A method and system for SLM forming parameter optimization of an aeronautical component

Abstract

The application discloses an SLM forming parameter optimization method and system for an aviation component, and relates to the technical field of SLM forming.The SLM forming parameter optimization system for the aviation component comprises an aviation component machining prediction module and an aviation component forming parameter optimization module.The application realizes accurate prediction of multiple performance indexes such as size deviation, surface roughness and internal defect degree by fusing aviation component machining allowance, local geometric features, material properties and historical forming data into a component machining prediction model and combining with thermal-mechanical coupling finite element simulation; and the application realizes hierarchical management of the importance of forming parameters by establishing a quantitative analysis mechanism of primary forming influence parameters and secondary forming influence parameters, so that the optimization process can be preferentially concentrated on core parameters which have the greatest influence on forming quality, and the optimization efficiency and stability of the result are improved.

Description

Technical Field

[0001] This invention relates to the field of SLM forming technology, and in particular to a method and system for optimizing SLM forming parameters for aerospace components. Background Technology

[0002] Selective laser melting (SLM) forming technology has been widely used in the manufacturing of aerospace components. However, traditional process parameter optimization methods generally rely on empirical settings or single-factor experiments, lacking the ability to systematically model the coupling relationships between machining allowances, geometric features, material properties, and various forming defects. In existing technologies, parameter optimization typically employs single-objective or static optimization methods, making it difficult to simultaneously consider multiple performance indicators such as dimensional accuracy, surface quality, and internal density, and also lacking the ability to dynamically adjust for uncertainties in the forming process. Furthermore, traditional methods do not fully utilize historical forming data for predictive modeling during the optimization process and lack efficient search and iterative optimization mechanisms, resulting in low parameter optimization efficiency and poor result stability.

[0003] Therefore, there is a need for an efficient dynamic optimization method for SLM forming parameters that integrates multi-source predictive modeling and hierarchical parameter optimization to achieve a balance between dimensional accuracy, surface quality, and internal density. Summary of the Invention

[0004] The present invention aims to provide a method and system for optimizing SLM forming parameters for aerospace components, which improves the adaptability and controllability of the manufacturing process while ensuring dimensional accuracy, surface roughness and internal quality.

[0005] A method for optimizing SLM forming parameters for aerospace components includes the following steps:

[0006] Acquire the preset 3D model data of the aerospace component to be formed; the preset 3D model data includes the aerospace component body and the aerospace component machining allowance; determine the material property data based on the basic characteristics of the aerospace component to be formed and generate the initial SLM forming parameters; the initial SLM forming parameters include the initial parameters for setting the aerospace component machining allowance.

[0007] Establish a component processing prediction model; based on the component processing prediction model, predict the aerospace component to be formed to obtain the predicted forming result of the aerospace component; the predicted forming result of the aerospace component includes the dimensional deviation corresponding to the aerospace component processing allowance, the surface roughness of the component, and the internal defect degree of the component;

[0008] Based on the predicted forming results of aerospace components, a quantitative analysis of the influencing factors on the initial SLM forming parameters is conducted to obtain the main forming influence parameters and the secondary forming influence parameters. Based on the main forming influence parameters and the secondary forming influence parameters, iterative prediction and adjustment are performed until the optimal SLM forming parameters are selected and new preset three-dimensional model data is obtained. Among them, the SLM forming parameters of the aerospace component to be formed are optimized based on the new preset three-dimensional model data and the optimal SLM forming parameters.

[0009] As a preferred embodiment of the present invention, the specific steps for predicting the aerospace component to be formed based on the component processing prediction model include:

[0010] For the prediction of dimensional deviation: obtain the local geometric feature parameters of the aerospace component body in the preset 3D model data; obtain historical SLM forming data; extract the historical forming parameter set with the highest similarity to the local geometric feature parameters from the historical SLM forming data; perform thermo-mechanical coupled finite element simulation based on the historical forming parameter set to obtain the predicted forming residual stress and predicted forming deformation; obtain the predicted dimensional deviation based on the predicted forming residual stress and predicted forming deformation.

[0011] For the prediction of component surface roughness: extract the initial forming scan parameters from the initial SLM forming parameters; calculate the historical roughness output with the highest similarity to the initial forming scan parameters from the historical SLM forming data; and make a prediction based on the historical roughness output to obtain the predicted component surface roughness.

[0012] For the prediction of internal defects in components: the forming set temperature and forming scan overlap rate are extracted from the initial SLM forming parameters; based on the forming set temperature, forming scan overlap rate and material property data, a predictive analysis is performed to obtain the predicted internal defects in the components.

[0013] As a preferred embodiment of the present invention, the specific steps for quantitative analysis of the influencing factors of initial SLM forming parameters based on the predicted forming results of aerospace components include:

[0014] A parameter-performance mapping relationship is constructed based on the initial SLM forming parameters and the predicted forming results of aerospace components; among them, the dimensional deviation, component surface roughness, and component internal defect degree are used as the prediction target variables; multiple feature importance evaluation sub-models are constructed by training the gradient boosting regression model and the parameter-performance mapping relationship.

[0015] Based on the target variable for prediction, the feature contribution weights corresponding to each feature importance assessment sub-model are output. An integrated feature ranking algorithm is used to normalize the feature contribution weights corresponding to each feature importance assessment sub-model to obtain the scoring results of different parameters in the initial SLM molding parameters.

