Particle generation method and system for pre-processing of complex geometry particle method for reactor

Abstract

The application discloses a particle generation method and system for pre-processing of a complex geometry particle method for a reactor, and relates to the technical field of nuclear reactor thermal hydraulic and fluid mechanics numerical simulation, and comprises the following steps: introducing a fluid-solid geometry model, and calculating pseudo-normal vectors of geometry boundaries; adopting a dynamic space downsampling fitting strategy of polynomial order self-adaptation and geometry topological form fission to construct a high-precision directed distance field; activating seed particles, and growing to generate a rough candidate particle cluster in a preset expansion direction; performing out-of-bound penetration detection by using the directed distance field, and performing particle overlap rejection by combining a Morton code one-dimensional space filling curve technology to obtain an initial particle distribution; the method effectively eliminates boundary normal misjudgment in geometry modeling, significantly reduces memory occupation of a background grid, avoids initial particle interference stacking, and improves the accuracy of particle method numerical modeling.

Description

Technical Field

[0001] This invention relates to the field of numerical simulation technology for nuclear reactor thermal-hydraulic and fluid dynamics, specifically to a particle generation method and system for preprocessing particles in complex reactor geometry. Background Technology

[0002] Meshless particle methods such as the Moving Semi-Implicit Particle Method (MPS) and Smooth Particle Hydrodynamics (SPH) offer irreplaceable advantages over traditional mesh methods when simulating complex flow boiling and core meltdown phenomena involving large deformations and free liquid surfaces within nuclear reactors. A high-quality initial particle spatial arrangement model is fundamental to ensuring the numerical stability and computational accuracy of particle method solvers.

[0003] In the preprocessing modeling of traditional particle methods, virtual particles or projection point methods are typically used to determine the internal and external relationships of geometric boundaries. However, nuclear reactors contain complex tube sheets, supports, and fuel rods, with numerous right-angled and acute-angled edges in their geometric topology. Traditional methods are prone to generating singularities when dealing with these sharp edges, leading to incorrect normal determination and consequently causing "particles passing through walls" or "boundary misalignment." Furthermore, traditional distance fields rely on the background mesh resolution, often requiring extremely high-density global background meshes to capture subtle local features, which consumes enormous memory resources. Simultaneously, when filling geometric surfaces, traditional particle generation algorithms are prone to localized multi-layered stacking interference in the generated particle clusters. This spatial stacking can generate extremely large repulsive forces at the start of fluid dynamics calculations, directly causing computational collapse and divergence.

[0004] In summary, existing particle method preprocessing modeling suffers from cusp singularity misjudgment, huge memory consumption, and initial particle stacking interference. It is prone to normal misjudgment, leading to "particles passing through walls" or "boundary misalignment," which reduces the accuracy of particle method numerical modeling and makes it difficult to accurately simulate the complex fluid behavior inside a nuclear reactor. Summary of the Invention

[0005] To address the shortcomings of existing particle method preprocessing modeling techniques, such as cusp singularity misjudgment, huge memory consumption, and initial particle stacking interference, this invention proposes a particle generation method and system for complex reactor geometry preprocessing, thereby solving the problems existing in the prior art.

[0006] A particle generation method for pretreatment of particles with complex reactor geometry includes the following steps: Acquire the initial geometric mesh data of the fluid and solid surfaces in the nuclear reactor, establish a bounding box, and calculate the pseudo normal vectors of the vertices and edges of the bounding box geometric model; Using the bounding box as the global search space and the pseudo normal vector as the criterion for determining whether a spatial point is located inside or outside the geometric model, a directed distance field containing polynomial order adaptive fitting and geometric topological morphological fission features is constructed using a dynamic spatial downsampling fitting strategy. The geometric normal pattern is determined based on the physical properties of the fluid or solid region of the nuclear reactor. The activation source point of the seed particle is located based on the pattern. The seed particle is activated in the spatial region with a positive scalar value in the directed distance field. The activated seed particle is used as the parent node. The outward growth direction is determined according to the gradient direction of the directed distance field. The particle grows outward along the preset growth direction to generate a coarse candidate particle cluster. Substitute the coordinates of each particle in the coarse candidate particle cluster into the constructed directed distance field, and remove out-of-bounds particles with negative directed distance values; discretize the coordinate system of the remaining particles in three-dimensional space, and calculate the Morton code of each remaining particle; by sorting and comparing the Morton codes, identify and remove stacked particles within the same Morton code index, and obtain the final initial particle distribution of the nuclear reactor.

