Axial flux motor stator impregnation rapid setting method, device and apparatus

Abstract

The application relates to a kind of axial flux motor stator impregnation rapid setting method, device and equipment, including the surface of stator end portion is divided into a plurality of solidification micro area according to the three-dimensional grid model data of stator end portion;Multi-objective optimization model is constructed, and the multi-objective optimization model is with the ultraviolet irradiation intensity, irradiation time and thermal catalytic temperature of each solidification micro area as decision variable, with surface curing time, resin temperature rise peak value and total energy consumption of irradiation as optimization goal, obtain pareto optimal solution set, each pareto optimal solution corresponds a group of the ultraviolet irradiation intensity, irradiation time and thermal catalytic temperature of multiple solidification micro area;From pareto optimal solution set, select the solution with the shortest surface curing time as target solution;According to the ultraviolet irradiation intensity, irradiation time and thermal catalytic temperature of multiple solidification micro area corresponding to target solution, control ultraviolet lamp array to carry out partition scanning irradiation to stator end portion, and control infrared heater to carry out partition heating to stator end portion.

Description

Technical Field

[0001] This application relates to the field of motor manufacturing technology, and in particular to a method, apparatus and equipment for rapid shaping of axial flux motor stator by impregnation. Background Technology

[0002] In the impregnation process of axial flux motor stators, the resin needs to be cured after impregnation. Traditional curing methods, such as overall heating or overall ultraviolet irradiation, have the following technical problems: First, the surface resin and the internal resin cure simultaneously, causing resin to drip before the end surface has set, affecting the curing quality and appearance consistency; second, overall heating causes the internal resin temperature to be too high, which can easily lead to heat accumulation and premature resin curing, hindering subsequent flow and filling; third, curing parameters (such as irradiation intensity, duration, and temperature) can be set based on experience, lacking differentiated control methods for different areas of the stator end, resulting in uneven curing. Summary of the Invention

[0003] The purpose of this invention is to provide a method, apparatus, and equipment for rapid curing of axial flux motor stators through impregnation, aiming to improve the curing uniformity and quality of axial flux motor stators.

[0004] To achieve the above objectives, a first aspect of this disclosure provides a rapid shaping method for impregnating an axial flux motor stator, comprising: dividing the surface of the stator end into multiple micro-regions to be cured based on three-dimensional mesh model data of the stator end of the axial flux motor; constructing a multi-objective optimization model, wherein the multi-objective optimization model uses the ultraviolet irradiation intensity, irradiation duration, and thermocatalytic temperature of each micro-region to be cured as decision variables, and the surface curing time, resin temperature rise peak, and total irradiation energy consumption as optimization objectives, to obtain a Pareto optimal solution set, wherein each Pareto optimal solution corresponds to a set of ultraviolet irradiation intensity, irradiation duration, and thermocatalytic temperature of the multiple micro-regions to be cured; selecting the solution with the shortest surface curing time from the Pareto optimal solution set as the target solution; and controlling an ultraviolet lamp array to perform partitioned scanning irradiation of the stator end and controlling an infrared heater to perform partitioned heating of the stator end according to the ultraviolet irradiation intensity, irradiation duration, and thermocatalytic temperature of the multiple micro-regions to be cured corresponding to the target solution.

[0005] A second aspect of this disclosure provides a rapid shaping device for impregnating an axial flux motor stator. The device includes: a partitioning module configured to partition the surface of the stator end of the axial flux motor into multiple micro-regions to be cured based on three-dimensional mesh model data of the stator end; a construction module configured to construct a multi-objective optimization model, wherein the multi-objective optimization model uses the ultraviolet irradiation intensity, irradiation duration, and thermocatalytic temperature of each micro-region to be cured as decision variables, and the surface curing time, resin temperature rise peak, and total irradiation energy consumption as optimization objectives, to obtain a Pareto optimal solution set, wherein each Pareto optimal solution corresponds to a set of ultraviolet irradiation intensity, irradiation duration, and thermocatalytic temperature of the multiple micro-regions to be cured; a selection module configured to select the solution with the shortest surface curing time from the Pareto optimal solution set as the target solution; and a control module configured to control an ultraviolet lamp array to perform partitioned scanning irradiation of the stator end and control an infrared heater to perform partitioned heating of the stator end based on the ultraviolet irradiation intensity, irradiation duration, and thermocatalytic temperature of the multiple micro-regions to be cured corresponding to the target solution.

[0006] A third aspect of this disclosure provides an electronic device, comprising: A memory having a computer program stored thereon; a processor for executing the computer program in the memory to implement the steps of the method of any one of the first aspects.

[0007] This invention provides a method, apparatus, and equipment for rapid shaping of an axial flux motor stator through impregnation. Compared with the prior art, it has the following advantages: The above technical solution divides the stator end surface into multiple micro-regions to be cured. This refined division allows for accurate identification of geometric differences at the stator end, and independent optimization of UV irradiation intensity, duration, and thermocatalytic temperature for each micro-region. This achieves selective curing of complex curved surfaces and solves the problem of resin dripping. A multi-objective optimization model is constructed, with surface curing time, peak resin temperature rise, and total irradiation energy consumption as optimization objectives. The obtained Pareto optimal solution set clearly shows the trade-offs between these objectives, finding the optimal balance between surface curing time, peak resin temperature rise, and total irradiation energy consumption. In particular, by selecting the objective solution with the shortest surface curing time, production efficiency and anti-dripping effect are maximized.

[0008] Furthermore, selecting the solution with the shortest surface curing time minimizes surface setting time while ensuring controllable internal temperature rise and energy consumption, thereby improving production cycle time. By using zoned scanning irradiation and zoned heating, high-curvature and edge regions receive higher intensity UV irradiation and assisted thermal catalysis, while parameters are appropriately reduced in flat regions, ensuring consistent curing and surface quality. This eliminates the need for repeated manual trial and error, reducing reliance on human experience and lowering process development costs. Moreover, the optimized model can be transferred to different stator models, demonstrating good versatility and scalability.

[0009] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0010] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a rapid shaping method for an axial flux motor stator by impregnation, as shown in the embodiments of the specification.

[0011] Figure 2 This is one implementation shown in the embodiment of the specification. Figure 1 A flowchart of step S11.

[0012] Figure 3 This is one implementation shown in the embodiment of the specification. Figure 1 A flowchart illustrating step S12.

[0013] Figure 4 A block diagram of a rapid shaping device for axial flux motor stator impregnation is shown in the embodiment of the specification.

[0014] Figure 5 This is a block diagram of another axial flux motor stator impregnation and rapid shaping device according to the embodiments in the specification. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0016] This disclosure provides a rapid shaping method for the stator of an axial flux motor through impregnation. Figure 1 This is a flowchart illustrating a rapid shaping method for the stator of an axial flux motor by impregnation, according to one embodiment. The method includes: In step S11, based on the three-dimensional mesh model data of the stator end of the axial flux motor, the surface of the stator end is divided into multiple micro-regions to be cured; The three-dimensional mesh model data can be digital model data representing the geometry of the stator end obtained through three-dimensional scanning equipment or computer-aided design software. For example, it can include the spatial coordinate information of all coordinate points on the stator end surface. The micro-region to be cured can be a small region unit with independent geometric features formed by discretizing the continuous stator end surface.

[0017] In this embodiment of the disclosure, the process of dividing the micro-regions to be cured may include: based on the three-dimensional mesh model data of the stator end, and according to the geometric features such as surface curvature changes and normal vector directions, clustering the point cloud data into different candidate micro-regions, and merging micro-regions with too small an area into neighboring micro-regions, ultimately forming multiple micro-regions to be cured. Each micro-region to be cured is assigned a unique index number and its spatial coordinates, normal vector, and radius of curvature are associated and stored.

[0018] For example, regions with high curvature at the stator end edge can be divided into independent micro-regions to be cured, allowing for the application of specific curing parameters tailored to these high curvature features. Conversely, flat regions can be merged into larger micro-regions to reduce computational load. Through refined partitioning based on a 3D mesh model, geometric differences at different locations on the stator end can be accurately captured. This transforms the macroscopic stator end surface into independently controllable micro-units, thus addressing the problem that traditional methods cannot differentiate between regions with different geometric features.

[0019] In step S12, a multi-objective optimization model is constructed. The multi-objective optimization model takes the ultraviolet irradiation intensity, irradiation duration and thermocatalytic temperature of each micro-region to be cured as decision variables, and the surface curing time, resin temperature rise peak and total irradiation energy consumption as optimization objectives to obtain a Pareto optimal solution set. Each Pareto optimal solution corresponds to a set of ultraviolet irradiation intensity, irradiation duration and thermocatalytic temperature of the multiple micro-regions to be cured. The multi-objective optimization model is a mathematical model used to solve for the optimal equilibrium point of multiple conflicting objectives. Decision variables include the UV irradiation intensity, irradiation duration, and thermocatalytic temperature of each micro-region to be cured, which determine the curing process parameters of that micro-region.

[0020] Among the optimization objectives, including surface curing time, peak resin temperature rise, and total irradiation energy consumption, shorter surface curing time is better, peak resin temperature rise must be controlled below the threshold to prevent heat accumulation, and total irradiation energy consumption must be minimized. The Pareto optimal solution set can be the set of all solutions in a multi-objective optimization space that cannot further improve one objective without worsening any other objective.

[0021] In this embodiment of the disclosure, the multi-objective optimization model can treat the optimization objectives as individuals in a population, i.e., a set of optimization objectives as individuals in the population. After initializing the population, i.e., initializing the surface curing time, resin temperature rise peak and total irradiation energy consumption of the optimization objectives, the dominance relationship between individuals is calculated, the non-dominated frontier is divided, and the crowding distance is combined to perform tournament selection, simulated binary crossover and polynomial mutation operations, and iterative evolution until the termination condition is met, thereby obtaining the Pareto optimal solution set.

