A rotor coating additive method based on amorphous nanocrystalline composite

Abstract

The application discloses a kind of based on amorphous nanocrystalline composite rotor coating additive method, belong to material preparation field, the additive method specific steps are as follows: I, the rotor surface is treated with multiple scale topological structure, and construct three-level bionic transition layer on the rotor surface after pretreatment;II, before coating formal deposition, construct rotor virtual simulation model, and forecast optimization is carried out to temperature field, stress distribution and organization evolution;The application significantly improves the bonding strength and anti-peeling capacity of coating and matrix, realizes strong interface and high toughness, meets the wear resistance and fatigue resistance requirement under complex working condition, effectively considers efficiency, quality and cost, significantly improves the flexibility and intelligent level of method, ensures that deposition quality is stable, realizes traceable, reproducible manufacturing process, greatly shortens development cycle, reduces test cost, improves the research and development efficiency of new coating system.

Description

Technical Field

[0001] This invention relates to the field of materials preparation, and in particular to an additive manufacturing method for rotor coating based on amorphous nanocrystalline composites. Background Technology

[0002] With the increasing demands on rotor component performance in aerospace, gas turbine, nuclear energy, and high-end manufacturing, traditional coating manufacturing processes (such as plasma spraying, sputtering deposition, or hot-dip galvanizing) have shown significant limitations in handling high-temperature, high-stress, and complex curved surface deposition. Conventional coatings are prone to problems such as insufficient interfacial adhesion, coarsening of coating structure, residual stress concentration, and limited functionality, making it difficult to meet the stable service requirements of rotors under high-speed and thermal shock conditions. In recent years, amorphous-nanocrystalline composite coatings have become a development direction for next-generation key functional protective materials due to their combination of high hardness, high toughness, excellent corrosion resistance, and oxidation resistance. However, achieving high-quality deposition, controllable evolution of microstructure, and synergistic optimization of multiple properties in such composite coatings on complex rotor surfaces still faces challenges such as high process complexity, narrow parameter windows, and difficulties in quality control. Therefore, inventing an additive manufacturing method for rotor coatings based on amorphous-nanocrystalline composites has become particularly important.

[0003] Existing rotor coating additive methods based on amorphous and nanocrystalline composites exhibit poor bonding strength and peel resistance between the coating and the substrate, failing to achieve a balance between strong interface and high toughness. Furthermore, they cannot meet the wear resistance and fatigue resistance requirements under complex working conditions, and the stability of deposition quality cannot be guaranteed. This increases development cycles, experimental costs, and reduces the efficiency of developing novel coating systems. Therefore, we propose a rotor coating additive method based on amorphous and nanocrystalline composites. Summary of the Invention

[0004] The purpose of this invention is to address the deficiencies in the prior art by proposing a rotor coating additive method based on amorphous nanocrystalline composites.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: An additive manufacturing method for rotor coating based on amorphous nanocrystalline composites, the specific steps of which are as follows: I. Perform multi-scale topological pretreatment on the rotor surface, and construct a three-level biomimetic transition layer on the pretreated rotor surface; II. Before the formal deposition of the coating, a virtual simulation model of the rotor is constructed to predict and optimize the temperature field, stress distribution and microstructure evolution; III. Based on the rotor shape and simulation optimization results, multi-axis dynamic path planning is performed, and the attitude and trajectory of the deposition head are intelligently adjusted through a multi-axis linkage control system; IV. During the coating deposition process, the frequency and intensity of the dynamically modulated energy field induce in-situ recombination of amorphous phase and nanocrystal at the moment of droplet solidification, forming a dense composite structure; V. Real-time monitoring of temperature field, deposition rate and stress changes in the coating area, and automatic adjustment of deposition power, feed rate and scanning speed through closed-loop feedback; VI. During the deposition process, the ratio of amorphous to nanocrystalline materials is adjusted layer by layer. At the same time, the composition and pressure of the cavity atmosphere are adjusted in real time to control the oxygen content and the ratio of inert gas. VII. Analyze the morphology and performance of different regions of the rotor, and automatically select and switch the optimal deposition mode based on the analysis results.

[0006] As a further aspect of the present invention, the specific steps of performing multi-scale topological preprocessing on the rotor surface in step I, and constructing a three-level biomimetic transition layer on the preprocessed rotor surface, are as follows: S1.1: Determine the overall layout of the grooves according to the rotor stress or coating bearing requirements, and select the groove depth range and equal or variable spacing layout. Then, define the start and end points, depth curves and cross-sectional shapes of each groove on the rotor surface in the local coordinate system using 3D CAD. S1.2: Perform finite element local verification on each groove on the rotor surface. If the verification exceeds the limit, adjust the width or curvature of the corresponding groove, output the manufacturing executable tool or laser path and generate a processing process card. The processing process card specifically includes the processing sequence, processing power and cooling requirements. S1.3: Pre-divide the texture unit array area on the surface of the trench or slope, and determine the texture period, depth and orientation. Select the ultrashort pulse laser parameter set according to the required texture scale, including pulse width, frequency, energy and focal position. Then, adjust the energy density through trial cutting or simulation to simulate single hole or single texture processing, and establish the corresponding design template according to the processing results. S1.4: A line-by-line or point-by-point etching strategy is adopted to control the overlap rate and pulse count. After the machining is completed, the rotor surface is scanned and compared with the design template. If the deviation between the machining result and the design template exceeds the preset value, secondary fine-tuning is performed to determine the target function of the three-level bionic transition layer. Then, based on the determined target function, a layered model is established according to the thickness of the bionic transition layer to determine the target component or phase ratio of each layer. Then, according to the feasibility of the process, the continuous gradient is discretized into finite layers through the continuous component gradient function to obtain the relative content of the target main phase at different depths from the matrix interface in the transition layer. S1.5: Discretize the relative content of the target main phase into multiple layers, and specify the material ratio, powder feeding rate and energy input for each layer. Then, establish a parameter mapping table to record the powder feeding ratio, spraying speed, energy density and residence time of different layers. Then, based on the preset small area sample, conduct trial laying to verify the interlayer bonding force and structure, and adjust the powder feeding ratio and energy in real time, while recording the self-consistent parameter set. S1.6: Set the energy input strategy for each deposition layer, then verify the real-time temperature field by temperature measurement or thermocouple array, passively or actively cool each layer, adjust the energy density according to the test results, and record the interlayer temperature history as the basis for subsequent quality tracking and process reproduction. After the biomimetic transition layer is completed, use optical profile or white light interferometry to measure the layer thickness and surface roughness. S1.7: Compare the deviation between the design value and the measured value, and perform real-time compensation according to the type of deviation. After the entire biomimetic transition layer is deposited, perform corresponding mechanical trimming and output the final test report as a quality control record. The test report specifically includes the layer thickness distribution, surface roughness and local deviations.

