Aerospace engine 3D printing thermal stress distribution control method

Abstract

The invention discloses an airspace engine 3D printing thermal stress distribution control method which comprises the following steps: S1, importing a three-dimensional CAD model of an airspace engine structure, and generating a printing layer set; s2, constructing a three-dimensional node graph structure of a printing layer; s3, inputting the thermodynamic prediction input tensor into the heterogeneous attention map thermodynamic network to obtain a node-level thermal stress tensor map; s4, analyzing the node-level thermal stress tensor diagram, and extracting an optimized region; s5, calling an improved chameleon search algorithm in the optimization region; s6, optimal printing parameters of the current printing layer are determined; s7, the 3D printing device is controlled based on the optimized printing parameters, and real-time thermal inductance data are collected; and S8, mapping the thermal inductance feedback set of the printing layer to the corresponding node, and repeatedly executing the steps S2 to S7 until the whole airspace engine structure is printed. According to the method, the heterogeneous graph network and the improved chameleon search algorithm are fused, and precise control over the printing thermal stress of the airspace engine is achieved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent thermal control technology, and in particular to a method for controlling thermal stress distribution in 3D printing of aerospace engines. Background Art

[0002] With the in-depth application of metal additive manufacturing technology in the field of high-end equipment manufacturing, especially in the manufacturing process of complex structural components of aerospace engines, 3D laser melting printing has become the core manufacturing method for realizing key structures such as multi-channel cooling shells, gradient wall thickness nozzles and high-temperature cavities due to its advantages such as high forming freedom, high material utilization rate and short prototype cycle. However, the structure of aerospace engines usually has complex curved surfaces, non-uniform heat conduction paths and multi-material gradient structures, which makes the problem of thermal stress accumulation during the printing process particularly prominent. Due to the drastic thermal strain response of metals during the laser layer-by-layer melting and rapid cooling process, defects such as warping, cracks, and residual deformation are easily formed, which seriously affects the structural strength and service reliability.

[0003] In the prior art, the methods for controlling printing thermal stress mainly include two categories: one is the printing path optimization method, and the other is the thermal field prediction and simulation adjustment method. The path optimization method usually controls the uniformity of the heat input distribution by adjusting the scanning direction, scanning interval, jump strategy or partition scanning, thereby alleviating the accumulation of local thermal gradients. However, such methods are mostly based on empirical rules or single preset paths, and cannot be dynamically adjusted according to the thermal feedback in the actual printing process. On the other hand, the thermal field prediction method generally uses a simple heat conduction model of finite element simulation to simulate the thermal stress field before printing, and adjusts the parameters based on the results. Although this method can theoretically reflect the stress concentration trend, in the face of complex geometry, multi-heat source interaction and real-time thermal fluctuation scenarios, its simulation accuracy and timeliness are far from meeting the needs of multi-layer real-time control in aerospace engine additive manufacturing.

[0004] In addition, most existing methods use a static control mechanism, that is, the path design and power configuration are set once before printing, and do not have the ability to adaptively update during the process. This static control strategy is often unable to correct parameters in a timely manner when faced with problems such as thermal field disturbances, laser power fluctuations, and nonlinear material thermal responses during the printing process, resulting in the accumulation of deviations between the model and the actual situation, which in turn causes stress over-limit, geometric distortion, and even printing failure. In some studies, attempts have been made to introduce graph neural networks or deep learning models for thermal stress prediction, but these methods generally have two problems: first, the feature extraction dimension is single, and heterogeneous features such as printing process parameters, material thermophysical properties, and historical thermal responses cannot be effectively integrated; second, the prediction model usually runs independently of the path optimization and fails to form a deep coupling with the path generation process, resulting in improved prediction accuracy, but the overall printing quality control is still limited.

[0005] Furthermore, most of the current printing path optimization algorithms are based on heuristic algorithms, such as genetic algorithms, ant colony algorithms, etc., which have achieved certain optimization effects on static structures, but lack sufficient coping ability for multi-objective constraints in multi-dimensional dynamic thermal fields. In terms of bionic optimization algorithms that simulate natural behaviors, such as grey wolves, whales, fireflies, etc., they have been tried to be applied to path planning, but these algorithms have not considered the search strategy driven by thermal response, especially the search jump mechanism in three-dimensional thermal anomaly regions has not formed a structural model with transferability and interpretability.

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

[0007] An object of the present invention is to propose a method for controlling the thermal stress distribution in 3D printing of aerospace engines. The present invention integrates a heterogeneous attention graph neural network and an improved chameleon search algorithm to construct a multi-dimensional thermal prediction and three-stage path optimization model, realizing accurate prediction, dynamic optimization and real-time feedback control of thermal stress during the 3D printing process of complex structures of aerospace engines. It has the advantages of high prediction accuracy, reasonable path planning, strong processing stability and strong feedback adaptability, effectively solving problems such as thermal field simulation lag, static path and stress out of control in traditional methods, and significantly improving the consistency, forming quality and engineering repeatability of printed structures.