[0022] For example, each Pareto optimal solution might correspond to a parameter combination of high UV intensity, short irradiation duration, and moderate temperature, which can complete surface curing in a very short time with low energy consumption; while another solution might correspond to a combination of low UV intensity, long irradiation duration, and low temperature, which, although time-consuming, results in extremely low temperature rise. By constructing a multi-objective optimization model, the complex nonlinear relationship between UV irradiation, thermocatalysis, and curing effects can be scientifically quantified, avoiding uneven curing or thermal damage caused by relying on manual experience to set parameters. In this way, a set of candidate solutions that achieve the best balance in the three dimensions of time, temperature, and energy consumption can be selected.

[0023] In step S13, the solution with the shortest surface curing time is selected from the Pareto optimal solution set as the target solution; The objective solution is a specific solution selected from the Pareto optimal solution set, with the selection criterion being the minimum value of the surface curing time, which is the optimization objective. After obtaining the Pareto optimal solution set, each solution in the set can be traversed, and its corresponding surface curing time value can be compared to directly lock in the solution with the shortest time.

[0024] For example, if the Pareto optimal solution set contains three solutions with surface curing times of 30 seconds, 45 seconds, and 50 seconds, respectively, then the solution with a curing time of 30 seconds will be selected as the target solution. The core logic of this step is to prioritize production efficiency, that is, to pursue the fastest surface setting speed while meeting the constraints of keeping the resin temperature rise peak within limits and energy consumption under control. By selecting the solution with the shortest surface curing time, it is possible to prevent the impregnated resin from flowing or dripping under gravity, ensuring the appearance consistency and dimensional accuracy of the stator ends.

[0025] In step S14, based on the UV irradiation intensity, irradiation duration, and thermocatalytic temperature of the multiple micro-regions to be cured corresponding to the target solution, the UV lamp array is controlled to perform partitioned scanning irradiation on the stator end, and the infrared heater is controlled to perform partitioned heating on the stator end.

[0026] The UV lamp array is an illumination device composed of multiple independently controllable UV emitting units, while the infrared heater is a heating device composed of multiple independently adjustable infrared radiating units. Zoned scanning irradiation allows control of the corresponding emitting units in the UV lamp array to perform point-to-point or scanning irradiation of a specified micro-region based on the UV irradiation intensity and duration corresponding to each micro-region to be cured in the target solution. Zoned heating allows control of the corresponding radiating units in the infrared heater to precisely heat a specified micro-region based on the thermocatalytic temperature corresponding to each micro-region to be cured in the target solution.

[0027] Specifically, the parameters of each micro-region in the target solution are read. For high curvature or easily dripping regions, the ultraviolet lamp array is controlled to output high-intensity ultraviolet light and perform short-term focused irradiation. At the same time, an infrared heater is used to apply an appropriate thermal catalytic temperature to accelerate the crosslinking reaction. For flat or near-liquid pool regions, the irradiation intensity is reduced or the irradiation time is extended, and the heating temperature is controlled to prevent the resin from curing prematurely.

[0028] For example, when a micro-area to be cured is detected at the lower edge of the stator end, the corresponding UV lamp unit can be irradiated at 120% rated intensity for 15 seconds according to the target solution command, while the infrared heater maintains the temperature of the area at 60°C, thereby quickly forming a surface curing layer to lock in the resin. For the upper edge area, it can be irradiated at 80% intensity for 25 seconds, with the temperature maintained at 40°C. Through the coordinated zonal control of the UV lamp array and the infrared heater, precise spatial and temporal matching is achieved, allowing the surface resin to preferentially and quickly solidify, preventing dripping, while avoiding heat accumulation and premature curing caused by internal overheating, ensuring the resin's flow and filling capabilities.

[0029] The above technical solution divides the stator end surface into multiple micro-regions to be cured. This refined division allows for accurate identification of geometric differences at the stator end, and independent optimization of UV irradiation intensity, duration, and thermocatalytic temperature for each micro-region. This achieves selective curing of complex curved surfaces and solves the problem of resin dripping. A multi-objective optimization model is constructed, with surface curing time, peak resin temperature rise, and total irradiation energy consumption as optimization objectives. The obtained Pareto optimal solution set clearly shows the trade-offs between these objectives, finding the optimal balance between surface curing time, peak resin temperature rise, and total irradiation energy consumption. In particular, by selecting the objective solution with the shortest surface curing time, production efficiency and anti-dripping effect are maximized.

[0030] Furthermore, selecting the solution with the shortest surface curing time minimizes surface setting time while ensuring controllable internal temperature rise and energy consumption, thereby improving production cycle time. By using zoned scanning irradiation and zoned heating, high-curvature and edge regions receive higher intensity UV irradiation and assisted thermal catalysis, while parameters are appropriately reduced in flat regions, ensuring consistent curing and surface quality. This eliminates the need for repeated manual trial and error, reducing reliance on human experience and lowering process development costs. Moreover, the optimized model can be transferred to different stator models, demonstrating good versatility and scalability.

[0031] In this disclosure embodiment, see Figure 2 As shown, in step S11, the surface of the stator end of the axial flux motor is divided into multiple micro-regions to be cured based on the three-dimensional mesh model data of the stator end, including: In step S111, based on the three-dimensional mesh model data of the stator end of the axial flux motor, the point cloud corresponding to the three-dimensional mesh model data is classified into different candidate micro-regions; Point cloud is a set of points obtained by discrete sampling of the surface of a three-dimensional mesh model. Each point contains spatial coordinate information and is used to characterize the geometry of the stator end.

[0032] In this embodiment, point cloud data is extracted from the three-dimensional mesh model at the stator end, and the normal vector of each point is calculated. A region growing algorithm based on the normal angle threshold is used to group adjacent points with a normal direction change of less than 15 degrees into the same candidate micro-region. The growing process starts from the point of minimum curvature and expands layer by layer to neighboring points until no new points can be absorbed. Unlabeled seed points are repeatedly selected until all points are assigned, resulting in several candidate micro-regions.

[0033] In step S112, the area of ​​each candidate micro-region is calculated, and the candidate micro-regions with areas smaller than a preset area are merged into the adjacent candidate micro-regions that are closest to their normal direction, thus obtaining multiple micro-regions to be cured. In this embodiment, the surface area of ​​each candidate micro-region is calculated, and micro-regions with an area smaller than a preset area threshold are marked as objects to be merged. For each micro-region to be merged, the angle between its normal and the normals of all adjacent micro-regions is calculated, and the adjacent micro-region with the smallest angle is selected for merging. After merging, the point set and boundary of the micro-region are updated, and the area is recalculated. The above process is repeated until there are no small micro-regions with an area smaller than the threshold, resulting in the final micro-region to be solidified.

[0034] In step S113, the spatial coordinates, normal vector and radius of curvature of the micro-region to be cured are determined based on the coordinates of all points in the plurality of micro-regions to be cured. In this embodiment of the disclosure, for each micro-region to be solidified, the average of the spatial coordinates of all points contained therein is calculated to obtain the spatial coordinates of the center of the micro-region; the normal vectors of all points are normalized and summed to obtain the representative normal vector; the covariance matrix of the micro-region point set is constructed, and three eigenvalues ​​are solved. The ratio of the smallest eigenvalue to the sum of the eigenvalues ​​is taken as the curvature, and its reciprocal is the radius of curvature.

[0035] In step S114, a unique index number is assigned to each micro-region to be cured, and the number is associated with and stored in relation to the corresponding spatial coordinates, normal vector and radius of curvature.

[0036] In this embodiment, each micro-region to be cured is assigned an integer index number starting from zero, following the order of increasing circumferential angle of the micro-region center point and increasing radial distance from inside to outside. This number is associated with the micro-region's spatial coordinates, normal vector, and radius of curvature and stored in the process database. A binary mask is generated, with a length equal to the total number of micro-regions, and only the index bit is set to one, for fast addressing of the UV lamp array.

[0037] The above-mentioned technical solution achieves refined division of micro-regions at the stator end through point cloud segmentation, small region merging, geometric feature calculation, and index storage. This method ensures that each micro-region has a reasonable area and accurate geometric parameters, providing structured input for subsequent multi-objective optimization, improving the spatial resolution of curing process parameters, thereby enhancing surface shaping uniformity and reducing internal thermal impact.

[0038] In one possible implementation, step S111, which involves classifying the point cloud corresponding to the three-dimensional mesh model data of the stator end of the axial flux motor into different candidate micro-regions, includes: In step S1111, the normal vector and curvature of all points are calculated based on the three-dimensional mesh model data of the stator end of the axial flux motor. The calculation of the normal vector and curvature of all points can be considered as a process of geometric feature extraction from the discretized point cloud data of the stator end surface. The normal vector is used to characterize the vertical direction of the point cloud surface at a certain point, reflecting the orientation of the local plane where that point is located; the curvature is used to quantify the degree of curvature of the surface at that point. The smaller the curvature value, the flatter the surface, and the larger the curvature value, the more severe the surface curvature or the more it is located at an edge.

[0039] Specifically, based on the acquired 3D mesh model data, this is achieved by constructing local neighborhoods and calculating the eigenvalue decomposition of the covariance matrix. For example, for each point Pi in the point cloud, neighborhood points with a radius of R or the k nearest neighbors are selected, and the centroid coordinates of these points are calculated. This allows for the construction of the covariance matrix. The normal vector is obtained by solving for the eigenvector corresponding to the smallest eigenvalue of this matrix, and the ratio of the smallest eigenvalue to the sum of the eigenvalues ​​is used as an estimate of the curvature. This allows for the accurate capture of geometrical differences in complex surfaces at the stator ends (such as straight edges and rounded corner transition areas), providing accurate measurement data for subsequent region growing.

[0040] In step S1112, the point with the smallest curvature is selected from all the points as the seed point, and the point closest to the seed point is selected from the point cloud as the neighboring point according to the coordinates of all the points; Selecting the point with the minimum curvature as the seed point can employ a greedy strategy, prioritizing the region with the most stable and flattest geometric features as the starting center for cluster growth. Since the point with the minimum curvature can be located at the center of a planar region or the center of a large-radius arc, using it as the seed point can effectively avoid region divergence caused by starting clustering from the edges or noisy points. Selecting the point closest to the seed point as a neighbor can establish the topological connectivity between the seed point and the unlabeled points in its spatial neighborhood, which can be determined by calculating the Euclidean distance.