[0007] As a further aspect of the present invention, the topography scanning described in S1.4 specifically includes white light contouring and SEM; The target function of the three-level biomimetic transition layer described in S1.4 is to transition from the base metal to the amorphous / nanocrystalline main coating, while taking into account chemical compatibility, CTE (coefficient of thermal expansion) matching and hardness transition. The deviation types mentioned in S1.6 specifically include excessive thickness, insufficient thickness, and local unevenness.

[0008] As a further aspect of the present invention, the specific steps of constructing the rotor virtual simulation model in step II to predict and optimize the temperature field, stress distribution, and microstructure evolution are as follows: S2.1: Import the rotor shape through CAD drawings and scanned data, and remove noise data in the imported data by filling holes and merging patches. At the same time, maintain high-resolution point clouds for key surfaces such as grooves and edges, define local coordinate system and region of interest (ROI), implement local mesh refinement strategy for ROI, and use coarse mesh for other areas. S2.2: After the mesh type is generated, select hexahedron or modified tetrahedron as the element type, and then adaptively reduce the element size according to the curvature and the estimated thermal gradient. Check the minimum inscribed sphere radius of each mesh. If the minimum inscribed sphere radius exceeds the preset threshold, mark the corresponding mesh as inferior and correct it. Then output the corrected final mesh. S2.3: Determine the coupled physical field according to the simulation target, establish the corresponding rotor virtual simulation model, select the constitutive relation for each material, and prepare experimental calibration data. Then, adopt the phase field model form for phase transformation or microstructure evolution, determine various dynamic parameters, define the heat input area, convection boundary, contact and constraint boundary conditions according to the process scenario, and map the energy to the grid in the form of time-space distribution function. S2.4: Arrange virtual sensor points and sampling frequencies in the virtual simulation model, formulate initial conditions for the virtual experiment, prepare multiple sets of different excitation spectra, set the time step, nonlinear solution strategy and iterative convergence criteria for the simulation experiment, estimate resources for a single simulation experiment, construct a parameter scanning matrix, run batch simulations of experiments and record temperature, stress, phase distribution time series and key section data, extract the extreme values, gradient fields, cooling rate curves and microstructure phase fraction distribution with depth of each simulation experiment from the simulation results to establish a structured dataset; S2.5: Based on the spatial-temporal proxy model structure of convolutional network, an evolution prediction model is established, and the input features and input targets of the evolution prediction model are set. Then, the structured dataset is divided into training set, validation set and test set, and feature scaling and spatiotemporal dimensionality reduction are performed. The training set is then input into the evolution prediction model. The evolution prediction model processes each data in the training set layer by layer based on the forward propagation algorithm and outputs the predicted evolution results at different times. S2.6: Calculate the loss value between the predicted evolution result and the actual evolution result through the regularized mean square error loss function. If the loss value is higher than the preset threshold, update the parameters of the evolution prediction model through the Adam optimizer and recalculate the corresponding loss value using the training set until the loss value converges to within the preset threshold. Then, evaluate the performance of the evolution prediction model using the validation set and test the prediction accuracy of the evolution prediction model on unknown data using the test set. S2.7: If the performance and prediction accuracy of the evolution prediction model reach the preset expected value, the training ends; otherwise, the evolution prediction model is retrained and optimized. The latest simulation data generated by the rotor virtual simulation model is input into the evolution prediction model. Then, the evolution prediction model processes the received simulation data layer by layer and outputs the temperature field, stress distribution and tissue evolution state within the preset time interval in the future. S2.8: Based on the temperature field, stress distribution and microstructure evolution state within the obtained future preset time interval, if the evolution state exceeds the preset threshold, multiple sets of candidate process parameters are output through Bayesian optimization. Then, each set of candidate process parameters is fed back to the rotor virtual simulation model for back-checking, and the set of candidate process parameters with the lowest evolution state change is selected as the optimal set of process parameters, and the original process parameters are replaced.