[0008] A method for controlling the thermal stress distribution in 3D printing of aerospace engines according to an embodiment of the present invention includes the following steps: S1. Import a three-dimensional CAD model of the aerospace engine structure, perform layer division on the three-dimensional CAD model based on a preset printing layer thickness, and generate a set of printing layers; S2. Construct a three-dimensional node graph structure for each printing layer, and extract node thermophysical properties, printing process parameters and historical thermal field responses from the three-dimensional node graph structure to generate a thermal prediction input tensor; S3. Input the thermal prediction input tensor into a heterogeneous attention graph thermal network to obtain a node-level thermal stress tensor map of the current printing layer; S4. Analyze the node-level thermal stress tensor map, and extract a node subgraph containing a thermal gradient anomaly region as an optimization region; S5. Invoke an improved chameleon search algorithm within the optimization region, perform three-stage jump search in the environmental scanning period, thermal response jump period and path curing period, and output a set of candidate laser scanning paths; S6. Perform a path energy evaluation function calculation on the set of candidate laser scanning paths, determine the optimal laser scanning path and the corresponding power configuration as the optimized printing parameters for the current printing layer; S7. Control the 3D printing device based on the optimized printing parameters to complete the processing operation of the current printing layer, and collect real-time thermal data during the processing to form a thermal feedback set of the printing layer; S8. Map the thermal feedback set of the printing layer to the corresponding nodes in the three-dimensional node graph structure, and repeat steps S2 to S7 until the entire aerospace engine structure is printed.

[0009] Optionally, the S2 specifically includes: S21. Divide each printing layer into several spatial voxel units, and each spatial voxel unit serves as a node in the three-dimensional node graph structure; the three-dimensional node graph structure is mainly used for modeling the physical properties and spatial layout at the node level of the printing layer, without involving the definition of graph edge connections or adjacency weights, nor relying on the adjacency matrix for graph propagation operations; S22. Establish a thermophysical property table for each node, and the thermophysical property table includes the thermal conductivity, specific heat capacity, density, and melting point parameters of the material corresponding to the node; S23. Configure a printing process parameter vector for each node, and the printing process parameter vector includes scanning speed, laser power, scanning spacing, and layer thickness; S24. Extract the thermal field response value corresponding to the node position from the historical printing data, and the thermal field response value includes historical temperature value, stress value, and gradient change rate; S25. Perform normalization and tensor encoding processing on the thermophysical properties, printing process parameter vector, and thermal field response value respectively to generate a node-level multi-channel feature vector; S26. Assemble the node-level multi-channel feature vectors according to the node number and spatial layout method to construct a thermal prediction input tensor corresponding to the current printing layer.

[0010] Optionally, the S3 specifically includes: S31. Input the thermal prediction input tensor into the heterogeneous attention graph thermal network, decompose it according to the node dimension, and extract the spatial coordinate feature sub-tensor, thermophysical property feature sub-tensor, printing process parameter feature sub-tensor, and thermal response feature sub-tensor; S32. Input each feature sub-tensor into the heterogeneous feature processing branch of the corresponding heterogeneous attention graph thermal network, and the heterogeneous feature processing branch includes a group of linear transformation layers and normalization layers for unifying the feature scales of different dimensions; S33. Perform a feature splicing operation on the output of each heterogeneous feature processing branch to generate a fused feature map; S34. Calculate the node-level attention weighting coefficient based on the heterogeneous attention mechanism in the heterogeneous attention graph thermal network: ; Wherein, Denote the attention weighted coefficient of the th node, Denote the three-dimensional spatial coordinates of the th node, Denote the thermal history response feature of the th node, Denote the three-dimensional spatial coordinates of the th node, Denote the thermal history response feature of the th node, Denote the exponential function, Denote the activation function, Denote the hyperbolic tangent function, , and Denote the weight matrix, Denote the bias vector, Denote the total number of nodes; S35. Use the node-level attention weighted coefficient to weight the corresponding node features in the fusion feature map, obtain the aggregated feature representation of each node in the current printing layer, and generate the node-level thermal stress tensor map of the current printing layer based on the fully connected mapping.

[0011] Optionally, the S4 specifically includes: S41. Calculate the thermal stress gradient value for each node in the node-level thermal stress tensor map, where the thermal stress gradient value is the thermal stress change rate between the node and its spatially adjacent nodes; S42. Based on the distribution of the thermal stress gradient values of the whole layer, statistically calculate the mean value of the thermal stress gradient and the standard deviation , and set the abnormal gradient threshold : ; where, Denote the positive real number control parameter; S43. Mark all nodes with thermal stress gradient values greater than the abnormal gradient threshold as the initial abnormal node set; S44. Perform a connectivity expansion operation based on the node spatial topology relationship, merge the nodes continuously distributed within the set radius with the initial abnormal nodes to form a thermal gradient abnormal region, and use the node subgraph containing the thermal gradient abnormal region as the optimization region.

[0012] Optionally, the S5 specifically includes: S51. Call the improved chameleon search algorithm within the obtained optimized region, initialize the search individual set, and each search individual defines the current position state vector and the initial observation angle according to the thermal stress gradient value and the historical scanning position; S52. During the environmental scanning period, each search individual performs multi-angle thermal field inspections, simulates the independent movement behavior of the chameleon's eyeball by rotating the scan, collects multiple candidate jump directions and records the thermal stress gradient values in the corresponding directions; S53. During the thermal jump period, based on the set thermal perturbation sensitivity threshold, when the thermal stress gradient value in a certain candidate jump direction exceeds the thermal perturbation sensitivity threshold, trigger the jump action in the candidate jump direction and calculate the jump increment: ; Where, represents the jump increment of the th candidate jump direction, represents the jump amplification coefficient, represents the th candidate jump direction of the thermal stress gradient value, represents the th candidate jump direction of the perturbation adjustment factor, represents the sign function; S54. During the path curing period, calculate the thermal stress adaptability scoring function for the paths in all candidate jump directions: ; Where, represents the value of the thermal stress adaptability scoring function, , and represent the weighting coefficients, represents the number of path sampling points, represents the th node of the thermal stress gradient value, represents the path length, represents the maximum allowable path length, represents the power standard deviation, represents the average laser power; S55. According to the minimization criterion of the thermal stress adaptability scoring function, sort all jump paths, screen out the paths with the thermal stress adaptability scoring function value lower than the set fitness threshold, and satisfy the following constraints: the average thermal stress gradient value on the path is not greater than the average thermal stress gradient value of the previous printing layer, the path length does not exceed the maximum limit ratio, and the ratio of the power variance to the average power is lower than the power stability upper limit, and generate and output the candidate laser scan path set.

