Three-dimensional modeling method and system for automobile battery plastic injection mold

Abstract

The application discloses a kind of three-dimensional modeling method and system of automobile battery plastic injection mold, it is related to three-dimensional modeling technical field, comprising: deployment first-order matrix anchor network, obtain multiple reference anchor points, and construct lightweight component model;With mold demolding direction as the orthogonal projection direction of virtual camera carries out depth rendering, obtains global demolding occlusion depth map;Multiple potential processing feature points are projected into global demolding occlusion depth map and carry out multi-view demolding visibility test, filter out conflict feature points and obtain clear interference anchor network;According to the multiple filling error values of each cavity, high-risk cavity identification is carried out, and multiple defect cavity grid areas are obtained, back projection to blank virtual three-dimensional space, and construct three-dimensional production model.The present application solves the technical problems of low modeling efficiency and difficulty in accurately identifying demolding occlusion interference in the prior art, and achieves the technical effects of improving the three-dimensional modeling efficiency and demolding interference identification accuracy of multi-cavity injection mold.

Description

Technical Field

[0001] This invention relates to the field of 3D modeling technology, specifically to a 3D modeling method and system for injection molds of automotive battery plastic parts. Background Technology

[0002] As automotive battery plastic parts become more mass-produced and complex, injection molds often adopt multi-cavity layouts to improve production efficiency. However, the 3D modeling process for multi-cavity injection molds requires the simultaneous arrangement of multiple cavities, configuration of machining features, and verification of demolding direction. Due to the large number of cavities, their dense spatial distribution, and the mutual influence of local structures, relying on cavity-by-cavity modeling and manual inspection can lead to a long modeling cycle. Furthermore, issues such as local feature occlusion, projection overlap, or demolding path interference can easily occur in the demolding direction, making it difficult to quickly and accurately identify the occlusion and interference positions. This affects the efficiency and manufacturability of the 3D production model of the mold. Summary of the Invention

[0003] This application provides a three-dimensional modeling method and system for injection molds of automotive battery plastic parts, which is used to solve the technical problems of low modeling efficiency and difficulty in accurately identifying demolding obstruction interference in the prior art for multi-cavity injection molds.

[0004] In view of the above problems, this application provides a three-dimensional modeling method and system for injection molds of automotive battery plastic parts.

[0005] A first aspect of this application provides a method for three-dimensional modeling of injection molds for automotive battery plastic parts, the method comprising: In the virtual 3D space of the mold blank, a first-order matrix anchor point network is deployed according to the production cavity number requirements to obtain multiple reference anchor points, and a lightweight component model of a single automotive battery plastic part cavity is constructed. The demolding direction of the mold is defined as the orthogonal projection direction of the virtual camera. The hardware rasterization pipeline of the graphics processor is used to perform depth rendering on the multiple reference anchor points and the lightweight component model in a single-channel depth buffer to obtain a global demolding occlusion depth map. Multiple potential processing feature points in the multiple reference anchor points are projected onto the global demolding occlusion depth map to perform multi-view demolding visibility testing, filter out conflicting feature points that cause occlusion interference, and obtain an interference clearing anchor point network. Multiple filling error values ​​of each cavity in the interference clearing anchor point network are obtained under multi-cavity injection fluid simulation analysis. Based on the multiple filling error values, high-risk cavities are identified in the interference clearing anchor point network to obtain multiple defective cavity mesh regions. The multiple defective cavity mesh regions are back-projected onto the virtual 3D space of the mold blank to construct a 3D production model.

[0006] A second aspect of this application provides a three-dimensional modeling system for injection molds of automotive battery plastic parts, the system comprising: The network deployment module is used to deploy a first-order matrix anchor point network in the virtual 3D space of the mold blank according to the production cavity number requirements, obtain multiple reference anchor points, and construct a lightweight component model of a single automotive battery plastic part cavity; the depth rendering module is used to define the mold release direction of the mold as the orthogonal projection direction of the virtual camera, and use the hardware rasterization pipeline of the graphics processor to perform depth rendering on the multiple reference anchor points and the lightweight component model in a single-channel depth buffer to obtain a global release occlusion depth map; the visibility testing module is used to test multiple potential additions among the multiple reference anchor points. The feature points are projected onto the global demolding occlusion depth map for multi-view demolding visibility testing, filtering out conflicting feature points that cause occlusion interference to obtain a network of interference-clearing anchor points; the identification module is used to obtain multiple filling error values ​​of each cavity in the interference-clearing anchor point network under multi-cavity injection fluid simulation analysis, and to identify high-risk cavities based on the multiple filling error values ​​to obtain multiple defective cavity mesh regions; the model construction module is used to back-project the multiple defective cavity mesh regions onto the virtual three-dimensional space of the mold blank to construct a three-dimensional production model.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application deploys a first-order matrix anchor point network in the virtual 3D space of the mold blank according to the production cavity number requirements to obtain multiple reference anchor points and constructs a lightweight component model of a single automotive battery plastic part cavity. The demolding direction of the mold is defined as the orthogonal projection direction of the virtual camera. The hardware rasterization pipeline of the graphics processor is used to perform depth rendering on the multiple reference anchor points and the lightweight component model in a single-channel depth buffer to obtain a global demolding occlusion depth map. Multiple potential processing feature points in the multiple reference anchor points are projected onto the global demolding occlusion depth map to perform multi-view demolding visibility testing, filtering out conflicting feature points that cause occlusion interference to obtain an interference clearing anchor point network. Multiple filling error values ​​of each cavity in the interference clearing anchor point network are obtained under multi-cavity injection fluid simulation analysis. Based on the multiple filling error values, the interference clearing anchor point network is used to identify high-risk cavities to obtain multiple defective cavity mesh regions. The multiple defective cavity mesh regions are back-projected onto the virtual 3D space of the mold blank to construct a 3D production model. This invention addresses the technical problems of low modeling efficiency and difficulty in accurately identifying demolding interference in existing multi-cavity injection molds. By constructing an anchor point network and combining it with depth rendering for demolding visibility testing, the invention achieves the technical effect of improving the efficiency of 3D modeling of multi-cavity injection molds and the accuracy of demolding interference identification. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A schematic diagram of a three-dimensional modeling method for injection molds for automotive battery plastic parts provided in this application embodiment; Figure 2 A multilayer perceptron training convergence diagram in a three-dimensional modeling method for automotive battery plastic injection molds provided in this application embodiment; Figure 3 This is a schematic diagram of a three-dimensional modeling system for an injection mold of automotive battery plastic parts, provided as an embodiment of this application.

[0010] Figure labeling: Network deployment module 11, depth rendering module 12, visibility testing module 13, recognition module 14, model building module 15. Detailed Implementation

[0011] This application provides a 3D modeling method and system for automotive battery plastic injection molds, addressing the technical problems of low modeling efficiency and difficulty in accurately identifying demolding interference in existing multi-cavity injection molds. By constructing an anchor point network and combining it with depth rendering for demolding visibility testing, the method achieves the technical effect of improving the efficiency of 3D modeling of multi-cavity injection molds and the accuracy of demolding interference identification.

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0013] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0014] Example 1, as Figure 1 As shown, this application provides a three-dimensional modeling method for injection molds of automotive battery plastic parts, the method comprising: Step S100: In the virtual three-dimensional space of the mold blank, deploy a first-order matrix anchor point network according to the production cavity number requirement to obtain multiple reference anchor points and construct a lightweight component model of a single automotive battery plastic part cavity.

