Additive manufacturing intelligent repair and surface optimization method for complex special-shaped curved surface

Abstract

The invention discloses an additive manufacturing intelligent repairing and surface optimizing method for a complex special-shaped curved surface, and particularly relates to the technical field of additive manufacturing. A three-dimensional digital model is obtained, and a complex special-shaped curved surface defect area is recognized; extracting a local morphology parameter set; constructing a repair path generation model based on a graph neural network, and performing collaborative optimization of forming precision, residual stress and deposition stability on the path through a multi-objective optimization algorithm; additive manufacturing equipment is controlled to perform material deposition according to the optimized path; carrying out online error detection by adopting a structured light system, judging repair quality through a fuzzy decision tree model, and triggering path reconstruction if the repair quality exceeds a threshold value; after repairing is completed, surface micro texture factors are extracted, and functional texture regulation and control are conducted through the laser micromachining technology; according to the method, high-precision intelligent repair of the complex curved surface area and collaborative improvement of surface function performance are achieved, and the method is suitable for the fields of aerospace, energy, high-end equipment manufacturing and the like.

Description

Technical Field

[0001] The present invention relates to the field of additive manufacturing technology, and in particular to an additive manufacturing intelligent repair and surface optimization method for complex special-shaped curved surfaces. Background Art

[0002] With the continuous improvement of the requirements for the precision and service life of complex components in the fields of aerospace, high-end equipment manufacturing, energy, etc., traditional manufacturing and repair technologies have problems such as low repair accuracy, low processing efficiency, and difficulty in generating process paths when facing components with complex special-shaped surfaces (such as discontinuous curvature, multi-axis bending and torsion, multi-scale textures, etc.), making it difficult to meet the stringent requirements of high-performance equipment for surface quality and mechanical properties.

[0003] While existing surface repair solutions based on technologies like laser cladding, cold spraying, and additive manufacturing exist, they still rely heavily on manual modeling and empirical path planning, making it difficult to achieve high-precision, intelligent repair of complex, irregularly shaped surfaces. Furthermore, traditional additive manufacturing methods often fail to balance structural mechanical properties with surface topography quality, leading to potential failure risks such as microcracks and stress concentration in the repaired area.

[0004] Therefore, there is an urgent need for an intelligent additive manufacturing repair and surface optimization method for complex special-shaped surfaces, which can integrate intelligent recognition of complex surfaces, path adaptive planning and multi-objective optimization strategies to achieve coordinated optimization of the mechanical properties and surface quality of the repair area, improve the repair accuracy, efficiency and reliability, and become a key solution to break through the existing technical bottlenecks. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent repair and surface optimization method for additive manufacturing of complex special-shaped surfaces to address the shortcomings of the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solutions: an additive manufacturing intelligent repair and surface optimization method for complex special-shaped surfaces, comprising:

[0007] S100, obtaining a three-dimensional digital model of a workpiece to be repaired, and identifying a complex irregular curved surface area with defects on the surface thereof, wherein the area has discontinuous curvature or multi-axis free-form features;

[0008] S200, performing curvature analysis and morphological feature extraction on the irregular curved surface area to generate a corresponding local morphological parameter set P;

[0009] S300, based on the parameter set P, constructing an additive manufacturing repair path generation model for special-shaped curved surfaces, and performing multi-objective optimization of the path using a machine learning algorithm, with the objectives including maximizing forming accuracy, minimizing residual stress, and improving cladding stability;

[0010] S400: Based on the optimized repair path, the additive manufacturing equipment is controlled to deposit materials to complete the precise repair of the defective area.

[0011] S500, performing online inspection and error evaluation on the repaired surface. If the error exceeds a preset threshold, returning to S200 to re-optimize the repair path;

[0012] S600: After the repair is completed, the surface of the repaired area is microscopically reconstructed and the texture is regulated.

[0013] Preferably, the S200 includes:

[0014] S201, generating a local parameterized surface model based on the three-dimensional point cloud data of the irregular curved surface area by using an adaptive high-order fitting surface reconstruction method;

[0015] S202, jointly extracting multi-scale discrete Gaussian curvature and mean curvature from the parameterized surface model to construct a curvature distribution matrix;

[0016] S203: Utilize a multi-channel shape and texture coding network to fuse the curvature distribution matrix with the surface gradient texture image, generate a local shape description vector, and construct a local shape parameter set P.

