layer;

[0012] S4, inputting the stress shadow mapping tensor and the printing parameters of the current printing layer into an inverse stress propagation neural network model, and outputting a deformation prediction vector field of the current printing layer;

[0013] S5. Performing geometric compensation processing on a geometric region corresponding to the printing layer in the three-dimensional CAD model according to the deformation prediction vector field, adjusting control points of the geometric region corresponding to the printing layer in the three-dimensional CAD model using a nonlinear geometric reparameterization method, and generating a compensated geometric model;

[0014] S6. Generate printing code according to the compensated geometric model and control the printing device to print;

[0015] S7. After printing is completed, the printed product is three-dimensionally scanned to obtain an actual deformation vector field. The actual deformation vector field is compared with the overall predicted deformation vector field formed by superimposing the deformation prediction vector fields of each printing layer. If the error vector field is greater than a preset tolerance threshold, the error information is fed back to step S5, and geometric compensation processing is re-performed on the corresponding printing layer geometric area in the 3D CAD model. Steps S5 to S7 are repeated until the error vector field is less than or equal to the preset tolerance threshold.

[0016] Optionally, the printing parameters include laser power, scanning speed, scanning path, scanning spacing, powder preheating temperature, laser spot diameter, and thermal conductivity, specific heat capacity, density, thermal expansion coefficient and Young's modulus of the printing material.

[0017] Optionally, the S2 specifically includes:

[0018] S21. Establishing a thermal-mechanical coupling finite element simulation model using the printing parameters and geometric information of the corresponding printed layer, wherein the establishment of the thermal-mechanical coupling finite element simulation model includes geometric discretization, mesh generation, material property assignment, and boundary condition setting;

[0019] S22, simulating the heat input process of the laser scanning path in the actual 3D printing process layer by layer based on the thermal-mechanical coupling finite element simulation model, and calculating the dynamic change process of the temperature field of each printed layer over time to obtain instantaneous temperature distribution and temperature gradient information;

[0020] S23, using the instantaneous temperature distribution and temperature gradient information to calculate the transient thermal stress field distribution of the current printing layer during the printing process, and after each layer is laser scanned and cooled to room temperature, calculate and obtain the residual stress tensor field of the printing layer after printing;

[0021] S24, performing cross-layer stress accumulation calculation on the residual stress tensor fields of several printing layers before the current printing layer, and establishing a residual stress tensor field history database, wherein the residual stress tensor field of each printing layer in the residual stress tensor field history database is associated with the current layer in an inter-layer accumulation manner;

[0022] S25. Construct a residual stress input data set based on the residual stress tensor field of the current printing layer and the residual stress tensor fields of several previous historical printing layers in the residual stress tensor field history database.

[0023] Optionally, the S3 specifically includes:

[0024] S31. Based on the residual stress tensor field of each printing layer in the residual stress input data set, a three-dimensional space coordinate system and a printing time axis are established with the current printing layer as the reference. The three-dimensional coordinate points are recorded as ,in Indicates the spatial coordinates of the printed part in the width direction, Indicates the spatial coordinates of the printed part in the length direction, Indicates the spatial coordinates of the printed part in the height direction, The direction corresponds to the stacking direction of layer-by-layer printing;

[0025] S32. In the three-dimensional spatial coordinate system, spatially register the residual stress tensor fields of each historical printing layer by three-dimensional bilinear interpolation to obtain a tensor expression with consistent spatial resolution and coordinate frame with the current printing layer;

[0026] S33. Based on the thermo-elastoplastic mechanics propagation theory, a stress propagation function model is constructed:

[0027] ;

[0028] in, Indicates the Layer The propagation of layers affects the tensor, Indicates the Layer at 3D coordinate point The residual stress tensor on Indicates the The material coupling coefficient of the layer, represents the exponential decay factor, Indicates the Layer to Layer The vertical interlayer distance in the direction;

[0029] S34. Through the time weight decay algorithm, the propagation influence tensor of the historical printing layer is time-weighted to construct the stress shadow mapping tensor , used to describe the spatial position of the historical printing layer in the current printing layer The residual stress affected area formed by the upper projection is:

[0030] ;

[0031] in, represents the time decay control parameter, Indicates the The printing time interval from the first layer to the current printing layer, Indicates the The printing time interval from the first layer to the current printing layer, The value of , Indicates the total number of printing layers;

[0032] S35, output the stress shadow mapping tensor .

[0033] Optionally, the S4 specifically includes:

[0034] S41, constructing a reverse stress propagation neural network model, wherein the reverse stress propagation neural network model includes an input layer, a spatiotemporal tensor feature fusion layer, a bidirectional gated recurrent unit layer, a tensor attention mechanism layer, and an output prediction layer;

[0035] S42, inputting the stress shadow mapping tensor and the printing parameters of the current printing layer into the input layer of the inverse stress propagation neural network model, the input layer converting the three-dimensional tensor data into a high-dimensional feature vector and performing data preprocessing through linear normalization;

[0036] S43, the spatiotemporal tensor feature fusion layer performs a multi-scale convolution operation on the high-dimensional feature vector, extracts spatial local features and inter-layer temporal features through multiple convolution kernels, and generates a spatiotemporal coupling feature expression by using a feature splicing operation;

[0037] S44, the bidirectional gated recurrent unit layer processes the spatiotemporal coupling feature expression, using a gated unit mechanism to simultaneously capture the forward propagation of the impact of the historical printing layer on the current printing layer and the reverse tracing of the stress state of the current layer on the historical layer, thereby obtaining a forward and reverse coupled bidirectional stress propagation feature expression;

[0038] S45, the tensor attention mechanism layer uses the attention calculation method of the spatial dimension and channel dimension based on the bidirectional stress propagation feature representation to generate a space-channel joint attention weight mask tensor ,in Represents a three-dimensional coordinate point, Indicates the feature channel number;

[0039] S46, the output prediction layer fuses the spatial-channel joint attention weight mask tensor with the bidirectional stress propagation feature by tensor multiplication, and uses The activation function performs nonlinear mapping and outputs the deformation prediction vector field of the current printing layer. The deformation prediction vector field is for each three-dimensional coordinate point The corresponding spatial deformation prediction vector represents the deformation prediction of the current printing layer.