[0041] For example, if there are multiple unlabeled points in the current point cloud, the curvature values ​​of all unlabeled points are first traversed, and the point with the smallest absolute curvature value, Pseed, is locked as the initial seed point. Then, within the preset search radius of Pseed, the nearest unlabeled point, Pneighbor, is found as the neighbor to be determined. This ensures that region growth always expands outward from the most stable geometric core, improving the robustness of clustering.

[0042] In step S1113, the angle between the normal vector of the adjacent point and the normal vector of the seed point is determined; Determining the included angle can be achieved by calculating the angle difference between the normal vectors of adjacent points and the normal vector of the current seed point in three-dimensional space. This angle difference is a key geometric constraint for determining whether two points belong to the same smooth surface.

[0043] Specifically, the angle between the two normal vectors can be obtained by calculating the dot product of their normal vectors and taking the inverse cosine function. This angle reflects the degree of surface inclination of adjacent points relative to the seed point. For example, if the seed point is located on a flat side of the stator end, its normal vector is perpendicular to that side. If an adjacent point is located on a chamfered surface intersecting that side, its normal vector will be significantly deflected, resulting in a larger calculated angle. Conversely, if the adjacent points are still located on the same flat side, the angle approaches zero.

[0044] In step S1114, if the angle between the normal vector of the adjacent point and the normal vector of the seed point is less than a preset angle and the adjacent point is not marked, then the adjacent point is included in the candidate micro-region to which the current seed point belongs and is used as a new seed point. The preset angle is a threshold set based on the maximum allowable rate of change of the continuous surface at the stator end, used to define the range of geometric consistency. Including the adjacent point in the candidate micro-region to which the current seed point belongs can be achieved by performing a region merging operation, marking the point as already assigned, and adding its attributes (such as coordinates, normal, and curvature) to the statistical characteristics of the current micro-region. Using it as a new seed point can be achieved by performing a growth iteration, dynamically migrating the cluster center to the newly added point.

[0045] For example, if a preset angle is set to 15 degrees, and the calculated included angle is 10 degrees and the point is not occupied by other micro-regions, it is determined that the point has the same geometric orientation as the current micro-region. Therefore, it is marked and added to the set, and the current growth front is updated to the new point. Through the mechanism of dynamically updating seed points, continuous growth from a single point to a patch is achieved, ensuring the consistency of the normals within the micro-region and avoiding the erroneous merging of patches with different orientations.

[0046] In step S1115, the steps from "based on the coordinates of all the points" to "adding the adjacent point to the candidate micro-region to which the current seed point belongs and using it as a new seed point" are repeated until the conditions for reselecting a seed point are met. The repeated execution can be a process of iteratively performing neighbor search, angle determination, and seed point update within a complete region growth cycle. The condition for reselecting a seed point can be met if all neighboring points of the current seed point have been marked, or if the angle between the normals of all neighboring points and the current seed point exceeds a preset angle, preventing further outward expansion. At this point, it indicates that an independent connected region (i.e., a candidate micro-region) based on the current geometric features has been fully formed.

[0047] For example, when the growth process extends to the edge at the end of the stator, the growth stops because the normals on both sides of the edge change abruptly, and the included angle exceeds the threshold, thus naturally dividing the two sides of the edge into different regions. In this way, it is ensured that each candidate micro-region is a maximally connected subgraph composed of points with continuous geometric features.

[0048] In step S1116, the next seed point with the smallest curvature is selected from the unmarked points, and the steps from the coordinates of all the points to the inclusion of the adjacent point into the candidate micro-region to which the current seed point belongs and the inclusion of it as the new seed point are repeated until all points have been included into the candidate micro-region to which the seed point belongs.

[0049] The process of selecting the next seed point with the smallest curvature from unlabeled points can be achieved by restarting a new growth cycle in the remaining unlabeled point cloud after the division of a micro-region, still following the priority strategy of flattening first and then curving. Re-execution can reuse the above logic to generate the next independent candidate micro-region. The process of traversing until all points have been included can be the termination condition, ensuring that the point cloud data is completely segmented without any omissions.

[0050] For example, after processing the largest flat side at the stator end, the algorithm searches for the point with the second smallest curvature among the remaining unlabeled points (which may be another smaller plane or a large-radius surface) as a new seed to begin the growth of the second micro-region, until all complex high-curvature regions such as chamfers and hole edges have been divided one by one. Through an iterative full-coverage strategy, the complex stator end point cloud can be adaptively decomposed into several candidate micro-regions with clear geometric boundaries.

[0051] The aforementioned technical solution prioritizes points with the lowest curvature as seed points, ensuring the initial stability of region growth and avoiding erroneous clustering caused by noise points. Combined with the normal vector angle judgment and dynamic seed point update mechanism, it ensures the continuity and consistency of the generated candidate micro-regions in terms of geometry, allowing for natural segmentation along the physical boundaries of the stator end (such as edges and curved transition areas) without crossing different functional surfaces. The region growth algorithm guided by geometric features not only effectively solves the problem of fragmented partitioning caused by uneven point distribution but also accurately identifies different curvature regions at the stator end. This allows subsequent processes to formulate differentiated UV irradiation and heating strategies for flat areas, transition areas, and edge areas, significantly improving the uniformity and shaping quality of resin curing. Through synergistic effects, efficient and high-precision clustering of point cloud data at the stator end of an axial flux motor is achieved.

[0052] In one possible implementation, step S1111, determining the normal vector and curvature of all points based on the three-dimensional mesh model data of the stator end of the axial flux motor, includes: In step S1111a, the coordinates of all points are determined based on the three-dimensional mesh model data of the stator end of the axial flux motor, and neighborhood points of each point are selected based on the coordinates of all points. The neighborhood points are points with a preset radius of the corresponding point or the k points closest to the corresponding point. Neighborhood points can be a set of surrounding points used to characterize the local geometric features of the target point. The source of neighborhood points is determined based on the point cloud distribution in the aforementioned obtained three-dimensional mesh model data of the stator end. Neighborhood points can provide a data basis for calculating local statistical characteristics, and their setting method can be flexibly selected according to the density distribution of the actual point cloud: one is the fixed radius method, that is, setting a spatial search radius R, and taking all points falling into the sphere with the target point as the center and R as the radius as neighborhood points; the other is the k nearest neighbor method, that is, searching in the point cloud for the k points that are closest to the target point in Euclidean distance as neighborhood points.

[0053] For example, when the scanning accuracy of the stator end surface is high and the point cloud distribution is uniform, a fixed radius of 2mm can be used to select neighboring points; however, when the surface has complex curves that cause uneven point cloud density, the 15 closest points (k=15) are selected as neighboring points to ensure the stability of local feature extraction. Through the adaptive neighbor selection mechanism, different scanning accuracy scenarios can be effectively adapted.

[0054] In step S1111b, the mean value of the coordinates in each dimension is calculated based on the coordinates of each point and the coordinates of the corresponding neighborhood point to obtain the coordinates of the centroid of each point. The centroid's coordinates can be the geometric center of the local point set consisting of the current point and its neighbors in three-dimensional space. The centroid's coordinates are obtained by summing the coordinates of the target point and all its neighbors in the X, Y, and Z dimensions and then taking the average. The centroid can serve as the origin of the local coordinate system, eliminating the influence of translation on shape analysis and thus more accurately reflecting the distribution pattern of the point set.

[0055] Specifically, if point P and its neighborhood set are {P1, P2, ..., Pn}, then the X-coordinate of the centroid C is the sum of the X-coordinates of all points divided by (n+1), and the Y and Z coordinates are calculated similarly. For example, for a local region containing 10 neighborhood points, the calculated centroid coordinates are (12.5, 8.3, 4.1), which represent the center position of this micro-region. By translating the local point set to a position with the centroid as the origin, the calculation process of the covariance matrix can be simplified, and the stability of numerical calculation can be improved.

[0056] In step S1111c, the covariance matrix corresponding to each point is determined based on the coordinates of each point, the coordinates of the corresponding neighborhood point, and the coordinates of the centroid. The covariance matrix is ​​a 3x3 symmetric matrix used to describe the dispersion and correlation of a local point set across various dimensions. It is calculated from the decentralized point coordinates (i.e., each point's coordinates minus the centroid's coordinates), and its elements reflect the variance of the point cloud in the X, Y, and Z directions, as well as the covariance between any two points in each direction. The covariance matrix reveals the statistical characteristics of the local shape of the point cloud and is a core mathematical tool for extracting normal vectors and curvature.

[0057] Specifically, the diagonal elements of the matrix represent the variance of each dimension, while the off-diagonal elements represent the correlation between dimensions. For example, if a local point set approximates a plane, the variance perpendicular to that plane will be smaller than the variances of the two directions within the plane; this characteristic will be recorded in the covariance matrix. By constructing the covariance matrix, the discrete point cloud geometric information is transformed into an analyzable algebraic form.

[0058] In step S1111d, the normal vector and curvature of all the points are determined based on the covariance matrix of each point.

[0059] Here, the normal vector can be a unit vector perpendicular to the local tangent plane of the point cloud, and the curvature can be a scalar value characterizing the degree of curvature of the local surface. The normal vector and curvature are obtained by eigenvalue decomposition of the covariance matrix obtained above. Specifically, the three eigenvalues ​​of the covariance matrix are ordered from smallest to largest as λ1≤λ2≤λ3, and the corresponding eigenvectors are v1, v2, v3. The eigenvector v1 corresponding to the smallest eigenvalue λ1 is the normal vector of that point, because the variance of the point cloud distribution is smallest in this direction, representing the normal direction of the surface; while the curvature can be determined by the ratio of the smallest eigenvalue to the sum of the eigenvalues ​​(λ1 / (λ1+λ2+λ3)). The smaller this ratio, the flatter the surface; the larger it is, the more severe the curvature.