[0013] Optionally, the path energy evaluation function is defined as: ; Among them, represents the path energy evaluation function value, and represent the weighting coefficients, represents the number of path sampling points, represents the th thermal stress value of the node, represents the length of the path segment corresponding to the th node, represents the th laser power of the node; According to the minimum value of the path energy evaluation function value , determine the candidate laser scanning path with the lowest comprehensive thermal stress energy consumption and laser energy consumption as the optimal laser scanning path, and output the corresponding power configuration as the optimized printing parameter of the current printing layer.

[0014] Optionally, the S7 specifically includes: S71. Load the optimal laser scanning path and the corresponding power configuration in the optimized printing parameters into the control module of the 3D printing device, and drive the laser to perform processing operations point by point according to the path coordinate sequence; S72. Collect real-time thermal data during the processing, specifically including using the infrared thermal imaging unit and the thermocouple array installed on the 3D printing device to perform real-time temperature monitoring on the current printing layer, and obtain the thermal field image data and the node temperature sampling values; S73. Match the thermal field image data and the node temperature sampling values according to the node numbers to form the original dataset of the printing layer thermal sense; S74. Perform denoising, normalization, and structuring processing on the original dataset of the printing layer thermal sense to form the feedback set of the printing layer thermal sense.

[0015] Optionally, the S8 specifically includes: S81. Perform spatial mapping and matching between each data point in the feedback set of the printing layer thermal sense and the corresponding node in the three-dimensional node map structure according to the spatial coordinates and number indexes to establish a feedback mapping table; S82. For each matching node, extract the thermal sense feedback values, including real-time temperature, in-layer temperature difference, thermal stress change direction, and cooling rate; S83. Update the state of the corresponding node in the three-dimensional node map structure according to the thermal sense feedback values; S84. Re-encode the state of the nodes in the updated three-dimensional node map structure into the auxiliary feature channels of the thermal prediction input tensor of the next printing layer; S85. Repeat steps S2 to S7 until the entire aerospace engine structure is printed.

[0016] The beneficial effects of the present invention are as follows: Firstly, by constructing a three-dimensional node graph structure and extracting node-level multi-channel features, including thermophysical properties, printing process parameters, and historical thermal field responses, and then inputting them into a heterogeneous attention graph thermal network for processing, the present invention can achieve thermal stress prediction with node resolution at the printing level, greatly improving the perception accuracy of internal thermal behavior changes in complex structures. The introduced heterogeneous attention mechanism can not only dynamically adjust the feature fusion weights according to spatial positions and historical thermal responses but also support adaptive modeling of different physical dimensions, significantly enhancing the network's processing ability for the heterogeneity of multi-source inputs, and making the prediction output more stable and physically reasonable.

[0017] Secondly, based on the thermal stress prediction results, the present invention constructs a thermal gradient abnormal area extraction mechanism, which can accurately identify high-risk areas according to the statistical characteristics of thermal stress gradients and spatial connectivity, thereby serving as the target subgraph area for path optimization. In this area, an improved chameleon search algorithm with bionic characteristics is used to generate paths. The algorithm is divided into three stages: environmental scanning period, thermal-sensitive jumping period, and path solidification period, and has a clear bio-inspired structure. Especially in the thermal-sensitive jumping period, directional jumps are triggered according to the local thermal perturbation sensitivity to effectively avoid thermal stress concentration areas; while the thermal stress adaptability scoring function introduced in the path solidification period integrates multi-dimensional indicators such as thermal stress gradients, path compactness, and laser power stability, realizing a comprehensive quantitative evaluation of candidate paths, and making the path generation process rise from heuristic search to a self-regulating optimization process driven by thermal control.

[0018] In addition, the present invention constructs a real-time thermal sense acquisition and graph node state feedback mechanism during the printing process. While performing the printing task, the thermal field data of the printing layer is collected in real time through an infrared thermal imager and a thermocouple array, and mapped back to the corresponding nodes in the three-dimensional node graph structure to form a thermal sense feedback set. This feedback set is used to dynamically update the thermal response state of the nodes and serve as an auxiliary channel for the input tensor of the next layer's thermal stress prediction, thus forming a full-process closed-loop control logic of prediction - optimization - execution - feedback. Compared with the existing schemes that rely on one-way prediction and fixed paths, the scheme of the present invention can achieve adaptive learning of the printing state of each layer and dynamic adjustment of paths, and has stronger steady-state response ability and thermal stress control ability, showing excellent forming consistency and stress stability in the additive manufacturing scenario of aerospace engines with complex structures and strong thermal coupling. Description of the Drawings

[0019] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation to the present invention. In the drawings: Figure 1The overall flowchart of a method for controlling the thermal stress distribution in 3D printing of a space engine proposed by the present invention; Figure 2 The flowchart of layer - dividing a 3D CAD structural model of a space engine and constructing a 3D node - graph structure for a method for controlling the thermal stress distribution in 3D printing of a space engine proposed by the present invention; Figure 3 The three - stage jump - search flowchart of an improved chameleon search algorithm for a method for controlling the thermal stress distribution in 3D printing of a space engine proposed by the present invention. Detailed implementation manners