[0040] Optionally, the S5 specifically includes:

[0041] S51, extracting a corresponding geometric region of the current printing layer in the three-dimensional CAD model based on the deformation prediction vector field, and locating the CAD surface segments and boundary range contained in the geometric region;

[0042] S52, perform parametric modeling on the geometric area of ​​the printing layer, convert the original CAD surface segment expression into a re-parameterized surface form based on non-uniform rational B-spline, and construct a two-dimensional control point array ,in Indicates the first segment of the original CAD surface Rank The spatial coordinate vector of the column control point;

[0043] S53. Construct a control point compensation function based on the distribution of the deformation prediction vector field in the control point space:

[0044] ;

[0045] in, Represents the spatial coordinate vector of the control point after geometric compensation, represents the global compensation adjustment coefficient, represents the direction weight matrix, express The deformation prediction vector at the location, represents the deformation adjustment coefficient, represents the total number of multi-scale layers, Representation scale The weighting factor of express Location on scale The average deformation prediction vector of the lower adjacent region;

[0046] S54, performing surface reconstruction based on the spatial coordinate vectors of the control points after geometric compensation to form an updated expression of the area corresponding to the current printing layer in the three-dimensional CAD model after geometric compensation;

[0047] S55. Perform a geometric validity check, which includes constraint judgments on boundary closure, surface continuity, and the maximum compensation amplitude not exceeding the corresponding preset tolerance. If the constraint judgment conditions are met, the compensated geometric model is output.

[0048] Optionally, the S6 specifically includes:

[0049] S61, converting the compensated geometric model into an intermediate model file in a processing format, wherein the intermediate model adopts a faceting format for geometric approximation, and the file format is STL;

[0050] S62, performing path planning processing on the intermediate model, adjusting the scanning spacing and path density according to the curvature distribution and structure thickness of the geometric compensation area, and generating a printing path set;

[0051] S63: Generate a print control parameter table based on the print path set through print parameter mapping;

[0052] S64: Generate a print control code recognized by a standard printing device based on the print control parameter table, wherein the print control code is an additive manufacturing control program based on a G-code structure and includes layer-by-layer control instructions and compensation correction amount annotation information;

[0053] S65: Transmit the printing control code to the printing device control module, and control the printing device to perform additive manufacturing operations layer by layer according to the printing control code.

[0054] Optionally, the S7 specifically includes:

[0055] S71. After printing is completed, a laser scanner is used to perform a three-dimensional full-field scan of the printed product to obtain point cloud data of the complete surface and construct a scan model;

[0056] S72, aligning the scanned model with the compensated 3D CAD model, using an iterative closest point algorithm to achieve rigid registration, and establishing a point-to-point spatial mapping relationship;

[0057] S73, calculating an actual deformation vector field based on the registration result, where the actual deformation vector field is the spatial difference between corresponding points of the scanned model and the compensated 3D CAD model in a unified coordinate system;

[0058] S74, accumulating the deformation prediction vector field of each layer during the printing process layer by layer to form an overall predicted deformation vector field, and comparing it with the actual deformation vector field to obtain an error vector field;

[0059] S75. Extract the error peak area based on the distribution characteristics of the error vector field in the three-dimensional space, and obtain the control point indexes in the corresponding printing layer and geometric area;

[0060] S76. The sampled value of the error vector field at the control point position is used as an error correction term and introduced into the control point update mechanism. The control point is compensated for the second time based on the previous round of compensation results:

[0061] ;

[0062] in, Represents the spatial coordinate vector of the control point after geometric compensation, represents the spatial coordinate vector of the control point after error feedback adjustment, for The error correction term at position, is the error adjustment coefficient;

[0063] S77. Use the control points adjusted by error feedback to reconstruct the corresponding printing layer geometric area in the 3D CAD model to generate a new geometric model, and repeat steps S5 to S7 until the error vector field is less than or equal to the preset tolerance threshold.

[0064] The beneficial effects of the present invention are:

[0065] First, the present invention introduces a residual stress tensor field simulation mechanism based on thermal-mechanical coupled finite element simulation and constructs a residual stress input dataset for historical printed layers. This establishes a high-dimensional, continuous physical data foundation for subsequent stress propagation and deformation prediction. This residual stress tensor not only considers the thermal stress distribution of the current printed layer but also fully models the coupling of historical layers to the current layer in both space and time, resolving the problem that traditional methods cannot capture the nonlinear stress transfer between multiple printed layers.

[0066] Secondly, this paper introduces the concept of a stress shadow mapping tensor for the first time. Combining a thermo-elastoplastic propagation function with a time decay algorithm, this method maps and time-weights the residual stresses of historical printed layers layer by layer in three-dimensional space, visualizing cross-layer stress effects and enhancing structural perception. This stress shadow mapping tensor serves as a key input to the inverse stress propagation neural network model, enabling a high-dimensional representation of the spatiotemporal evolution of the historical stress field, improving the accuracy and robustness of deformation prediction.