[0016] The parameters are classified according to their scores in the initial SLM forming parameters; the parameters whose scores are greater than a set threshold are classified as primary forming parameters, and the rest are classified as secondary forming parameters.

[0017] As a preferred embodiment of the present invention, the specific steps for iterative prediction and adjustment based on the main forming influence parameters and the secondary forming influence parameters include:

[0018] The main forming parameters are mapped to a high-dimensional feature space to obtain a high-dimensional main forming parameter feature space; in the high-dimensional main forming parameter feature space, downsampling is performed based on the sparse grid approximation method to obtain a search subset of the main forming parameters;

[0019] Within the main forming parameter search subset, single-parameter simulation prediction is performed using the control variable method to obtain several single-parameter aerospace component prediction forming results; a variational Bayesian prediction model is constructed based on the main forming parameter search subset and several single-parameter aerospace component prediction forming results; in the variational Bayesian prediction model, the objective function is used to provide the posterior distribution characteristics corresponding to different influencing parameters;

[0020] The optimal SLM forming parameters are obtained by performing parameter optimization search based on the variational Bayesian prediction model and the master forming parameter search subset.

[0021] As a preferred embodiment of the present invention, the specific steps for parameter optimization search based on the variational Bayesian prediction model and the master forming parameter search subset include:

[0022] In the parameter optimization search process, global iterative search and local iterative search are used for parameter optimization. Specifically, the global iterative search uses an evolutionary algorithm to perform a global search in the principal forming parameter search subset to obtain a high-potential principal forming parameter search subset. The local iterative search performs a fine search in the vicinity of the high-potential principal forming parameter search subset based on the posterior distribution features provided by the variational Bayes prediction model to obtain a local principal forming parameter search subset.

[0023] Set the maximum number of iterations; introduce an interactive migration mechanism to replace the high-potential master forming parameter search subset with a local master forming parameter search subset when the high-potential master forming parameter search subset cannot be found in the global iterative search; set a set of interactive migration mechanisms as global-local loop;

[0024] When the maximum number of iterations is reached, the final optimal SLM forming parameters are output. Based on the optimal SLM forming parameters, the machining allowance of the aerospace component is adjusted to obtain new preset three-dimensional model data.

[0025] As a preferred technical solution of the present invention, after a round of global-local loop is completed, the weight coefficients corresponding to the objective function are adjusted based on preset rules.

[0026] A system for optimizing SLM forming parameters for aerospace components, comprising:

[0027] The aerospace component machining prediction module includes a model import unit and a machining prediction unit. The model import unit is used to acquire preset 3D model data of the aerospace component to be formed. The preset 3D model data includes the aerospace component body and the aerospace component machining allowance. Based on the basic characteristics of the aerospace component to be formed, material property data is determined and initial SLM forming parameters are generated. The initial SLM forming parameters include initial parameters for setting the aerospace component machining allowance. The machining prediction unit is used to establish a component machining prediction model. Based on the component machining prediction model, the aerospace component to be formed is predicted to obtain the predicted forming result of the aerospace component. The predicted forming result of the aerospace component includes the dimensional deviation, component surface roughness, and component internal defect degree corresponding to the aerospace component machining allowance.

[0028] The aerospace component parameter optimization module includes a parameter optimization unit. The parameter optimization unit is used to perform quantitative analysis of the influencing factors of the initial SLM forming parameters based on the predicted forming results of the aerospace component, and obtain the main forming influence parameters and secondary forming influence parameters. Iterative prediction and adjustment are performed based on the main forming influence parameters and secondary forming influence parameters until the optimal SLM forming parameters are selected and new preset three-dimensional model data is obtained. The SLM forming parameters of the aerospace component to be formed are optimized based on the new preset three-dimensional model data and the optimal SLM forming parameters.

[0029] The present invention has the following advantages:

[0030] 1. This invention integrates machining allowances, local geometric features, material properties, and historical forming data of aerospace components into a component machining prediction model. Combined with thermo-mechanical coupled finite element simulation, it achieves accurate prediction of multiple performance indicators such as dimensional deviation, surface roughness, and internal defect rate, effectively avoiding the problems of insufficient prediction accuracy and low data utilization in traditional trial-and-error methods. By establishing a quantitative analysis mechanism for primary and secondary forming influence parameters, and using gradient boosting regression and integrated feature ranking methods, it achieves hierarchical management of the importance of forming parameters, enabling the optimization process to prioritize the core parameters that have the greatest impact on forming quality, thereby improving optimization efficiency and the stability of results.

[0031] 2. This invention employs high-dimensional feature space mapping combined with sparse grid approximation for parameter search, and then uses a variational Bayesian prediction model for refined prediction. Through a global-local interactive iterative search strategy, it achieves efficient exploration of high-potential parameter regions in multi-objective optimization, significantly shortening parameter optimization time and improving multi-objective balance performance. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the structure of an SLM forming parameter optimization system for aerospace components used in an embodiment of the present invention. Detailed Implementation

[0033] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0034] Example 1: A method for optimizing SLM forming parameters for aerospace components, comprising the following steps:

[0035] Selective laser melting (SLM) is a powder bed-based metal additive manufacturing technology. Its working principle is to uniformly spread metal powder into a thin layer in a closed forming chamber filled with inert protective gas. Then, a high-energy-density laser beam is used to scan the designated area point by point and line by line according to the 3D model slice data, so that the powder melts instantly and metallurgically bonds with the underlying solidified material to complete the forming of one layer. After the laser scanning is completed, the forming platform descends by a set layer thickness, and a new layer of powder is spread and the above process is repeated until all layers are stacked, and finally a dense 3D metal component is obtained.