[0020] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0021] Reference Figures 1-3 , a method for controlling the thermal stress distribution in 3D printing of a space engine, includes the following steps: S1. Import a 3D CAD model of the space engine structure, perform layer - division on the 3D CAD model based on a preset printing layer thickness, and generate a set of printing layers; S2. Construct a 3D node - graph structure for each printing layer, and extract the node thermophysical properties, printing process parameters, and historical thermal field responses from the 3D node - graph structure to generate a thermal - force prediction input tensor; S3. Input the thermal - force prediction input tensor into a heterogeneous attention - map thermal - force network to obtain a node - level thermal stress tensor map of the current printing layer; S4. Analyze the node - level thermal stress tensor map, and extract a node sub - graph containing the thermal gradient abnormal region as the optimization region; S5. Call an improved chameleon search algorithm within the optimization region, perform a three - stage jump - search of an environment scanning period, a thermal - sensitive jump period, and a path solidification period, and output a set of candidate laser scanning paths; S6. Perform a path energy evaluation function calculation on the set of candidate laser scanning paths to determine the optimal laser scanning path and the corresponding power configuration as the optimized printing parameters for the current printing layer; S7. Control a 3D printing device to complete the processing operation of the current printing layer based on the optimized printing parameters, and collect real - time thermal - sense data during the processing to form a printing - layer thermal - sense feedback set; S8. Map the printing - layer thermal - sense feedback set to the corresponding nodes in the 3D node - graph structure, and repeat steps S2 to S7 until the printing of the entire space engine structure is completed.

[0022] The present invention establishes a 3D printing full-process thermal stress control process for the complex structure of aero-engine, and for the first time realizes a closed-loop control path from structural modeling, graphic data construction, thermal prediction, anomaly identification, path optimization, power configuration to thermal feedback. Its greatest benefit lies in that, by combining the node-level thermal stress tensor with the feedback mechanism, the process parameters of each printing layer have the ability of adaptive update, significantly reducing problems such as thermal stress accumulation and crack generation caused by the traditional static path strategy. Compared with the single preset path and fixed thermal simulation method, this solution integrates intelligent prediction and bionic search algorithms, showing significant advantages in dynamic regulation and local thermal anomaly suppression, ensuring that the printed structure has stronger deformation control ability and structural stability under high complexity and extreme thermal coupling conditions.

[0023] In this embodiment, the S2 specifically includes: S21. Divide each printing layer into a number of spatial voxel units, and each spatial voxel unit serves as a node in the three-dimensional node graph structure; the three-dimensional node graph structure is mainly used for modeling the node-level physical properties and spatial layout of the printing layer, without involving the definition of graph edges or adjacency weights, nor relying on the adjacency matrix for graph propagation operations; S22. Establish a thermophysical property table for each node, and the thermophysical property table includes the thermal conductivity, specific heat capacity, density and melting point parameters of the material corresponding to the node; S23. Configure a printing process parameter vector for each node, and the printing process parameter vector includes scanning speed, laser power, scanning spacing and layer thickness; S24. Extract the thermal field response values corresponding to the node positions from the historical printing data, and the thermal field response values include historical temperature values, stress values and gradient change rates; S25. Perform normalization and tensor encoding processing on the thermophysical properties, printing process parameter vector and thermal field response values respectively to generate a node-level multi-channel feature vector; S26. Assemble the node-level multi-channel feature vectors according to the node numbers and spatial layout methods to construct a thermal prediction input tensor corresponding to the current printing layer.

[0024] By dividing each printing layer voxel into graph structure nodes and introducing multi-source feature information such as thermophysical properties, process parameters and historical responses in the node dimension, the problem of single thermal input dimension and weak information expression ability in the prior art is effectively solved. The normalization and quantization methods are used to encode the node features, so that the input tensor not only retains the physical meaning, but also has high compatibility and learnability, providing sufficient support for subsequent neural network prediction. This method realizes the unified modeling of the printing layer structure, material behavior and historical thermal field information, improves the fine-grained accuracy of thermal stress prediction on the basis of multi-channel feature aggregation, and lays a foundation for the local thermal anomaly regulation of complex printing areas.

[0025] In this embodiment, step S3 specifically includes: S31. Input the thermal prediction input tensor into the heterogeneous attention map thermal network, decompose it according to the node dimension, and extract the spatial coordinate feature sub-tensor, thermophysical property feature sub-tensor, printing process parameter feature sub-tensor, and thermal response feature sub-tensor; S32. Input each feature sub-tensor into the heterogeneous feature processing branch of the corresponding heterogeneous attention map thermal network. The heterogeneous feature processing branch includes a group of linear transformation layers and normalization layers for unifying the feature scales of different dimensions; S33. Perform a feature splicing operation on the output of each heterogeneous feature processing branch to generate a fused feature map; S34. Based on the heterogeneous attention mechanism in the heterogeneous attention map thermal network, calculate the node-level attention weighting coefficient: ; where, represents the attention weighting coefficient of the th node, represents the three-dimensional spatial coordinates of the th node, represents the thermal history response feature of the th node, represents the three-dimensional spatial coordinates of the th node, represents the thermal history response feature of the th node, represents the exponential function, represents the activation function, represents the hyperbolic tangent function, 、 and represent the weight matrices, represents the bias vector, represents the total number of nodes; S35. Use the node-level attention weighting coefficient to weight the corresponding node features in the fused feature map to obtain the aggregated feature representation of each node in the current printing layer, and generate the node-level thermal stress tensor map of the current printing layer based on the fully connected mapping.