[0067] In addition, the present invention designs an improved multi-level coupled neural network structure, which integrates spatiotemporal convolution, bidirectional gated recurrent unit and tensor attention mechanism, significantly enhancing the network's ability to model complex nonlinear stress-deformation relationships. The nonlinear activation function and the spatial channel joint attention mechanism improve the deformation prediction resolution of key areas. The output deformation prediction vector field has stronger local expression ability and global coordination ability, providing highly reliable data support for subsequent compensation.

[0068] Furthermore, this invention employs a geometric reparameterization compensation method based on non-uniform rational B-splines, combined with the multi-scale deformation prediction response of control point positions, to construct a control point compensation function, breaking through traditional linear displacement compensation methods. By adjusting the directional weights of local control points, it achieves refined compensation for complex boundaries and high-curvature areas, ensuring that the local geometric response of the 3D CAD model more closely matches the actual deformation trend.

[0069] Ultimately, by constructing a feedback mechanism based on an error vector field, this paper establishes a closed-loop control process for fine-tuning geometric errors toward control points after printing. After each printing cycle, the error vector is sampled, mapped, and fed back to the control point update process, enabling adaptive optimization of deformation prediction and geometric compensation strategies. This significantly improves the consistency and accuracy of printing complex structural parts, while reducing the number of iterations and resource waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0071] Figure 1 This is an overall flow chart of a method for compensating for deformation in 3D printing of aerospace engines proposed by the present invention;

[0072] Figure 2 This is a schematic diagram of the structure of a reverse stress propagation neural network model for a method of compensating for deformation in 3D printing of aerospace engines proposed in the present invention;

[0073] Figure 3 This is a flow chart of the control point error feedback compensation mechanism of the aerospace engine 3D printing deformation compensation method proposed by the present invention. DETAILED DESCRIPTION

[0074] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0075] refer to Figure 1-Figure 3 A method for compensating for deformation in 3D printing of an aerospace engine comprises the following steps:

[0076] S1. Constructing a three-dimensional CAD model of an aerospace engine component, and slicing the three-dimensional CAD model to generate a set of printing layers;

[0077] S2. Obtain the printing parameters of each printing layer, and obtain the residual stress tensor field of the current printing layer and several previous historical printing layers based on finite element simulation, and construct a residual stress input data set;

[0078] S3. The residual stress is input into the data set to perform propagation modeling in the time and space directions, and a stress shadow mapping tensor is constructed. The stress shadow mapping tensor is used to describe the residual stress influence area formed by the projection of the historical printing layer on the current layer;

[0079] S4, inputting the stress shadow mapping tensor and the printing parameters of the current printing layer into an inverse stress propagation neural network model, and outputting a deformation prediction vector field of the current printing layer;

[0080] S5. Performing geometric compensation processing on a geometric region corresponding to the printing layer in the three-dimensional CAD model according to the deformation prediction vector field, adjusting control points of the geometric region corresponding to the printing layer in the three-dimensional CAD model using a nonlinear geometric reparameterization method, and generating a compensated geometric model;

[0081] S6. Generate printing code according to the compensated geometric model and control the printing device to print;

[0082] S7. After printing is completed, the printed product is three-dimensionally scanned to obtain an actual deformation vector field. The actual deformation vector field is compared with the overall predicted deformation vector field formed by superimposing the deformation prediction vector fields of each printing layer. If the error vector field is greater than a preset tolerance threshold, the error information is fed back to step S5, and geometric compensation processing is re-performed on the corresponding printing layer geometric area in the 3D CAD model. Steps S5 to S7 are repeated until the error vector field is less than or equal to the preset tolerance threshold.

[0083] By establishing a full-process 3D printing deformation compensation system from CAD modeling, stress analysis, predictive modeling, geometric compensation to printing verification and feedback closed loop, active control of the printing accuracy of complex components of aerospace engines has been achieved. The entire process takes into account the coupling characteristics of physical modeling and intelligent prediction. It not only fully considers the cross-layer propagation effect of residual stress during the printing process, but also can provide real-time feedback on printing errors and perform iterative geometric adjustments. This method connects the three key links of prediction, compensation and feedback verification, has good engineering applicability, systematicness and scalability, and significantly improves the dimensional consistency and forming stability of metal 3D printed structures. It is particularly suitable for the manufacture of high-precision and high-complexity aerospace parts.

[0084] In this embodiment, the printing parameters include laser power, scanning speed, scanning path, scanning spacing, powder preheating temperature, laser spot diameter, and thermal conductivity, specific heat capacity, density, thermal expansion coefficient and Young's modulus of the printing material.

[0085] In this embodiment, S2 specifically includes:

[0086] S21. Establishing a thermal-mechanical coupling finite element simulation model using the printing parameters and geometric information of the corresponding printed layer, wherein the establishment of the thermal-mechanical coupling finite element simulation model includes geometric discretization, mesh generation, material property assignment, and boundary condition setting;

[0087] S22, simulating the heat input process of the laser scanning path in the actual 3D printing process layer by layer based on the thermal-mechanical coupling finite element simulation model, and calculating the dynamic change process of the temperature field of each printed layer over time to obtain instantaneous temperature distribution and temperature gradient information;

[0088] S23, using the instantaneous temperature distribution and temperature gradient information to calculate the transient thermal stress field distribution of the current printing layer during the printing process, and after each layer is laser scanned and cooled to room temperature, calculate and obtain the residual stress tensor field of the printing layer after printing;

[0089] S24, performing cross-layer stress accumulation calculation on the residual stress tensor fields of several printing layers before the current printing layer, and establishing a residual stress tensor field history database, wherein the residual stress tensor field of each printing layer in the residual stress tensor field history database is associated with the current layer in an inter-layer accumulation manner;

[0090] S25. Construct a residual stress input data set based on the residual stress tensor field of the current printing layer and the residual stress tensor fields of several previous historical printing layers in the residual stress tensor field history database.