[0009] As a further aspect of the present invention, the specific steps of the multi-axis dynamic path planning in step III are as follows: S3.1: According to the process requirements, deploy laser scanning or industrial camera arrays in the actual processing scene, and set the sampling frequency and field of view coverage strategy. Continuously acquire original point cloud frames within the set sampling period, then use fast feature to perform coarse registration on point clouds from different sensors or different times, and then use the ICP algorithm to perform local fine registration to generate a rigid transformation matrix and merge it into a frame of global point cloud. S3.2: Statistically identify and remove outliers in the global point cloud, retain high-density points in key local areas as needed, and perform voxel lattice downsampling on the remaining areas. Then, perform surface reconstruction processing on the processed global point cloud to generate a point cloud with normal estimation and output the corresponding coverage. If the coverage is lower than a preset threshold, perform regional supplementary sampling. Select the neighborhood point set of each deposition trajectory point based on a preset radius, and then perform sampling based on the selected neighborhood point set. S3.3: Fit the corresponding local plane using the least squares method, calculate the unit normal vector of each local plane and unify its direction, then calculate the corresponding principal curvature and average curvature based on the covariance matrix of each local plane, and mark the region with principal curvature higher than the preset threshold as a high curvature region. Then identify and classify the geometric feature points of rotor boundary, acute angle, and groove root, and provide labels for attitude constraints and path refinement in different regions to establish a complete local reference system and curvature index matrix. S3.4: Based on the established local reference frame, a sequence of solution points is generated on the rotor surface according to the design layer thickness and desired line width. Each point includes position, desired normal and tangential. The IK solver is then used to calculate the joint angle vector corresponding to each trajectory point of the multi-axis linkage control system and record the multiple solutions. Collision, limit and speed limit detection are then performed on each joint solution. Based on the detection results, local trajectory rearrangement is performed on unreachable or conflict points. S3.5: Generate a preliminary draft of the joint-level trajectory for the multi-axis linkage control system based on the rearrangement results. Divide the preliminary draft of the joint-level trajectory into multiple segments according to physical accessibility, geometric features and task priority, and generate time parameters for each segment. When a collision warning occurs, perform local replanning on the affected segment, generate the corresponding joint time series, and replace the original trajectory with the replanned new trajectory.

[0010] As a further aspect of the present invention, the specific steps for adjusting the ratio of amorphous to nanocrystalline materials layer by layer in step VI are as follows: S4.1: Set the target value range for hardness, toughness, and wear resistance for the outer and inner layers of the coating, and determine the total coating thickness. In the form of a distribution function, set the target component or phase ratio profile within the total coating thickness range. At the same time, map the coating performance requirements to the target curve of component value or phase fraction. Then, quantify the continuous target curve and save it as process input. Record special functional areas and mark the priority and minimum layer thickness constraints. S4.2: Select the lower limit of single-layer thickness based on the equipment's feeding and process accuracy capabilities, and use it as the layer resolution of the corresponding coating. Discretize the total thickness into multiple layers according to the layer resolution of different coatings, calculate the center depth of each layer, and extract the target component value at that location from the target profile. Based on the target component value, layer thickness, priority, and subsequent feeding channel requirements of each layer, establish the corresponding process layer table. S4.3: Count the number of component channels and define the physical properties and unit powder volume mass ratio table for each channel. Decompose the target components of different coatings to obtain the component vector required for each channel. Set the total powder feeding rate requirement for different coatings and calculate the initial powder feeding rate of each channel. Then write the initial powder feeding rate of each channel into the feeding time series table and synchronize it with the switching sequence of the deposition head and the interlayer cooling window. S4.4: Online component sensors are placed at the deposition head or close to the deposition point, and the proxy index values ​​of the current mixture or deposition product are collected periodically. The measured proxy index values ​​are mapped to the actual component estimates and compared with the target mass fraction. If the deviation exceeds the allowable error, the feeding correction is triggered. At the same time, a smooth gain correction law is used to gradually adjust the powder feeding rate of each channel, and corresponding saturation and rate limits are applied during the correction process. The closed-loop correction log of each layer is recorded in real time.

[0011] As a further embodiment of the present invention, the special functional area described in S4.1 includes a stress buffer layer near the substrate, a surface hardening layer, etc.

[0012] As a further aspect of the present invention, the specific steps for automatically selecting and switching the optimal deposition mode in step VII are as follows: S5.1: Divide the three-dimensional surface of the rotor into multiple groups of regional units according to function and geometric features, and establish a unique identifier for each regional unit. Obtain the geometric feature vector corresponding to each regional unit by collecting data and point cloud data of each region, and define the working condition drive performance requirements of each region. S5.2: Based on the geometric characteristics and performance requirements of each regional unit, establish corresponding demand vectors and store them in the regional demand library, mark the priority and tolerance zone, then evaluate the processing accessibility of each regional unit and record it as a constraint condition, and then establish a deposition mode library, in which each mode includes mode ID, mode type label, controllable parameter range, typical energy input and process constraints. S5.3: Obtain the capability vector of each model in the sedimentation model library in multiple sets of simulation experiments, transform different capability vectors into vectors of the same type, standardize the capability vector of each model, and establish a model-region feasibility matrix based on the preprocessed capability vectors of each model. S5.4: Select a set of candidate models from the deposition model library that meet the feasibility matrix of each region and do not violate hard constraints. Set multiple sets of performance criteria and assign engineering weights to each criterion according to process priority. Calculate the aggregation score of each candidate model with each region. Then sort the candidate models from high to low score and output the top five as switching candidates. At the same time, record the estimated switching cost between models. S5.5: Based on the aggregated score results, formulate a pattern assignment sequence for the entire rotor surface, design a transition strategy for each proposed switching point, evaluate the feasibility of the switching plan in the time dimension and align it with the production cycle constraints. If it is not feasible, fall back to the second-best score but lower switching cost solution, and generate an executable switching instruction sequence.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This rotor coating additive manufacturing method based on amorphous and nanocrystalline composites parametrically arranges macroscopic grooves according to stress and load requirements, and etches submicroscopic textures on their surfaces using ultrashort pulse lasers to generate a processing card. Subsequently, a continuous composition profile of a three-level biomimetic transition layer is delineated according to the designed thickness and discretized into multiple layers. A mapping table for the material ratio, powder feeding rate, and energy of each layer is established. The method is first validated using small-scale sampling, and energy / feeding and cooling strategies are adjusted. Temperature history and quality control data are recorded. High-fidelity meshes are constructed in parallel using CAD / point cloud, and physical constitutive and phase transition models are set. Batch simulations generate datasets, and a spatiotemporal surrogate model is trained to predict temperature-stress-microstructure evolution. Bayesian optimization is used to review candidate processes. Finally, multi-view point cloud real-time registration generates trajectory points with normals for on-site testing. Kinematic reachability and collision verification are implemented to generate multi-axis joint trajectories and replan in real time. During deposition, multi-channel feeding commands are issued layer by layer and corrected using online component sensing closed loop. Regional requirements are mapped to a pattern library. Multi-pattern fusion sequences are planned based on multi-criteria scoring and switching costs. Smooth transition and online verification / degradation strategies are designed to form a traceable manufacturing and quality control closed loop. This significantly improves the bonding strength and anti-peeling ability between the coating and the substrate, achieving a balance between strong interface and high toughness. It meets the wear resistance and fatigue resistance requirements under complex working conditions, effectively balancing efficiency, quality, and cost. This significantly improves the flexibility and intelligence of the method, ensures stable deposition quality, and enables a traceable and reproducible manufacturing process. It greatly shortens the development cycle, reduces experimental costs, and improves the R&D efficiency of new coating systems. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0015] Figure 1 This is a flowchart of an additive manufacturing method for rotor coating based on amorphous nanocrystalline composites proposed in this invention. Detailed Implementation