[0015] In this embodiment, a three-dimensional coordinate environment consistent with the actual mold blank size is first established in the virtual three-dimensional space of the mold blank. The mold blank length, width, height, parting surface height, demolding direction, cavity layout area boundary, mold blank edge reserved distance, runner reserved distance, and the external dimensions of a single automotive battery plastic part cavity are recorded. The total number of cavities is determined according to the production cavity requirements, and the number of rows and columns of the first-order matrix anchor point network is set according to the total number of cavities. Then, based on the lateral dimension, longitudinal dimension, safety distance between adjacent cavities, and runner reserved distance of a single automotive battery plastic part cavity, the center-to-center distance between adjacent anchor points in the lateral and longitudinal directions is calculated. Subsequently, the first placement point in the virtual 3D space of the mold blank is selected as the starting position of the first-order matrix anchor point network. Anchor point positions are generated point by point in row and column order. For each anchor point position generated, the anchor point position is matched and verified with the outer dimensions of the cavity to confirm that the outer envelope of the placed cavity is within the boundary of the cavity placement area and does not enter the reserved distance of the mold blank edge and the safe distance range between adjacent cavities. The verified anchor point position is recorded as the reference anchor point, and the 3D coordinates, row number, column number, corresponding cavity number, demolding direction parameters and the distance between adjacent reference anchor points are recorded until multiple reference anchor points consistent with the number of cavities required for production are obtained.

[0016] After obtaining multiple reference anchor points, the original three-dimensional geometric data of a single automotive battery plastic part cavity is imported, and structural filtering and geometric simplification are performed on this original three-dimensional geometric data. Specifically, the cavity forming surface, parting surface boundary, outer contour of the demolding direction, gate connection area, runner connection area, mounting positioning surface, and machining feature boundary are retained; small chamfers, small fillets, shallow textures, small grooves, and non-molded decorative surfaces with dimensions below a preset threshold are deleted, and adjacent coplanar surfaces are merged. Continuous curved surfaces are resampled according to the preset surface accuracy to generate a cavity model with reduced geometric data. Subsequently, local reference positions are set on this cavity model, and the local reference positions are set as the cavity center points. The local demolding direction is adjusted to be consistent with the demolding direction in the virtual three-dimensional space of the mold blank. Then, the outer envelope dimensions, local reference positions, gate connection positions, runner connection positions, and parting surface boundary data of the cavity model are recorded to form a lightweight component model of a single automotive battery plastic part cavity.

[0017] Step S200: Define the demolding direction of the mold as the orthogonal projection direction of the virtual camera, and use the hardware rasterization pipeline of the graphics processor to perform depth rendering on the multiple reference anchor points and lightweight component models in a single-channel depth buffer to obtain a global demolding occlusion depth map.

[0018] In this embodiment, when defining the demolding direction of the mold as the orthogonal projection direction of the virtual camera, the three directional components of the demolding direction in three-dimensional coordinates are first read. The three directional components are squared and added together, and the square root of the sum is taken to obtain the length of the demolding direction. The three directional components are then divided by the demolding direction length to obtain the normalized demolding direction. Next, a reference direction that is not parallel to the normalized demolding direction is selected. The reference direction is cross-producted with the normalized demolding direction and normalized to obtain the first projection direction in the virtual camera image. The normalized demolding direction is cross-producted with the first projection direction and normalized to obtain the second projection direction in the virtual camera image. The orthogonal projection parameters of the virtual camera are determined by the normalized demolding direction, the first projection direction, and the second projection direction.

[0019] Read the 3D coordinates of multiple reference anchor points and all vertex coordinates of the lightweight component model. For each 3D coordinate, perform a dot product with the first projection direction, the second projection direction, and the normalized demodulation direction to obtain the horizontal, vertical, and depth values ​​of that 3D coordinate in the orthographic projection image. Calculate the minimum and maximum values ​​among all horizontal projection values, and use the difference as the orthographic projection width; calculate the minimum and maximum values ​​among all vertical projection values, and use the difference as the orthographic projection height; calculate the minimum and maximum values ​​among all depth values, and use them as the near and far values ​​of the depth range. Divide the orthographic projection width by the preset pixel precision and round up to obtain the pixel width of the single-channel depth buffer; divide the orthographic projection height by the preset pixel precision and round up to obtain the pixel height of the single-channel depth buffer, and initialize each pixel position within the single-channel depth buffer to the far-end depth value.

[0020] Multiple reference anchor points are converted into point geometry data, and lightweight component models are converted into vertex and triangle facet data, which are then submitted to the graphics processor's hardware rasterization pipeline. The hardware rasterization pipeline orthogonally projects the vertex data, mapping each vertex to a pixel position in a single-channel depth buffer. When rasterizing the triangle facet data, the pixel range it covers is determined based on the projected positions of the three vertices of the triangle facet. For any pixel position within this range, the centroid weights of that pixel position relative to the three vertices of the triangle facet are calculated. Specifically, by connecting the three vertices of the triangle facet to the pixel position, three sub-triangles are obtained. The area of ​​each sub-triangle is divided by the area of ​​the original triangle facet to obtain the centroid weight of the corresponding vertex. The sum of the three centroid weights is 1. The depth values ​​of the three vertices are then multiplied by their corresponding centroid weights and summed to obtain the current depth value of the pixel position. The current depth value is compared with the depth value already saved at the same pixel position in the single-channel depth buffer. If the current depth value is closer to the virtual camera, it is written to the current depth value; otherwise, the original depth value is retained. After all the reference anchor points and lightweight component models have completed depth rendering, the final depth values ​​in the single-channel depth buffer are read in pixel row and column order to obtain the global demodulation occlusion depth map.

[0021] Step S300: Project multiple potential processing feature points from the multiple reference anchor points onto the global demolding occlusion depth map to perform multi-view demolding visibility test, filter out conflicting feature points that cause occlusion interference, and obtain the interference-clearing anchor point network.

[0022] In this embodiment, when projecting multiple potential processing feature points from multiple reference anchor points onto a global demodulation occlusion depth map for multi-view demodulation visibility testing, a single CUDA kernel synchronously performs parallel fusion calculations of view frustum culling and occlusion culling on multiple potential processing feature points, converting the potential processing feature points within the effective projection field of view into multiple device pixel coordinates; then, based on the mapping of multiple device pixel coordinates in the global demodulation occlusion depth map, multiple hardware linear camera spatial depths at corresponding positions are read; then, the linear depth of each potential processing feature point is compared point by point with the corresponding hardware linear camera spatial depth, filtering out conflicting feature points whose linear depth is greater than the hardware linear camera spatial depth, thereby obtaining a network of cleared interference anchor points.

[0023] Furthermore, in the method provided in the application embodiment, projecting multiple potential processing feature points from the plurality of reference anchor points onto the global demolding occlusion depth map to perform multi-view demolding visibility testing, filtering out conflicting feature points that generate occlusion interference, and obtaining an interference-clearing anchor point network, further includes: The frustum culling and occlusion culling of the multiple potential processing feature points are performed in parallel and fused using a single CUDA kernel to obtain multiple device pixel coordinates. Multiple hardware linear camera space depths are then mapped and extracted from the global demodulation occlusion depth map at the positions corresponding to the multiple device pixel coordinates. The linear depths of the multiple potential processing feature points are compared with the corresponding hardware linear camera space depths in the multiple hardware linear camera space depths, and conflicting feature points whose linear depths are greater than the hardware linear space depths are filtered out to obtain the interference-clearing anchor network.