[0091] By constructing a thermal-mechanical coupling finite element model to simulate the heat conduction, thermal gradient, and thermal stress evolution paths during the printing process, multi-layer dynamic acquisition of the printed residual stress tensor field is achieved. The residual stress of the historical printing layer is not only accurately quantified, but also preserved through accumulation and database storage, providing hierarchical data support for subsequent stress propagation modeling and deformation prediction. This modeling mechanism, from local to global, from transient to steady-state, effectively solves the problems of rough stress estimation and cross-layer information discontinuity in traditional methods, making the prediction results closer to the actual component behavior and greatly improving the reliability of the geometric compensation strategy.

[0092] In this embodiment, S3 specifically includes:

[0093] S31. Based on the residual stress tensor field of each printing layer in the residual stress input data set, a three-dimensional space coordinate system and a printing time axis are established with the current printing layer as the reference. The three-dimensional coordinate points are recorded as ,in Indicates the spatial coordinates of the printed part in the width direction, Indicates the spatial coordinates of the printed part in the length direction, Indicates the spatial coordinates of the printed part in the height direction, The direction corresponds to the stacking direction of layer-by-layer printing;

[0094] S32. In the three-dimensional spatial coordinate system, spatially register the residual stress tensor fields of each historical printing layer by three-dimensional bilinear interpolation to obtain a tensor expression with consistent spatial resolution and coordinate frame with the current printing layer;

[0095] S33. Based on the thermo-elastoplastic mechanics propagation theory, a stress propagation function model is constructed:

[0096] ;

[0097] in, Indicates the Layer The propagation of layers affects the tensor, Indicates the Layer at 3D coordinate point The residual stress tensor on Indicates the The material coupling coefficient of the layer, represents the exponential decay factor, Indicates the Layer to Layer The vertical interlayer distance in the direction;

[0098] S34. Through the time weight decay algorithm, the propagation influence tensor of the historical printing layer is time-weighted to construct the stress shadow mapping tensor , used to describe the spatial position of the historical printing layer in the current printing layer The residual stress affected area formed by the upper projection is:

[0099] ;

[0100] in, represents the time decay control parameter, Indicates the The printing time interval from the first layer to the current printing layer, Indicates the The printing time interval from the first layer to the current printing layer, The value of , Indicates the total number of printing layers;

[0101] S35, output the stress shadow mapping tensor .

[0102] By establishing a stress propagation function and introducing a time decay algorithm, they achieved the first three-dimensional mapping of the impact of historical printed layers on the stress of the current layer. They proposed the concept of a stress shadow mapping tensor, effectively describing the spatial distribution of the cross-layer coupling effect. This modeling approach not only considers the nonlinear relationship between spatial coordinates and interlayer distances but also incorporates weight decay in the time dimension, making the modeling results temporally causal and physically reasonable. The resulting tensor provides structured input for the neural network, making subsequent predictions more accurate, especially in complex components with significant multi-layer superposition effects.

[0103] In this embodiment, the S4 specifically includes:

[0104] S41, constructing a reverse stress propagation neural network model, wherein the reverse stress propagation neural network model includes an input layer, a spatiotemporal tensor feature fusion layer, a bidirectional gated recurrent unit layer, a tensor attention mechanism layer, and an output prediction layer;

[0105] S42, inputting the stress shadow mapping tensor and the printing parameters of the current printing layer into the input layer of the inverse stress propagation neural network model, the input layer converting the three-dimensional tensor data into a high-dimensional feature vector and performing data preprocessing through linear normalization;

[0106] S43, the spatiotemporal tensor feature fusion layer performs a multi-scale convolution operation on the high-dimensional feature vector, extracts spatial local features and inter-layer temporal features through multiple convolution kernels, and generates a spatiotemporal coupling feature expression by using a feature splicing operation;

[0107] S44, the bidirectional gated recurrent unit layer processes the spatiotemporal coupling feature expression, using a gated unit mechanism to simultaneously capture the forward propagation of the impact of the historical printing layer on the current printing layer and the reverse tracing of the stress state of the current layer on the historical layer, thereby obtaining a forward and reverse coupled bidirectional stress propagation feature expression;

[0108] S45, the tensor attention mechanism layer uses the attention calculation method of the spatial dimension and channel dimension based on the bidirectional stress propagation feature representation to generate a space-channel joint attention weight mask tensor ,in Represents a three-dimensional coordinate point, Indicates the feature channel number;

[0109] S46, the output prediction layer fuses the spatial-channel joint attention weight mask tensor with the bidirectional stress propagation feature by tensor multiplication, and uses The activation function performs nonlinear mapping and outputs the deformation prediction vector field of the current printing layer. The deformation prediction vector field is for each three-dimensional coordinate point The corresponding spatial deformation prediction vector represents the deformation prediction of the current printing layer.