[0036] Acquire the preset 3D model data of the aerospace component to be formed; the preset 3D model data includes the aerospace component body and the aerospace component machining allowance; determine the material property data based on the basic characteristics of the aerospace component to be formed and generate the initial SLM forming parameters; the initial SLM forming parameters include the initial parameters for setting the aerospace component machining allowance.

[0037] The preset 3D model data is constructed by professional technicians. The preset 3D model data includes a complete external outline, internal cavity channels, assembly reference surfaces, and annotation information of each functional feature. At the same time, it retains component partition labels for additive manufacturing to distinguish thin-walled, overhanging, heat-prone areas from high-precision functional surfaces.

[0038] Specifically, the aerospace component body refers to the target geometric entity itself that meets structural and functional requirements. That is, the part of the material and shape that needs to be retained according to the design intent, including the load-bearing frame, assembly datum, sealing surface, channels and connecting holes, and all areas that actually participate in stress, heat transfer, sealing and assembly in service. It embodies the final shape and dimensional chain of the product and is the direct carrier of performance and reliability. Therefore, when modeling, slicing and setting parameters, the dimensional accuracy, geometric tolerances, material structure and mechanical properties of the body are the core objects of evaluation and optimization.

[0039] Machining allowance for aerospace components is a layer of protective and corrective material intentionally added around the critical surfaces and features of the component, which can be subsequently removed. Its thickness is usually set separately for each region to offset dimensional and quality deviations caused by thermal deformation, shrinkage, and surface roughness during additive manufacturing, and to provide sufficient cutting space for subsequent processes such as machining, grinding, and polishing. The role of machining allowance in aerospace components is to transfer unavoidable errors in the printing stage to the controllable removal process. By adopting differentiated energy input and scanning strategies for the areas containing the allowance in the slicing and process parameters, boundary forming stability and thermal field balance are ensured. After forming is completed, the allowance is removed according to the planned removal amount, so that the critical surfaces achieve the target dimensions, tolerances, and surface quality, thereby improving assembly consistency and manufacturing yield while ensuring performance.

[0040] The initial SLM forming parameters specifically include initial control parameters for setting the machining allowance of aerospace components, which are used to link the matching relationship between geometric allowance and process window. These initial control parameters define the target size deviation margin, surface quality target, and corresponding fine-tuning coefficients for layer thickness and scanning overlap rate for different functional surfaces. By applying differentiated powder spreading and scanning sub-strategies to the area containing allowance during the slicing stage, the heat input balance and boundary forming stability of the allowance area are achieved.

[0041] It should be noted that, in addition to the initial control parameters used to set the machining allowance for aerospace components, key process parameters also include laser power, scanning speed, layer thickness, scanning spacing, scanning path strategy, substrate preheating temperature, and protective atmosphere control. Laser power determines the depth and width of powder melting, which is an important factor affecting density, melt pool stability, and heat input balance. The combination of scanning speed and power determines the energy density per unit area, thus affecting forming quality, deformation degree, and production efficiency. Layer thickness determines the amount of single-layer deposition and thermal accumulation effect, which directly affects the forming accuracy of details and the distribution of thermal stress. Scanning spacing controls the degree of overlap between melt channels, affecting overall density and surface uniformity. Scanning path strategy is used to optimize the thermal field distribution and stress release, reducing the risk of warpage and cracking. Substrate preheating temperature helps reduce thermal gradient and residual stress, improving metallurgical bonding. Protective atmosphere control ensures that the oxygen content in the forming area is within a safe threshold, preventing powder and melt pool oxidation and improving material properties. These parameters, together with the machining allowance control parameters, constitute an adjustable process window, used to achieve comprehensive optimization of forming quality and subsequent processing under different geometric features and material conditions.

[0042] Establish a component processing prediction model; based on the component processing prediction model, predict the aerospace component to be formed to obtain the predicted forming result of the aerospace component; the predicted forming result of the aerospace component includes the dimensional deviation corresponding to the aerospace component processing allowance, the surface roughness of the component, and the internal defect degree of the component;

[0043] When establishing a component processing prediction model, a pre-set 3D model, material properties, and initial forming parameters are used as input elements. Combined with historical component size, surface, and internal quality inspection data, a mapping model covering the relationship between geometric features, process settings, and forming results is constructed. Before model training, multi-source data undergoes unified coordinate and unit normalization, outlier removal, and missing value completion. Local feature quantities are extracted based on structural labels such as thin walls, overhangs, thick ribs, and internal cavities. Subsequently, a hierarchical modeling approach is adopted to fuse information from different physical scales: numerical features formed by geometric and process parameters and texture features from online monitoring or microscopic images are co-input into the model. The generalization ability of the model is evaluated through cross-validation and independent holdout sets. Finally, a multi-objective prediction framework that can simultaneously output dimensional deviation, surface roughness, and internal defect degree is obtained, providing interpretable response results for subsequent parameter optimization.

[0044] The specific steps for predicting the aerospace components to be formed based on the component processing prediction model include:

[0045] For the prediction of dimensional deviation: obtain the local geometric feature parameters of the aerospace component body in the preset 3D model data; obtain historical SLM forming data; extract the historical forming parameter set with the highest similarity to the local geometric feature parameters from the historical SLM forming data; perform thermo-mechanical coupled finite element simulation based on the historical forming parameter set to obtain the predicted forming residual stress and predicted forming deformation; obtain the predicted dimensional deviation based on the predicted forming residual stress and predicted forming deformation.