[0016] Example

[0017] Reference Figure 1 An additive manufacturing method for rotor coating based on amorphous nanocrystalline composites is disclosed, the specific steps of which are as follows: The rotor surface is pre-processed with a multi-scale topology, and a three-level biomimetic transition layer is constructed on the pre-processed rotor surface.

[0018] Specifically, the overall layout of the grooves is determined based on the rotor's stress or coating load-bearing requirements. The groove depth range and the spacing (equal or variable) are selected. Then, using 3D CAD, the start and end points, depth curves, and cross-sectional shapes of each groove on the rotor surface are defined in a local coordinate system. Finite element local verification is performed on each groove on the rotor surface. If the verification exceeds the limits, the width or curvature of the corresponding groove is adjusted. A manufacturing-executable tool or laser path is output, and a machining process card is generated. The machining process card includes the machining sequence, machining power, and cooling requirements. Texture unit array regions are pre-divided on the groove surface or slope, and the texture is determined. The period, depth, and orientation are determined by selecting an ultrashort pulse laser parameter set, including pulse width, frequency, energy, and focal position, based on the required texture scale. Then, the energy density is adjusted through trial cutting or simulation to model single-hole or single-texture processing. A corresponding design template is established based on the processing results. A row-by-row or point-by-point etching strategy is employed to control the overlap rate and pulse count. After processing, the rotor surface morphology is scanned and compared with the design template. If the deviation between the processing result and the design template exceeds a preset value, secondary fine-tuning is performed to determine the target function of the three-level biomimetic transition layer. Subsequently, based on the determined target function, layers are established according to the thickness of the biomimetic transition layer. The model determines the target composition or phase ratio for each layer. Then, based on process feasibility, the continuous gradient is discretized into finite layers using a continuous composition gradient function. The relative content of the target main phase at different depths from the matrix interface within the transition layer is obtained. The relative content of the target main phase is discretized into multiple layers, and material ratios, powder feeding rates, and energy inputs are specified for each layer. A parameter mapping table is then established to record the powder feeding ratio, spray rate, energy density, and residence time for different layers. Trial deployment is then conducted based on a pre-set small-area sample to verify interlayer bonding and microstructure, while the powder feeding ratio and energy are adjusted in real time. Simultaneously, a self-consistent parameter set is recorded, and the deposition parameters for each layer are set. The energy input strategy is followed by verification of the real-time temperature field through temperature measurement or thermocouple arrays. Each layer is passively or actively cooled, and the energy density is adjusted according to the experimental results. At the same time, the interlayer temperature history is recorded as a basis for subsequent quality tracking and process reproduction. After the biomimetic transition layer is completed, the layer thickness and surface roughness are measured using optical profilometry or white light interferometry. The deviation between the design value and the measured value is compared, and real-time compensation is performed according to the type of deviation. After the entire biomimetic transition layer is deposited, the corresponding mechanical trimming is performed, and the final test report is output as a quality control record. The test report specifically includes the layer thickness distribution, surface roughness, and local deviations.

[0019] It should be further explained that the morphology scanning specifically includes white light contouring and SEM; the target function of the three-level biomimetic transition layer is to transition from the base metal to the amorphous / nanocrystalline main coating, which needs to take into account chemical compatibility, CTE (coefficient of thermal expansion) matching and hardness transition; the deviation types specifically include ultra-thickness, thinness and local concavity / unevenness.

[0020] Before the coating is formally deposited, a virtual simulation model of the rotor is constructed to predict and optimize the temperature field, stress distribution and microstructure evolution.