[0017] Preferably, the S300 includes:

[0018] S301, constructing a repair path generation model based on a graph neural network, mapping the local morphology parameter set P into a graph structure input;

[0019] S302: Perform feature aggregation and edge weight optimization on the graph structure input, combine historical repair data labels, and train to generate a path candidate set with path feasibility prediction capabilities;

[0020] S303, introducing a joint objective function, integrating the forming accuracy evaluation index, the residual stress distribution prediction results based on thermal-mechanical coupling simulation, and the dynamic stability score of the material deposition process, to perform multi-objective optimization on the path candidate set;

[0021] S304: Use a reinforcement learning algorithm to update the optimization results and output a global optimal additive path sequence.

[0022] Preferably, the fusion multi-objective optimization parameters include:

[0023] S3031. Mapping the additive path candidate set to the parametric component model, and using the scanning contour error superposition algorithm to calculate the forming accuracy evaluation index of each path;

[0024] S3032. Based on the path geometry and process parameter input, a thermal-mechanical coupling finite element model is constructed to simulate and calculate the instantaneous temperature field and stress field of each path during the deposition process, and the residual stress peak value and its spatial distribution coefficient are extracted as path evaluation factors;

[0025] S3033: Collect the deposition rate changes, molten pool temperature fluctuations and interlayer overlap quality driven by the path, use the stability scoring function F(t), and output the path dynamic stability score;

[0026] S3034. The forming accuracy evaluation index, residual stress peak value, its spatial distribution coefficient and path dynamic stability score are standardized and weightedly integrated into the path objective function.

[0027] Preferably, the S500 includes:

[0028] S501. Use a structured light 3D reconstruction system to scan the repair area online, obtain high-precision surface point cloud data, and match it to the original design model in real time to establish an error comparison coordinate system;

[0029] S502, performing spatial vectorization on the deviation between the reconstructed surface and the original design surface, and calculating the maximum normal offset, curvature offset, and inter-surface transition rate;

[0030] S503: Construct an error assessment model based on a fuzzy decision tree, perform fuzzy fusion judgment on the maximum normal offset, curvature offset, and inter-surface transition rate, and output a binary judgment result of whether the error exceeds a preset error threshold;

[0031] S504: If the error determination result is out of standard, the process automatically triggers a return to S200, and updates the local morphology parameter set P based on the current error distribution.

[0032] Preferably, constructing an error assessment model based on a fuzzy decision tree includes:

[0033] S5031. Taking the maximum normal deviation, curvature deviation, and inter-surface transition rate as input variables, respectively, define the fuzzy membership function of each variable and divide it into three levels: low deviation, medium deviation, and high deviation;

[0034] S5032. Build a fuzzy rule base, use a fuzzy inference engine to match the input error indicators, and calculate the comprehensive error level through a weighted average membership aggregation method;

[0035] S5033. Convert the comprehensive error level into a binary judgment result through a threshold function. If the level is higher than a preset threshold, output an exceeding-standard signal.

[0036] Preferably, the S600 includes:

[0037] S601, obtaining a microscopic topography map of the repaired area and constructing its frequency domain texture feature matrix;

[0038] S602: Input the texture feature matrix into a multi-scale residual attention network model to extract texture factors associated with target wear resistance or hydrophobicity;

[0039] S603. Generate a micro-repair process parameter set based on the extracted factors to control the nanosecond laser scanning path and energy distribution.

[0040] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0041] 1. By introducing a graph neural network-based path generation mechanism and a multi-objective machine learning optimization algorithm, this invention enables intelligent design and adaptive adjustment of repair paths when processing complex, irregularly shaped surfaces. Compared with traditional fixed-rule path planning methods, this method demonstrates significant advantages in forming accuracy control, residual stress distribution management, and deposition process stability, effectively improving the overall quality and stability of the additive manufacturing repair process.

[0042] 2. This method combines online detection with a fuzzy decision tree evaluation model to construct a closed-loop control system for repair quality. After the repair is complete, a microtexture reconstruction strategy is used to enhance surface functional properties, such as abrasion resistance or hydrophobicity. This method not only overcomes the technical bottleneck of traditional methods for complex surface recognition and control, but also possesses high scalability and engineering practical value, making it particularly suitable for fields such as aerospace and high-end manufacturing, where repair precision and surface performance are extremely demanding. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0044] Figure 1 This is a mind map of the method of the present invention. DETAILED DESCRIPTION

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0046] Example 1, please refer to Figure 1 As shown, the additive manufacturing intelligent repair and surface optimization method for complex special-shaped surfaces described in this embodiment includes:

[0047] S100, obtaining a three-dimensional digital model of a workpiece to be repaired, and identifying a complex irregular curved surface area with defects on the surface thereof, wherein the area has discontinuous curvature or multi-axis free-form features;

[0048] S200, performing curvature analysis and morphological feature extraction on the irregular curved surface area to generate a corresponding local morphological parameter set P;

[0049] S300, based on the parameter set P, constructing an additive manufacturing repair path generation model for special-shaped curved surfaces, and performing multi-objective optimization of the path using a machine learning algorithm, with the objectives including maximizing forming accuracy, minimizing residual stress, and improving cladding stability;

[0050] S400: Based on the optimized repair path, the additive manufacturing equipment is controlled to deposit materials to complete the precise repair of the defective area.