[0026] By constructing a heterogeneous attention map heat network structure, the spatial coordinates, material properties, process parameters, and historical thermal responses are respectively input into different processing branches and fused, significantly enhancing the model's ability to express multi-dimensional heterogeneous features. The introduced spatial coordinate-thermal history joint attention mechanism in this model can dynamically identify the thermal influence sensitivity of nodes in different physical dimensions, improving the pertinence of feature weighted aggregation. Compared with traditional unified-channel neural networks, this method can capture the complex patterns of thermal stress changes more accurately, and the output node-level stress tensor is more in line with the actual thermal behavior, significantly enhancing the prediction stability and local risk identification ability under the printing path.

[0027] In this embodiment, S4 specifically includes: S41. Calculate the thermal stress gradient value for each node in the node-level thermal stress tensor map, where the thermal stress gradient value is the rate of change of thermal stress between a node and its spatially adjacent nodes; S42. Based on the distribution of the thermal stress gradient values of all layers, calculate the mean value and standard deviation , and set the abnormal gradient threshold : ; where represents a positive real control parameter; S43. Mark all nodes with thermal stress gradient values greater than the abnormal gradient threshold as the initial abnormal node set; S44. Perform a connectivity expansion operation based on the node spatial topological relationship, merge the nodes continuously distributed within the set radius of the initial abnormal nodes to form a thermal gradient abnormal region, and use the node subgraph containing the thermal gradient abnormal region as the optimization region.

[0028] An abnormal detection mechanism is constructed using the thermal stress gradient between nodes and statistical indicators, realizing the automatic identification and spatial aggregation of local high thermal gradient regions. By setting an adaptive threshold and combining the node spatial topological relationship, the potential thermal stress risk subgraph in the structure can be accurately extracted, avoiding misjudgment and missed judgment caused by relying on empirical judgment in traditional methods. This method effectively improves the recognition accuracy of the path optimization target region, enabling subsequent optimization algorithms to concentrate resources on processing high-risk regions, thereby improving the overall path quality and reducing the probability of structural cracks and stress distortion, and enhancing the inter-layer thermal stability and deformation control ability of printing.

[0029] In this embodiment, S5 specifically includes: S51. Call the improved chameleon search algorithm within the extracted optimization region, initialize the search individual set, and each search individual defines the current position state vector and the initial observation angle according to the thermal stress gradient value and the historical scanning position; S52. During the environmental scanning period, each search individual performs multi-angle thermal field inspections, simulates the independent movement behavior of the chameleon's eyeball by rotating the scan, collects multiple candidate jump directions, and records the thermal stress gradient values in the corresponding directions; S53. During the thermal jump period, based on the set thermal perturbation sensitivity threshold, when the thermal stress gradient value in a certain candidate jump direction exceeds the thermal perturbation sensitivity threshold, trigger the jump action in the candidate jump direction, and calculate the jump increment: ; where, represents the jump increment in the th candidate jump direction, represents the jump amplification factor, represents the th thermal stress gradient value in the candidate jump direction, represents the th perturbation adjustment factor in the candidate jump direction, represents the sign function; S54. During the path solidification period, calculate the thermal stress adaptability scoring function for the paths in all candidate jump directions: ; where, represents the value of the thermal stress adaptability scoring function, , and represent the weighting coefficients, represents the number of path sampling points, represents the th thermal stress gradient value of the node, represents the path length, represents the maximum allowable path length, represents the power standard deviation, represents the average laser power; S55. According to the minimization criterion of the thermal stress adaptability scoring function, sort all the jump paths, filter out the paths whose thermal stress adaptability scoring function values are lower than the set fitness threshold, and satisfy the following constraints: the average thermal stress gradient value on the path is not greater than the average thermal stress gradient value of the previous printing layer, the path length does not exceed the maximum limit ratio, and the ratio of the power variance to the average power is lower than the power stability upper limit, and generate and output the candidate laser scanning path set.

[0030] The improved chameleon search algorithm is used to divide the path optimization process into three stages: environmental scanning, thermosensitive jumping, and path solidification, making the search process more physically driven and behaviorally bionic. This mechanism uses the thermal stress gradient as the guiding variable to simulate multi-directional perception and dynamic jumping behavior, improving the response efficiency and avoidance ability of path search in high thermal gradient regions. In the path solidification stage, a thermal stress adaptability scoring function and multi-dimensional constraint screening rules are introduced to ensure that the generated path achieves an optimal balance among stress balance, length rationality, and power stability, and has stronger robustness and industrial practicability to adapt to complex thermal field environments.

[0031] In this embodiment, the path energy evaluation function is defined as: ; Where represents the value of the path energy evaluation function, and represent the weighting coefficients, represents the number of path sampling points, represents the th node's thermal stress value, represents the th node's corresponding path segment length, represents the th node's laser power; According to the minimum value of the path energy evaluation function value , the candidate laser scanning path with the lowest comprehensive thermal stress energy consumption and laser energy consumption is determined as the optimal laser scanning path, and the corresponding power configuration is output as the optimized printing parameter for the current printing layer.