[0007] Furthermore, the pseudo-normal vectors of the vertices and edges of the bounding box geometry model include vertex pseudo-normals and edge pseudo-normals; the vertex pseudo-normal is obtained by summing the normals of all adjacent triangular faces of the vertex according to the weighted summation of the interior angles of each face at the vertex; the edge pseudo-normal is obtained by summing the unit normals of two adjacent faces sharing the edge.

[0008] Furthermore, after acquiring the initial geometric mesh data of the nuclear reactor fluid and solid, an accelerated search topology is established using an octree and AABB bounding box tree structure; the bounding box is extended outward by 8 times the radius of the target particle based on the original volume mesh bounding box.

[0009] Furthermore, the polynomial-order adaptive fitting generates Gaussian integration points within the octree unit to construct a spatial distance lattice, and obtains the fitted distance value by solving the coefficients of the higher-order polynomial; the specific process is as follows: ; in, Represents coordinate point ( x , y , z The scalar value of the directed distance field obtained by fitting at point ); Indicates the highest order of the polynomial. The coordinates are respectively in x , y , z The exponent in the direction; This represents the spatial distance coefficient.

[0010] Furthermore, the geometric topology fission performs error back-verification on the polynomial fitting distance value. When the local error exceeds a preset threshold, the current octree unit is triggered to split into sub-mesh, and polynomial fitting is recursively performed until the error converges.

[0011] Furthermore, the coarse candidate particle cluster is generated by the parent node expanding outward in a specified direction to generate child node particles in a pre-allocated spatial grid memory. The child node particles are then translated based on a set particle radius distance to generate a coarse candidate particle cluster with translation coordinates.

[0012] Furthermore, the removal of out-of-bounds particles with negative directed distance values ​​specifically means that when a particle's directed distance value is negative or less than a preset boundary tolerance, it is determined to be an out-of-bounds particle and removed.

[0013] Furthermore, the discretization of the coordinate system of the remaining particles in three-dimensional space after removal, and the calculation of the Morton code for each remaining particle, are specifically expressed as follows: ; in, This represents the one-dimensional Morton code value obtained after dimensionality reduction of the particle. The bit width of the binary code representing integer coordinates; Representing the integer coordinates of the particles respectively X , Y , Z In binary representation m The value of the bit.

[0014] The present invention also includes a particle generation system for pretreatment of particles with complex reactor geometry, comprising: The acquisition module is used to acquire the initial geometric mesh data of the nuclear reactor fluids and solids, establish the bounding box, and calculate the pseudo normal vectors of the vertices and edges of the bounding box geometric model. The directed distance field construction module is used to construct a directed distance field that includes polynomial order adaptive fitting and geometric topological morphological fission features by using the bounding box as the global search space, the pseudo normal vector as the sign determination criterion for whether the spatial point is located inside or outside the geometric model, and adopting a dynamic space downsampling fitting strategy. The particle swarm generation module is used to determine the geometric normal pattern based on the physical properties of the fluid or solid region of the nuclear reactor, locate the activation source point of the seed particle based on the pattern, activate the seed particle in the spatial region with a positive scalar value in the directed distance field, take the activated seed particle as the parent node, determine the outward growth direction based on the gradient direction of the directed distance field, and grow outward along the preset growth direction to generate a coarse candidate particle cluster. The generation module is used to substitute the coordinates of each particle in the coarse candidate particle cluster into the constructed directed distance field, and remove out-of-bounds particles with negative directed distance values; the coordinate system of the remaining particles after removal is discretized in three-dimensional space, and the Morton code of each remaining particle is calculated; by sorting the Morton codes and comparing adjacent ones, stacked particles in the same Morton code index are identified and removed, and the final initial particle distribution of the nuclear reactor is obtained.

[0015] This invention provides a particle generation method for pretreatment of particles with complex reactor geometry, which has the following advantages: This invention eliminates the misjudgment of normals at geometric corners through pseudo-normal vector calculation, thereby fundamentally avoiding the directional distance field sign confusion and "wall-penetration" phenomenon. By employing the "P / H adaptive and dynamic spatial downsampling fitting" technique, it does not require a global high-density grid, but instead uses high-order polynomial fitting to extract distance field parameters and performs local octree splitting on error-boundary regions. Without increasing the memory burden of redundant grids, it achieves high-precision distance resolution for complex nuclear reactor boundaries, realizing efficient GPU parallel culling and defect-free modeling. By generating coarse candidate clusters through multi-directional growth of seed particles and combining a dual screening mechanism of directional distance field penetration detection and Morton code overlap culling, it obtains a high-precision initial particle model without stacking or wall-penetration, thus simultaneously solving the three major technical problems of wall-penetration, memory explosion, and stacking collapse, significantly improving the accuracy of simulating the complex fluid behavior process inside a nuclear reactor. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the pretreatment particle generation method for complex reactor geometry in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] This invention proposes a particle generation method for preprocessing in complex reactor geometry. The method analyzes the fluid and solid geometry model of a nuclear reactor, constructs a high-precision adaptive directional distance field, and initializes the initial fluid particle distribution characteristics of the reactor.