[0051] S500, performing online inspection and error evaluation on the repaired surface. If the error exceeds a preset threshold, returning to S200 to re-optimize the repair path;

[0052] S600: After the repair is completed, the surface of the repaired area is microscopically reconstructed and the texture is regulated.

[0053] In this embodiment, a three-dimensional digital model of a workpiece to be repaired is obtained, and complex irregular curved surface areas with defects on the surface are identified, specifically including the following steps:

[0054] First, a high-precision industrial-grade blue light scanner is used to perform non-contact full-area scanning of the surface of the workpiece to be repaired, obtaining point cloud data with a coverage rate greater than 99%. The multi-angle scanning results are then fused through an adaptive registration algorithm to generate a three-dimensional digital model with an accuracy better than 0.02mm.

[0055] Then, the three-dimensional model is imported into the geometric feature extraction module, and the curvature field analysis of each small area on the surface of the model is performed using the Gaussian curvature and mean curvature combined calculation method based on the second-order derivative. -1 ) are marked as discontinuous curvature areas; at the same time, combined with the analysis of the principal curvature direction field and the normal vector offset rate, areas with multi-axis free deformation characteristics, such as multiple torsion edges and non-uniform gradient transition surfaces, are screened out and marked as multi-axis free form feature areas.

[0056] To further identify surface defects, a density-based defect detection algorithm is used to compare the local density of the point cloud with the theoretical design model. After identifying areas of point cloud defects, warping, or local collapse, the aforementioned geometric analysis results are combined to determine the complex, irregularly shaped surface areas to be repaired, which serve as key input areas for subsequent additive path generation.

[0057] Through this step, not only is high-precision modeling of the geometric features of complex special-shaped surfaces achieved, but the local adaptability and targeting of subsequent repair paths are also guaranteed, laying the foundation for precise repair and surface optimization.

[0058] In this embodiment, based on the local morphology parameter set P, a model for additive manufacturing repair path generation for complex special-shaped surfaces is constructed, and a globally optimal additive path sequence is obtained through a multi-objective optimization algorithm. The path generation and optimization method specifically includes the following steps:

[0059] First, the local morphological parameter set P extracted in S200 is used as the input basis. This parameter set P includes but is not limited to descriptors of multiple geometric dimensions, such as local Gaussian curvature, mean curvature, surface gradient direction, curvature gradient distribution density, and normal perturbation distribution. To efficiently capture the non-Euclidean relationships between these complex surface features, the set P is converted into a graph structure input G = (V, E), where V represents the set of feature points and E represents the edge weight relationship constructed based on surface connectivity and geometric similarity.

[0060] In constructing the graph neural network model, a structure based on a graph attention mechanism is employed to achieve regional aggregation and non-uniform weighting of geometric features within irregularly shaped surfaces. This model adjusts the weights of edges between nodes, resulting in greater sensitivity to characteristic regions such as transitions and sudden changes within irregular surfaces.

[0061] The graph structure G is input into a trained path generation model. During the training phase, the model incorporates path trajectories and restoration results from historical restoration projects as label information, forming a supervised learning mechanism. By mapping and learning the aggregated features between graph nodes, the model outputs multiple candidate sets of additive manufacturing paths. Each path consists of a continuous sequence of spatial trajectory points and a matching set of process parameters (such as scanning speed, power, and layer thickness). This set of candidate paths is capable of scoring path feasibility, filtering out solutions that are structurally unstable, cross-boundary, or unformable.

[0062] In order to achieve a synergistic improvement in forming quality and service performance, a joint objective function integrating three types of performance evaluation indicators is introduced to perform multi-objective optimization on the path candidate set. This process is divided into the following four sub-steps:

[0063] The candidate paths are mapped one by one onto the target parametric component model, and a spatial comparison mechanism is established between the theoretical component surface and the path stacking reconstruction model. A scanning contour error superposition algorithm is used to calculate the cumulative normal error between the projected contour of each path on the component surface and the original contour, generating a geometric error surface graph. This graph is used to extract multiple accuracy parameters, such as the maximum local deviation, the overall average deviation, and the continuity disturbance index, as a basis for evaluating path forming accuracy.