[0110] A reverse stress propagation neural network was constructed, consisting of an input layer, a spatiotemporal tensor feature fusion layer, a bidirectional gated recurrent unit layer, a tensor attention mechanism layer, and an output prediction layer, enabling deep learning-based coupled modeling of the historical stress field and printing parameters. The bidirectional GRU architecture employed captures both the forward propagation and reverse feedback of stress, while the tensor attention mechanism further enhances feature focus in key regions. The resulting deformation prediction vector field exhibits not only high spatial resolution and sufficient channel representation, but also excellent adjustability and interpretability, significantly outperforming traditional single-layer perception-based neural network prediction methods.

[0111] In this embodiment, the S5 specifically includes:

[0112] S51, extracting a corresponding geometric region of the current printing layer in the three-dimensional CAD model based on the deformation prediction vector field, and locating the CAD surface segments and boundary range contained in the geometric region;

[0113] S52, perform parametric modeling on the geometric area of ​​the printing layer, convert the original CAD surface segment expression into a re-parameterized surface form based on non-uniform rational B-spline, and construct a two-dimensional control point array ,in Indicates the first segment of the original CAD surface Rank The spatial coordinate vector of the column control point;

[0114] S53. Construct a control point compensation function based on the distribution of the deformation prediction vector field in the control point space:

[0115] ;

[0116] in, Represents the spatial coordinate vector of the control point after geometric compensation, represents the global compensation adjustment coefficient, represents the direction weight matrix, express The deformation prediction vector at the location, represents the deformation adjustment coefficient, represents the total number of multi-scale layers, Representation scale The weighting factor of express Location on scale The average deformation prediction vector of the lower adjacent region;

[0117] S54, performing surface reconstruction based on the spatial coordinate vectors of the control points after geometric compensation to form an updated expression of the area corresponding to the current printing layer in the three-dimensional CAD model after geometric compensation;

[0118] S55. Perform a geometric validity check, which includes constraint judgments on boundary closure, surface continuity, and the maximum compensation amplitude not exceeding the corresponding preset tolerance. If the constraint judgment conditions are met, the compensated geometric model is output.

[0119] The proposed geometric compensation method, based on the control point mechanism of non-uniform rational B-spline surfaces, achieves refined compensation in localized areas by constructing a compensation function strongly coupled to the deformation prediction field. Multi-scale adjustment and directional weighting ensure that control point adjustments not only conform to deformation trends but also suppress geometric mutations caused by local overcompensation. This compensation strategy is adaptable to complex CAD geometric fragments of arbitrary shapes, resulting in smooth and stable compensation results and naturally transitioning boundaries between the compensation areas. This fundamentally addresses the issues of loss of control and discontinuity often encountered with traditional displacement methods.

[0120] In this embodiment, S6 specifically includes:

[0121] S61, converting the compensated geometric model into an intermediate model file in a processing format, wherein the intermediate model adopts a faceting format for geometric approximation, and the file format is STL;

[0122] S62, performing path planning processing on the intermediate model, adjusting the scanning spacing and path density according to the curvature distribution and structure thickness of the geometric compensation area, and generating a printing path set;

[0123] S63: Generate a print control parameter table based on the print path set through print parameter mapping;

[0124] S64: Generate a print control code recognized by a standard printing device based on the print control parameter table, wherein the print control code is an additive manufacturing control program based on a G-code structure and includes layer-by-layer control instructions and compensation correction amount annotation information;

[0125] S65: Transmit the printing control code to the printing device control module, and control the printing device to perform additive manufacturing operations layer by layer according to the printing control code.

[0126] By establishing a fully automated print path planning and parameter mapping process from CAD geometry to G-code, we ensure that the compensated model can be losslessly converted into a print control program recognizable by the device. During the print path generation process, the scan density is dynamically adjusted based on the local complexity of the geometric area, effectively improving print quality and material utilization efficiency. In addition, compensation information is embedded in the print instruction stream as an annotation, facilitating subsequent traceability and parameter correction.

[0127] In this embodiment, the S7 specifically includes:

[0128] S71. After printing is completed, a laser scanner is used to perform a three-dimensional full-field scan of the printed product to obtain point cloud data of the complete surface and construct a scan model;

[0129] S72, aligning the scanned model with the compensated 3D CAD model, using an iterative closest point algorithm to achieve rigid registration, and establishing a point-to-point spatial mapping relationship;

[0130] S73, calculating an actual deformation vector field based on the registration result, where the actual deformation vector field is the spatial difference between corresponding points of the scanned model and the compensated 3D CAD model in a unified coordinate system;

[0131] S74, accumulating the deformation prediction vector field of each layer during the printing process layer by layer to form an overall predicted deformation vector field, and comparing it with the actual deformation vector field to obtain an error vector field;

[0132] S75. Extract the error peak area based on the distribution characteristics of the error vector field in the three-dimensional space, and obtain the control point indexes in the corresponding printing layer and geometric area;

[0133] S76. The sampled value of the error vector field at the control point position is used as an error correction term and introduced into the control point update mechanism. The control point is compensated for the second time based on the previous round of compensation results:

[0134] ;

[0135] in, Represents the spatial coordinate vector of the control point after geometric compensation, represents the spatial coordinate vector of the control point after error feedback adjustment, for The error correction term at position, is the error adjustment coefficient;

[0136] S77. Use the control points adjusted by error feedback to reconstruct the corresponding printing layer geometric area in the 3D CAD model to generate a new geometric model, and repeat steps S5 to S7 until the error vector field is less than or equal to the preset tolerance threshold.