[0024] In this embodiment, when performing parallel fusion calculations of frustum culling and occlusion culling for multiple potential processing feature points using a single CUDA kernel, multiple potential processing feature points are first extracted from the lightweight component models corresponding to multiple reference anchor points. These potential processing feature points include cavity forming surface boundary points, gate connection points, runner connection points, venting groove preset points, cooling water channel adjacent points, and processing hole / groove boundary points. The three-dimensional coordinates, the reference anchor point number, and the feature type of each potential processing feature point are recorded. Subsequently, the three-dimensional coordinates, virtual camera position, orthogonal projection lateral direction, orthogonal projection longitudinal direction, mold demolding direction, orthogonal projection lateral range, orthogonal projection longitudinal range, depth range determined during the generation of the global demolding occlusion depth map, and the pixel width and pixel height of the global demolding occlusion depth map are written to the graphics processor's video memory. A CUDA thread is allocated for each potential processing feature point within a single CUDA kernel. Each CUDA thread processes the three-dimensional coordinates of the potential processing feature points... Subtracting the virtual camera position from the dimensional coordinates yields the position difference relative to the virtual camera. This position difference is then multiplied by the horizontal direction of the orthographic projection, the vertical direction of the orthographic projection, and the mold's demolding direction to obtain the horizontal projection coordinates, vertical projection coordinates, and the mold's linear depth. During frustum culling, the horizontal projection coordinates are compared with the horizontal range of the orthographic projection, the vertical projection coordinates are compared with the vertical range of the orthographic projection, and the mold's linear depth is compared with the depth range. When all three are within their corresponding ranges, the device pixel coordinates are calculated. The horizontal device pixel coordinates are the horizontal projection coordinates minus the initial value of the horizontal range of the orthographic projection, divided by the length of the horizontal range of the orthographic projection, multiplied by the pixel width of the global demolding occlusion depth map, and rounded to the nearest integer. The vertical device pixel coordinates are the vertical projection coordinates minus the initial value of the vertical range of the orthographic projection, divided by the length of the vertical range of the orthographic projection, multiplied by the pixel height of the global demolding occlusion depth map, and rounded to the nearest integer. This process yields multiple device pixel coordinates. Next, when mapping and extracting multiple hardware linear camera space depths corresponding to multiple device pixel coordinates in the global occlusion depth map, each CUDA thread in a single CUDA kernel accesses the global occlusion depth map using the corresponding horizontal and vertical device pixel coordinates. Specifically, the vertical device pixel coordinate is multiplied by the pixel width of the global occlusion depth map, and then the horizontal device pixel coordinate is added to obtain the one-dimensional read position of that device pixel coordinate in the global occlusion depth map. Subsequently, the CUDA thread reads the depth value stored in the single-channel depth buffer according to this one-dimensional read position. This depth value is the hardware linear camera space depth written by the graphics processor's hardware rasterization pipeline at the corresponding device pixel coordinate. After multiple CUDA threads perform the above reading process in parallel on multiple device pixel coordinates, multiple hardware linear camera space depths corresponding one-to-one with multiple device pixel coordinates are obtained.

[0025] After obtaining the linear depths of multiple potential processing feature points and the corresponding hardware linear camera space depths, a preset demolding safety margin is first obtained, and the preset demolding safety margin is superimposed on the hardware linear camera space depths corresponding to the pixel coordinates of each device to obtain the limit visible depth value of each device pixel coordinate. Then, the multiple linear depths of multiple potential processing feature points in the camera space are compared with the limit visible depth values ​​of the corresponding device pixel coordinates. If the linear depth is greater than the corresponding limit visible depth value, it is determined that the potential processing feature point has demolding occlusion interference and is cleared and removed. If the linear depth is not greater than the corresponding limit visible depth value, the potential processing feature point is retained. Finally, the retained potential processing feature points are used to update and form the interference clearing anchor point network.

[0026] Furthermore, in the method provided in the application embodiments, comparing the linear depths of multiple potential processing feature points with the corresponding hardware linear camera spatial depths in multiple hardware linear camera spatial depths, filtering out conflicting feature points whose linear depths are greater than the hardware linear spatial depths, and obtaining an interference-clearing anchor point network, further includes: Obtain a preset demolding safety margin, and add multiple hardware linear camera space depths to it to obtain the limit visible depth value of each device pixel coordinate; if the multiple linear depths of the multiple potential processing feature points in the camera space are greater than the limit visible depth value of the corresponding device pixel coordinate position, then it is determined that the current potential processing feature point causes demolding occlusion interference and is cleared and removed; otherwise, it is retained.

[0027] In this embodiment, when obtaining the preset demolding safety margin, the safety allowance corresponding to the demolding direction is first read from the mold design parameters. This preset demolding safety margin is the minimum depth interval at which potential processing feature points are allowed to approach the occluding surface in the demolding direction. Subsequently, according to the index order of multiple device pixel coordinates, the hardware linear camera spatial depth corresponding to each device pixel coordinate is read, and the preset demolding safety margin is added to the corresponding hardware linear camera spatial depth to obtain the limit visible depth value of each device pixel coordinate. The hardware linear camera spatial depth represents the depth of the occluding surface first observed by the virtual camera at the device pixel coordinate in the global demolding occlusion depth map, and the limit visible depth value represents the maximum depth position at which potential processing feature points are allowed to exist at the device pixel coordinate after considering the preset demolding safety margin.

[0028] After obtaining the limit visible depth values ​​of each device pixel coordinate, according to the correspondence between potential processing feature points and device pixel coordinates, the linear depths of multiple potential processing feature points in camera space are read one by one, and the linear depth of each potential processing feature point is compared with the limit visible depth value of the corresponding device pixel coordinate position. If the linear depth of the current potential processing feature point is greater than the limit visible depth value of the corresponding device pixel coordinate position, it means that the potential processing feature point is located behind the occlusion surface in the demolding direction and exceeds the preset demolding safety margin. It is determined that the current potential processing feature point causes demolding occlusion interference, and it is removed from the set of potential processing feature points associated with the corresponding reference anchor point. If the linear depth of the current potential processing feature point is less than or equal to the limit visible depth value of the corresponding device pixel coordinate position, the potential processing feature point is retained. After completing the above comparison and removal for multiple potential processing feature points, the retained potential processing feature points continue to be attached to the corresponding multiple reference anchor points, thereby obtaining the interference removal anchor point network.

[0029] Step S400: Obtain multiple filling error values ​​of each cavity in the interference removal anchor point network under multi-cavity injection fluid simulation analysis, and identify high-risk cavities based on the multiple filling error values ​​to obtain multiple defective cavity mesh regions.

[0030] In this embodiment, to obtain multiple filling error values ​​for each cavity in the interference-clearing anchor point network under multi-cavity injection molding fluid simulation analysis, firstly, according to the positional relationship of each reference anchor point in the interference-clearing anchor point network, the lightweight component model corresponding to each reference anchor point is determined as a cavity, and each cavity is assigned a cavity number; then, the geometric data of each cavity, main runner, branch runner, and gate are imported, and the melt temperature, mold temperature, injection pressure, injection speed, material viscosity, holding pressure, and holding time are set, and the main runner, branch runner, gate, and each cavity are meshed, so that each mesh unit records the cavity number to which it belongs; during the multi-cavity injection molding fluid simulation analysis, the flow state of the molten plastic after entering the branch runner from the main runner and then entering each cavity through the gate is calculated according to the set time step, and the inlet pressure, end pressure, volume shrinkage rate, and shear stress of each cavity are recorded; after the simulation is completed, the local pressure drop error of each cavity is calculated, first by subtracting the cavity inlet pressure from the local pressure drop error. The pressure at the end of the cavity is used to obtain the local pressure drop of the cavity. Then, the average local pressure drop of all cavities is calculated. The difference between the local pressure drop of the cavity and the average value is taken as the absolute value to obtain the local pressure drop error of the cavity. When calculating the volume shrinkage rate error for each cavity, the average volume shrinkage rate of all grid elements in the cavity is first calculated to obtain the volume shrinkage rate of the cavity. Then, the average volume shrinkage rate of all cavities is calculated. The difference between the volume shrinkage rate of the cavity and the average value is taken as the absolute value to obtain the volume shrinkage rate error of the cavity. When calculating the shear stress imbalance for each cavity, the maximum and minimum values ​​of the shear stress of all grid elements in the cavity are extracted. The maximum value is subtracted from the minimum value to obtain the shear stress imbalance of the cavity. Finally, according to the cavity number, the local pressure drop error, volume shrinkage rate error, and shear stress imbalance are written into the data record of the corresponding cavity in the interference clearing anchor point network to obtain multiple filling error values ​​of each cavity in the multi-cavity injection molding fluid simulation analysis in the interference clearing anchor point network.