[0064] For each path, a thermal-mechanical coupled finite element simulation model was constructed, combining its spatial geometry and matching process parameters. This model utilizes a moving heat input model (e.g., a double ellipsoid heat source) for the deposition heat source to simulate temperature field changes during deposition and calculate the thermal stress distribution generated during cooling. The maximum residual stress peak value, stress gradient change rate, and stress concentration factor along the path along the Z-axis were extracted as quantitative parameters of structural integrity risk. To improve the model's response efficiency, parameterized order reduction was employed to reduce the simulation dimensionality.

[0065] Data from the path-driven deposition process is collected, including metrics such as instantaneous deposition velocity change Δv(t), melt pool center temperature fluctuation ΔT(t), and interlayer overlap error rate r_lap. This function, F(t), integrates the dynamic characteristics of three physical fields and generates a single-value score using a variable-weighted fuzzy integral model. This score reflects the path's comprehensive performance on material transport continuity, interlayer bonding consistency, and melt pool thermal stability. Higher scores indicate a more stable deposition process and can be used to help eliminate potential risk paths for discontinuities or defects.

[0066] The three performance indicators mentioned above—forming accuracy evaluation index, residual stress peak value and its spatial distribution coefficient, and path dynamic stability score—are normalized and then a weighted fusion strategy is used to construct a joint path objective function. The objective function is as follows: F_total = w1·E_geo + w2·S_res + w3·D_stab, where E_geo represents the comprehensive forming accuracy error; S_res represents the composite residual stress index; and D_stab represents the dynamic stability score (higher is better). w1, w2, and w3 are preset weight coefficients that are adaptively adjusted based on the application scenario (e.g., E_geo is prioritized for precision parts, while S_res is prioritized for structural parts).

[0067] Finally, using F_total as the reward function in reinforcement learning, a path policy optimization module based on Deep Q Learning (DQN) is introduced to iteratively select candidate paths. The system automatically evaluates the path quality corresponding to the current strategy and converges the strategy through multiple rounds of simulation evaluation and actual feedback until the optimal path sequence with minimum F_total (or maximum D_stab) is found. The output path contains a set of three-dimensional coordinate trajectory points, a set of process control parameters, and a score record of multiple rounds of optimization, which serves as the path control instruction for the actual additive manufacturing process.

[0068] In this embodiment, the globally optimal additive path sequence obtained in S300 is used to precisely repair defective areas on complex, irregularly shaped surfaces. The material deposition process is performed by additive manufacturing equipment with an integrated multi-axis linkage control system, a laser directed energy deposition system also applicable to controllable energy field-driven material deposition systems such as laser cladding and cold spraying.

[0069] First, the optimized 3D additive path sequence (including the spatial trajectory point set and process parameter set) is input into the path parsing module. This module reconstructs the path based on the standard G-code framework and embeds dynamic control instructions for process parameters at key points, such as laser power P(t), scanning speed v(t), powder feed rate f(t), and layer thickness h(t).

[0070] These control commands are transmitted via an industrial Ethernet interface to the additive manufacturing system's motion control and process control modules, enabling synchronized trajectory and process control. For multi-axis areas with irregularly shaped surfaces, the system automatically activates a five-axis or six-axis linkage control strategy to ensure the nozzle is always aligned with the deposition surface at a normal angle, preventing over-melting or melt pool drift.

[0071] During the actual deposition process, the control system executes material deposition layer by layer according to the path point instructions. A composite strategy of Z-direction rise and rotation angle adjustment is used between each layer path to ensure uniform overlap between layers and a stable melt pool.

[0072] For path segments with abrupt changes in curvature or discontinuous curvature, the system employs a highly dynamic response strategy, dynamically adjusting laser power and scanning speed to offset uneven energy density caused by sudden changes in curvature. For example, in areas where the curvature radius is less than a set threshold (e.g., R < 5mm), laser power is reduced by 10% and scanning speed is increased by 15% to ensure that penetration depth does not exceed the specified value.

[0073] In the multi-axis transition area, the control system preferentially calls the "transition section fine-tuning control library" to automatically fine-tune the powder feeding amount and trajectory node density based on the surface normal change rate Δn and the angle θ between the front and rear trajectory vectors to achieve a fine transition in the edge area.