[0046] In predicting dimensional deviations, local geometric features of the aerospace component body, such as wall thickness distribution, radius of curvature, overhang angle, and hot spot heat accumulation path, are automatically identified and quantified from the preset 3D model data. Then, the process combinations and boundary conditions most similar to these features are retrieved from the historical forming database to form a reference parameter set for simulation. Based on this reference set, thermal and mechanical coupled finite element calculations are performed, applying heat input and cooling cycles layer by layer to solve for the residual stress field and interlayer displacement field, obtaining the forming deformation trend under the current geometric and process settings. Finally, the predicted residual stress and deformation are projected onto the key reference surface and functional surface, the difference between the predicted and theoretical nominal dimensions is calculated, and compared with the set machining allowance, thereby outputting the dimensional deviation of each region and the judgment result of whether the subsequent machining removal amount is met.

[0047] For the prediction of component surface roughness: extract the initial forming scan parameters from the initial SLM forming parameters; calculate the historical roughness output with the highest similarity to the initial forming scan parameters from the historical SLM forming data; and make a prediction based on the historical roughness output to obtain the predicted component surface roughness.

[0048] In predicting the surface roughness of components, scanning settings directly related to surface forming are extracted from the initial SLM forming parameters. These include the ratio of laser power and speed, the combination of layer thickness and scanning spacing, path angle rotation, and edge finishing strategies. Simultaneously, the geometric surface orientation and slope are segmented to differentiate the energy differences between the upper surface, lower surface, and sidewalls. Subsequently, the set of samples with the highest similarity to the current scanning settings is calculated from the historical database. The corresponding roughness measurements and microscopic melt channel morphology features are summarized, establishing a correspondence between energy input density, melt channel overlap rate, and surface microtexture. Based on this, regional inferences are performed on the different orientation surfaces of the target component, outputting predicted roughness values ​​in micrometers and their distribution heatmaps. Areas that may require secondary processing or surface strengthening are also marked, providing a basis for the coordinated adjustment of processing allowance and sweeping strategies.

[0049] For the prediction of the internal defect degree of the component: the forming set temperature and forming scan overlap rate in the initial SLM forming parameters are extracted; based on the forming set temperature, forming scan overlap rate and material property data, a predictive analysis is performed to obtain the predicted internal defect degree of the component.

[0050] In predicting the internal defects of components, the forming set temperature and scanning overlap rate in the initial SLM forming parameters are read first, and combined with the material's thermal conductivity, specific heat capacity, powder particle size distribution, and upper limit of oxygen content in the atmosphere, the risk factors of molten pool stability and porosity formation are estimated. The forming set temperature refers to the temperature value that is preset and maintained for the substrate or forming area during selective laser melting, usually provided by the heating system of the forming equipment or an external heating source. It is used to reduce the temperature gradient of the metal material during forming, reduce residual stress and warping deformation caused by rapid heating and cooling, improve the interlayer metallurgical bonding quality, and inhibit crack formation. The forming scanning overlap rate refers to the ratio of the overlap width between adjacent scanning tracks to the effective width of the scanning track. It reflects the uniformity of single-layer powder coverage and the degree of fusion between molten pools.

[0051] By comparing the X-ray or tomographic scan data of historical components with the corresponding process settings, statistical indicators such as average porosity, maximum defect size, and defect location concentration are extracted. An empirical-physical hybrid model is established, which ranges from temperature field gradient, molten pool size fluctuation, and scan overlap imbalance to defect measurement. Subsequently, this model is applied to the zonal evaluation of the target component, providing a quantitative prediction of internal defect degree and spatial indication of high-risk areas. It is also linked with the processing allowance distribution for analysis, so as to make targeted corrections to energy input and path coverage before printing, reducing the pressure and cost of subsequent densification or hot isostatic pressing.

[0052] Based on the predicted forming results of aerospace components, a quantitative analysis of the influencing factors on the initial SLM forming parameters is conducted to obtain the main forming influencing parameters and the secondary forming influencing parameters. Based on the main forming influencing parameters and the secondary forming influencing parameters, iterative prediction and adjustment are performed until the optimal SLM forming parameters are selected and new preset three-dimensional model data is obtained. Among them, the SLM forming parameters of the aerospace component to be formed are optimized based on the new preset three-dimensional model data and the optimal SLM forming parameters.

[0053] The specific steps for quantitatively analyzing the influencing factors of initial SLM forming parameters based on the predicted forming results of aerospace components include:

[0054] A parameter-performance mapping relationship is constructed based on the initial SLM forming parameters and the predicted forming results of aerospace components; among them, the dimensional deviation, component surface roughness, and component internal defect degree are used as the prediction target variables; multiple feature importance evaluation sub-models are constructed by training the gradient boosting regression model and the parameter-performance mapping relationship.

[0055] Based on the initial SLM forming parameters and the predicted forming results of aerospace components, a mapping relationship covering the input parameters to the quality performance is established. This mapping relationship uses process parameters, geometric features and material properties as independent variables, and dimensional deviation, component surface roughness and component internal defect degree as output variables. Through joint processing and cleaning of historical samples and current prediction data, the dimensional unification, outlier removal and feature enhancement are completed. On this basis, a parameter-performance dataset that can be used for supervised learning is formed for subsequent model training and interpretation.