[0060] For example, at a stator end edge point, if the calculated eigenvalues ​​are 0.01, 0.5, and 0.6, then the normal vector corresponds to the direction of the eigenvector with a value of 0.01, and the curvature value is 0.01 / 1.11 ≈ 0.009, indicating a significant bending characteristic at that location. Therefore, using the spatial distribution statistical characteristics of a local point set to infer geometric properties is more robust than simple interpolation or triangular patch normal averaging, and can effectively suppress noise interference.

[0061] The aforementioned technical solution, by selecting neighborhood points with a preset radius or k-nearest neighbors and combining centroid calculation and covariance matrix construction, can transform discrete 3D mesh data into a statistical matrix with clear physical meaning. Then, using eigenvalue decomposition technology, the normal vector and curvature are directly calculated from the algebraic properties of the matrix. This processing method not only eliminates noise interference at individual measurement points but also quantifies the surface curvature through the strong correlation between the minimum eigenvalue and the normal vector. Accurate normal vector and curvature data directly serve subsequent candidate micro-region partitioning, allowing the region growing algorithm to merge regions based on true geometric continuity rather than simple distance thresholds. This significantly improves the accuracy of micro-region partitioning to be cured, ensures the precision of UV irradiation and thermocatalytic control, and effectively solves the problem of uneven curing caused by geometric feature estimation bias in traditional methods. High-precision point cloud geometric attribute extraction is achieved.

[0062] In one possible implementation, step S113, determining the spatial coordinates, normal vector, and radius of curvature of the micro-region to be cured based on the coordinates of all points in the plurality of micro-regions to be cured, includes: In step S1131, the average coordinate of all points in the plurality of micro-regions to be cured is determined based on the coordinates of all points in the micro-regions to be cured. The average coordinate value can be the geometric center position of all discrete points within the micro-region to be solidified in three-dimensional space. It is obtained by summing the X, Y, and Z coordinates of all points within the micro-region and dividing by the total number of points. This average coordinate value is used as a reference point for calculating the covariance matrix, eliminating the influence of data translation on directional feature extraction.

[0063] For example, if a micro-region to be solidified contains N points, and the coordinates of the i-th point are (xi, yi, zi), then the X component of the average coordinate is the sum of all xi divided by N. Through a centralization process, the geometric distribution of the micro-region can be transformed into a local coordinate system with the centroid as the origin.

[0064] In step S1132, the average coordinate value is used as the spatial coordinate of the micro-region to be cured; The spatial coordinates are unique identifiers representing the position of the micro-region to be cured within the overall three-dimensional model of the stator end, and they are derived from the average coordinates calculated above. These spatial coordinates can be used to locate the scanning target center of the ultraviolet lamp array and the infrared heater, ensuring that the energy application device can be accurately aligned with the micro-region.

[0065] Specifically, spatial coordinates are converted into motion commands for the robotic arm or scanning galvanometer, guiding the light source to move above those coordinate points. For example, when the spatial coordinates are determined to be (Xc, Yc, Zc), the control module drives a three-axis motion platform to move the UV lamp head to that position, ensuring that the center of the light spot coincides with the geometric center of the micro-area. This significantly improves the positioning accuracy of zonal scanning and avoids overlapping or omissions in the curing of adjacent micro-areas due to positional deviations.

[0066] In step S1133, the covariance matrix of the micro-region to be cured is determined based on the coordinates of all points in each micro-region to be cured and the corresponding average coordinate value. The covariance matrix describes the dispersion and correlation of point cloud data within the micro-region to be solidified across various dimensions. It is obtained by averaging the outer product of the differences between the coordinate vector of each point and the average coordinate value. The covariance matrix quantifies the distribution pattern of points within the micro-region.

[0067] Specifically, for each point within the micro-region, its offset relative to spatial coordinates is calculated, an offset vector is constructed, and then the covariance of these vectors in each dimension is statistically analyzed. For example, if the surface of the micro-region is relatively flat, the coordinate fluctuations perpendicular to the surface direction are small, resulting in smaller element values ​​in the corresponding direction in the covariance matrix; if the surface is severely curved, the fluctuations in each direction are large.

[0068] In step S1134, the covariance matrix of each micro-region to be cured is decomposed into eigenvalues ​​to obtain multiple eigenvalues ​​and corresponding eigenvectors for each micro-region to be cured. Eigenvalue decomposition can be a mathematical process of diagonalizing the covariance matrix. The resulting eigenvalues ​​represent the variance of the point cloud data along the principal component directions, and the eigenvectors represent the corresponding principal component directions. The magnitude of the eigenvalues ​​reflects the extent of the micro-region's extension along that direction, and the direction of the eigenvectors indicates the main axis of the data distribution. In this way, the complex point cloud distribution can be decoupled into three mutually orthogonal principal directions and their corresponding dispersion.

[0069] For example, for a micro-region that approximates a plane, we obtain two large eigenvalues ​​(corresponding to the two extending directions within the plane) and one extremely small eigenvalue (corresponding to the normal direction of the plane). Through decomposition, key parameters characterizing the macroscopic geometry of the micro-region can be extracted from discrete point cloud data.

[0070] In step S1135, the eigenvector corresponding to the smallest eigenvalue in each of the micro-regions to be cured is selected, and the normal vector corresponding to the micro-region to be cured is determined; the ratio between the smallest eigenvalue in each of the micro-regions to be cured and the sum of the plurality of eigenvalues ​​corresponding to the micro-region to be cured is determined as the radius of curvature of the micro-region to be cured.

[0071] The normal vector is a unit vector perpendicular to the average surface of the micro-region to be cured, derived from the eigenvector corresponding to the minimum eigenvalue obtained from the aforementioned eigenvalue decomposition. Since the minimum eigenvalue represents the direction with the least fluctuation in point cloud data, i.e., the direction with the least significant change in micro-region thickness, this direction is the normal direction of the micro-region. The function of this normal vector can be to guide the irradiation angle of the ultraviolet lamp, ensuring that the light is incident on the micro-region surface as perpendicularly as possible, thereby maximizing energy absorption efficiency.

[0072] For example, when the eigenvector corresponding to the minimum eigenvalue is calculated to be (0,0,1), it indicates that the surface of the micro-region is mainly parallel to the XY plane, with the normal pointing upwards. Based on this, the pitch angle of the UV lamp array can be adjusted to ensure that the beam is projected perpendicularly. By estimating the normal based on statistical characteristics, the orientation changes caused by the complex curved surface at the stator end can be effectively adapted, thus improving the curing uniformity.

[0073] The radius of curvature is defined as a quantitative indicator characterizing the degree of local bending in a micro-region. It is obtained by calculating the ratio of the minimum eigenvalue to the sum of all eigenvalues. This ratio reflects the proportion of the micro-region's flatness in the normal direction relative to the overall distribution size. A smaller ratio means the micro-region is closer to a plane, while a larger ratio means the micro-region is more severely bent or has a spherical distribution. This radius of curvature can be used as a basis for adjusting the thermocatalytic temperature and irradiation time. High curvature areas often have the risk of stress concentration and require more precise control of curing parameters.

[0074] For example, if the three characteristic values ​​of a certain micro-region are 0.1, 5.0, and 5.2, and their sum is 10.3, then the radius of curvature index is approximately 0.0097, indicating that the region is very flat. Conversely, if the characteristic values ​​are 3.0, 3.5, and 4.0, the ratio is approximately 0.286, indicating that the region has significant curvature characteristics. Based on this index, a preset process strategy can be matched to appropriately reduce the heating rate for high-curvature micro-regions to prevent resin cracking. This achieves differentiated curing control for regions with different geometric characteristics, effectively solving the problem of uneven stress caused by traditional integral curing.

[0075] The aforementioned technical solution establishes spatial coordinates by calculating the average coordinate value, providing precise target points for equipment positioning. Utilizing the eigenvalue decomposition of the covariance matrix, the eigenvector corresponding to the smallest eigenvalue is identified as the normal vector, and the ratio of the smallest eigenvalue to the sum of eigenvalues ​​is defined as the radius of curvature. This allows for rapid quantification of the orientation and curvature of the micro-region without the need for complex surface fitting. Based on this, the spatial coordinates guide the equipment to the designated position, the normal vector guides the light source to adjust the incident angle to optimize energy coupling, and the radius of curvature serves as a feedback variable to dynamically adjust the curing process parameters. This multi-dimensional geometric characterization mechanism enables the UV lamp array and infrared heater to adaptively adjust the geometric characteristics of different regions at the stator end. This not only improves the energy utilization rate of the curing process but also effectively avoids curing defects caused by excessive local curvature or unfavorable orientation, thus enhancing the impregnation and shaping quality of the axial flux motor stator.

[0076] In this disclosure embodiment, see Figure 3 As shown, in step S12, the multi-objective optimization model uses the UV irradiation intensity, irradiation duration, and thermocatalytic temperature of each micro-region to be cured as decision variables, and the surface curing time, resin temperature rise peak, and total irradiation energy consumption as optimization objectives to obtain a Pareto optimal solution set, including: In step S121, the multi-objective optimization model uses a set of the surface curing time, the peak temperature rise of the resin, and the total irradiation energy consumption as individuals to obtain a population; In this context, an individual can be a candidate solution in a multi-objective optimization problem, and its encoding form contains specific numerical combinations of three decision variables: UV irradiation intensity, irradiation duration, and thermocatalytic temperature for all micro-regions to be cured. Surface curing duration, peak resin temperature rise, and total irradiation energy consumption are the three optimization objective function values ​​corresponding to this individual, representing the efficiency, thermal safety, and economy of the process, respectively. The population is a set of a predetermined number of individuals, representing the distribution state in the solution space at the current iteration number. The population can be initialized by random generation or uniform sampling within the feasible region, ensuring that the initial solution covers different parameter combinations.

[0077] For example, if the stator end is divided into 100 micro-regions to be cured, then the chromosome length of an individual is 300 (100 micro-regions × 3 variables). Each gene locus corresponds to the specific process parameters of a micro-region. By decoding the chromosome, the total curing time, maximum temperature rise, and total energy consumption under this scheme can be calculated. By constructing a population containing multiple individuals, multiple potential optimal regions can be searched in parallel in the solution space.