[0074] To ensure deposition quality and stability, the additive manufacturing equipment integrates a multi-sensor online monitoring module to collect the following parameters in real time:

[0075] Molten pool temperature distribution diagram (based on infrared thermal imager);

[0076] Material deposition height and layer thickness (based on laser displacement meter);

[0077] Molten pool shape boundary contour (based on high-speed vision sensor);

[0078] Feedback of actual process execution parameters (laser power, current, voltage, etc.)

[0079] The system's built-in quality control model compares and analyzes this sensor data in real time. If an anomaly is detected (such as excessive deposit thickness or unusual melt pool temperature fluctuations), the deposition process is paused, an alarm is issued, and the trajectory of the abnormal section is recorded. The repair path can then be recalculated to complete the repair loop.

[0080] After material deposition is complete, a digital image correlation (DIC) system and a 3D optical scanning system are used to perform a preliminary comparison of the repaired area's forming quality. If the scanned surface matches the original 3D design model within ±0.05mm, the repair is deemed qualified and proceeds to the subsequent surface microtexture control phase (corresponding to S600).

[0081] Through the closed-loop control strategy of path control - equipment linkage - real-time monitoring - quality feedback described in this embodiment, high-precision and high-consistency additive repair can be achieved in complex special-shaped surface areas, significantly improving the service reliability and structural consistency of the repaired workpiece.

[0082] In this embodiment, to ensure that the repaired complex irregular curved surface area meets the design requirements in terms of morphological accuracy and structural continuity, a quality control mechanism based on online detection and closed-loop error evaluation is proposed, which specifically includes the following steps:

[0083] After the additive manufacturing repair is complete, the repaired area is first scanned online using a structured light 3D reconstruction system. This system, which includes a multi-frequency grating projector and a binocular camera array, uses phase-shift encoding technology to acquire high-density point cloud data, achieving a point cloud accuracy better than 10μm and a spatial resolution exceeding 1,000 points per square millimeter.

[0084] The acquired scanned point cloud data is globally registered with the original design model. This registration process is based on the Iterative Closest Point (ICP) algorithm, incorporating normal angle constraints and curvature gradient penalties to enhance robustness on irregularly curved surfaces. After registration, an error reference coordinate system is established between the scanned coordinate system and the design model for subsequent error mapping and analysis.

[0085] In the error reference coordinate system, the geometric deviation between the reconstructed surface and the original design surface is calculated. The error expression adopts the spatial vector modeling method. For each point cloud node Pi, the normal vector offset Δn(Pi) from the nearest point Q on the target design surface is calculated, and the following error index is constructed:

[0086] Maximum normal deviation (Δn_max): The maximum value of all node normal deviations, reflecting the most serious deviation degree;

[0087] Curvature deviation (Δκ): Calculate the deviation distribution between the mean curvature κm and the target curvature κd on the surface, and extract its mean and standard deviation to characterize the degree of micromorphological change;

[0088] Inter-surface transition rate (τ): Analyzes the normal continuity and boundary gradient of the local surface splicing area, and calculates the inter-surface transition smoothness index by statistically analyzing the distribution of normal angles between adjacent path segments.

[0089] The above three error indicators together constitute the input feature vector E = [Δn_max, Δκ, τ] for quality judgment, which will serve as the input of the subsequent fuzzy evaluation model.

[0090] In order to realize the fuzzy correlation judgment between complex error indicators, an error evaluation model based on fuzzy decision tree is constructed, which is divided into the following sub-steps:

[0091] The maximum normal offset Δn_max, the curvature offset Δκ and the inter-surface transition rate τ are set as fuzzy input variables respectively. Three membership levels are defined for each variable: "low deviation", "medium deviation" and "high deviation". The corresponding membership function is a combination of trapezoidal function and Gaussian function to improve the transition smoothness of the boundary fuzzy interval.

[0092] For example, the fuzziness level of Δn_max is defined as follows:

[0093] Low deviation: 0-20μm, fully subordinate;

[0094] Medium deviation: 15-50 μm, partially affiliated;

[0095] High deviation: ≥45μm, high full membership.

[0096] The membership function parameters are trained and optimized through the deviation distribution of historical repair data to ensure high consistency between the input variables and the actual quality perception.

[0097] Based on the actual industrial repair case experience, at least 15 groups of fuzzy rules are constructed to cover typical error pattern combinations. Each rule has the following form:

[0098] Rule Example 1: If (Δn_max is high) and (Δκ is medium) and (τ is high), then the overall error level is "high";

[0099] Rule Example 2: If (Δn_max is medium) and (Δκ is low) and (τ is medium), then the overall error level is "medium";

[0100] Rule Example 3: If (Δn_max is low) and (Δκ is low) and (τ is low), then the overall error level is “low”.