[0032] In this embodiment, the S7 specifically includes: S71. Load the optimal laser scanning path and the corresponding power configuration in the optimized printing parameters into the control module of the 3D printing device, and drive the laser to perform processing operations point by point according to the path coordinate sequence; S72. Collect real-time thermal sensing data during the processing, specifically including using the infrared thermal imaging unit and thermocouple array installed on the 3D printing device to perform real-time temperature monitoring on the current printing layer to obtain thermal field image data and node temperature sampling values; S73. Match the thermal field image data and the node temperature sampling values according to the node numbers to form the original dataset of the printing layer thermal sensing; S74. Denoise, normalize, and structure the original dataset of the printing layer thermal sensing to form the feedback set of the printing layer thermal sensing.

[0033] Quantify the cumulative thermal stress influence and energy input of the path through the path energy evaluation function. The energy sum is composed of multiplying the node-level stress by the path segment length and then adding the average power, providing a clear and computable optimization goal for path selection. Different from traditional path evaluation that only focuses on path length or thermal peak value, this function accurately reflects the spatial distribution of thermal work during the printing process with low computational complexity, effectively screening out the path plan with the minimum heat load and the optimal energy efficiency. This not only improves the thermal stability of the formed structure but also reduces the redundant heat input in laser processing, achieving a double improvement in printing efficiency and forming quality.

[0034] In this embodiment, the specific steps of S8 include: S81. Perform spatial mapping and matching between each data point with concentrated thermal sensation feedback of the printing layer and the corresponding nodes in the three-dimensional node map structure according to spatial coordinates and number indexes to establish a feedback mapping table; S82. For each matching node, extract the thermal sensation feedback values, including the real-time temperature, in-layer temperature difference, thermal stress change direction, and cooling rate; S83. Update the state of the corresponding nodes in the three-dimensional node map structure according to the thermal sensation feedback values; S84. Re-encode the state of the nodes in the updated three-dimensional node map structure as the auxiliary feature channels of the thermal prediction input tensor for the next printing layer; S85. Repeat steps S2 to S7 until the printing of the entire aerospace engine structure is completed.

[0035] By loading the optimized laser path and power configuration into the control system, driving the laser for precise processing, and synchronously collecting infrared thermal image and thermocouple point data during the processing, the parallel coordination of the printing task and thermal sensation acquisition is achieved. The data processing link fuses, denoises, and normalizes the image and numerical signals to generate a structured thermal sensation feedback set, providing high-quality input for the update of the map state after the printing layer. This mechanism significantly enhances the timeliness and data quality of thermal response monitoring during the printing process, provides a true and reliable closed-loop feedback basis for subsequent thermal control prediction, and effectively avoids the problems of thermal anomaly accumulation and path deviation.

[0036] Example 1

[0037] To verify the feasibility of the present invention in implementation, the present invention is applied to the additive manufacturing project of a multi-channel annular cooling nozzle section being carried out in a certain aerospace research center. In an aerospace engine, the nozzle is used to accelerate and discharge high-temperature gas, and its inner surface directly bears the high-temperature gas flow of thousands of degrees Celsius from the combustion chamber. Therefore, to ensure the thermal protection performance and structural strength of the nozzle, the inner layer of the nozzle is often designed as a multi-channel cooling structure, in which the cooling channels are arranged in a ring, spiral or serpentine shape inside the nozzle wall to form a multi-channel annular cooling nozzle section. This nozzle section belongs to a typical double-layer thin-walled shell structure, with complex serpentine cooling channels arranged between the outer shell and the inner cavity, the wall thickness varying from 0.8 mm to 3.2 mm, and the material being Inconel 718 alloy with high thermal stress sensitivity. The project objective requires that the formed structure achieve minimum warping, minimum residual stress and continuous interlayer forming stability control without adding a support strategy. The traditional printing process has serious thermal stress concentration problems on this structure, often resulting in cracks at the trailing edge and collapse of the nozzle inner cavity, and the rejection rate exceeding 27%.

[0038] In this embodiment, we first divide the CAD model of the nozzle section into printing layers, with each layer thickness set to 40 μm, and construct a three-dimensional node map structure during the printing process by combining voxelization. Each node corresponds to a spatial voxel unit and embeds thermophysical properties (such as thermal conductivity, density, specific heat), process parameters (scanning speed 300 mm / s, power 200 W, layer thickness 40 μm) and historical thermal field response data. The node tensor thus constructed is input into the heterogeneous attention map thermal network to generate a thermal stress tensor map corresponding to each layer. Subsequently, we perform a thermal stress gradient analysis on the tensor map to identify the rear wall of the nozzle inner cavity, the trailing edge and the curvature inversion region as thermal stress mutation regions.

[0039] Based on the above regions, we call an improved chameleon search algorithm. During the environmental scanning period, the search individuals cover the path with the most drastic gradient change from the outer ring to the inner wall of the nozzle outlet; during the thermal jump period, the algorithm judges the gradient direction in real time and jumps to the low-stress region; during the path curing period, cross-evaluation is carried out through the thermal stress adaptability scoring function and the path energy evaluation function, and finally an optimal path set is selected with an average thermal gradient reduction of 13.5% and a power change stability improvement of 18.2% for the laser path. Subsequently, the control system executes the processing task according to the optimal path and power configuration, and uses a thermocouple array and an infrared thermal imager for real-time acquisition during the printing process. The acquisition frequency is 10 Hz, covering the key node temperatures and heat flux densities of the whole layer, and generating a thermal sense feedback set for updating the state of the next layer's map structure.