[0021] Specifically, the rotor shape is imported from CAD drawings and scanned data. Noise in the imported data is removed by filling holes and merging surfaces. High-resolution point clouds are maintained for key surfaces such as grooves and edges. A local coordinate system and Region of Interest (ROI) are defined, and a local mesh refinement strategy is applied to the ROI, while a coarse mesh is used for other areas. After mesh generation, hexahedrons or modified tetrahedrons are selected as the element type. The element size is then adaptively reduced based on curvature and predicted thermal gradient. The minimum inscribed sphere radius of each mesh is checked. If the minimum inscribed sphere radius exceeds a preset threshold, the corresponding mesh is marked as inferior and corrected. The corrected final mesh is then output. The coupled physical field is determined according to the simulation objective, and a corresponding virtual rotor is established. A simulation model is created, constitutive relations are selected for each material, and experimental calibration data is prepared. Then, a phase-field model is used to assess phase transitions or microstructure evolution, while various dynamic parameters are determined. Based on the process scenario, heat input regions, convection boundaries, and contact and constraint boundary conditions are defined. Energy is mapped to the mesh in the form of a time-space distribution function. Virtual sensor points and sampling frequencies are arranged in the virtual simulation model. Initial conditions are then established for the virtual experiment, and multiple sets of different excitation spectra are prepared. The time step, nonlinear solution strategy, and iterative convergence criteria for the simulation experiment are set. Resource estimation is then performed on individual simulation experiments. Subsequently, a parameter scanning matrix is ​​constructed, and batch simulations are run, recording temperature, stress, phase distribution time series, and key section data. The simulation results extract extreme values, gradient fields, cooling rate curves, and tissue phase fraction distributions with depth from each simulation experiment to establish a structured dataset. Based on the spatial-temporal surrogate model structure of a convolutional network, an evolution prediction model is established, and the input features and input targets of the evolution prediction model are set. The structured dataset is then divided into training, validation, and test sets, and feature scaling and spatiotemporal dimensionality reduction are performed. The training set is then input into the evolution prediction model, which processes each data point in the training set layer by layer using the forward propagation algorithm and outputs the predicted evolution results at different time points. The loss value between the predicted evolution result and the actual evolution result is calculated using a regularized mean squared error loss function. If the loss value exceeds a preset threshold, the Adam algorithm is used to determine the outcome. The optimizer updates the parameters of the evolution prediction model and recalculates the corresponding loss values ​​using the training set until the loss values ​​converge to a pre-threshold. Then, the performance of the evolution prediction model is evaluated using the validation set, and the prediction accuracy on unknown data is tested using the test set. If the performance and prediction accuracy of the evolution prediction model meet the preset expectations, training ends; otherwise, the evolution prediction model is retrained and optimized. The latest simulation data generated by the rotor virtual simulation model is input into the evolution prediction model. The evolution prediction model then processes the received simulation data layer by layer and outputs the temperature field, stress distribution, and tissue evolution state within a preset future time interval. Based on the obtained temperature field, stress distribution, and tissue evolution state within the preset future time interval...If the evolution state exceeds a preset threshold, Bayesian optimization is used to output multiple sets of candidate process parameters. These sets are then fed back into the rotor virtual simulation model for review. The candidate set with the lowest evolution state change is selected as the optimal set and replaces the original process parameters.

[0022] Based on the rotor shape and simulation optimization results, multi-axis dynamic path planning is performed, and the attitude and trajectory of the deposition head are intelligently adjusted through a multi-axis linkage control system.

[0023] Specifically, according to process requirements, laser scanning or industrial camera arrays are deployed in the actual processing scenario, and sampling frequency and field-of-view coverage strategies are set. Raw point cloud frames are continuously acquired within the set sampling period. Fast feature mapping is then used to perform coarse registration of point clouds from different sensors or different times. Next, the ICP algorithm is used for local fine registration to generate a rigid transformation matrix, which is then merged into a single global point cloud frame. Outliers in the global point cloud are statistically analyzed and removed. High-density points in key local areas are retained as needed, and voxel lattice downsampling is performed on the remaining areas. Afterward, surface reconstruction processing is performed on the processed global point cloud to generate a point cloud with normal estimation, and the corresponding coverage rate is output. If the coverage rate is lower than a preset threshold, regional supplementary sampling is performed. Neighborhood point sets for each deposition trajectory point are selected based on a preset radius. Based on the selected neighborhood point sets, the corresponding local planes are fitted using the least squares method. Simultaneously, the unit normal vectors of each local plane are calculated and their directions are unified. Then, based on the covariance matrix of each local plane, the corresponding principal curvature and mean curvature are calculated. Points with principal curvature higher than a preset value are selected. High curvature regions are marked with threshold values. Then, geometric feature points at rotor boundaries, acute angles, and groove roots are identified and classified. Labels are provided for attitude constraints and path refinement in different regions to establish a complete local reference system and curvature index matrix. Based on the established local reference system, a sequence of solution points is generated on the rotor surface according to the design layer thickness and desired linewidth. Each point includes position, desired normal, and tangential. The IK solver is then used to calculate the joint angle vector corresponding to each trajectory point of the multi-axis linkage control system and record multiple solutions. Collision, limit, and speed limit detection are then performed on each joint solution. Based on the detection results, local trajectory rearrangement is performed on unreachable or conflict points. Based on the rearrangement results, a preliminary draft of the joint-level trajectory of the multi-axis linkage control system is generated. According to physical accessibility, geometric features, and task priority, the preliminary draft of the joint-level trajectory is divided into multiple segments, and time parameters for each segment are generated. When a collision warning occurs, local replanning is performed on the affected segment, and the corresponding joint time sequence is generated. The original trajectory is replaced with the replanned new trajectory. Example

[0024] Reference Figure 1 An additive manufacturing method for rotor coating based on amorphous nanocrystalline composites is described, and the specific steps of the additive manufacturing method are as follows: During the coating deposition process, the frequency and intensity of the dynamically modulated energy field induce in-situ recombination of amorphous phase and nanocrystals at the moment of droplet solidification, forming a dense composite structure.

[0025] The temperature field, deposition rate, and stress changes in the coating area are monitored in real time, and the deposition power, feed rate, and scanning speed are automatically adjusted through closed-loop feedback.