[0101] A fuzzy inference engine is used to perform rule matching and membership aggregation on the input feature vector E. The output levels that meet the rules are fused using the weighted average method to calculate the comprehensive error level E_level∈[0,1], where a larger value indicates a more serious error.

[0102] The comprehensive error level E_level is input to the threshold function Θ(E_level), which is defined as follows:

[0103] If E_level≤0.4, the output judgment result is "qualified";

[0104] If E_level>0.4, the output judgment result is "exceeding the standard".

[0105] The threshold value of 0.4 is calculated based on the cluster centers of qualified and unqualified parts in historical samples and is statistically significant.

[0106] Ultimately, the judgment result output by the system will serve as the trigger condition for the closed-loop control logic and enter the next step of the path optimization process.

[0107] When the evaluation model outputs an "exceeding the standard" result, the control system automatically triggers the repair path reconstruction mechanism. This mechanism extracts the spatial location and geometric features of the abnormal area based on the current error point cloud data and updates the local morphological parameter set P.

[0108] The update process includes:

[0109] Convert the current error data into additional constraints for the local repair area;

[0110] Re-enter the morphology parameter extraction process in S200 and update the curvature analysis results;

[0111] Based on the newly generated P set, the path optimization process S300 is executed to obtain the second round of optimized paths.

[0112] Through this closed-loop control strategy, the system can autonomously adjust the repair strategy after actual processing errors occur, achieve regional precision compensation and process iterative optimization, and greatly improve the consistency and yield of repairing complex special-shaped surfaces.

[0113] After completing the additive manufacturing repair process for complex and irregularly shaped surfaces, in order to further improve the service performance (such as wear resistance and liquid adhesion resistance) of the repaired surface, this embodiment provides a surface optimization method based on micro-morphology reconstruction and texture control. The method includes the following steps:

[0114] First, an in-situ confocal interference microscopy system is used to perform microscopic scanning of the repaired target area. This system has nanometer-level vertical resolution (<10nm) and micrometer-level lateral resolution (<1μm), enabling the acquisition of high-precision surface topology maps without damaging the sample.

[0115] A dual-channel spatial-frequency domain conversion algorithm is used to perform a fast Fourier transform (FFT) on the acquired raw 3D topography data to extract frequency domain texture features. The converted spectrum data reflects key information such as surface roughness periodicity, directionality, and spatial texture density. Subsequently, a frequency domain texture feature matrix M_f is constructed, which contains the following sub-dimension indicators:

[0116] The center value and distribution width of the main frequency band (reflecting the periodicity);

[0117] anisotropy parameter (a measure of texture directionality);

[0118] The proportion of high-frequency energy (indicating the degree of microscopic disturbance);

[0119] Spatial frequency gradient tensor (captures texture transition characteristics).

[0120] The matrix M_f serves as the basic input for subsequent texture intelligent recognition and control strategies.

[0121] The frequency domain texture feature matrix M_f is input into the constructed multi-scale residual attention network model (MSRAN) to automatically extract micro-texture factors that are highly correlated with specific target functional performance (such as wear resistance and hydrophobicity).

[0122] The MSRAN model architecture includes multiple scale channels, each of which performs convolutional extraction on the response features of the texture map at different resolutions, and retains the original texture details through residual connections. An attention mechanism is introduced to achieve weighted recognition of texture regions, enabling the model to automatically focus on microstructural features that are significant under specific orientations, frequencies, or densities.

[0123] During the training phase, the model is supervised by using a dataset of functional performance labels. The labels include surface contact angle, friction coefficient, wetting speed, and other indicators, which are derived from standard tests (such as ASTM D1894, GB / T 4745, etc.). After training, the model outputs a set of texture factors T = {t1, t2, ..., tn} that are highly correlated with the target function, where each factor corresponds to a specific texture feature response, such as "a transverse ridge texture with a period of 30-50μm" or "a frequency of 10 4 -10 5 Hz range of isotropic texture perturbations"; n is the total number of texture factors.

[0124] In this way, semantic mapping of texture data to functional indicators can be achieved, providing an operational data basis for subsequent control strategies.

[0125] Based on the extracted texture factor set T, a micro-repair process parameter set is constructed to control the nanosecond laser processing system to perform orderly surface reconstruction. The parameter set includes:

[0126] Laser wavelength λ (commonly 1064nm or 532nm);

[0127] Single pulse energy E_p;

[0128] repetition frequency f_r;

[0129] Scan speed v_s;

[0130] Focus spot diameter d;

[0131] Path design modes (grid, spiral, adaptive function trajectory);

[0132] Overlap ratio r_ov.