[0056] After the mapping relationship is constructed, multiple feature importance assessment sub-models based on gradient boosting regression are trained, with dimensional deviation, component surface roughness, and component internal defect degree as separate prediction targets. During training, sectional cross-validation and independent holdout set evaluation methods are used to ensure the robustness of each sub-model under different component structures and material conditions. After the model converges, the contribution score of each input feature to the corresponding target variable is output, forming three sets of importance measurement results for the three types of quality indicators, thus providing an interpretable basis for parameter priority division.

[0057] Based on the target variable for prediction, the feature contribution weights corresponding to each feature importance assessment sub-model are output. An integrated feature ranking algorithm is used to normalize the feature contribution weights corresponding to each feature importance assessment sub-model to obtain the scoring results of different parameters in the initial SLM molding parameters.

[0058] After obtaining the three sets of feature contribution scores, the outputs of each sub-model are normalized and integrated for ranking, so that the importance of parameters under different objectives is on the same evaluation scale. Specifically, the contribution weight under each objective is first standardized, and then the comprehensive score of each parameter is obtained by weighted summarization or voting fusion. The comprehensive score takes into account the requirements of dimensional accuracy, surface quality and internal density, avoiding the situation of over-optimizing only a single index at the expense of other indicators. After integration, the unified score result of each parameter in the initial SLM forming parameters can be obtained.

[0059] The parameters are classified according to their scores in the initial SLM molding parameters. Parameters with scores greater than a set threshold are classified as primary molding influence parameters, while the rest are classified as secondary molding influence parameters. The set threshold is manually set by professional technicians.

[0060] The specific steps for iterative prediction and adjustment based on the primary forming influence parameters and secondary forming influence parameters include:

[0061] The main forming parameters are mapped to a high-dimensional feature space to obtain a high-dimensional main forming parameter feature space; in the high-dimensional main forming parameter feature space, downsampling is performed based on the sparse grid approximation method to obtain a search subset of the main forming parameters;

[0062] In the step of mapping the main forming parameters to a high-dimensional feature space, the parameters are first standardized according to their physical dimensions, value ranges, and mutual constraints. Auxiliary features that reflect the differences in geometric partitions and material properties are introduced, so that each main parameter not only exists in numerical form but also carries semantic information about heat input, melt pool stability, and stress release. Subsequently, kernel mapping or orthogonal basis expansion is used to embed the parameters into a high-dimensional space that can characterize nonlinear correlations. In this space, a sparse mesh approximation method is used to downsample the parameter domain with adaptive density. Higher sampling density is assigned to dimensions that are sensitive to changes or have large historical errors, while sampling is reduced in regions with weak influence or monotonicity. This results in a search subset of main forming parameters that is controllable in scale and representative.

[0063] Within the main forming parameter search subset, single-parameter simulation prediction is performed using the control variable method to obtain several single-parameter aerospace component prediction forming results; a variational Bayesian prediction model is constructed based on the main forming parameter search subset and several single-parameter aerospace component prediction forming results; in the variational Bayesian prediction model, the objective function is used to provide the posterior distribution characteristics corresponding to different influencing parameters;

[0064] When performing single-parameter simulation prediction using the controlled variable method within the search subset of master forming parameters, the remaining master parameters and all secondary forming influence parameters are fixed within the baseline value or narrow range. Rapid forming prediction is then performed only for the target master parameter within its permissible range, taking stratified points. During the prediction process, response curves for three types of indicators—dimensional deviation, component surface roughness, and internal defect rate—are output, while local sensitivity and uncertainty estimates for different geometric regions are recorded. Based on these single-parameter response data and the point information of the search subset, a variational Bayesian prediction model with approximate posterior as the objective is constructed. The objective function defined by the three quality indicators is used as the optimization object of the evidence lower bound, enabling the model to provide the posterior distribution characteristics of each master parameter with respect to the objective function, including mean, variance, and correlation, without significantly increasing the number of simulations. This provides an interpretable prior with uncertainty for multi-parameter joint optimization.

[0065] The optimal SLM forming parameters are obtained by performing parameter optimization search based on the variational Bayesian prediction model and the master forming parameter search subset.

[0066] The specific steps for parameter optimization search based on the variational Bayesian prediction model and the master forming parameter search subset include:

[0067] In the parameter optimization search process, global iterative search and local iterative search are used for parameter optimization. Specifically, the global iterative search uses an evolutionary algorithm to perform a global search in the principal forming parameter search subset to obtain a high-potential principal forming parameter search subset. The local iterative search performs a fine search in the vicinity of the high-potential principal forming parameter search subset based on the posterior distribution features provided by the variational Bayes prediction model to obtain a local principal forming parameter search subset.

[0068] A collaborative mechanism of global and local two-level iterations is employed, enabling the search to possess both cross-regional exploration capabilities and the ability to finely approximate high-quality solutions. The global iterative search uses a subset of master forming parameters as the initial population and adopts an evolutionary search with adaptive mutation and optimal retention strategies. Under the premise of satisfying physical feasibility constraints and equipment safety boundaries, it continuously evaluates the comprehensive performance of candidate parameter combinations in three indicators: dimensional deviation, surface roughness, and internal defects. As the generations progress, candidate solutions are sorted and reorganized based on comprehensive fitness, dynamically screening out high-potential master forming parameter search subsets that contain diversity and excellence, serving as the starting point for subsequent local iterations.