[0078] In step S122, for each individual in the population, the number of times it is dominated by other individuals is calculated; wherein, domination means that any individual is not inferior to another individual on all optimization objectives and is strictly superior on at least one optimization objective; Domination is the core criterion for evaluating the superiority or inferiority of individuals. Specifically, for any two individuals A and B in a population, if individual A's values ​​for surface curing time, peak resin temperature rise, and total irradiation energy consumption are all less than or equal to individual B's corresponding values, and individual A is strictly less than individual B in at least one of these targets, then individual A is said to dominate individual B. Calculating the number of times an individual is dominated means counting how many other individuals in the population can dominate it. A number of times an individual is dominated means that no other individual in the current population can simultaneously outperform it in all targets; such individuals can be located on the Pareto front.

[0079] For example, if individual X has a curing time of 10 minutes, a temperature rise of 50°C, and an energy consumption of 2 kWh, and individual Y has a curing time of 12 minutes, a temperature rise of 50°C, and an energy consumption of 2.5 kWh, then X dominates Y, and Y's dominance count increases by 1. If individual Z has a curing time of 9 minutes but a temperature rise of 60°C, then X and Z do not dominate each other. By accurately calculating the dominance count of each individual, the relative dominance position of an individual in a multidimensional target space can be quantified.

[0080] In step S123, all individuals whose number of dominations is zero are assigned to the first non-dominated front; the individuals in the first non-dominated front are temporarily stored and removed from the population; the number of times each individual in the population is dominated by other individuals in the current population is repeatedly calculated; and all individuals whose number of dominations is zero are assigned to the second non-dominated front. The first non-dominated frontier is the set of all individuals in the current population that are not dominated by any other individual, representing the optimal solution layer in the current generation. These individuals have achieved an optimal balance among the three optimization objectives and cannot further improve the other objectives without sacrificing the performance of one objective. Classifying such individuals into the first level signifies that they have the highest priority and will be preferentially retained to the next generation or as final output candidates.

[0081] For example, after one round of dominance calculation, if 20 individuals in the population have a dominance count of 0, these 20 individuals collectively constitute the first non-dominated frontier, which forms the optimal boundary under the current combination of process parameters. This achieves the function of rapidly extracting high-quality solution sets from the mixed population, ensuring that the optimization process always evolves in the direction of convergence.

[0082] In step S124, this process is repeated layer by layer until all individuals are assigned to the non-dominated front, thus obtaining the Pareto front level for each individual. The population is stratified using a layer-by-layer stripping strategy. After identifying and removing individuals from the first non-dominated front, the dominance relationships among the remaining individuals change; previously suboptimal individuals dominated by individuals in the first layer may become non-dominated in the remaining subset. By repeatedly calculating the dominance count and performing zero-value screening, the remaining individuals can be classified into the second non-dominated front. This process is repeated until the population is empty.

[0083] For example, after removing the first layer, the previously suboptimal individual M is no longer constrained by comparisons with individuals in the first layer and performs best among the remaining population, thus being relegated to the second layer. This hierarchical mechanism not only distinguishes between absolutely optimal solutions but also establishes a complete quality hierarchy of solutions, allowing the algorithm to utilize information from different layers to maintain population diversity and avoid premature convergence to a single local optimum.

[0084] In this context, the Pareto front rank is a hierarchical label for an individual, with the first non-dominated front having a rank of 1, the second non-dominated front having a rank of 2, and so on. A smaller rank value indicates better overall performance. Through this hierarchical process, each individual in the population is assigned a distinct rank, forming a complete order relation based on non-dominated ranking. This rank directly determines the survival probability of an individual during selection; higher-ranked (lower-valued) individuals are more likely to be selected for reproduction. For example, a group of individuals with a rank of 1 represents the solution set closest to the true Pareto front, while individuals with a rank of 5 are relatively worse. Obtaining the Pareto front rank for each individual ensures that evolutionary pressure primarily affects the convergence of solutions.

[0085] In step S125, the crowding distance for each individual is calculated based on the optimization objective of all individuals in the same Pareto front level; Crowding distance measures the density of solutions around individuals within the same frontier level, aiming to maintain population diversity. For individuals at the same non-dominated frontier, the sum of their distances to neighboring individuals across all optimization objective dimensions is calculated.

[0086] For example, firstly, individuals in this layer are sorted for each optimization objective. The crowding distance of boundary individuals is set to infinity or a preset maximum value, and the crowding distance of middle individuals is the normalized sum of the differences in function values ​​between their preceding and following individuals on that objective. For instance, in the first non-dominated frontier, if individual P is adjacent to individuals O and Q on the solidification time objective, and the solidification times of O, P, and Q increase sequentially, then the distance component of P on this dimension is (T imeQ -T imeO ) / (Timemax -T imemin ), where T imeQ T represents the solidification time of two adjacent intermediate individuals O of individual P. imeO T is the solidification time of two adjacent intermediate individuals Q of individual P. imemax T represents the maximum solidification time within the Pareto frontier levels for individuals P, O, and Q. imemin This represents the minimum curing time within the Pareto frontier levels for individuals P, O, and Q.

[0087] Furthermore, the distance components of the other two optimization objectives are calculated sequentially, and the total crowding distance is obtained by summing the components of the three objective dimensions. The larger the crowding distance, the sparser the solutions around the individual, and the stronger the representativeness.

[0088] By using crowding distance, the optimal solution set can be effectively prevented from being overly concentrated in a narrow region, ensuring that the final Pareto solution set is uniformly distributed in the target space.

[0089] In step S126, based on the Pareto front level and the corresponding crowding distance of each individual, tournament selection, simulated binary crossover, and polynomial mutation operations are performed, iteratively evolving until the termination condition is met, to obtain the Pareto optimal solution set.

[0090] The tournament selection is based on a screening mechanism that considers rank and crowding distance. Several individuals are randomly selected for comparison each time, with priority given to those with higher ranks (lower values). If the ranks are the same, those with greater crowding distance are selected. This process is used to build a mating pool and ensure that high-quality and dispersed genes are introduced into the next generation.

[0091] Simulated binary crossover involves linearly combining the decision variables of paired individuals in the mating pool to generate offspring according to a preset distribution exponent, aiming to explore new solution spaces between parent solutions. Multinomial mutation, on the other hand, applies random perturbations to the decision variables of offspring individuals with a certain probability, and the perturbation amplitude decays exponentially with the number of generations, used for fine-tuning in the later stages of evolution and escaping local optima.

[0092] The termination condition can be reaching the maximum number of generations or the change in population fitness being less than a preset threshold. For example, after 500 iterations, the individual ranks in the population are stable and the crowding distance is evenly distributed. At this point, evolution stops, and the set of individuals at the first non-dominated frontier is the Pareto optimal solution set. It can efficiently search in complex nonlinear solution spaces, ensuring comprehensive optimality with short solidification time, low temperature rise, and low energy consumption, while also providing multiple parameter combinations for flexible selection in actual production.

[0093] The aforementioned technical solution treats each set of curing parameters as an individual and constructs a population. Utilizing a non-dominated sorting mechanism, it progressively divides the frontier levels, ensuring the discovery of a truly non-dominated solution among the three conflicting objectives of surface curing time, peak resin temperature rise, and total irradiation energy consumption. This avoids the suboptimal problem caused by subjective weighting settings in traditional single-objective weighted methods. Furthermore, a crowding distance calculation mechanism is introduced to effectively maintain the diversity of solution distribution within the same frontier level, preventing the algorithm from getting trapped in local optima or solution set clustering. This provides a rich library of solutions covering different focuses (such as rapid curing, low-temperature safety, and energy-saving). Through tournament selection, simulated binary crossover, and polynomial mutation co-evolutionary operations, the population is driven to globally search and gradually converge in the solution space. The resulting Pareto optimal solution set not only significantly shortens the curing time but also effectively controls resin temperature rise to prevent thermal damage, while reducing total energy consumption. This solves the problems of uneven curing and heat accumulation caused by traditional empirical parameter settings, improving the overall quality and stability of the stator impregnation process. By constructing a multi-objective optimization model and employing a non-dominated sorting genetic algorithm, intelligent optimization of the stator impregnation process parameters of an axial flux motor was achieved.

[0094] In one possible implementation, step S125, calculating the crowding distance for each individual based on the optimization objective of all individuals in the same Pareto frontier level, includes: In step S1251, for all individuals in the same Pareto front level, the surface curing time is taken as the first optimization target, and the first optimization target is sorted from smallest to largest. The crowding distance between the first individual and the last individual after sorting is set to a preset maximum value. Among them, the surface curing time can be the time required for the resin at the stator end to change from a liquid to a solid state. As the first optimization objective in multi-objective optimization, it directly reflects production efficiency. In this way, the sparseness of the distribution of individuals in the first optimization objective dimension can be quantified.

[0095] Specifically, the surface curing time values ​​of all individuals in the current Pareto front level are first extracted and arranged in ascending order. For boundary individuals in the sorted sequence, i.e., the first individual (curing fastest) and the last individual (curing slowest), their crowding distance component in that direction is directly assigned a preset maximum value, such as infinity or a very large constant, to ensure that boundary solutions are not discarded in subsequent selection operations, thereby maintaining the coverage of the solution set. For each individual in the middle of the sequence, the calculation of its crowding distance component depends on the numerical difference between its two adjacent individuals on that target.

[0096] In step S1252, for each individual in the middle, the difference between its two adjacent individuals on the first optimization target is calculated, and then divided by the difference between the maximum and minimum values ​​of the target within the front edge to obtain the crowding distance component in the direction of the first optimization target. For example, the difference in surface curing time between the preceding and following individuals is calculated, and this difference is divided by the difference between the maximum and minimum values ​​of the target within the current frontal line to achieve normalization, thereby eliminating the influence of dimensions and obtaining the congestion distance component in the first optimization target direction. For instance, if the maximum surface curing time within a certain frontal line is 120 seconds and the minimum is 60 seconds, with a difference of 60 seconds; for an individual in the middle, the curing time of the preceding individual is 80 seconds and that of the following individual is 90 seconds, then the component of this individual in the first target direction is (90-80) / 60≈0.167. Through normalization calculation, the local density of this individual in the curing time dimension can be objectively reflected; the larger the interval, the sparser the solution in that region.