[0137] A geometric model recompensation mechanism based on error vector field feedback is proposed, forming a closed-loop control path of 3D scanning, error identification, and control point reoptimization. Point-by-point comparison of the actual and predicted deformation fields provides a precise basis for error localization, while the extraction of error peak regions and control point index mapping significantly improve compensation efficiency. Furthermore, a secondary iterative update of control points driven by error correction terms imbues the model compensation capability with adaptive and progressive optimization characteristics. This mechanism is particularly suitable for complex surface multi-print correction tasks, significantly improving compensation accuracy and system convergence.

[0138] Example 1:

[0139] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a certain type of aerospace engine nozzle lining component as the test object of this embodiment. During the actual printing process, this component is prone to significant deformation due to temperature gradients and accumulated thermal stresses, affecting the precision control of key dimensions.

[0140] First, a 3D CAD model of the target nozzle lining was constructed using a 3D CAD modeling tool. Slicing was then performed using an industrial-grade slicing engine, generating a set of 120 print layers. Each layer was set to 0.1mm thick, resulting in a total print height of 12mm.

[0141] For each layer, the relevant printing parameters were extracted, including: laser power 350W, scanning speed 800mm / s, scanning path using a rotation partitioning strategy, scanning spacing 0.08mm, powder preheating temperature 150℃, laser spot diameter 70μm. The printing material is a high-temperature alloy with a thermal conductivity of 18W / (m·K), a specific heat capacity of 435J / (kg·K), a density of 8250kg / m³, and a thermal expansion coefficient of , Young's modulus is 135GPa.

[0142] Based on the aforementioned printing parameters, a thermal-mechanical coupled finite element model was constructed using multiphysics simulation software. Taking into account the effects of the laser scanning path and heat input, dynamic time stepping was used to obtain the instantaneous temperature field and its gradient for each printed layer. The thermal-elastic-plastic stress calculation module then extracted the residual stress tensor field for each layer after cooling to room temperature. The stress results for historical printed layers were accumulated to construct a residual stress history database.

[0143] Then, based on the spatial coordinates A unified tensor field framework is established, and all residual stress tensor fields are spatially registered using three-dimensional bilinear interpolation. Then, the exponential decay function and time weighting function are used to construct the stress propagation model and stress shadow mapping tensor. By introducing time decay control parameters , all layers are weighted superimposed to obtain the stress projection influence area for the current printing layer.

[0144] Will The combined input, along with the printing parameters, is fed into the proposed improved reverse stress propagation neural network. This network architecture comprises five components, responsible for feature extraction, spatiotemporal fusion, gated state tracking, attention weighting, and output prediction. During training, approximately 2,000 sets of historical printing data were used, and the Adam optimizer was used for 30 rounds of training, resulting in a prediction accuracy of 94.7%.

[0145] Output deformation prediction vector field It is used to compensate the control points of the CAD model area corresponding to the printing layer. The CAD model is represented by NURBS. Adjustments are made according to the following compensation function:

[0146] ;

[0147] in, Represents the space coordinate vector of the control point after compensation, Indicates the global compensation adjustment coefficient, the value is 0.7, Represents the multi-scale adjustment coefficient, with a value of 0.3. represents the direction weight matrix; represents the deformation prediction vector, Indicates the The weighting factor at each scale, Indicates the The average deformation prediction vector within the neighborhood at this scale, Indicates the total number of scales, and its value is 3.

[0148] The compensated model is converted into an STL file, and the scanning path is generated using the printing path planning module. The G-code control program is automatically generated based on the compensated path and parameters and loaded into the printing device for printing.

[0149] After printing, a laser scanner was used to perform a full-surface 3D scan of the sample. Iterative closest point registration was performed with the pre-compensated CAD model to determine the actual deformation vector field. The error vector field was primarily concentrated in the Z direction, with a maximum height deviation of +0.14mm and a minimum of -0.08mm. Based on this error vector field, a secondary compensation iteration was performed on the control points. After reprinting, the maximum error of the sample was reduced to ±0.035mm, meeting the required tolerance of ±0.05mm.

[0150] Table 1 Comparison of deformation compensation data for 3D printing of aerospace engines

[0151]

[0152] By analyzing Table 1 above, it can be seen that during the initial printing process without compensation, the component has obvious spatial deformation errors, with the maximum positive deviation reaching +0.14mm, the maximum negative deviation being -0.08mm, and the average prediction error being 0.067mm, which significantly exceeds the set tolerance threshold of ±0.05mm, indicating that traditional printing methods are difficult to effectively control local warping and dimensional deviation caused by residual stress. Through the aerospace engine 3D printing deformation compensation method proposed in the present invention, combined with stress shadow mapping tensor modeling and reverse stress propagation neural network prediction mechanism, in the second printing after implementing one compensation, the overall spatial deviation of the component is significantly reduced. The data in Table 1 show that after compensation, the maximum positive deviation drops to +0.032mm, the maximum negative deviation is -0.035mm, and the overall deformation is controlled within the tolerance range of ±0.05mm. At the same time, the average prediction error is reduced to 0.021mm, and the prediction accuracy is significantly improved, verifying the synergistic correction effect between the deformation prediction vector field and the nonlinear reparameterization of the control points. From the adjustment of the compensation parameters in Table 1, it can be seen that by introducing the error vector field to perform feedback correction on the spatial position of the control point and reasonably adjusting the global compensation adjustment coefficient (from 0.7 to 0.6) and the multi-scale adjustment coefficient (from 0.3 to 0.4), the local adaptability of the geometric response is enhanced while avoiding the sudden change or continuity destruction of the geometric morphology, reflecting the stability and adaptability of the compensation model.