[0069] In the local iterative search phase, the posterior distribution features given by the variational Bayesian prediction model are invoked around the solution points of the high-potential master forming parameter search subset. For parameters with high uncertainty, sampling is intensified, while for dimensions with small posterior variance and already stable, the step size and boundaries are reduced. This improves the effectiveness of the refined search while ensuring efficiency. The local search uses multi-objective comprehensive loss as the evaluation criterion, incorporating dimensional deviation constraints, surface quality targets, and defect risk thresholds into the evaluation. Rapid prediction and verification are performed on each neighborhood solution to form a local master forming parameter search subset. The corresponding confidence interval and sensitivity are recorded for adaptive adjustment of the search strategy in subsequent rounds.

[0070] Set the maximum number of iterations; introduce an interactive transfer mechanism to replace the local master forming parameter search subset when a high-potential master forming parameter search subset cannot be found in the global iterative search; set a set of interactive transfer mechanisms as global-local loop; the maximum number of iterations is manually set by professional technicians.

[0071] After a global-local loop is completed, the weight coefficients corresponding to the objective function are adjusted based on preset rules;

[0072] Specifically, the predicted values ​​of the optimal and candidate solutions in this round are summarized for three indicators: dimensional deviation, surface roughness, and internal defects. These values ​​are then compared item by item with the corresponding target values ​​and allowable tolerances. The normalized deviation of each indicator and the penalty term for whether the constraint boundary is touched are calculated. At the same time, the uncertainty given by the variational Bayesian model is read to measure the credibility of the current evaluation. Subsequently, according to the preset weight update rules, the three weight coefficients in the objective function are adaptively adjusted: the weights of indicators with large deviations or exceeding limits are increased; the weights of indicators that are significantly better than the target, have limited room for improvement, and have low uncertainty are decreased; and the weights of indicators with high uncertainty are moderately increased to promote improvement. Further identification; to avoid back-and-forth oscillations, the update adopts a moving average with a smoothing factor and upper and lower bound constraints, and sets a minimum step size and a maximum step size to control the adjustment speed. At the same time, the priority of key areas (such as assembly reference surface, sealing surface, stress hot spot, etc.) is introduced to add weights to the relevant areas, so that the global objective will automatically tilt towards the true bottleneck indicator in the next round of search. After completing the weight recalibration, the objective function is normalized to keep the dimensions of each item consistent. The new weight coefficients are archived along with the search statistics and uncertainty information of this round and used as the starting setting for the next round of global-local search, so as to achieve dynamic balance and robust convergence among multiple objectives at different stages.

[0073] When the maximum number of iterations is reached, the final optimal SLM forming parameters are output. Based on the optimal SLM forming parameters, the machining allowance of the aerospace component is adjusted to obtain new preset three-dimensional model data.

[0074] To prevent search stagnation and maintain solution diversity, the algorithm sets a maximum number of iterations as an upper limit and introduces a global-local interactive migration mechanism: when the global iteration fails to generate new high-potential subsets or the overall fitness improvement falls below a set threshold within several iterations, the low-quality part of the global population is replaced by the best-performing and most confident solution cluster in the current local subset, thereby breaking the local optimum trap and reactivating global exploration; conversely, when the local search exhibits excessive convergence or insufficient marginal improvement, highly diverse candidate solutions are introduced from the global population to expand the local neighborhood; when the maximum number of iterations is reached or the convergence criterion is met, the final optimal selective laser melting forming parameters are output, and these optimal parameters, along with the latest response of the prediction model, are written back to the component model. The machining allowances corresponding to each key functional surface and machining surface are quantitatively adjusted and updated, generating new preset 3D model data, providing a consistent and executable process configuration for subsequent slicing, forming, and subsequent machining.

[0075] Meanwhile, it should be noted that during the parameter optimization search process, different search strategies with different step sizes are adopted for the primary forming influence parameters and the secondary forming influence parameters to balance the needs of global exploration and local refinement. The primary forming influence parameters, as core variables affecting dimensional deviation, surface roughness, and internal defects, are searched across intervals within the parameter space using a larger step size to quickly cover the global scope and capture the distribution trend of potential high-quality solutions. The secondary forming influence parameters, on the other hand, are mainly used to fine-tune forming stability, refine the thermal field distribution, and improve local quality; their variation has a relatively mild impact on the overall process window. Therefore, a smaller step size is used to perform fine-grained scanning under the condition of fixed or narrow fluctuation of the master parameters, so as to improve the quality of details and the accuracy of boundary control without destroying the optimization direction of the master parameters. The setting and adjustment of the two types of step sizes rely on the posterior variance and sensitivity analysis results given by the variational Bayes prediction model. The step size is dynamically increased in areas with greater uncertainty and decreased in areas with obvious convergence trend, thus forming a dual-scale search mechanism that can quickly locate global high-potential areas and achieve stable convergence in key neighborhoods, so that the optimal forming parameters obtained in the end have both global applicability and local accuracy.