[0097] In step S1253, the above sorting and difference normalization calculations are repeated sequentially with the peak temperature rise of the resin as the second optimization target and the total irradiation energy consumption as the third optimization target, to obtain the crowding distance component of each individual in the direction of the second optimization target and the third optimization target respectively. The peak resin temperature rise can be the highest temperature reached inside the resin during the curing process, used to assess the risk of heat accumulation; the total irradiation energy consumption can be the total energy consumption of the UV lamp array and infrared heater throughout the curing process, used to assess economic efficiency. This further evaluates the distribution of individual components from two dimensions: thermal safety and energy efficiency. Specifically, the sorting and normalization logic described above is reused, and the same operations are performed for both the second optimization objective (peak resin temperature rise) and the third optimization objective (total irradiation energy consumption).

[0098] For example, firstly, all individuals are sorted from smallest to largest according to the peak resin temperature rise, and the components of the first and last individuals are set to the maximum value. The normalized result of the adjacent difference of the middle individuals is calculated to obtain the crowding distance component of the second optimization target direction. Then, all individuals are sorted from smallest to largest according to the total irradiation energy consumption, and the components of the first and last individuals are set to the maximum value. The normalized result of the adjacent difference of the middle individuals is calculated to obtain the crowding distance component of the third optimization target direction.

[0099] For example, when calculating the component of the resin temperature rise peak direction, if an individual is located in a dense region with low temperature rise and surrounding individuals also having low temperature rises, the difference between its adjacent values ​​is small, resulting in a lower calculated component value. Conversely, if an individual has unique temperature rise characteristics and no similar solutions exist around it, its component value will be higher. This multi-dimensional independent evaluation mechanism ensures that the algorithm can comprehensively capture the distribution characteristics of individuals under different performance indicators, avoiding evaluation bias caused by a single indicator.

[0100] In step S1254, the sum of the crowding distance components in the three directions is calculated to obtain the crowding distance of the individual.

[0101] Crowding distance is a comprehensive indicator that measures the relative sparsity of an individual within the Pareto front, and its value directly determines the individual's retention priority in population evolution. Thus, the multidimensional distribution information obtained above can be integrated to generate the final evaluation scalar.

[0102] Specifically, the first optimized target direction component calculated in the first step, the second optimized target direction component calculated in the second step, and the third optimized target direction component calculated in the second step are arithmetically summed, and the sum is the final crowding distance of that individual. By linearly superimposing three orthogonal or approximately orthogonal target direction components, a volume concept in a multidimensional space can be constructed. The larger this volume, the fewer other solutions there are around that individual, and the stronger its uniqueness in the solution set.

[0103] For example, if an individual's component in the surface curing time direction is 0.5, its component in the resin temperature rise peak direction is 0.3, and its component in the total irradiation energy consumption direction is 0.4, then its total crowding distance is 1.2. This value will serve as an important basis for subsequent tournament selection operations, working together with the Pareto front level to determine the individual's survival, thereby guiding the population towards a more even distribution and greater diversity.

[0104] The aforementioned technical solution, by ranking and normalizing the difference calculations of three key indicators—surface curing time, peak resin temperature rise, and total irradiation energy consumption—and combining a head-and-tail protection strategy, not only preserves the extreme optimal solutions in each objective dimension but also significantly improves the diversity of solutions in the intermediate region. The cumulative synthesis of the three directional components allows the final crowding distance to comprehensively characterize the sparsity of individual solutions in the three-dimensional target space. This guides the algorithm to prioritize solutions that achieve a unique balance between curing efficiency, thermal safety, and energy consumption, preventing optimization results from getting trapped in local dense areas and providing more diverse and high-quality process parameter combinations for actual production. By utilizing the crowding distance mechanism in multi-objective optimization algorithms, the problem of uneven distribution of curing parameter solution sets in the axial flux motor stator impregnation process is effectively solved.

[0105] In one possible implementation, the step of performing tournament selection, simulated binary crossover, and polynomial mutation operations based on the Pareto front rank and crowding distance of each individual in the population, iteratively evolving until a termination condition is met to obtain a Pareto optimal solution set, includes: The tournament selection operation includes: randomly selecting multiple individuals from the current population each time, comparing their Pareto front levels, and the one with the highest Pareto front level wins; if the Pareto front levels are the same, comparing their crowding distance, and the one with the larger crowding distance wins; and placing the winning individual into the mating pool until the number of individuals in the mating pool reaches a preset number. Tournament selection is a key screening mechanism for constructing a high-quality offspring population. Individuals can be a combination of decision variables in a multi-objective optimization model, specifically including the UV irradiation intensity, irradiation duration, and thermocatalytic temperature of each micro-region to be cured. The Pareto front rank is determined based on the individual's hierarchy in the non-dominated ordination; a smaller rank value indicates better overall performance across the three optimization objectives: surface curing time, peak resin temperature rise, and total irradiation energy consumption. Crowding distance characterizes the sparsity of individual distribution within the same front rank and is obtained by calculating the sum of normalized distances between adjacent individuals in each optimization objective direction.

[0106] Specifically, a predetermined number (e.g., 2 or 3) of individuals are randomly selected from the population generated in the current iteration as candidate groups. First, the Pareto front ranks of each individual within the candidate group are compared, prioritizing the selection of the individual with the highest rank. When multiple individuals within a candidate group have the same highest rank, their crowding distances are further compared, selecting the individual with the largest crowding distance. This hierarchical comparison strategy ensures that high-quality genes (high ranks) are preferentially inherited, while preserving individuals with good diversity (large crowding distance) at the same quality level, preventing the algorithm from prematurely converging to a local optimum. The selected individuals are replicated and placed into a mating pool. This process is repeated until the total number of individuals in the mating pool reaches a predetermined number of parents, providing a basis for subsequent crossover and mutation operations.

[0107] Performing the simulated binary crossover operation includes: each time, linearly combining the decision variables of two individuals in the mating pool with a preset crossover probability, generating two corresponding offspring individuals according to a preset distribution index, until the number of offspring individuals reaches the preset number of offspring; Simulated binary crossover is a biomimetic recombination technique for optimizing continuous variables, aiming to generate new, potentially better solutions through the exchange of parental information. The decision variables are the aforementioned ultraviolet irradiation intensity, irradiation duration, and thermocatalytic temperature parameters. The preset crossover probability can be set to a relatively high value (e.g., 0.9) to ensure that most individuals participate in recombination. A preset distribution index (denoted as ηc) is used to control the dispersion of offspring individuals relative to their parents; the larger the distribution index, the closer the generated offspring are to their parents.

[0108] Specifically, two parent individuals are randomly paired from the mating pool. For each parent pair, a linear combination operation is performed on each of their corresponding decision variables with a preset probability. This operation is based on a simulated binary crossover formula, using a distribution index to generate random factors and calculate the decision variable values ​​for two new offspring individuals. For example, if parent A's irradiation intensity in a certain micro-region is IA and parent B's is IB, the crossover operation can generate two new intensity values, Ichild1 and Ichild2, that are between or slightly outside this range.

[0109] In this way, the parameter characteristics of rapid solidification strategies and low-energy consumption strategies can be integrated across different microregions to explore new solutions that take into account multiple performance indicators. This process is repeated until the total number of offspring individuals generated meets the preset offspring number requirement.

[0110] Performing the polynomial mutation operation includes: applying a random perturbation to each decision variable of the offspring individual with a mutation probability, the distribution exponent being one hundred, and the perturbation amplitude decreasing exponentially with the number of generations; iterating until the termination condition is met to obtain the Pareto optimal solution set.

[0111] The multinomial mutation operation is used to maintain population diversity and provide local fine-tuning capabilities, preventing the algorithm from stagnating. The mutation probability can be set to a low value (e.g., 1 / n, where n is the dimension of the decision variable) to ensure that only a small number of variables are fine-tuned. The distribution exponent is fixed at one hundred here, determining the initial distribution of the mutation perturbation. A larger distribution exponent means that the offset produced by mutation can be smaller, which is beneficial for fine-tuning near the parent solution. The perturbation amplitude decays exponentially with the number of generations, which allows for larger parameter fluctuations in the early stages of evolution to explore the solution space extensively, while reducing the fluctuation range in the later stages of evolution to focus on fine-tuning near the optimal solution.

[0112] Specifically, for each decision variable of each offspring individual, a random number is generated and compared with the mutation probability. If the mutation condition is met, the decay coefficient is calculated based on the current generation number, and a random perturbation value is applied to the technical parameter in combination with the distribution index. For example, in the 10th generation of evolution, the thermocatalytic temperature of a certain micro-region may undergo a significant jump from its original value to escape the local extremum; while in the 500th generation of evolution, the same mutation operation will only produce a small temperature correction, thereby achieving precise convergence of the curing process parameters.

[0113] The termination condition is the criterion for determining whether the evolutionary algorithm has ended its operation. It can include reaching the maximum number of generations, the change in population fitness being less than a preset threshold, or the computation time exceeding the limit. The Pareto optimal solution set is the set of non-dominated solutions that are finally retained after multiple rounds of selection, crossover, and mutation. Each solution represents the optimal balance between surface curing time, peak resin temperature rise, and total irradiation energy consumption.

[0114] Specifically, the offspring population after mutation is merged with the original parent population, and non-dominated sorting and crowding distance calculation are performed again to select a new generation of population, completing one full iteration cycle. This cycle is repeated continuously, with the quality of each generation of population gradually improving and the distribution becoming more uniform. When the termination condition is met, the iteration stops, and all non-dominated individuals in the current population are output as the final Pareto optimal solution set. Each solution in this solution set corresponds to a complete set of UV irradiation intensity, irradiation duration, and thermocatalytic temperature control parameters for all micro-regions to be cured at the stator end, in order to select the optimal scheme for execution.