[0031] Next, when identifying high-risk cavities based on multiple filling error values ​​for the interference anchor point network, the local pressure drop error, volume shrinkage rate error, and shear stress imbalance corresponding to each cavity are first extracted from the output results of the multi-cavity injection fluid simulation analysis. The three types of errors are then weighted according to preset weights to obtain multiple comprehensive filling error values ​​for each cavity. Subsequently, the global average error value of the multiple comprehensive filling error values ​​is calculated, and the comprehensive filling error value of each cavity is compared with three times the global average error value. When the comprehensive filling error value is greater than three times the global average error value, the corresponding cavity three-dimensional mesh range is marked as a high-risk cavity mesh, thereby obtaining multiple defective cavity mesh regions.

[0032] Furthermore, in the method provided in the application embodiment, high-risk cavity identification is performed on the interference removal anchor point network based on the multiple filling error values ​​to obtain multiple defect cavity mesh regions, which further includes: Local pressure drop error, volume shrinkage rate error, and shear stress imbalance are extracted from the output results of multi-cavity injection molding fluid simulation analysis and weighted to obtain multiple comprehensive filling error values ​​for each cavity. The global average error value of the multiple comprehensive filling error values ​​is calculated, and the three-dimensional mesh range of the cavity with a comprehensive filling error value greater than 3 times the global average error value is marked as a high-risk cavity mesh, thus obtaining multiple defective cavity mesh regions.

[0033] In this embodiment, when extracting the local pressure drop error, volume shrinkage rate error, and shear stress imbalance from the output of multi-cavity injection molding fluid simulation analysis, the three types of error data corresponding to each cavity are first read according to the cavity number, and the three types of error data are then organized into error records under the same cavity. Since the local pressure drop error, volume shrinkage rate error, and shear stress imbalance have different dimensions, the three types of errors are first processed by the same type of data. That is, the local pressure drop error of each cavity is divided by the maximum value of the local pressure drop errors of all cavities to obtain the pressure drop error processing value; the volume shrinkage rate error of each cavity is divided by the maximum value of the volume shrinkage rate errors of all cavities to obtain the volume shrinkage rate error processing value; and the shear stress imbalance of each cavity is divided by the maximum value of the shear stress imbalance of all cavities to obtain the shear stress imbalance processing value. Subsequently, weights for local pressure drop error, volume shrinkage rate error, and shear stress imbalance are set, with the sum of these three weights being 1. In one embodiment, the local pressure drop error weight is 0.4, the volume shrinkage rate error weight is 0.3, and the shear stress imbalance weight is 0.3. For the same cavity, the pressure drop error is processed by multiplying it by the local pressure drop error weight, the volume shrinkage rate error is processed by multiplying it by the volume shrinkage rate error weight, and the shear stress imbalance is processed by multiplying it by the shear stress imbalance weight. The products of these three factors are then added together to obtain the comprehensive filling error value for the cavity. The above calculation is performed on each cavity in the interference anchor point network to obtain multiple comprehensive filling error values ​​for each cavity.

[0034] When calculating the global average error value of multiple comprehensive filling error values, the number of cavities involved in the calculation is first counted. Then, the comprehensive filling error values ​​of each cavity are summed, and the sum is divided by the number of cavities to obtain the global average error value. Subsequently, the comprehensive filling error value of each cavity is compared with three times the global average error value. When the comprehensive filling error value of a cavity is greater than three times the global average error value, the corresponding three-dimensional mesh range of that cavity in the multi-cavity injection molding fluid simulation analysis is read, including the mesh element number, mesh node coordinates, and mesh boundary data. This three-dimensional mesh range is then marked as a high-risk cavity mesh. When the comprehensive filling error value of a cavity is not greater than three times the global average error value, the three-dimensional mesh range of that cavity is not marked as high-risk. After completing the comparison of all cavities, the three-dimensional mesh ranges of all cavities marked as high-risk are summarized to obtain multiple defective cavity mesh regions.

[0035] Step S500: Back-project the multiple defect cavity mesh regions onto the virtual three-dimensional space of the mold blank to construct a three-dimensional production model.

[0036] In this embodiment, when back-projecting multiple defect cavity mesh regions onto the virtual three-dimensional space of the mold blank, high-error mesh nodes in the multiple defect cavity mesh regions are first extracted, and their three-dimensional coordinates are back-projected back into the virtual three-dimensional space of the mold blank, so that the high-error mesh nodes and the corresponding defect cavity reference anchor points are spatially correlated; then, the mold repair features are densified and grown around the corresponding defect cavity reference anchor points to form geometric fine-tuning results for the local area of ​​the defect cavity, and an adaptive fine-tuning cavity model is obtained; then, a multi-resolution level of detail (LOD) mechanism is introduced to perform hierarchical dynamic assembly of the adaptive fine-tuning cavity model, and Boolean operations are used to synthesize each level model and related mold structures into a unified three-dimensional production-level model of the automotive battery plastic injection mold.

[0037] Furthermore, in the method provided in the application embodiments, the method of back-projecting the plurality of defect cavity mesh regions onto the virtual three-dimensional space of the mold blank to construct a three-dimensional production model further includes: The three-dimensional coordinates of the high-error grid nodes in the multiple defect cavity grid regions are back-projected back into the virtual three-dimensional space of the mold blank. The mold repair features are densified and grown around the corresponding defect cavity reference anchor points to obtain an adaptive fine-tuning cavity model. A multi-resolution level of detail (LOD) mechanism is introduced to perform hierarchical dynamic assembly of the adaptive fine-tuning cavity model, and a three-dimensional production-level model of the automotive battery plastic injection mold is uniformly output through Boolean operations.

[0038] In this embodiment, when back-projecting the three-dimensional coordinates of high-error mesh nodes in multiple defect cavity mesh regions back to the virtual three-dimensional space of the mold blank, the high-error mesh nodes in multiple defect cavity mesh regions are first extracted. Combined with the hardware depth value corresponding to the high-error mesh nodes in the global demolding occlusion depth map, the spatial inverse transformation matrix constructed based on the virtual camera intrinsic and extrinsic rotation matrices is used to back-project them back to the virtual three-dimensional space point coordinates in the virtual three-dimensional space of the mold blank. Then, the virtual three-dimensional space point coordinates are introduced as new densification anchor points and associated with the corresponding defect cavity reference anchor points. Densification growth of the mold repair features is performed around the corresponding defect cavity reference anchor points. The cavity flow channel cross-section diameter associated with the anchor point is increased, the gate width is enlarged, or the spatial distance between the cooling water pipe wall and the cavity surface is shortened through the feature fine-tuning script, thereby obtaining an adaptive fine-tuning cavity model.

[0039] When introducing a multi-resolution level of detail (LOD) mechanism for hierarchical dynamic assembly of the adaptive fine-tuning cavity model, a hierarchical control framework for the injection mold 3D model is first established based on an explicit octree structure, and the adaptive fine-tuning cavity model is organized according to spatial hierarchy. Then, the view zoom event of the design interface is monitored, and model data of different LOD resolutions are called according to the view zoom state. When the view is in global viewing state, the lightweight simplified entity of the cavity with low LOD resolution is called for display. When the view is zoomed to a specific cavity area, based on the specific reference anchor point corresponding to the cavity, the high-precision machining-level 3D topological features are decoded and read in real time through a multilayer perceptron for display, thereby completing the hierarchical dynamic assembly of the adaptive fine-tuning cavity model.

[0040] Finally, when outputting the three-dimensional production-level model of the automotive battery plastic injection mold through Boolean operations, the hierarchical dynamically assembled adaptive fine-tuning cavity model, mold blank entity, main runner entity, branch runner entity, gate entity, cooling water channel entity, and mold repair feature entity are first loaded into the same mold blank virtual three-dimensional space, and their positions are aligned according to the reference anchor point coordinates, demolding direction, and assembly reference surface recorded by each entity. Then, the closure of each entity is checked, the boundary stitching of the face with broken edges is performed, and the duplicate faces are merged, so that the objects participating in the Boolean operation are all closed entities. Next, using the mold blank entity as the base entity, the adaptive fine-tuning cavity model is first subjected to Boolean difference set from the mold blank entity to obtain the cavity space corresponding to the shape of the automotive battery plastic part. Then, the main runner entity, branch runner entity, and gate entity are sequentially subjected to Boolean difference set from the mold blank entity to obtain the melt flow channels connecting the cavity space. Then, the cooling water channel entity is subjected to Boolean difference set from the mold blank entity to obtain the cooling water channel. For the entities in the mold repair feature entity used to expand the runner cross-sectional diameter and the gate width, they are treated as bodies to be removed and Boolean difference set is performed again. For the entities used to reinforce local structures, they are subjected to Boolean union set with the mold blank entity. After completing the above Boolean operations, the Boolean operation results are cleaned up for intersecting edges, merged for overlapping surfaces, deleted for small fragments, and corrected for normal consistency. The connectivity boundaries between the cavity, runner, gate, and cooling water channel are recalculated. Finally, the processed closed entity is output as a 3D production-level model of the automotive battery plastic part injection mold.