[0026] During the deposition process, the ratio of amorphous to nanocrystalline materials is adjusted layer by layer. At the same time, the composition and pressure of the cavity atmosphere are adjusted in real time to control the oxygen content and the ratio of inert gas.

[0027] Specifically, target value ranges for hardness, toughness, and wear resistance are set for the outer and inner layers of the coating, and the total coating thickness is determined. Target component or phase ratio profiles are set within the total coating thickness range using a distribution function. Coating performance requirements are mapped to target curves for component values ​​or phase fractions. Continuous target curves are then numerically processed and saved as process input. Special functional areas are recorded and their priorities and minimum layer thickness constraints are marked. A lower limit for single-layer thickness is selected based on equipment feeding and process accuracy capabilities, and this is used as the layer resolution for the corresponding coating. The total thickness is discretized into multiple layers based on the layer resolution of different coatings, and the center depth of each layer is calculated. The target component value at that location is extracted from the target profile. Based on the target component value, layer thickness, priority, and subsequent feeding channel requirements for each layer, a corresponding process layer table is established, and the number of component channels is counted. A table of physical properties and unit powder volume mass ratios was defined for each channel. The target components of different coatings were decomposed to obtain the required component vectors for each channel. The total powder feeding rate requirements for different coatings were set, and the initial powder feeding rate for each channel was calculated. The initial powder feeding rate of each channel was then written into the feeding time series table and synchronized with the switching sequence of the deposition head and the interlayer cooling window. Online component sensors were placed at the deposition head or close to the deposition point, and surrogate index values ​​of the current mixture or deposition product were collected periodically. The measured surrogate index values ​​were mapped to the actual component estimates and compared with the target mass fraction. If the deviation exceeded the allowable error, the feeding correction was triggered. At the same time, a smooth gain correction law was used to gradually adjust the powder feeding rate of each channel, and corresponding saturation and rate limits were applied during the correction process. The closed-loop correction log of each layer was recorded in real time.

[0028] It should be further explained that the special functional areas include stress buffer layers near the substrate and surface hardening layers.

[0029] The morphology and performance of different regions of the rotor are analyzed, and the optimal deposition mode is automatically selected and switched based on the analysis results.

[0030] Specifically, the three-dimensional surface of the rotor is divided into multiple regions based on function and geometric features, and a unique identifier is established for each region. Geometric feature vectors corresponding to each region are obtained through point cloud acquisition. Simultaneously, the driving performance requirements for each region are defined, and corresponding requirement vectors are established based on the geometric features and performance requirements of each region. These vectors are then stored in a region requirement library, with priority and tolerance bands marked. The processability of each region is then evaluated and recorded as constraints. A deposition model library is then established, where each model includes a model ID, model type label, controllable parameter range, typical energy input, and process constraints. The capability vectors of each model in the deposition model library are obtained from multiple sets of simulation experiments, and different capability vectors are converted into vectors of the same type. Each model capability vector is then further processed. Standardization is performed, and a mode-region feasibility matrix is ​​established based on the preprocessed model capability vectors. A set of candidate modes that meet the feasibility matrix of each region and do not violate hard constraints are selected from the deposition mode library. Multiple performance criteria are set, and engineering weights are assigned to each criterion according to process priority. The aggregate score of each candidate mode and each region is calculated. Then, the candidate modes are sorted from high to low score, and the top five are output as switching candidates. At the same time, the switching cost estimate between modes is recorded. Based on the aggregate score results, a mode allocation sequence is formulated for the entire rotor surface. At the same time, a transition strategy is designed for each proposed switching point. The feasibility of the switching plan in the time dimension is evaluated and aligned with the production cycle constraints. If it is not feasible, it is backed to the second-best solution with a lower switching cost. At the same time, an executable switching instruction sequence is generated.