[0115] The aforementioned technical solution, through a tournament selection mechanism that prioritizes comparing Pareto front ranks and, when ranks are equal, compares crowding distances, ensures that superior individuals (i.e., parameter combinations with good curing effects and low energy consumption) can quickly enter the next generation, while maintaining population diversity and preventing premature convergence. Building upon this, a simulated binary crossover operation is used to linearly combine the decision variables of parent individuals, promoting information exchange and fusion between different optimization strategies (such as high speed and low energy consumption), generating new solutions with smooth transitions and potential. A polynomial mutation operation with a distribution exponent of 100 and perturbation amplitude decaying exponentially with generation provides broad exploration capabilities in the early stages of evolution to break through local optima, and in the later stages, it provides refined local search capabilities to approximate the true Pareto front. A Pareto optimal solution set that achieves the best balance across the three dimensions of surface curing time, peak resin temperature rise, and total irradiation energy consumption can be obtained. Through the synergistic cooperation of tournament selection, simulated binary crossover, and polynomial mutation operations, an efficient and robust multi-objective optimization process is achieved.

[0116] See Figure 4As shown in the figure, this disclosure also provides an axial flux motor stator impregnation rapid shaping device 400, wherein the axial flux motor stator impregnation rapid shaping device 400 includes: a division module 410, configured to divide the surface of the stator end of the axial flux motor into multiple micro-regions to be cured based on the three-dimensional mesh model data of the stator end of the axial flux motor; and a construction module 420, configured to construct a multi-objective optimization model, wherein the multi-objective optimization model uses the ultraviolet irradiation intensity, irradiation duration, and thermocatalytic temperature of each micro-region to be cured as decision variables, and the surface curing time, resin temperature rise peak, and The total energy consumption of irradiation is used as the optimization objective to obtain a Pareto optimal solution set. Each Pareto optimal solution corresponds to a set of ultraviolet irradiation intensities, irradiation durations, and thermocatalytic temperatures of the plurality of micro-regions to be cured. The selection module 430 is configured to select the solution with the shortest surface curing time from the Pareto optimal solution set as the target solution. The control module 440 is configured to control the ultraviolet lamp array to perform partitioned scanning irradiation on the stator end and to control the infrared heater to perform partitioned heating on the stator end according to the ultraviolet irradiation intensities, irradiation durations, and thermocatalytic temperatures of the plurality of micro-regions to be cured corresponding to the target solution.

[0117] Optionally, the partitioning module 410 is configured to: classify the point cloud corresponding to the three-dimensional mesh model data of the stator end of the axial flux motor into different candidate micro-regions based on the three-dimensional mesh model data of the stator end of the axial flux motor; calculate the area of ​​each candidate micro-region and merge the candidate micro-regions with areas smaller than a preset area into the adjacent candidate micro-regions closest to their normal direction to obtain multiple micro-regions to be solidified; determine the spatial coordinates, normal vector and radius of curvature of the micro-region to be solidified based on the coordinates of all points in the multiple micro-regions to be solidified; assign a unique index number to each micro-region to be solidified and store the number in association with the corresponding spatial coordinates, normal vector and radius of curvature.

[0118] Optionally, the partitioning module 410 is configured to: calculate the normal vector and curvature of all points based on the three-dimensional mesh model data of the stator end of the axial flux motor; select the point with the smallest curvature from all the points as the seed point, and select the point closest to the seed point from the point cloud as the neighboring point based on the coordinates of all the points; determine the angle between the normal vector of the neighboring point and the normal vector of the seed point; if the angle between the normal vector of the neighboring point and the normal vector of the seed point is less than a preset angle and the neighboring point is not marked, then include the neighboring point in the candidate micro-region to which the current seed point belongs and use it as a new seed point; repeat the steps from the coordinates of all the points to the addition of the neighboring point to the candidate micro-region to which the current seed point belongs and use it as a new seed point until the condition for reselecting the seed point is met; select the next seed point with the smallest curvature from the unmarked points, and repeat the steps from the coordinates of all the points to the addition of the neighboring point to the candidate micro-region to which the current seed point belongs and use it as a new seed point until all points have been included in the candidate micro-region to which the seed point belongs.

[0119] Optionally, the partitioning module 410 is configured to: determine the coordinates of all points based on the three-dimensional mesh model data of the stator end of the axial flux motor; select neighborhood points for each point based on the coordinates of all points, wherein the neighborhood points are points with a preset radius corresponding to the point or the k points closest to the corresponding point; calculate the mean of the coordinates in each dimension based on the coordinates of each point and the coordinates of the corresponding neighborhood points to obtain the coordinates of the centroid of each point; determine the covariance matrix corresponding to each point based on the coordinates of each point, the coordinates of the corresponding neighborhood points, and the coordinates of the centroid; and determine the normal vector and curvature of all points based on the covariance matrix of each point.

[0120] Optionally, the partitioning module 410 is configured to: determine the average coordinate of all points in the plurality of micro-regions to be cured based on the coordinates of all points in the plurality of micro-regions to be cured; use the average coordinate as the spatial coordinate of the micro-region to be cured; determine the covariance matrix of the micro-region to be cured based on the coordinates of all points in each micro-region to be cured and the corresponding average coordinate; perform eigenvalue decomposition on the covariance matrix of each micro-region to be cured to obtain a plurality of eigenvalues ​​and a corresponding eigenvector of each micro-region to be cured; select the eigenvector corresponding to the smallest eigenvalue in each micro-region to be cured to determine the normal vector of the micro-region to be cured; and determine the radius of curvature of the micro-region to be cured as the ratio between the smallest eigenvalue in each micro-region to the sum of the plurality of eigenvalues ​​of the micro-region to be cured.

[0121] Optionally, the construction module 420 is configured such that the multi-objective optimization model uses a set of surface curing time, resin temperature rise peak, and total irradiation energy consumption as individuals to obtain a population; for each individual in the population, the number of times it is dominated by other individuals is calculated; wherein, domination means that any individual is not inferior to another individual on all optimization objectives and is strictly superior on at least one optimization objective; all individuals with zero dominations are assigned to the first non-dominated frontier; the individuals in the first non-dominated frontier are temporarily stored and removed from the population, and the remaining individuals in the population are... The number of times an individual is dominated by other individuals in the current population is repeatedly calculated, and all individuals with zero domination counts are assigned to the second non-dominated front. This process is repeated layer by layer until all individuals are assigned to the non-dominated front, resulting in the Pareto front level for each individual. Based on the optimization objectives of all individuals in the same Pareto front level, the crowding distance for each individual is calculated. Based on the Pareto front level and the corresponding crowding distance for each individual, tournament selection, simulated binary crossover, and polynomial mutation operations are performed, iteratively evolving until the termination condition is met, to obtain the Pareto optimal solution set.

[0122] Optionally, the construction module 420 is configured to: for all individuals in the same Pareto front level, sort the first optimization targets from smallest to largest using the surface curing time as the first optimization target, and set the crowding distance between the first and last individuals after sorting to a preset maximum value; for each individual in the middle, calculate the difference between its two adjacent individuals on the first optimization target, and then divide it by the difference between the maximum and minimum values ​​of the target within the front to obtain the crowding distance component in the first optimization target direction; repeat the above sorting and difference normalization calculation with the resin temperature rise peak as the second optimization target and the total irradiation energy consumption as the third optimization target, respectively, to obtain the crowding distance component of each individual in the second and third optimization target directions; calculate the sum of the crowding distance components in the three directions to obtain the crowding distance of the individual.

[0123] Optionally, the construction module 420 is configured to: perform the tournament selection operation including: randomly selecting multiple individuals from the current population each time, comparing their Pareto front levels, with the one with the highest Pareto front level winning; if the Pareto front levels are the same, comparing their crowding distances, with the one with the larger crowding distance winning; placing the winning individual into the mating pool until the number of individuals in the mating pool reaches a preset number of individuals; performing the simulated binary crossover operation including: linearly combining the decision variables of two individuals in the mating pool each time with a preset crossover probability, generating two corresponding offspring individuals according to a preset distribution index, until the number of offspring individuals reaches a preset number of offspring; performing the polynomial mutation operation including: applying a random perturbation to each decision variable of the offspring individuals with a mutation probability, the distribution index being one hundred, and the perturbation amplitude decreasing exponentially with the number of generations; iterating until the termination condition is met to obtain the Pareto optimal solution set.

[0124] Specific limitations regarding the rapid shaping device for axial flux motor stator impregnation can be found in the above-described limitations of the rapid shaping method for axial flux motor stator impregnation, and will not be repeated here. Each module in the aforementioned rapid shaping device for axial flux motor stator impregnation can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0125] This disclosure also provides an electronic device, including: a memory storing a computer program thereon; and a processor for executing the computer program in the memory to implement the steps of any of the methods described in the foregoing embodiments.

[0126] Figure 5 The axial flux motor stator impregnation rapid shaping device 100 shown includes a processor 1001 and a memory 1003. The processor 1001 and memory 1003 are connected, for example, via a bus 1002. Optionally, the axial flux motor stator impregnation rapid shaping device 100 may further include a communication component, which can be used for data interaction between the device 100 and other devices, such as sending or receiving data. It should be noted that in actual scheduling, the communication component is not limited to one, and the structure of this axial flux motor stator impregnation rapid shaping device 100 does not constitute a limitation on the embodiments of this application.

[0127] Processor 1001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 1001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0128] Bus 1002 may include a pathway for transmitting information between the aforementioned components. Bus 1002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 1002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0129] The memory 1003 may be ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium that can be used to carry or store program code and can be read by a computer, without limitation herein.

[0130] The memory 1003 is used to store program code for executing embodiments of the present disclosure, and its execution is controlled by the processor 1001. The processor 1001 is used to execute the program code stored in the memory 1003 to implement the steps shown in the aforementioned embodiments of the rapid shaping method for impregnating the stator of an axial flux motor.

[0131] This disclosure also provides a computer-readable storage medium storing program code. When the program code is executed by a processor, it can implement the steps and corresponding content of the aforementioned embodiment of the rapid shaping method for impregnating the stator of an axial flux motor.