[0041] Furthermore, in the method provided in the application embodiment, the three-dimensional coordinates of the plurality of defect cavity mesh regions are back-projected back into the virtual three-dimensional space of the mold blank, and the densification growth of the mold repair features is performed around the corresponding defect cavity reference anchor points to obtain an adaptive fine-tuning cavity model, which also includes: Construct a spatial inverse transformation matrix based on the rotation matrices of the virtual camera's intrinsic and extrinsic parameters; using the spatial inverse transformation matrix, back-project the multiple defect cavity mesh regions, combined with their corresponding hardware depth values ​​in the global demolding occlusion depth map, to restore them to virtual three-dimensional spatial point coordinates; introduce the virtual three-dimensional spatial point coordinates as new densification anchor points, activate the feature fine-tuning script, and automatically increase the cavity flow channel cross-sectional diameter associated with the anchor point, expand the gate width, or shorten the spatial distance between the cooling water pipe wall and the cavity surface to obtain an adaptive fine-tuning cavity model.

[0042] In this embodiment, when constructing the spatial inverse transformation matrix based on the virtual camera's intrinsic and extrinsic rotation matrices, the virtual camera's intrinsic parameters, saved during the generation of the global demodulation occlusion depth map, are first read. These intrinsic parameters include the pixel width, pixel height, orthogonal projection lateral starting value, orthogonal projection lateral length, orthogonal projection longitudinal starting value, orthogonal projection longitudinal length, near-plane depth, and far-plane depth of the global demodulation occlusion depth map. Then, the virtual camera position and extrinsic rotation matrix are read. The virtual camera position includes X, Y, and Z coordinates, and the extrinsic rotation matrix is ​​a 3x3 matrix. When transposing the extrinsic rotation matrix, the elements in the first row and second column are swapped with those in the second row and first column; the elements in the first row and third column are swapped with those in the third row and first column; and the elements in the second row and third column are swapped with those in the third row and second column. The elements in the first row and first column, the second row and second column, and the third row and third column remain unchanged, resulting in the orientation transformation matrix. The orientation transformation matrix is ​​then combined with the virtual camera position to form the spatial inverse transformation matrix. Specifically, the nine elements of the orientation transformation matrix are written into the top three rows and three columns of a four-row, four-column matrix. The X, Y, and Z coordinates of the virtual camera position are written into the first row and fourth column, the second row and fourth column, and the third row and fourth column, respectively. The first row and first column, the second row and second column, and the third row and third column are set to 0, and the fourth row and fourth column is set to 1, thus obtaining the spatial inverse transformation matrix.

[0043] Next, using the inverse spatial transformation matrix, multiple defect cavity mesh regions are back-projected to restore virtual 3D space point coordinates by combining their corresponding hardware depth values ​​in the global demolding occlusion depth map. First, high-error mesh nodes in multiple defect cavity mesh regions are read according to the cavity number and mesh node number. Then, the device pixel coordinates of each high-error mesh node in the global demolding occlusion depth map are read. For any device pixel coordinate, the horizontal pixel number is incremented by 0.5 and divided by the pixel width of the global demolding occlusion depth map to obtain the horizontal scale value. This horizontal scale value is then multiplied by the horizontal length of the orthogonal projection and added to the horizontal starting value of the orthogonal projection to obtain the horizontal coordinates in camera space. Similarly, the vertical pixel number is incremented by 0.5 and divided by the pixel height of the global demolding occlusion depth map to obtain the vertical scale value. This vertical scale value is then multiplied by the vertical length of the orthogonal projection and added to the vertical starting value of the orthogonal projection to obtain the vertical coordinates in camera space. The hardware depth value corresponding to this device pixel coordinate is read from the global demolding occlusion depth map and used as the camera space depth coordinate. Subsequently, the camera space horizontal coordinates, camera space vertical coordinates, and camera space depth coordinates are input into the inverse spatial transformation matrix. Multiplication and addition calculations are performed according to the first, second, and third rows of the inverse spatial transformation matrix to obtain the X, Y, and Z coordinates of the virtual three-dimensional space points. The above calculations are then performed sequentially on the high-error mesh nodes in multiple defect cavity mesh regions to obtain the coordinates of multiple virtual three-dimensional space points.

[0044] Finally, when introducing virtual three-dimensional spatial point coordinates as new densification anchor points, the coordinates of each virtual three-dimensional spatial point are first written into the data record of the corresponding defect cavity grid region, and the new densification anchor point is associated with the corresponding defect cavity reference anchor point according to the cavity number of the defect cavity grid region. Subsequently, the feature fine-tuning script is activated. The script reads the local pressure drop error, volume shrinkage rate error, and shear stress imbalance associated with the new densification anchor point, and reads the cavity flow channel cross-section diameter, gate width, and spatial distance between the cooling water pipe wall and the cavity surface near the new densification anchor point. When the local pressure drop error reaches the preset adjustment condition, the cavity flow channel cross-section diameter is increased by the increment of the original diameter multiplied by the preset flow channel adjustment coefficient, and the flow channel cross-section is regenerated along the original flow channel centerline. When the volume shrinkage rate error reaches the preset adjustment condition, the gate width is expanded by the increment of the original width multiplied by the preset gate adjustment coefficient, and the two sides of the gate boundary are shifted outward by 1 / 2 of the corresponding increment. When the shear stress imbalance reaches the preset adjustment condition, the cooling water channel centerline is moved along the shortest direction to the cavity surface by the preset water channel adjustment distance, so that the spatial distance between the cooling water pipe wall and the cavity surface is reduced to no less than the preset minimum wall thickness. After adjusting the cavity flow channels, gates, and cooling water channels corresponding to multiple new densification anchor points, the local cavity structure around the defect cavity reference anchor point is regenerated to obtain an adaptive fine-tuning cavity model.

[0045] Furthermore, in the method provided in the application embodiment, after introducing the virtual three-dimensional space point coordinates as new densification anchor points, it further includes: In the virtual 3D space of the mold blank, a spatial 3D proxy mesh with a specified spatial origin and fixed cell size is established. The spatial density of the newly introduced compacted anchor points is constrained using the spatial 3D proxy mesh. Based on the current 3D position coordinates of the newly introduced compacted anchor point, the specific cell in which the compacted anchor point falls is calculated by referring to the spatial 3D proxy mesh, and the corresponding cell index is obtained. The total number of existing anchor points in the cell corresponding to the current cell index is counted. If the number of existing anchor points is less than the preset maximum capacity threshold, the operation of inserting the current compacted anchor point into the cell is performed.

[0046] In this embodiment, when establishing a spatial 3D proxy mesh in the virtual 3D space of the mold blank, the boundary range of the virtual 3D space of the mold blank is first read, and a specified spatial origin is selected as the starting position of the spatial 3D proxy mesh. Then, a fixed cell size is set, for example, a fixed length is set in the X, Y, and Z directions respectively, and cells are divided sequentially according to the range of the virtual 3D space of the mold blank in the three directions to form a spatial 3D proxy mesh composed of multiple regular cells. Each cell records its own sequence number, cell index, cell boundary range, and list of existing anchor points in the X, Y, and Z directions, thereby enabling newly introduced dense anchor points to be assigned to corresponding cells according to their 3D position coordinates, and spatial density constraints are achieved through the number of anchor points in the cell.