Claims

1. A rotor coating additive manufacturing method based on amorphous nanocrystalline composites, characterized in that, The specific steps of this additive manufacturing method are as follows: I. Perform multi-scale topological preprocessing on the rotor surface, and construct a three-level biomimetic transition layer on the preprocessed rotor surface; the specific steps are as follows: S1.1: Determine the overall layout of the grooves according to the rotor stress or coating bearing requirements, and select the groove depth range and equal or variable spacing layout. Then, define the start and end points, depth curves and cross-sectional shapes of each groove on the rotor surface in the local coordinate system using 3D CAD. S1.2: Perform finite element local verification on each groove on the rotor surface. If the verification exceeds the limit, adjust the width or curvature of the corresponding groove, output the manufacturing executable tool or laser path and generate a processing process card. The processing process card specifically includes the processing sequence, processing power and cooling requirements. S1.3: Pre-divide the texture unit array area on the surface of the trench or slope, and determine the texture period, depth and orientation. Select the ultrashort pulse laser parameter set according to the required texture scale, including pulse width, frequency, energy and focal position. Then, adjust the energy density through trial cutting or simulation to simulate single hole or single texture processing, and establish the corresponding design template according to the processing results. S1.4: A line-by-line or point-by-point etching strategy is adopted to control the overlap rate and pulse count. After the machining is completed, the rotor surface is scanned and compared with the design template. If the deviation between the machining result and the design template exceeds the preset value, secondary fine-tuning is performed to determine the target function of the three-level bionic transition layer. Then, based on the determined target function, a layered model is established according to the thickness of the bionic transition layer to determine the target component or phase ratio of each layer. Then, according to the feasibility of the process, the continuous gradient is discretized into finite layers through the continuous component gradient function to obtain the relative content of the target main phase at different depths from the matrix interface in the transition layer. S1.5: Discretize the relative content of the target main phase into multiple layers, and specify the material ratio, powder feeding rate and energy input for each layer. Then, establish a parameter mapping table to record the powder feeding ratio, spraying speed, energy density and residence time of different layers. Then, based on the preset small area sample, conduct trial laying to verify the interlayer bonding force and structure, and adjust the powder feeding ratio and energy in real time, while recording the self-consistent parameter set. S1.6: Set the energy input strategy for each deposition layer, then verify the real-time temperature field by temperature measurement or thermocouple array, passively or actively cool each layer, adjust the energy density according to the test results, and record the interlayer temperature history as the basis for subsequent quality tracking and process reproduction. After the biomimetic transition layer is completed, use optical profile or white light interferometry to measure the layer thickness and surface roughness. S1.7: Compare the deviation between the design value and the measured value, and perform real-time compensation according to the type of deviation. After the entire biomimetic transition layer is deposited, perform the corresponding mechanical trimming and output the final test report as a quality control record. The test report specifically includes the layer thickness distribution, surface roughness and local deviation. II. Before the formal deposition of the coating, a virtual simulation model of the rotor is constructed to predict and optimize the temperature field, stress distribution, and microstructure evolution; the specific steps are as follows: S2.1: Import the rotor shape through CAD drawings and scanned data, and remove noise data in the imported data by filling holes and merging patches. At the same time, maintain high-resolution point clouds for key surfaces such as grooves and edges, define local coordinate system and region of interest (ROI), implement local mesh refinement strategy for ROI, and use coarse mesh for other areas. S2.2: After the mesh type is generated, select hexahedron or modified tetrahedron as the element type, and then adaptively reduce the element size according to the curvature and the estimated thermal gradient. Check the minimum inscribed sphere radius of each mesh. If the minimum inscribed sphere radius exceeds the preset threshold, mark the corresponding mesh as inferior and correct it. Then output the corrected final mesh. S2.3: Determine the coupled physical field according to the simulation target, establish the corresponding rotor virtual simulation model, select the constitutive relation for each material, and prepare experimental calibration data. Then, adopt the phase field model form for phase transformation or microstructure evolution, determine various dynamic parameters, define the heat input area, convection boundary, contact and constraint boundary conditions according to the process scenario, and map the energy to the grid in the form of time-space distribution function. S2.4: Arrange virtual sensor points and sampling frequencies in the virtual simulation model, formulate initial conditions for the virtual experiment, prepare multiple sets of different excitation spectra, set the time step, nonlinear solution strategy and iterative convergence criteria for the simulation experiment, estimate resources for a single simulation experiment, construct a parameter scanning matrix, run batch simulations of experiments and record temperature, stress, phase distribution time series and key section data, extract the extreme values, gradient fields, cooling rate curves and microstructure phase fraction distribution with depth of each simulation experiment from the simulation results to establish a structured dataset; S2.5: Based on the spatial-temporal proxy model structure of convolutional network, an evolution prediction model is established, and the input features and input targets of the evolution prediction model are set. Then, the structured dataset is divided into training set, validation set and test set, and feature scaling and spatiotemporal dimensionality reduction are performed. The training set is then input into the evolution prediction model. The evolution prediction model processes each data in the training set layer by layer based on the forward propagation algorithm and outputs the predicted evolution results at different times. S2.6: Calculate the loss value between the predicted evolution result and the actual evolution result through the regularized mean square error loss function. If the loss value is higher than the preset threshold, update the parameters of the evolution prediction model through the Adam optimizer and recalculate the corresponding loss value using the training set until the loss value converges to within the preset threshold. Then, evaluate the performance of the evolution prediction model using the validation set and test the prediction accuracy of the evolution prediction model on unknown data using the test set. S2.7: If the performance and prediction accuracy of the evolution prediction model reach the preset expected value, the training ends; otherwise, the evolution prediction model is retrained and optimized. The latest simulation data generated by the rotor virtual simulation model is input into the evolution prediction model. Then, the evolution prediction model processes the received simulation data layer by layer and outputs the temperature field, stress distribution and tissue evolution state within the preset time interval in the future. S2.8: Based on the temperature field, stress distribution and microstructure evolution state within the future preset time interval, if the evolution state exceeds the preset threshold, multiple sets of candidate process parameters are output through Bayesian optimization. Then, each set of candidate process parameters is fed back to the rotor virtual simulation model for back-checking, and the set of candidate process parameters with the lowest evolution state change is selected as the optimal set of process parameters and the original process parameters are replaced. III. Based on the rotor shape and simulation optimization results, multi-axis dynamic path planning is performed, and the attitude and trajectory of the deposition head are intelligently adjusted through a multi-axis linkage control system; IV. During the coating deposition process, the frequency and intensity of the dynamically modulated energy field induce in-situ recombination of amorphous phase and nanocrystal at the moment of droplet solidification, forming a dense composite structure; V. Real-time monitoring of temperature field, deposition rate and stress changes in the coating area, and automatic adjustment of deposition power, feed rate and scanning speed through closed-loop feedback; VI. During the deposition process, the ratio of amorphous to nanocrystalline materials is adjusted layer by layer. At the same time, the composition and pressure of the cavity atmosphere are adjusted in real time to control the oxygen content and the ratio of inert gas. VII. Analyze the morphology and performance of different regions of the rotor, and automatically select and switch the optimal deposition mode based on the analysis results.