[0076] In this embodiment, a selective laser melting forming parameter optimization process was carried out for a turbine casing component of an aero-engine. The component material is the high-temperature alloy Inconel 718. In the preset three-dimensional model, a machining allowance of 0.35 mm was reserved on all assembly reference surfaces and key sealing surfaces, and a machining allowance of 0.20 mm was reserved on non-critical outer contours. The initial forming parameters included a laser power of 340 watts, a scanning speed of 900 mm / s, a layer thickness of 30 micrometers, a scanning spacing of 0.11 mm, a scanning overlap rate of 32%, a substrate preheating temperature of 200 degrees Celsius, and an oxygen content of less than 100 ppm in the argon protective atmosphere. After obtaining the sample with the most similar geometric features from the historical forming database, a thermo-mechanical coupled finite element simulation model was established to predict residual stress and deformation. The results showed that the initial dimensional deviation could reach +0.28 mm on the key reference surface, the local surface roughness Ra was 12.5 micrometers, and the average porosity of internal defects was 0.32%. Based on gradient boosting regression model analysis of influencing factors, laser power, scanning speed, and scanning overlap rate were classified as primary forming influencing parameters, while scanning path strategy and substrate preheating temperature were classified as secondary forming influencing parameters. During the optimization search process, the step size of the primary forming influencing parameters was set to laser power ±10 watts, scanning speed ±40 mm / s, and scanning overlap rate ±2%, while the step size of the secondary forming influencing parameters was set to ±5% for relative adjustment. A global-local cyclic and variational Bayesian prediction model was used to guide the search. The objective function was to optimize dimensional deviation ≤ ±0.05 mm, local surface roughness Ra ≤ 6.3 μm, and internal defect rate ≤ 0.10%. Finally, the optimal process combination of laser power 355 watts, scanning speed 880 mm / s, scanning overlap rate 34%, and substrate preheating temperature 210 degrees Celsius was found. Simulation verification showed that the dimensional deviation was reduced to ±0.04 mm, the local surface roughness Ra was 5.9 μm, and the internal defect rate was 0.08%. Based on the optimization results, a new preset 3D model processing allowance distribution was updated for subsequent actual forming and machining stages.

[0077] Example 2, an SLM forming parameter optimization system for aerospace components, see [link to example]. Figure 1 As shown, it includes:

[0078] The aerospace component machining prediction module includes a model import unit and a machining prediction unit. The model import unit is used to acquire preset 3D model data of the aerospace component to be formed. The preset 3D model data includes the aerospace component body and the aerospace component machining allowance. Based on the basic characteristics of the aerospace component to be formed, material property data is determined and initial SLM forming parameters are generated. The initial SLM forming parameters include initial parameters for setting the aerospace component machining allowance. The machining prediction unit is used to establish a component machining prediction model. Based on the component machining prediction model, the aerospace component to be formed is predicted to obtain the predicted forming result of the aerospace component. The predicted forming result of the aerospace component includes the dimensional deviation, component surface roughness, and component internal defect degree corresponding to the aerospace component machining allowance.

[0079] The aerospace component parameter optimization module includes a parameter optimization unit. The parameter optimization unit is used to perform quantitative analysis of the influencing factors of the initial SLM forming parameters based on the predicted forming results of the aerospace component, and obtain the main forming influence parameters and secondary forming influence parameters. Iterative prediction and adjustment are performed based on the main forming influence parameters and secondary forming influence parameters until the optimal SLM forming parameters are selected and new preset three-dimensional model data is obtained. The SLM forming parameters of the aerospace component to be formed are optimized based on the new preset three-dimensional model data and the optimal SLM forming parameters.

[0080] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for optimizing SLM forming parameters for aerospace components, characterized in that, Includes the following steps: Acquire the preset 3D model data of the aerospace component to be formed; the preset 3D model data includes the aerospace component body and the aerospace component machining allowance; determine the material property data based on the basic characteristics of the aerospace component to be formed and generate the initial SLM forming parameters; the initial SLM forming parameters include the initial parameters for setting the aerospace component machining allowance. Establish a component processing prediction model; based on the component processing prediction model, predict the aerospace component to be formed to obtain the predicted forming result of the aerospace component; the predicted forming result of the aerospace component includes the dimensional deviation corresponding to the aerospace component processing allowance, the surface roughness of the component, and the internal defect degree of the component; Based on the predicted forming results of aerospace components, a quantitative analysis of the influencing factors on the initial SLM forming parameters was conducted to obtain the main forming influencing parameters and the secondary forming influencing parameters. Iterative prediction and adjustment are performed based on the primary and secondary forming influence parameters until the optimal SLM forming parameters are selected and new preset 3D model data is obtained. Specifically, the SLM forming parameters of the aerospace component to be formed are optimized based on the new preset 3D model data and the optimal SLM forming parameters. The specific steps for predicting the aerospace component to be formed based on the component processing prediction model include: For the prediction of dimensional deviation: obtain the local geometric feature parameters of the aerospace component body in the preset 3D model data; obtain historical SLM forming data; extract the historical forming parameter set with the highest similarity to the local geometric feature parameters from the historical SLM forming data; perform thermo-mechanical coupled finite element simulation based on the historical forming parameter set to obtain the predicted forming residual stress and predicted forming deformation; obtain the predicted dimensional deviation based on the predicted forming residual stress and predicted forming deformation. For the prediction of component surface roughness: extract the initial forming scan parameters from the initial SLM forming parameters; calculate the historical roughness output with the highest similarity to the initial forming scan parameters from the historical SLM forming data; and make a prediction based on the historical roughness output to obtain the predicted component surface roughness. For the prediction of internal defects in components: the forming set temperature and forming scan overlap rate are extracted from the initial SLM forming parameters; based on the forming set temperature, forming scan overlap rate and material property data, a predictive analysis is performed to obtain the predicted internal defects in the components.