[0047] Based on the current 3D position coordinates of the newly introduced compacted anchor point, when calculating the specific cell the compacted anchor point falls into in relation to the spatial 3D proxy mesh, the X, Y, and Z coordinates of the compacted anchor point are first read. Then, the X, Y, and Z coordinates of the specified spatial origin are subtracted from each of these coordinates to obtain the offsets of the compacted anchor point relative to the specified spatial origin in three directions. Subsequently, the X-direction offset is divided by the X-direction cell size and rounded down to obtain the X-direction cell number. The Y-direction offset is divided by the Y-direction cell size and rounded down to obtain the Y-direction cell number. The Z-direction offset is divided by the Z-direction cell size and rounded down to obtain the Z-direction cell number. The cell into which the compacted anchor point falls is determined based on the X, Y, and Z-direction cell numbers, and the corresponding cell index is calculated according to the indexing rules of the spatial 3D proxy mesh. For example, the X, Y, and Z-direction cell numbers are linearly numbered according to the number of cells in the X and Y directions of the mesh to obtain the current cell index.

[0048] After obtaining the current cell index, the list of existing anchor points in the cell corresponding to that index is read, and the total number of anchor points in the list is counted. Then, the total number of existing anchor points is compared with the preset maximum capacity threshold. If the number of existing anchor points is less than the preset maximum capacity threshold, the 3D position coordinates of the current densified anchor point, the reference anchor point number of the defect cavity to which it belongs, the corresponding defect cavity mesh region number, and the current cell index are written into the anchor point list of that cell. At the same time, the total number of existing anchor points in that cell is updated, thus completing the operation of inserting the current densified anchor point into the cell.

[0049] Furthermore, the method provided in the application embodiments also includes: Otherwise, the current densification anchor point will be discarded.

[0050] In this embodiment, the total number of existing anchor points in the cell corresponding to the current cell index is read, and this number is compared with a preset maximum capacity threshold. When the number of existing anchor points is greater than or equal to the preset maximum capacity threshold, it is determined that the cell has reached the upper limit of spatial density constraint, and the current densified anchor point is no longer written into the anchor point list of the cell. Subsequently, the record to be inserted for the current densified anchor point is deleted, and its corresponding three-dimensional position coordinates, the reference anchor point number of the defect cavity to which it belongs, the corresponding defect cavity mesh region number, and the temporary cache data of the current cell index are cleared. The current densified anchor point is marked as discarded, so that it does not participate in the densification growth of subsequent model repair features and the generation of adaptive fine-tuning cavity models.

[0051] Furthermore, the method provided in the application embodiments, which introduces a multi-resolution level of detail (LOD) mechanism to perform hierarchical dynamic assembly of the adaptive fine-tuning cavity model, also includes: A hierarchical control framework for the 3D model of the injection mold is established based on an explicit octree structure; the view zoom event of the design interface is monitored; when the view is in a global viewing state, a lightweight simplified entity of the cavity with low LOD resolution is called for display; when the view is zoomed in to a specific cavity area, the high-precision machining-grade 3D topological features are decoded and read in real time through a multilayer perceptron based on the specific reference anchor point corresponding to the cavity for display.

[0052] In this embodiment, when establishing the hierarchical control framework of the injection mold 3D model based on an explicit octree structure, the minimum bounding area of ​​the injection mold 3D model in the virtual 3D space of the mold blank is first read. The minimum and maximum coordinates of this bounding area in the X, Y, and Z directions are obtained, and this minimum bounding area is used as the root node space range of the explicit octree structure. Then, the center point of the root node space range is calculated. Using this center point as the dividing datum, the root node is divided into 8 child nodes along the X, Y, and Z directions respectively. For each child node, the geometric data of the cavity, runner, gate, cooling water channel, and mold repair features contained within it are read. When the size of the child node is greater than the preset minimum space size and the number of geometric data in the child node is greater than the preset number, it is further divided into 8 equal parts according to the center point. After the recursive partitioning is completed, the node number, parent node number, child node number, node spatial range, node level, cavity number contained in the node, corresponding reference anchor point, low LOD resolution lightweight simplified entity index of cavity, and high precision machining level three-dimensional topological feature index are recorded for each octree node, thereby obtaining the hierarchical control framework of the injection mold three-dimensional model.

[0053] When monitoring viewpoint zoom events in the design interface, the system first reads the virtual viewing angle's position, viewing direction, viewing distance, zoom ratio, and visible range during each frame refresh of the interface. It then compares the viewing distance and zoom ratio of the current frame with those of the previous frame. When the change in viewing distance or zoom ratio reaches a preset threshold, it is recorded as a viewpoint zoom event. Next, the system intersects the current visible range with the node space of each octree node in the explicit octree structure to filter out the octree nodes covered by the current viewpoint. When the current viewpoint covers the root node or a higher-level octree node, and the covered area contains multiple cavity numbers, the viewpoint is determined to be in a global viewing state. At this point, according to the low-LOD resolution cavity lightweight simplified entity index recorded in the covered octree node, the corresponding cavity lightweight simplified entity is read and assembled into the injection mold 3D model for display based on the corresponding reference anchor point position.

[0054] Before use, the multilayer perceptron is trained. During training, a 3D model of an automotive battery plastic injection mold that has already been modeled at the machining level is selected as the training sample. Training input data is extracted according to the reference anchor points corresponding to each cavity. The training input data includes the 3D coordinates of the reference anchor points, the cavity number, the LOD level parameters, the diameter of the runner section, the gate width, the spatial distance between the cooling water pipe wall and the cavity surface, and the mold repair feature parameters. At the same time, training output data is extracted from the corresponding machining-level 3D model. The training output data includes the coordinates of the cavity forming surface nodes, the coordinates of the gate local structure nodes, the coordinates of the runner section boundary nodes, the coordinates of the adjacent structure nodes of the cooling water channel, and the coordinates of the mold repair feature boundary nodes. During training, the training input data is fed into a multilayer perceptron. The multilayer perceptron performs weight multiplication, bias addition, and activation function calculation layer by layer, outputting predicted high-precision machining-grade 3D topological features. Then, the coordinate differences between the predicted results and the training output data are calculated point by point, and the squared differences are summed to obtain the training error. Based on the training error, the weights and biases of each layer in the multilayer perceptron are updated using backpropagation. This process of input, prediction, error calculation, and weight update is repeated until the training error is less than a preset error threshold, resulting in a fully trained multilayer perceptron. Figure 2As shown in the figure, the horizontal axis represents the training rounds in the multilayer perceptron training process, and the vertical axis represents the error value between the predicted high-precision machining-grade 3D topological features and the training output data. Specifically, the training error curve represents the error change between the predicted node coordinates and the actual machining-grade 3D model node coordinates after the multilayer perceptron learns features based on the reference anchor point 3D coordinates, cavity number, LOD level parameters, runner cross-section diameter, gate width, spatial distance between the cooling water pipe wall and the cavity surface, and mold repair feature parameters. The validation error curve represents the error change obtained by validating the multilayer perceptron's output cavity forming surface node coordinates, gate local structure node coordinates, runner cross-section boundary node coordinates, cooling water adjacent structure node coordinates, and mold repair feature boundary node coordinates using sample data that did not participate in weight updates. The preset error threshold represents the error limit condition required for the multilayer perceptron to complete training. As the training rounds increase, the multilayer perceptron continuously adjusts the weights and bias parameters of each layer through backpropagation, gradually reducing the training error and validation error. When the training error is less than the preset error threshold, a trained multilayer perceptron capable of outputting high-precision machining-grade 3D topological features is obtained.