2. The rotor coating additive manufacturing method based on amorphous nanocrystalline composites according to claim 1, characterized in that, The specific steps of the multi-axis dynamic path planning described in step III are as follows: S3.1: According to the process requirements, deploy laser scanning or industrial camera arrays in the actual processing scene, and set the sampling frequency and field of view coverage strategy. Continuously acquire original point cloud frames within the set sampling period, then use fast feature to perform coarse registration on point clouds from different sensors or different times, and then use the ICP algorithm to perform local fine registration to generate a rigid transformation matrix and merge it into a frame of global point cloud. S3.2: Statistically identify and remove outliers in the global point cloud, retain high-density points in key local areas as needed, and perform voxel lattice downsampling on the remaining areas. Then, perform surface reconstruction processing on the processed global point cloud to generate a point cloud with normal estimation and output the corresponding coverage. If the coverage is lower than a preset threshold, perform regional supplementary sampling. Select the neighborhood point set of each deposition trajectory point based on a preset radius, and then perform sampling based on the selected neighborhood point set. S3.3: Fit the corresponding local plane using the least squares method, calculate the unit normal vector of each local plane and unify its direction, then calculate the corresponding principal curvature and average curvature based on the covariance matrix of each local plane, and mark the region with principal curvature higher than the preset threshold as a high curvature region. Then identify and classify the geometric feature points of rotor boundary, acute angle, and groove root, and provide labels for attitude constraints and path refinement in different regions to establish a complete local reference system and curvature index matrix. S3.4: Based on the established local reference frame, a sequence of solution points is generated on the rotor surface according to the design layer thickness and desired line width. Each point includes position, desired normal and tangential. The IK solver is then used to calculate the joint angle vector corresponding to each trajectory point of the multi-axis linkage control system and record the multiple solutions. Collision, limit and speed limit detection are then performed on each joint solution. Based on the detection results, local trajectory rearrangement is performed on unreachable or conflict points. S3.5: Generate a preliminary draft of the joint-level trajectory for the multi-axis linkage control system based on the rearrangement results. Divide the preliminary draft of the joint-level trajectory into multiple segments according to physical accessibility, geometric features and task priority, and generate time parameters for each segment. When a collision warning occurs, perform local replanning on the affected segment, generate the corresponding joint time series, and replace the original trajectory with the replanned new trajectory.

3. The rotor coating additive manufacturing method based on amorphous nanocrystalline composites according to claim 1, characterized in that, The specific steps for adjusting the ratio of amorphous to nanocrystalline materials layer by layer as described in step VI are as follows: S4.1: Set the target value range for hardness, toughness, and wear resistance for the outer and inner layers of the coating, and determine the total coating thickness. In the form of a distribution function, set the target component or phase ratio profile within the total coating thickness range. At the same time, map the coating performance requirements to the target curve of component value or phase fraction. Then, quantify the continuous target curve and save it as process input. Record special functional areas and mark the priority and minimum layer thickness constraints. S4.2: Select the lower limit of single-layer thickness based on the equipment's feeding and process accuracy capabilities, and use it as the layer resolution of the corresponding coating. Discretize the total thickness into multiple layers according to the layer resolution of different coatings, calculate the center depth of each layer, and extract the target component value of the corresponding layer from the target profile. Based on the target component value, layer thickness, priority and subsequent feeding channel requirements of each layer, establish the corresponding process layer table. S4.3: Count the number of component channels and define the physical properties and unit powder volume mass ratio table for each channel. Decompose the target components of different coatings to obtain the component vector required for each channel. Set the total powder feeding rate requirement for different coatings and calculate the initial powder feeding rate of each channel. Then write the initial powder feeding rate of each channel into the feeding time series table and synchronize it with the switching sequence of the deposition head and the interlayer cooling window. S4.4: Online component sensors are placed at the deposition head or close to the deposition point, and the proxy index values ​​of the current mixture or deposition product are collected periodically. The measured proxy index values ​​are mapped to the actual component estimates and compared with the target mass fraction. If the deviation exceeds the allowable error, the feeding correction is triggered. At the same time, a smooth gain correction law is used to gradually adjust the powder feeding rate of each channel, and corresponding saturation and rate limits are applied during the correction process. The closed-loop correction log of each layer is recorded in real time.

4. The rotor coating additive manufacturing method based on amorphous nanocrystalline composites according to claim 1, characterized in that, The specific steps for automatically selecting and switching the optimal deposition mode described in step VII are as follows: S5.1: Divide the three-dimensional surface of the rotor into multiple groups of regional units according to function and geometric features, and establish a unique identifier for each regional unit. Obtain the geometric feature vector corresponding to each regional unit by collecting data and point cloud data of each region, and define the working condition drive performance requirements of each region. S5.2: Based on the geometric characteristics and performance requirements of each regional unit, establish corresponding demand vectors and store them in the regional demand library, mark the priority and tolerance zone, then evaluate the processing accessibility of each regional unit and record it as a constraint condition, and then establish a deposition mode library, in which each mode includes mode ID, mode type label, controllable parameter range, typical energy input and process constraints. S5.3: Obtain the capability vector of each model in the sedimentation model library in multiple sets of simulation experiments, transform different capability vectors into vectors of the same type, standardize the capability vector of each model, and establish a model-region feasibility matrix based on the preprocessed capability vectors of each model. S5.4: Select a set of candidate models from the deposition model library that meet the feasibility matrix of each region and do not violate hard constraints. Set multiple sets of performance criteria and assign engineering weights to each criterion according to process priority. Calculate the aggregation score of each candidate model with each region. Then sort the candidate models from high to low score and output the top five as switching candidates. At the same time, record the estimated switching cost between models. S5.5: Based on the aggregated score results, formulate a pattern assignment sequence for the entire rotor surface, design a transition strategy for each proposed switching point, evaluate the feasibility of the switching plan in the time dimension and align it with the production cycle constraints. If it is not feasible, fall back to the second-best score but lower switching cost solution, and generate an executable switching instruction sequence.

Images