[0133] The process parameters are determined by matching the texture factors in T with the texture result data in the historical laser processing library, selecting the optimal combination within the constraint space. For example, if the target texture factor is "transverse corrugation type, 30μm period", the corresponding process parameters may be: scanning speed 80mm / s, frequency 20kHz, spot diameter 15μm, and overlap ratio 80%.

[0134] The laser processing is performed by a multi-axis nanosecond laser system, supporting a five-axis linkage processing head, enabling high-precision micro-texturing on any free-form surface. A dynamic focus compensation algorithm embedded in the control system adjusts the focal length in real time to accommodate complex curvature variations, ensuring that each texture element is in the optimal processing focal plane.

[0135] To ensure control quality, the system integrates a synchronous confocal reflectance monitoring module, which measures the texture topography of the processed area in real time and compares it with the expected model. If the texture size deviation exceeds ±2μm, the system automatically adjusts subsequent laser parameters, forming a closed-loop control mechanism.

[0136] Example 2: To verify the beneficial effects of the "Additive Manufacturing Intelligent Repair and Surface Optimization Method for Complex Special-Shaped Surfaces" described in this invention, aircraft engine blade components were selected as test objects. Typical characteristics of such components include complex free-form surfaces, multi-axis transitions, and discontinuous edge curvature. Performance evaluation and comparative analysis were conducted using the following experimental steps:

[0137] Experimental setup

[0138]

[0139] The reconstructed point cloud of the repaired area is obtained by structured light scanning and compared with the original CAD model. The maximum normal error and global mean square error (RMSE) are statistically calculated as shown in the following table:

[0140] Method Maximum normal deviation (μm) Global RMSE (μm)

[0141] The present invention 28.4 12.6

[0142] Control group 73.8 34.5

[0143] The results show that the maximum error of the present invention is reduced by about 61.5%, and the overall repaired morphology is significantly closer to the original design.

[0144] ANSYS was used to perform thermal-mechanical coupling simulation to calculate the residual stress peak and stress gradient changes in the repaired area:

[0145] Method Peak residual stress (MPa) Stress gradient (MPa / mm)

[0146] Invention 221.4 54.2

[0147] Method Peak residual stress (MPa) Stress gradient (MPa / mm)

[0148] Control group 384.7 102.3

[0149] The results show that the peak residual stress is reduced by 42%, the stress distribution is more uniform, and the risk of stress concentration is effectively reduced.

[0150] The melt pool monitoring system was used to collect temperature fluctuations and deposition height changes, and a stability scoring function F(t) was constructed. The results are as follows:

[0151] Method F(t) (stability score, full score 100)

[0152] The present invention 91.6

[0153] Control group 67.3

[0154] Description: The path planning of the present invention improves deposition consistency and energy control stability, and reduces the probability of processing defects.

[0155] Use a contact angle meter to test the surface contact angle (hydrophobicity index) of the repaired area:

[0156] Method contact angle (°)

[0157] The present invention 127.5

[0158] Control group 86.3

[0159] Note: After multi-scale laser microtexturing, the surface acquires significant hydrophobic properties, which is of great significance to the service reliability under high-temperature condensation conditions.

[0160] In summary, this embodiment verified the multi-dimensional performance improvement of the present invention compared with the traditional method through physical repair, process monitoring, simulation analysis and surface performance testing, specifically: repair accuracy was improved by more than 60%; residual stress was reduced by more than 40%; dynamic stability score was improved by about 36%; and surface functionality was significantly enhanced.

[0161] The above data fully demonstrate that the present invention has significant technical effects in structural restoration, performance regulation and intelligent control, and solves the problems of rigidity of existing repair paths, residual stress concentration and difficulty in regulating functional surfaces.

[0162] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. An intelligent repair and surface optimization method for additive manufacturing of complex and irregularly shaped surfaces, characterized by: include: S100, obtaining a three-dimensional digital model of a workpiece to be repaired, and identifying a complex irregular curved surface area with defects on the surface thereof, wherein the area has discontinuous curvature or multi-axis free-form features; S200, performing curvature analysis and morphological feature extraction on the irregular curved surface area to generate a corresponding local morphological parameter set P; S300, based on the parameter set P, constructing an additive manufacturing repair path generation model for special-shaped curved surfaces, and performing multi-objective optimization of the path using a machine learning algorithm, with the objectives including maximizing forming accuracy, minimizing residual stress, and improving cladding stability; S400: Based on the optimized repair path, the additive manufacturing equipment is controlled to deposit materials to complete the precise repair of the defective area. S500, performing online inspection and error evaluation on the repaired surface. If the error exceeds a preset threshold, returning to S200 to re-optimize the repair path; S600: After the repair is completed, the surface of the repaired area is microscopically reconstructed and the texture is regulated.