[0055] When the viewpoint is zoomed in to a specific cavity region, the currently zoomed-in octree node is determined first based on the intersection of the current visible range and the lower-level octree nodes in the explicit octree structure. Then, the cavity number and the specific reference anchor point corresponding to the cavity recorded in the octree node are read. Subsequently, the 3D coordinates of the specific reference anchor point, the cavity number, the LOD level parameters, and the mold modification feature parameters corresponding to the cavity are input into the trained multilayer perceptron. The multilayer perceptron calculates layer by layer according to the weights and biases obtained during training. Each layer first multiplies the input data with the weight matrix, adds the bias value, and passes it through an activation function to obtain the input for the next layer, until the output layer generates a high-precision machining-grade 3D topological feature. Then, this high-precision machining-grade 3D topological feature replaces the original low-LOD resolution lightweight simplified cavity entity displayed in the specific cavity region, while other un-zoomed areas continue to use the low-LOD resolution lightweight simplified cavity entity, thereby completing the hierarchical dynamic assembly of the adaptive fine-tuning cavity model.

[0056] In summary, the embodiments of this application have at least the following technical effects: This application deploys a first-order matrix anchor point network in the virtual 3D space of the mold blank according to the production cavity number requirements to obtain multiple reference anchor points and constructs a lightweight component model of a single automotive battery plastic part cavity. The demolding direction of the mold is defined as the orthogonal projection direction of the virtual camera. The hardware rasterization pipeline of the graphics processor is used to perform depth rendering on the multiple reference anchor points and the lightweight component model in a single-channel depth buffer to obtain a global demolding occlusion depth map. Multiple potential processing feature points in the multiple reference anchor points are projected onto the global demolding occlusion depth map to perform multi-view demolding visibility testing, filtering out conflicting feature points that cause occlusion interference to obtain an interference clearing anchor point network. Multiple filling error values ​​of each cavity in the interference clearing anchor point network are obtained under multi-cavity injection fluid simulation analysis. Based on the multiple filling error values, the interference clearing anchor point network is used to identify high-risk cavities to obtain multiple defective cavity mesh regions. The multiple defective cavity mesh regions are back-projected onto the virtual 3D space of the mold blank to construct a 3D production model. This invention addresses the technical problems of low modeling efficiency and difficulty in accurately identifying demolding interference in existing multi-cavity injection molds. By constructing an anchor point network and combining it with depth rendering for demolding visibility testing, the invention achieves the technical effect of improving the efficiency of 3D modeling of multi-cavity injection molds and the accuracy of demolding interference identification.

[0057] Example 2, based on the same inventive concept as the three-dimensional modeling method for an automotive battery plastic injection mold described in the foregoing examples, such as... Figure 3 As shown, this application provides a 3D modeling system for injection molds of automotive battery plastic parts. The system and method embodiments in this application are based on the same inventive concept. The system includes: The network deployment module 11 is used to deploy a first-order matrix anchor point network in the virtual three-dimensional space of the mold blank according to the production cavity number requirement, obtain multiple reference anchor points, and construct a lightweight component model of a single automotive battery plastic part cavity; the depth rendering module 12 is used to define the mold release direction of the mold as the orthogonal projection direction of the virtual camera, and use the hardware rasterization pipeline of the graphics processor to perform depth rendering on the multiple reference anchor points and the lightweight component model in a single-channel depth buffer to obtain a global release occlusion depth map; the visibility testing module 13 is used to test multiple potential... The processing feature points are projected onto the global demolding occlusion depth map for multi-view demolding visibility testing, filtering out conflicting feature points that cause occlusion interference, and obtaining a network of interference-clearing anchor points; the identification module 14 is used to obtain multiple filling error values ​​of each cavity in the interference-clearing anchor point network under multi-cavity injection fluid simulation analysis, and to identify high-risk cavities based on the multiple filling error values, thereby obtaining multiple defective cavity mesh regions; the model construction module 15 is used to back-project the multiple defective cavity mesh regions onto the virtual three-dimensional space of the mold blank to construct a three-dimensional production model.

[0058] Furthermore, the system is also used to implement the following functions: The frustum culling and occlusion culling of the multiple potential processing feature points are performed in parallel and fused using a single CUDA kernel to obtain multiple device pixel coordinates. Multiple hardware linear camera space depths are then mapped and extracted from the global demodulation occlusion depth map at the positions corresponding to the multiple device pixel coordinates. The linear depths of the multiple potential processing feature points are compared with the corresponding hardware linear camera space depths in the multiple hardware linear camera space depths, and conflicting feature points whose linear depths are greater than the hardware linear space depths are filtered out to obtain the interference-clearing anchor network.

[0059] Furthermore, the system is also used to implement the following functions: Obtain a preset demolding safety margin, and add multiple hardware linear camera space depths to it to obtain the limit visible depth value of each device pixel coordinate; if the multiple linear depths of the multiple potential processing feature points in the camera space are greater than the limit visible depth value of the corresponding device pixel coordinate position, then it is determined that the current potential processing feature point causes demolding occlusion interference and is cleared and removed; otherwise, it is retained.

[0060] Furthermore, the system is also used to implement the following functions: Local pressure drop error, volume shrinkage rate error, and shear stress imbalance are extracted from the output results of multi-cavity injection molding fluid simulation analysis and weighted to obtain multiple comprehensive filling error values ​​for each cavity. The global average error value of the multiple comprehensive filling error values ​​is calculated, and the three-dimensional mesh range of the cavity with a comprehensive filling error value greater than 3 times the global average error value is marked as a high-risk cavity mesh, thus obtaining multiple defective cavity mesh regions.

[0061] Furthermore, the system is also used to implement the following functions: The three-dimensional coordinates of the high-error grid nodes in the multiple defect cavity grid regions are back-projected back into the virtual three-dimensional space of the mold blank. The mold repair features are densified and grown around the corresponding defect cavity reference anchor points to obtain an adaptive fine-tuning cavity model. A multi-resolution level of detail (LOD) mechanism is introduced to perform hierarchical dynamic assembly of the adaptive fine-tuning cavity model, and a three-dimensional production-level model of the automotive battery plastic injection mold is uniformly output through Boolean operations.

[0062] Furthermore, the system is also used to implement the following functions: Construct a spatial inverse transformation matrix based on the rotation matrices of the virtual camera's intrinsic and extrinsic parameters; using the spatial inverse transformation matrix, back-project the multiple defect cavity mesh regions, combined with their corresponding hardware depth values ​​in the global demolding occlusion depth map, to restore them to virtual three-dimensional spatial point coordinates; introduce the virtual three-dimensional spatial point coordinates as new densification anchor points, activate the feature fine-tuning script, and automatically increase the cavity flow channel cross-sectional diameter associated with the anchor point, expand the gate width, or shorten the spatial distance between the cooling water pipe wall and the cavity surface to obtain an adaptive fine-tuning cavity model.

[0063] Furthermore, the system is also used to implement the following functions: In the virtual 3D space of the mold blank, a spatial 3D proxy mesh with a specified spatial origin and fixed cell size is established. The spatial density of the newly introduced compacted anchor points is constrained using the spatial 3D proxy mesh. Based on the current 3D position coordinates of the newly introduced compacted anchor point, the specific cell in which the compacted anchor point falls is calculated by referring to the spatial 3D proxy mesh, and the corresponding cell index is obtained. The total number of existing anchor points in the cell corresponding to the current cell index is counted. If the number of existing anchor points is less than the preset maximum capacity threshold, the operation of inserting the current compacted anchor point into the cell is performed.

[0064] Furthermore, the system is also used to implement the following functions: Otherwise, the current densification anchor point will be discarded.

[0065] Furthermore, the system is also used to implement the following functions: A hierarchical control framework for the 3D model of the injection mold is established based on an explicit octree structure; the view zoom event of the design interface is monitored; when the view is in a global viewing state, a lightweight simplified entity of the cavity with low LOD resolution is called for display; when the view is zoomed in to a specific cavity area, the high-precision machining-grade 3D topological features are decoded and read in real time through a multilayer perceptron based on the specific reference anchor point corresponding to the cavity for display.