[0153] This embodiment fully demonstrates the practical application effect of the present invention in the additive manufacturing of key components of aerospace engines. By constructing a multi-physics field simulation model, establishing a stress propagation mechanism, introducing an inverse stress propagation neural network for deformation prediction, and combining nonlinear geometric reparameterization technology to perform fine compensation for the control points of the three-dimensional CAD model, the deformation prediction before printing and the error feedback closed-loop control after printing are effectively achieved. After actual testing, without changing the manufacturing process path, this method significantly reduced the maximum geometric error of the printed part from ±0.14mm to ±0.035mm, comprehensively improving the printing accuracy and part quality, and verifying the feasibility, advancement and engineering practical value of the method in the manufacture of high-precision complex aerospace components.

[0154] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for compensating for deformation in 3D printing of aerospace engines, characterized in that: The steps include: S1. Constructing a three-dimensional CAD model of an aerospace engine component, and slicing the three-dimensional CAD model to generate a set of printing layers; S2. Obtain the printing parameters of each printing layer, and obtain the residual stress tensor field of the current printing layer and several previous historical printing layers based on finite element simulation, and construct a residual stress input data set; S3. The residual stress is input into the data set to perform propagation modeling in the time and space directions, and a stress shadow mapping tensor is constructed. The stress shadow mapping tensor is used to describe the residual stress influence area formed by the projection of the historical printing layer on the current layer; S4, inputting the stress shadow mapping tensor and the printing parameters of the current printing layer into an inverse stress propagation neural network model, and outputting a deformation prediction vector field of the current printing layer; S5. Performing geometric compensation processing on a geometric region corresponding to the printing layer in the three-dimensional CAD model according to the deformation prediction vector field, adjusting control points of the geometric region corresponding to the printing layer in the three-dimensional CAD model using a nonlinear geometric reparameterization method, and generating a compensated geometric model; S6. Generate printing code according to the compensated geometric model and control the printing device to print; S7. After printing is completed, the printed product is three-dimensionally scanned to obtain an actual deformation vector field. The actual deformation vector field is compared with the overall predicted deformation vector field formed by superimposing the deformation prediction vector fields of each printing layer. If the error vector field is greater than a preset tolerance threshold, the error information is fed back to step S5, and geometric compensation processing is re-performed on the corresponding printing layer geometric area in the 3D CAD model. Steps S5 to S7 are repeated until the error vector field is less than or equal to the preset tolerance threshold.

2. The method for compensating for deformation in 3D printing of an aerospace engine according to claim 1, characterized in that: The printing parameters include laser power, scanning speed, scanning path, scanning spacing, powder preheating temperature, laser spot diameter, and thermal conductivity, specific heat capacity, density, thermal expansion coefficient and Young's modulus of the printing material.

3. The aerospace engine 3D printing deformation compensation method according to claim 1, characterized in that: The S2 specifically includes: S21. Establishing a thermal-mechanical coupling finite element simulation model using the printing parameters and geometric information of the corresponding printed layer, wherein the establishment of the thermal-mechanical coupling finite element simulation model includes geometric discretization, mesh generation, material property assignment, and boundary condition setting; S22, simulating the heat input process of the laser scanning path in the actual 3D printing process layer by layer based on the thermal-mechanical coupling finite element simulation model, and calculating the dynamic change process of the temperature field of each printed layer over time to obtain instantaneous temperature distribution and temperature gradient information; S23, using the instantaneous temperature distribution and temperature gradient information to calculate the transient thermal stress field distribution of the current printing layer during the printing process, and after each layer is laser scanned and cooled to room temperature, calculate and obtain the residual stress tensor field of the printing layer after printing; S24, performing cross-layer stress accumulation calculation on the residual stress tensor fields of several printing layers before the current printing layer, and establishing a residual stress tensor field history database, wherein the residual stress tensor field of each printing layer in the residual stress tensor field history database is associated with the current layer in an inter-layer accumulation manner; S25. Construct a residual stress input data set based on the residual stress tensor field of the current printing layer and the residual stress tensor fields of several previous historical printing layers in the residual stress tensor field history database.

4. The method for compensating for deformation in 3D printing of an aerospace engine according to claim 1, characterized in that: The S3 specifically includes: S31. Based on the residual stress tensor field of each printing layer in the residual stress input data set, a three-dimensional space coordinate system and a printing time axis are established with the current printing layer as the reference. The three-dimensional coordinate points are recorded as ,in Indicates the spatial coordinates of the printed part in the width direction, Indicates the spatial coordinates of the printed part in the length direction, Indicates the spatial coordinates of the printed part in the height direction, The direction corresponds to the stacking direction of layer-by-layer printing; S32. In the three-dimensional spatial coordinate system, spatially register the residual stress tensor fields of each historical printing layer by three-dimensional bilinear interpolation to obtain a tensor expression with consistent spatial resolution and coordinate frame with the current printing layer; S33. Based on the thermo-elastoplastic mechanics propagation theory, a stress propagation function model is constructed: ； in, Indicates the Layer The propagation of layers affects the tensor, Indicates the Layer at 3D coordinate point The residual stress tensor on Indicates the The material coupling coefficient of the layer, represents the exponential decay factor, Indicates the Layer to Layer The vertical interlayer distance in the direction; S34. Through the time weight decay algorithm, the propagation influence tensor of the historical printing layer is time-weighted to construct the stress shadow mapping tensor , used to describe the spatial position of the historical printing layer in the current printing layer The residual stress affected area formed by the upper projection; S35, output the stress shadow mapping tensor .