[0132] The preferred embodiments of the present disclosure have been described in detail above with reference to the accompanying drawings. However, the present disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present disclosure, various changes, modifications, substitutions and variations can be made to these embodiments, and all such changes, modifications, substitutions and variations fall within the protection scope of the present disclosure.

[0133] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction, and such combinations should also be considered as part of this disclosure. To avoid unnecessary repetition, this disclosure will not further describe the various possible combinations. The technical scope of this application is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A rapid shaping method for the stator of an axial flux motor by impregnation, characterized in that, The method includes: Based on the three-dimensional mesh model data of the stator end of the axial flux motor, the surface of the stator end is divided into multiple micro-regions to be cured; A multi-objective optimization model is constructed, wherein the ultraviolet irradiation intensity, irradiation duration and thermocatalytic temperature of each micro-region to be cured are used as decision variables, and the surface curing time, resin temperature rise peak and total irradiation energy consumption are used as optimization objectives to obtain a Pareto optimal solution set. Each Pareto optimal solution corresponds to a set of ultraviolet irradiation intensity, irradiation duration and thermocatalytic temperature of the multiple micro-regions to be cured. The solution with the shortest surface curing time is selected from the Pareto optimal solution set as the target solution; Based on the UV irradiation intensity, irradiation duration, and thermal catalytic temperature of the multiple micro-regions to be cured corresponding to the target solution, the UV lamp array is controlled to perform zoned scanning irradiation on the stator end, and the infrared heater is controlled to perform zoned heating on the stator end.

2. The method according to claim 1, characterized in that, The step involves dividing the surface of the stator end of the axial flux motor into multiple micro-regions to be cured based on the three-dimensional mesh model data of the stator end, including: Based on the three-dimensional mesh model data of the stator end of the axial flux motor, the point cloud corresponding to the three-dimensional mesh model data is classified into different candidate micro-regions; Calculate the area of ​​each candidate micro-region, and merge the candidate micro-regions with areas smaller than a preset area into the adjacent candidate micro-regions that are closest to their normal direction to obtain multiple micro-regions to be cured; Based on the coordinates of all points in the plurality of micro-regions to be cured, determine the spatial coordinates, normal vector, and radius of curvature of the corresponding micro-region to be cured; Each micro-region to be cured is assigned a unique index number, and the number is associated with and stored in relation to the corresponding spatial coordinates, normal vector, and radius of curvature.

3. The method according to claim 2, characterized in that, The step of classifying the point cloud corresponding to the three-dimensional mesh model data of the stator end of the axial flux motor into different candidate micro-regions includes: Based on the three-dimensional mesh model data of the stator end of the axial flux motor, calculate the normal vector and curvature of all points; Select the point with the smallest curvature from all the points as the seed point, and select the point closest to the seed point from the point cloud as the neighboring point based on the coordinates of all the points. Determine the angle between the normal vector of the adjacent point and the normal vector of the seed point; If the angle between the normal vector of the adjacent point and the normal vector of the seed point is less than a preset angle and the adjacent point is not marked, then the adjacent point is included in the candidate micro-region to which the current seed point belongs and is used as a new seed point. Repeat the steps from determining the coordinates of all the points to adding the adjacent point to the candidate micro-region to which the current seed point belongs and using it as a new seed point, until the condition for reselecting a seed point is met; Select the next seed point with the smallest curvature from the unmarked points, and repeat the steps from the coordinates of all the points to the inclusion of the adjacent point into the candidate micro-region to which the current seed point belongs and as the new seed point, until all points have been included into the candidate micro-region to which the seed point belongs.

4. The method according to claim 3, characterized in that, The step of determining the normal vector and curvature of all points based on the three-dimensional mesh model data of the stator end of the axial flux motor includes: Based on the three-dimensional mesh model data of the stator end of the axial flux motor, the coordinates of all points are determined, and based on the coordinates of all points, the neighborhood points of each point are selected, wherein the neighborhood points are points with a preset radius of the corresponding point or the k points closest to the corresponding point; Based on the coordinates of each point and the coordinates of the corresponding neighborhood point, the mean of the coordinates in each dimension is calculated to obtain the coordinates of the centroid of each point; Based on the coordinates of each point, the coordinates of the corresponding neighborhood point, and the coordinates of the centroid, determine the covariance matrix corresponding to each point; Based on the covariance matrix of each point, determine the normal vector and curvature of all points.

5. The method according to claim 2, characterized in that, The step of determining the spatial coordinates, normal vector, and radius of curvature of the corresponding micro-region based on the coordinates of all points in the plurality of micro-regions to be cured includes: Based on the coordinates of all points in the plurality of micro-regions to be cured, determine the average coordinate value of all points in the corresponding micro-regions to be cured; The average value of the coordinates is used as the spatial coordinates of the micro-region to be solidified. Based on the coordinates of all points in each micro-region to be cured and the corresponding average coordinates, determine the covariance matrix of the micro-region to be cured. Eigenvalue decomposition is performed on the covariance matrix of each micro-region to be cured to obtain multiple eigenvalues ​​and corresponding eigenvectors for each micro-region to be cured. Select the eigenvector corresponding to the smallest eigenvalue in each micro-region to be cured, and determine the normal vector corresponding to the micro-region to be cured. The ratio between the smallest feature value in each of the micro-regions to be cured and the sum of the plurality of feature values ​​of the corresponding micro-region is determined as the radius of curvature of the corresponding micro-region.

6. The method according to any one of claims 1-5, characterized in that, The multi-objective optimization model uses the UV irradiation intensity, irradiation duration, and thermocatalytic temperature of each micro-region to be cured as decision variables, and the surface curing time, peak resin temperature rise, and total irradiation energy consumption as optimization objectives, to obtain a Pareto optimal solution set, including: The multi-objective optimization model uses a set of the surface curing time, the peak temperature rise of the resin, and the total irradiation energy consumption as individuals to obtain a population. For each individual in the population, the number of times it is dominated by other individuals is calculated; wherein, domination is defined as any individual being no worse than another individual on all optimization objectives and being strictly better on at least one optimization objective. All individuals whose dominance frequency is zero are assigned to the first non-dominated frontier; The individuals in the first non-dominated front are temporarily stored and removed from the population. The number of times each individual in the population is dominated by other individuals in the current population is repeatedly calculated. All individuals whose number of dominations is zero are assigned to the second non-dominated front. This process continues layer by layer until all individuals are assigned to the non-dominated frontier, thus obtaining the Pareto frontier level for each individual. The crowding distance for each individual is calculated based on the optimization objective of all individuals in the same Pareto front level. Based on the Pareto front level and the corresponding crowding distance of each individual, tournament selection, simulated binary crossover, and polynomial mutation operations are performed, iteratively evolving until the termination condition is met to obtain the Pareto optimal solution set.

7. The method according to claim 6, characterized in that, The calculation of the crowding distance for each individual based on the optimization objective of all individuals in the same Pareto frontier level includes: For all individuals in the same Pareto front level, the surface curing time is taken as the first optimization target. The first optimization target is sorted from smallest to largest, and the crowding distance between the first and last individuals after sorting is set to a preset maximum value. For each individual in the middle, the difference between its two adjacent individuals on the first optimization target is calculated, and then divided by the difference between the maximum and minimum values ​​of the target within the front to obtain the crowding distance component in the direction of the first optimization target. The above sorting and difference normalization calculations were repeated, taking the peak temperature rise of the resin as the second optimization target and the total irradiation energy consumption as the third optimization target, to obtain the crowding distance component of each individual in the direction of the second and third optimization targets respectively. The sum of the crowding distance components in the three directions is calculated to obtain the crowding distance of the individual.

8. The method according to claim 6, characterized in that, The process involves performing tournament selection, simulated binary crossover, and polynomial mutation operations based on the Pareto front rank and crowding distance of each individual in the population, iteratively evolving until a termination condition is met to obtain a Pareto optimal solution set, including: The tournament selection operation includes: randomly selecting multiple individuals from the current population each time, comparing their Pareto front levels, and the one with the highest Pareto front level wins; if the Pareto front levels are the same, comparing their crowding distance, and the one with the larger crowding distance wins; and placing the winning individual into the mating pool until the number of individuals in the mating pool reaches a preset number. Performing the simulated binary crossover operation includes: each time, linearly combining the decision variables of two individuals in the mating pool with a preset crossover probability, generating two corresponding offspring individuals according to a preset distribution index, until the number of offspring individuals reaches the preset number of offspring; Performing the polynomial mutation operation includes: applying a random perturbation to each decision variable of the offspring individual with a mutation probability, the distribution exponent being one hundred, and the perturbation amplitude decreasing exponentially with the number of generations. The process iterates until the termination condition is met, thus obtaining the Pareto optimal solution set.

9. A rapid shaping device for axial flux motor stator impregnation, characterized in that, The device includes: The partitioning module is configured to divide the surface of the stator end of the axial flux motor into multiple micro-regions to be cured based on the three-dimensional mesh model data of the stator end. The construction module is configured to construct a multi-objective optimization model. The multi-objective optimization model takes the UV irradiation intensity, irradiation duration, and thermocatalytic temperature of each micro-region to be cured as decision variables, and takes the surface curing time, resin temperature rise peak, and total irradiation energy consumption as optimization objectives to obtain a Pareto optimal solution set. Each Pareto optimal solution corresponds to a set of UV irradiation intensity, irradiation duration, and thermocatalytic temperature of the multiple micro-regions to be cured. The selection module is configured to select the solution with the shortest surface curing time from the Pareto optimal solution set as the target solution; The control module is configured to control the ultraviolet lamp array to perform partitioned scanning irradiation on the stator end based on the ultraviolet irradiation intensity, irradiation duration and thermal catalytic temperature of the plurality of micro-regions to be cured corresponding to the target solution, and to control the infrared heater to perform partitioned heating on the stator end.

10. An electronic device, characterized in that, include: A memory on which computer programs are stored; A processor for executing the computer program in the memory to implement the steps of the method according to any one of claims 1-8.

Images