2. The method for optimizing SLM forming parameters for aerospace components according to claim 1, characterized in that, The specific steps for quantitatively analyzing the influencing factors of initial SLM forming parameters based on the predicted forming results of aerospace components include: A parameter-performance mapping relationship is constructed based on the initial SLM forming parameters and the predicted forming results of aerospace components; among them, the dimensional deviation, component surface roughness, and component internal defect degree are used as the prediction target variables; multiple feature importance evaluation sub-models are constructed by training the gradient boosting regression model and the parameter-performance mapping relationship. Based on the target variable for prediction, the feature contribution weights corresponding to each feature importance assessment sub-model are output. An integrated feature ranking algorithm is used to normalize the feature contribution weights corresponding to each feature importance assessment sub-model to obtain the scoring results of different parameters in the initial SLM molding parameters. The parameters are classified according to their scores in the initial SLM forming parameters; the parameters whose scores are greater than a set threshold are classified as primary forming parameters, and the rest are classified as secondary forming parameters.

3. The method for optimizing SLM forming parameters for aerospace components according to claim 2, characterized in that, The specific steps for iterative prediction and adjustment based on the primary forming influence parameters and secondary forming influence parameters include: The main forming parameters are mapped to a high-dimensional feature space to obtain a high-dimensional main forming parameter feature space; in the high-dimensional main forming parameter feature space, downsampling is performed based on the sparse grid approximation method to obtain a search subset of the main forming parameters; Within the main forming parameter search subset, single-parameter simulation prediction is performed using the control variable method to obtain several single-parameter aerospace component prediction forming results; a variational Bayesian prediction model is constructed based on the main forming parameter search subset and several single-parameter aerospace component prediction forming results; in the variational Bayesian prediction model, the objective function is used to provide the posterior distribution characteristics corresponding to different influencing parameters; The optimal SLM forming parameters are obtained by performing parameter optimization search based on the variational Bayesian prediction model and the master forming parameter search subset.

4. The method for optimizing SLM forming parameters for aerospace components according to claim 3, characterized in that, The specific steps for parameter optimization search based on the variational Bayesian prediction model and the master forming parameter search subset include: In the parameter optimization search process, global iterative search and local iterative search are used for parameter optimization. Specifically, the global iterative search uses an evolutionary algorithm to perform a global search in the principal forming parameter search subset to obtain a high-potential principal forming parameter search subset. The local iterative search performs a fine search in the vicinity of the high-potential principal forming parameter search subset based on the posterior distribution features provided by the variational Bayes prediction model to obtain a local principal forming parameter search subset. Set the maximum number of iterations; introduce an interactive migration mechanism to replace the high-potential master forming parameter search subset with a local master forming parameter search subset when the high-potential master forming parameter search subset cannot be found in the global iterative search; set a set of interactive migration mechanisms as global-local loop; When the maximum number of iterations is reached, the final optimal SLM forming parameters are output. Based on the optimal SLM forming parameters, the machining allowance of the aerospace component is adjusted to obtain new preset three-dimensional model data.

5. The method for optimizing SLM forming parameters for aerospace components according to claim 4, characterized in that, After a global-local loop is completed, the weight coefficients corresponding to the objective function are adjusted based on preset rules.

6. A system for optimizing SLM forming parameters for aerospace components, characterized in that, The system applies the SLM forming parameter optimization method for aerospace components as described in any one of claims 1-5, including: The aerospace component machining prediction module includes a model import unit and a machining prediction unit. The model import unit is used to acquire preset 3D model data of the aerospace component to be formed. The preset 3D model data includes the aerospace component body and the aerospace component machining allowance. Based on the basic characteristics of the aerospace component to be formed, material property data is determined and initial SLM forming parameters are generated. The initial SLM forming parameters include initial parameters for setting the aerospace component machining allowance. The machining prediction unit is used to establish a component machining prediction model. Based on the component machining prediction model, the aerospace component to be formed is predicted to obtain the predicted forming result of the aerospace component. The predicted forming result of the aerospace component includes the dimensional deviation, component surface roughness, and component internal defect degree corresponding to the aerospace component machining allowance. The aerospace component parameter optimization module includes a parameter optimization unit. The parameter optimization unit is used to perform quantitative analysis of the influencing factors of the initial SLM forming parameters based on the predicted forming results of the aerospace component, and obtain the main forming influence parameters and the secondary forming influence parameters. Iterative prediction and adjustment are performed based on the main forming influence parameters and the secondary forming influence parameters until the optimal SLM forming parameters are selected and new preset three-dimensional model data is obtained. The optimization of the SLM forming parameters of the aerospace component to be formed is achieved based on the new preset three-dimensional model data and the optimal SLM forming parameters. The specific steps for predicting the aerospace components to be formed based on the component processing prediction model include: For the prediction of dimensional deviation: obtain the local geometric feature parameters of the aerospace component body in the preset 3D model data; obtain historical SLM forming data; extract the historical forming parameter set with the highest similarity to the local geometric feature parameters from the historical SLM forming data; perform thermo-mechanical coupled finite element simulation based on the historical forming parameter set to obtain the predicted forming residual stress and predicted forming deformation; obtain the predicted dimensional deviation based on the predicted forming residual stress and predicted forming deformation. For the prediction of component surface roughness: extract the initial forming scan parameters from the initial SLM forming parameters; calculate the historical roughness output with the highest similarity to the initial forming scan parameters from the historical SLM forming data; and make a prediction based on the historical roughness output to obtain the predicted component surface roughness. For the prediction of internal defects in components: the forming set temperature and forming scan overlap rate are extracted from the initial SLM forming parameters; based on the forming set temperature, forming scan overlap rate and material property data, a predictive analysis is performed to obtain the predicted internal defects in the components.

Images