2. The method for intelligent repair and surface optimization of complex and irregularly shaped surfaces using additive manufacturing according to claim 1, characterized in that: The S200 includes: S201, generating a local parameterized surface model based on the three-dimensional point cloud data of the irregular curved surface area by using an adaptive high-order fitting surface reconstruction method; S202, jointly extracting multi-scale discrete Gaussian curvature and mean curvature from the parameterized surface model to construct a curvature distribution matrix; S203: Utilize a multi-channel shape and texture coding network to fuse the curvature distribution matrix with the surface gradient texture image, generate a local shape description vector, and construct a local shape parameter set P.

3. The method for intelligent repair and surface optimization of complex and irregularly shaped surfaces using additive manufacturing according to claim 1, characterized in that: The S300 includes: S301, constructing a repair path generation model based on a graph neural network, mapping the local morphology parameter set P into a graph structure input; S302: Perform feature aggregation and edge weight optimization on the graph structure input, combine historical repair data labels, and train to generate a path candidate set with path feasibility prediction capabilities; S303, introducing a joint objective function, integrating the forming accuracy evaluation index, the residual stress distribution prediction results based on thermal-mechanical coupling simulation, and the dynamic stability score of the material deposition process, to perform multi-objective optimization on the path candidate set; S304: Use a reinforcement learning algorithm to update the optimization results and output a global optimal additive path sequence.

4. The method for intelligent repair and surface optimization of complex and irregularly shaped surfaces using additive manufacturing according to claim 3, characterized in that: The fusion multi-objective optimization parameters include: S3031. Mapping the additive path candidate set to the parametric component model, and using the scanning contour error superposition algorithm to calculate the forming accuracy evaluation index of each path; S3032. Based on the path geometry and process parameter input, a thermal-mechanical coupling finite element model is constructed to simulate and calculate the instantaneous temperature field and stress field of each path during the deposition process, and the residual stress peak value and its spatial distribution coefficient are extracted as path evaluation factors; S3033: Collect the deposition rate changes, molten pool temperature fluctuations and interlayer overlap quality driven by the path, use the stability scoring function F(t), and output the path dynamic stability score; S3034. The forming accuracy evaluation index, residual stress peak value, its spatial distribution coefficient and path dynamic stability score are standardized and weightedly integrated into the path objective function.

5. The method for intelligent repair and surface optimization of complex and irregularly shaped surfaces using additive manufacturing according to claim 1, characterized in that: The S500 includes: S501. Use a structured light 3D reconstruction system to scan the repair area online, obtain high-precision surface point cloud data, and match it to the original design model in real time to establish an error comparison coordinate system; S502, performing spatial vectorization on the deviation between the reconstructed surface and the original design surface, and calculating the maximum normal offset, curvature offset, and inter-surface transition rate; S503: Construct an error assessment model based on a fuzzy decision tree, perform fuzzy fusion judgment on the maximum normal offset, curvature offset, and inter-surface transition rate, and output a binary judgment result of whether the error exceeds a preset error threshold; S504: If the error determination result is out of standard, the process automatically triggers a return to S200, and updates the local morphology parameter set P based on the current error distribution.

6. The method for intelligent repair and surface optimization of complex and irregularly shaped surfaces using additive manufacturing according to claim 5, characterized in that: Constructing an error assessment model based on fuzzy decision tree includes: S5031. Taking the maximum normal deviation, curvature deviation, and inter-surface transition rate as input variables, respectively, define the fuzzy membership function of each variable and divide it into three levels: low deviation, medium deviation, and high deviation; S5032. Build a fuzzy rule base, use a fuzzy inference engine to match the input error indicators, and calculate the comprehensive error level through a weighted average membership aggregation method; S5033. Convert the comprehensive error level into a binary judgment result through a threshold function. If the level is higher than a preset threshold, output an exceeding-standard signal.

7. The method for intelligent repair and surface optimization of complex and irregularly shaped surfaces using additive manufacturing according to claim 1, characterized in that: The S600 includes: S601, obtaining a microscopic topography map of the repaired area and constructing its frequency domain texture feature matrix; S602: Input the texture feature matrix into a multi-scale residual attention network model to extract texture factors associated with target wear resistance or hydrophobicity; S603. Generate a micro-repair process parameter set based on the extracted factors to control the nanosecond laser scanning path and energy distribution.

Images