5. The aerospace engine 3D printing deformation compensation method according to claim 1, characterized in that: The S4 specifically includes: S41, constructing a reverse stress propagation neural network model, wherein the reverse stress propagation neural network model includes an input layer, a spatiotemporal tensor feature fusion layer, a bidirectional gated recurrent unit layer, a tensor attention mechanism layer, and an output prediction layer; S42, inputting the stress shadow mapping tensor and the printing parameters of the current printing layer into the input layer of the inverse stress propagation neural network model, the input layer converting the three-dimensional tensor data into a high-dimensional feature vector and performing data preprocessing through linear normalization; S43, the spatiotemporal tensor feature fusion layer performs a multi-scale convolution operation on the high-dimensional feature vector, extracts spatial local features and inter-layer temporal features through multiple convolution kernels, and generates a spatiotemporal coupling feature expression by using a feature splicing operation; S44, the bidirectional gated recurrent unit layer processes the spatiotemporal coupling feature expression, using a gated unit mechanism to simultaneously capture the forward propagation of the impact of the historical printing layer on the current printing layer and the reverse tracing of the stress state of the current layer on the historical layer, thereby obtaining a forward and reverse coupled bidirectional stress propagation feature expression; S45, the tensor attention mechanism layer uses the attention calculation method of the spatial dimension and the channel dimension based on the bidirectional stress propagation feature representation to generate a space-channel joint attention weight mask tensor ,in Represents a three-dimensional coordinate point, Indicates the feature channel number; S46, the output prediction layer fuses the spatial-channel joint attention weight mask tensor with the bidirectional stress propagation feature by tensor multiplication, and uses The activation function performs nonlinear mapping and outputs the deformation prediction vector field of the current printing layer. The deformation prediction vector field is for each three-dimensional coordinate point The corresponding spatial deformation prediction vector represents the deformation prediction of the current printing layer.

6. The method for compensating for deformation in 3D printing of an aerospace engine according to claim 1, characterized in that: The S5 specifically includes: S51, extracting a corresponding geometric region of the current printing layer in the three-dimensional CAD model based on the deformation prediction vector field, and locating the CAD surface segments and boundary range contained in the geometric region; S52, perform parametric modeling on the geometric area of ​​the printing layer, convert the original CAD surface segment expression into a re-parameterized surface form based on non-uniform rational B-spline, and construct a two-dimensional control point array ,in Indicates the first segment of the original CAD surface Rank The spatial coordinate vector of the column control point; S53. Construct a control point compensation function based on the distribution of the deformation prediction vector field in the control point space: ； in, Represents the spatial coordinate vector of the control point after geometric compensation, represents the global compensation adjustment coefficient, represents the direction weight matrix, express The deformation prediction vector at the location, represents the deformation adjustment coefficient, represents the total number of multi-scale layers, Representation scale The weighting factor of express Location on scale The average deformation prediction vector of the lower adjacent region; S54, performing surface reconstruction based on the spatial coordinate vectors of the control points after geometric compensation to form an updated expression of the area corresponding to the current printing layer in the three-dimensional CAD model after geometric compensation; S55. Perform a geometric validity check, which includes constraint judgments on boundary closure, surface continuity, and the maximum compensation amplitude not exceeding the corresponding preset tolerance. If the constraint judgment conditions are met, the compensated geometric model is output.

7. The aerospace engine 3D printing deformation compensation method according to claim 1, characterized in that: The S6 specifically includes: S61, converting the compensated geometric model into an intermediate model file in a processing format, wherein the intermediate model adopts a faceting format for geometric approximation, and the file format is STL; S62, performing path planning processing on the intermediate model, adjusting the scanning spacing and path density according to the curvature distribution and structure thickness of the geometric compensation area, and generating a printing path set; S63: Generate a print control parameter table based on the print path set through print parameter mapping; S64: Generate a print control code recognized by a standard printing device based on the print control parameter table, wherein the print control code is an additive manufacturing control program based on a G-code structure and includes layer-by-layer control instructions and compensation correction amount annotation information; S65: Transmit the printing control code to the printing device control module, and control the printing device to perform additive manufacturing operations layer by layer according to the printing control code.

8. The method for compensating for deformation in 3D printing of an aerospace engine according to claim 1, characterized in that: The S7 specifically includes: S71. After printing is completed, a laser scanner is used to perform a three-dimensional full-field scan of the printed product to obtain point cloud data of the complete surface and construct a scan model; S72, aligning the scanned model with the compensated 3D CAD model, using an iterative closest point algorithm to achieve rigid registration, and establishing a point-to-point spatial mapping relationship; S73, calculating an actual deformation vector field based on the registration result, where the actual deformation vector field is the spatial difference between corresponding points of the scanned model and the compensated 3D CAD model in a unified coordinate system; S74, accumulating the deformation prediction vector field of each layer during the printing process layer by layer to form an overall predicted deformation vector field, and comparing it with the actual deformation vector field to obtain an error vector field; S75. Extract the error peak area based on the distribution characteristics of the error vector field in the three-dimensional space, and obtain the control point indexes in the corresponding printing layer and geometric area; S76: The sampled value of the error vector field at the control point position is introduced as an error correction term into the control point update mechanism, and the control point is compensated for a second time based on the previous round of compensation results; S77. Use the control points adjusted by error feedback to reconstruct the corresponding printing layer geometric area in the 3D CAD model to generate a new geometric model, and repeat steps S5 to S7 until the error vector field is less than or equal to the preset tolerance threshold.

Images