[0040] The entire structure was printed in 471 layers, with a printing time of 16 hours and 48 minutes, and the average layer processing time was approximately 2.15 seconds. After printing, a stress analysis was performed on typical parts using an XRD residual stress detection system. The results showed that in the traditional path scheme, the maximum residual tensile stress at the trailing edge of the nozzle reached 227 MPa, while after optimization by the present invention, it was only 154 MPa, a decrease of approximately 32.2%; in the deformation measurement, the roundness error at the bottom of the nozzle was controlled within ±0.12 mm, which was better than the project target value of ±0.2 mm. Finally, the overall structure was successfully formed in one step without the need for support rework, and the surface roughness Ra was maintained below 8.5 μm.

[0041] The deployment results of the present invention in the actual scenario verified its fine control ability of the thermal stress distribution in the additive manufacturing of complex aerospace structures. Through the whole-process linkage mechanism of prediction-identification-optimization-feedback, cracks and stress distortion caused by non-uniform thermal fields were effectively avoided, and the reliability of the printing process and the engineering yield rate were improved. Especially in multi-curved surface and multi-channel structures, compared with static path or empirical optimization schemes, the present invention demonstrated stronger local stress adaptation ability and model self-learning ability.

[0042] In this embodiment, by applying the present invention to the printing task of a typical multi-channel annular cooling nozzle section, its engineering adaptability and thermal stress regulation ability in the additive manufacturing of complex aerospace engine structures were successfully verified. Compared with the traditional fixed path strategy, the present invention has achieved significant improvements in the prediction accuracy of node-level thermal stress, the recognition rate of local high-gradient regions, the path thermal load balance, and the printing forming consistency. After printing, the risks of cracks and warping have decreased significantly, and the roundness and dimensional accuracy of the finished product are better than the design requirements, realizing the one-step printing and forming of complex structures without additional support, and verifying that this method has the practical application ability of stable, efficient, and self-learning closed-loop optimization. This result fully reflects the broad application prospects of the present invention in the thermal field control of complex aerospace components.

[0043] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered within the protection scope of the present invention.

Claims

1. A method for controlling the thermal stress distribution in 3D printing of a space engine, characterized in that, It includes the following steps: S1. Import the 3D CAD model of the aerospace engine structure, perform layer division on the 3D CAD model based on a preset printing layer thickness, and generate a set of printing layers; S2. Construct the 3D node graph structure of each printing layer, and extract the nodal thermophysical properties, printing process parameters, and historical thermal field responses from the 3D node graph structure to generate a thermal prediction input tensor; S3. Input the thermal prediction input tensor into the heterogeneous attention graph thermal network to obtain the nodal thermal stress tensor map of the current printing layer; S4. Analyze the nodal thermal stress tensor map, and extract the node subgraph containing the thermal gradient anomaly region as the optimization region; S5. Invoke the improved chameleon search algorithm within the optimization region, perform a three-stage jump search of the environmental scanning period, thermal jump period, and path curing period, and output a set of candidate laser scanning paths; S6. Perform a path energy evaluation function calculation on the set of candidate laser scanning paths to determine the optimal laser scanning path and the corresponding power configuration as the optimized printing parameters for the current printing layer; S7. Control the 3D printing device to complete the processing operation of the current printing layer based on the optimized printing parameters, and collect real-time thermal sensation data during the processing to form a printing layer thermal sensation feedback set; S8. Map the printing layer thermal sensation feedback set to the corresponding nodes in the 3D node graph structure, and repeat steps S2 to S7 until the entire aerospace engine structure is printed.

2. The method for controlling the thermal stress distribution in 3D printing of a space engine according to claim 1, characterized in that, The specific content of S2 includes: S21. Divide each printing layer into several spatial voxel units, and each spatial voxel unit serves as a node in the 3D node graph structure; S22. Establish a thermophysical property table for each node, and the thermophysical property table includes the thermal conductivity, specific heat capacity, density, and melting point parameters of the material corresponding to the node; S23. Configure a printing process parameter vector for each node, and the printing process parameter vector includes the scanning speed, laser power, scanning spacing, and layer thickness; S24. Extract the thermal field response value corresponding to the node position from the historical printing data, and the thermal field response value includes the historical temperature value, stress value, and gradient change rate; S25. Perform normalization and tensor encoding processing on the thermophysical properties, printing process parameter vector, and thermal field response value respectively to generate a nodal multi-channel feature vector; S26. Assemble the nodal multi-channel feature vectors according to the node numbers and spatial layout methods to construct the thermal prediction input tensor corresponding to the current printing layer.

3. The method for controlling the thermal stress distribution in 3D printing of a space engine according to claim 1, characterized in that, The specific content of S3 includes: S31. Input the thermal prediction input tensor into the heterogeneous attention graph thermal network, decompose it according to the node dimension, and extract the spatial coordinate feature sub-tensor, thermophysical property feature sub-tensor, printing process parameter feature sub-tensor, and thermal response feature sub-tensor; S32. Input each feature sub-tensor into the heterogeneous feature processing branch of the corresponding heterogeneous attention graph thermal network, and the heterogeneous feature processing branch includes a set of linear transformation layers and normalization layers for unifying the feature scales of different dimensions; S33. Perform a feature splicing operation on the output of each heterogeneous feature processing branch to generate a fused feature map; S34. Calculate the node-level attention weighting coefficient based on the heterogeneous attention mechanism in the heterogeneous attention map thermal network: ； Among them, represents the attention weighting coefficient of the th node, represents the three-dimensional spatial coordinate of the th node, represents the thermal history response feature of the th node, represents the three-dimensional spatial coordinate of the th node, represents the thermal history response feature of the th node, represents the exponential function, represents the activation function, represents the hyperbolic tangent function, , and represent the weight matrix, represents the bias vector, represents the total number of nodes; S35. Use the node-level attention weighting coefficient to weight the corresponding node features in the fused feature map to obtain the aggregated feature representation of each node in the current printing layer, and generate the node-level thermal stress tensor map of the current printing layer based on the fully connected mapping.