[0066] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0067] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for three-dimensional modeling of injection molds for automotive battery plastic parts, characterized in that, The method includes: In the virtual three-dimensional space of the mold blank, a first-order matrix anchor point network is deployed according to the production cavity number requirement to obtain multiple reference anchor points and construct a lightweight component model of a single automotive battery plastic part cavity. The demolding direction of the mold is defined as the orthogonal projection direction of the virtual camera. The hardware rasterization pipeline of the graphics processor is used to perform depth rendering on the multiple reference anchor points and lightweight component models in a single-channel depth buffer to obtain a global demolding occlusion depth map. Multiple potential processing feature points from the multiple reference anchor points are projected onto the global demolding occlusion depth map to perform multi-view demolding visibility test, filter out conflicting feature points that cause occlusion interference, and obtain the interference-clearing anchor point network. Multiple filling error values ​​of each cavity in the interference removal anchor point network are obtained under multi-cavity injection fluid simulation analysis. Based on the multiple filling error values, high-risk cavities are identified in the interference removal anchor point network to obtain multiple defective cavity mesh regions. The multiple defect cavity mesh regions are back-projected onto the virtual three-dimensional space of the mold blank to construct a three-dimensional production model.

2. The three-dimensional modeling method for injection molds of automotive battery plastic parts as described in claim 1, characterized in that, Multiple potential processing feature points from the plurality of reference anchor points are projected onto the global demolding occlusion depth map for multi-view demolding visibility testing. Conflicting feature points that cause occlusion interference are filtered out to obtain an interference-clearing anchor point network, including: The frustum culling and occlusion culling of the multiple potential processing feature points are performed in parallel using a single CUDA kernel to obtain multiple device pixel coordinates. Map and extract the spatial depth of multiple hardware linear cameras at positions corresponding to the pixel coordinates of the multiple devices in the global demodulation occlusion depth map; By comparing the linear depths of multiple potential processing feature points with the corresponding hardware linear camera spatial depths in multiple hardware linear camera spatial depths, conflicting feature points whose linear depths are greater than the hardware linear spatial depths are filtered out to obtain the interference-clearing anchor network.

3. The three-dimensional modeling method for injection molds of automotive battery plastic parts as described in claim 2, characterized in that, By comparing the linear depths of multiple potential processing feature points with the corresponding hardware linear camera spatial depths in multiple hardware linear camera spatial depths, conflicting feature points whose linear depths are greater than the hardware linear spatial depths are filtered out, resulting in an interference-clearing anchor network, including: Obtain the preset demolding safety margin, add multiple hardware linear camera spatial depths to it respectively, and obtain the limit visible depth value of each device pixel coordinate; If the linear depth of the multiple potential processing feature points in the camera space is greater than the limit visible depth value of the corresponding device pixel coordinate position, then it is determined that the current potential processing feature point has generated demolding occlusion interference and is cleared and removed; otherwise, it is retained.

4. The three-dimensional modeling method for injection molds of automotive battery plastic parts as described in claim 1, characterized in that, Based on the multiple filling error values, high-risk cavity identification is performed on the interference removal anchor point network to obtain multiple defect cavity mesh regions, including: The local pressure drop error, volume shrinkage rate error, and shear stress imbalance are extracted from the output results of multi-cavity injection molding fluid simulation analysis, and weighted calculations are performed to obtain multiple comprehensive filling error values ​​for each cavity. Calculate the global average error value of the multiple comprehensive filling error values, and mark the cavity three-dimensional mesh range where the comprehensive filling error value is greater than 3 times the global average error value as a high-risk cavity mesh, thereby obtaining multiple defective cavity mesh regions.

5. The three-dimensional modeling method for injection molds of automotive battery plastic parts as described in claim 1, characterized in that, The multiple defect cavity mesh regions are back-projected onto the virtual three-dimensional space of the mold blank to construct a three-dimensional production model, including: The three-dimensional coordinates of the high-error grid nodes in the multiple defect cavity grid regions are back-projected back into the virtual three-dimensional space of the mold blank, and the mold repair features are densified and grown around the corresponding defect cavity reference anchor point to obtain an adaptive fine-tuning cavity model. A multi-resolution level of detail (LOD) mechanism is introduced to perform hierarchical dynamic assembly of the adaptive fine-tuning cavity model, and Boolean operations are used to uniformly output a 3D production-level model of the automotive battery plastic injection mold.

6. The three-dimensional modeling method for injection molds of automotive battery plastic parts as described in claim 5, characterized in that, The three-dimensional coordinates of the multiple defect cavity mesh regions are back-projected back into the virtual three-dimensional space of the mold blank. Then, the mold-repairing features are densified and grown around the corresponding defect cavity reference anchor points to obtain an adaptive fine-tuning cavity model, including: Construct a spatial inverse transformation matrix based on the rotation matrices of the virtual camera's intrinsic and extrinsic parameters; Using the inverse spatial transformation matrix, the multiple defect cavity mesh regions, combined with their corresponding hardware depth values ​​in the global demolding occlusion depth map, are back-projected and restored to virtual three-dimensional spatial point coordinates. The virtual three-dimensional space point coordinates are introduced as new densification anchor points. The feature fine-tuning script is activated to automatically increase the cross-sectional diameter of the cavity flow channel associated with the anchor point, expand the gate width, or shorten the spatial distance between the cooling water pipe wall and the cavity surface, thereby obtaining an adaptive fine-tuning cavity model.

7. The three-dimensional modeling method for injection molds of automotive battery plastic parts as described in claim 6, characterized in that, The virtual three-dimensional space point coordinates are introduced as new compaction anchor points, followed by: In the virtual three-dimensional space of the mold blank, a spatial three-dimensional proxy mesh with a specified spatial origin and fixed cell size is established, and the spatial density of the newly introduced compacted anchor points is constrained by the spatial three-dimensional proxy mesh. Based on the current three-dimensional position coordinates of the newly introduced densified anchor point, and referring to the spatial three-dimensional proxy mesh, calculate the specific cell in which the densified anchor point falls, and obtain the corresponding cell index. Count the total number of existing anchor points in the cell corresponding to the current cell index. If the number of existing anchor points is less than the preset maximum capacity threshold, then insert the current dense anchor point into the cell.

8. The three-dimensional modeling method for injection molds of automotive battery plastic parts as described in claim 6, characterized in that, Otherwise, the current densification anchor point will be discarded.

9. A three-dimensional modeling method for injection molds of automotive battery plastic parts as described in claim 5, characterized in that, A multi-resolution level of detail (LOD) mechanism is introduced to perform hierarchical dynamic assembly of the adaptive fine-tuning cavity model, including: A hierarchical control framework for the three-dimensional model of the injection mold is established based on an explicit octree structure. Monitor the view zoom event of the design interface. When the view is in global view state, call the low LOD resolution cavity lightweight simplified entity for display. When the viewpoint is magnified to a specific cavity area, the high-precision machining-grade three-dimensional topological features are decoded and displayed in real time through a multilayer perceptron based on the specific reference anchor point corresponding to the cavity.

10. A three-dimensional modeling system for injection molds of automotive battery plastic parts, characterized in that, The system is used to execute a three-dimensional modeling method for an injection mold of an automotive battery plastic part as described in any one of claims 1-9, the system comprising: The network deployment module is used to deploy a first-order matrix anchor point network in the virtual three-dimensional space of the mold blank according to the production cavity number requirements, obtain multiple reference anchor points, and construct a lightweight component model of a single automotive battery plastic part cavity. The depth rendering module is used to define the mold release direction as the orthogonal projection direction of the virtual camera, and to perform depth rendering on the multiple reference anchor points and lightweight component models in a single-channel depth buffer using the hardware rasterization pipeline of the graphics processor to obtain a global mold release occlusion depth map. The visibility testing module is used to project multiple potential processing feature points from the multiple reference anchor points onto the global demolding occlusion depth map to perform multi-view demolding visibility testing, filter out conflicting feature points that cause occlusion interference, and obtain a network of cleared interference anchor points. The identification module is used to obtain multiple filling error values ​​of each cavity in the interference removal anchor point network under multi-cavity injection fluid simulation analysis, and to identify high-risk cavities in the interference removal anchor point network based on the multiple filling error values ​​to obtain multiple defective cavity mesh regions. The model building module is used to back-project the multiple defect cavity mesh regions onto the virtual three-dimensional space of the mold blank to build a three-dimensional production model.

Images