4. The method for controlling the thermal stress distribution in 3D printing of a space engine according to claim 1, characterized in that, The specific steps of S4 are as follows: S41. Calculate the thermal stress gradient value for each node in the node-level thermal stress tensor map. The thermal stress gradient value is the rate of change of thermal stress between a node and its spatially adjacent nodes; S42. Based on the distribution of the full-layer thermal stress gradient values, statistically calculate the mean value of the thermal stress gradient and the standard deviation , and set the abnormal gradient threshold : ； Among them, represents a positive real control parameter; S43. Mark all the nodes with thermal stress gradient values greater than the abnormal gradient threshold as the initial abnormal node set; S44. Perform a connectivity expansion operation based on the node space topological relationship, and merge the nodes continuously distributed within a set radius from the initial abnormal node to form a thermal gradient abnormal region, and use the node subgraph containing the thermal gradient abnormal region as the optimization region. The nodes continuously distributed within the set radius are merged to form a thermal gradient abnormal region, and the node subgraph containing the thermal gradient abnormal region is used as the optimization region.

5. The method for controlling the thermal stress distribution in 3D printing of a space engine according to claim 1, characterized in that,The specific steps of S5 are as follows: S51. Invoke the improved chameleon search algorithm within the extracted optimization region, initialize the search individual set, and each search individual defines the current position state vector and the initial observation angle according to the thermal stress gradient value and the historical scanning position; S52. During the environmental scanning period, each search individual performs multi-angle thermal field inspections, simulates the independent movement behavior of the chameleon's eyeball through rotational scanning, collects multiple candidate jump directions, and records the thermal stress gradient values corresponding to the directions; S53. During the thermal jump period, based on the set thermal perturbation sensitivity threshold, when the thermal stress gradient value in a certain candidate jump direction exceeds the thermal perturbation sensitivity threshold, trigger the jump action of the candidate jump direction and calculate the jump increment: ； Among them, represents the jump increment of the th candidate jump direction, represents the jump amplification factor, represents the thermal stress gradient value of the th candidate jump direction, represents the perturbation adjustment factor of the th candidate jump direction, represents the sign function; S54. During the path curing period, calculate the thermal stress adaptability scoring function for the paths in all candidate jump directions: ； Among them, represents the value of the thermal stress adaptability scoring function, , and represent weighting coefficients, represents the number of path sampling points, represents the thermal stress gradient value of the th node, represents the path length, represents the maximum allowable path length, represents the power standard deviation, represents the average laser power; S55. According to the minimization criterion of the thermal stress adaptability scoring function, sort all jump paths, screen out the paths with the thermal stress adaptability scoring function value lower than the set fitness threshold, and satisfy the following constraint conditions: the average thermal stress gradient value on the path is not greater than the average thermal stress gradient value of the previous printing layer, the path length does not exceed the maximum limit ratio, and the ratio of the power variance to the average power is lower than the power stability upper limit, and generate and output the candidate laser scanning path set.

6. A method for controlling the thermal stress distribution in 3D printing of a space engine according to claim 1, characterized in that, The path energy evaluation function is defined as: ； Among them, represents the path energy evaluation function value, and represent the weighting coefficients, represents the number of path sampling points, represents the thermal stress value of the -th node, represents the length of the path segment corresponding to the -th node, represents the laser power of the -th node; According to the minimum value of the path energy evaluation function value Determine the candidate laser scanning path with the lowest comprehensive thermal stress energy consumption and laser energy consumption as the optimal laser scanning path, and output the corresponding power configuration as the optimized printing parameters for the current printing layer.

7. A method for controlling the thermal stress distribution in 3D printing of a space engine according to claim 1, characterized in that, The specific steps of S7 are as follows: S71. Load the optimal laser scanning path and the corresponding power configuration in the optimized printing parameters into the control module of the 3D printing device, and drive the laser to perform processing operations point by point according to the path coordinate sequence; S72. Collect real-time thermal sensing data during the processing, specifically including using the infrared thermal imaging unit and the thermocouple array installed on the 3D printing device to perform real-time temperature monitoring on the current printing layer to obtain the thermal field image data and the node temperature sampling values; S73. Match the thermal field image data and the node temperature sampling values according to the node numbers to form the original dataset of the printing layer thermal sensing; S74. Perform denoising, normalization, and structuring processing on the original dataset of the printing layer thermal sensing to form the feedback set of the printing layer thermal sensing.

8. A method for controlling the thermal stress distribution in 3D printing of a space engine according to claim 1, characterized in that, The specific steps of S8 are as follows: S81. Perform spatial mapping matching on each data point in the feedback set of the printing layer thermal sensing with the corresponding node in the three-dimensional node map structure according to the spatial coordinates and number indexes to establish a feedback mapping table; S82. For each matching node, extract the thermal sensing feedback values, including the real-time temperature, the temperature difference within the layer, the direction of thermal stress change, and the cooling rate; S83. Update the state of the corresponding nodes in the three-dimensional node map structure according to the thermal feedback value; S84. Re-encode the state of the nodes in the updated three-dimensional node map structure into the auxiliary feature channels of the thermal prediction input tensor for the next printing layer; S85. Repeat steps S2 to S7 until the entire aerospace engine structure is printed.

Images