[0019] like Figure 1 As shown, the method specifically includes the following steps: S1. Geometric Information Preprocessing and Pseudo-Normal Vector Calculation: The initial geometric mesh data of the reactor fluid and solid are imported into the preprocessing module of the fluid dynamics numerical simulation system to establish a bounding box. Based on the geometric relationship between each vertex of the mesh and its shared adjacent triangular facets, the pseudo-normal vector of the geometric boundary is calculated to eliminate the cusp singularity at the corner of the geometric body.

[0020] First, the CAD mesh model of the nuclear reactor is imported, and geometric data is loaded using octree and AABB tree structures to accelerate spatial intersection testing. To accommodate possible splashing and overflow of particles in subsequent fluid calculations, the adaptive bounding box established by the system extends outward by 8 times the particle radius based on the original volume mesh bounding box, taking into account both the overflow range of fluid particle motion and the SDF background mesh boundary. To address the calculation errors caused by the traditional ray method at geometric corners, this invention introduces a pseudo-normal vector processing mechanism. For a geometric vertex, calculate the normals of all its neighboring triangles, and then perform a weighted summation using the corresponding included angles as weights to obtain the vertex pseudo-normal: (1) In the formula: Represents vertices v The pseudo-normal vector; Indicates the first i The adjacent triangular faces of the adjacent triangular faces at the vertex v The interior angle at the location; Indicates the first i The true unit normal vector of each adjacent triangular facet; This represents the total number of adjacent triangles sharing the same vertex.

[0021] For geometric edges, pseudo-normals are obtained by accumulating the unit normals of two adjacent faces. These pseudo-normals are used to replace the true geometric normals to determine whether a point in space is inside or outside the solid, thus avoiding the distortion of the SDF scalar field. The establishment of pseudo-normals provides a unique and smooth reference standard for determining the positive or negative attributes of a point in space within or outside the geometry. (2) In the formula: Representing an edge e The pseudo-normal vector; Indicates shared edge e The true unit normal vector of the adjacent face 1; Indicates shared edge e The true unit normal vector of the adjacent face 2.

[0022] S2. Construction of Adaptive Directed Distance Field (SDF): Based on bounding boxes and pseudo normal vectors, a dynamic spatial downsampling fitting strategy is adopted to construct a high-precision directed distance field with polynomial order adaptation and geometric topological morphological fission characteristics, which is used to record the directed distance of any coordinate point in space from the geometric boundary.

[0023] The directed distance field (SDF) is used to record spatial physical distances. The SDF value is defined as 0 for boundary surfaces, positive for the interior of the geometry, and negative for the external fluid space. This step uses the adaptive and dynamic spatial downsampling fitting strategy constructed in this invention to generate the SDF. This dynamic spatial downsampling fitting strategy includes a dual mechanism: polynomial-order adaptive fitting (P-adaptive) and geometric topological morphology fission (H-adaptive). The specific process for constructing the directed distance field is as follows: the bounding box is used as the global search space for constructing the distance field. The absolute distance from any spatial point to the geometric surface is calculated for subsequent determination of the distance relationship between particles and walls. The dot product projection relationship between this point and the pseudo-normal vector is used to determine whether the spatial point is located inside or outside the geometric entity, thus assigning a precise positive or negative sign to the distance value, thereby completing the construction of the directed distance field.

[0024] To obtain accurate distance scalar values, the specific fitting process is as follows: First, polynomial-order adaptive fitting (P-adaptive) is performed. Gaussian integration points are generated within the spatial octree mesh initially divided by the bounding box to construct a spatial distance lattice. The adaptive spatial coefficients of the higher-order polynomial are then solved. This allows the system to accurately obtain distance values ​​for arbitrary coordinates in space through polynomial algebraic solutions without infinitely subdividing the mesh. (3) In the formula: Represents coordinate point ( x , y , z The scalar value of the directed distance field obtained by fitting at point ); This indicates the highest order of the polynomial, which is set to 2. The coordinates are respectively in x , y , z The exponent in the direction; This represents the spatial distance coefficient.

[0025] Next, geometric topology fission (H-adaptive) is performed: after polynomial fitting, error back-verification is performed. Once it is found that the error between the SDF distance value fitted by the local polynomial and the distance between the actual facet exceeds the set threshold, local subdivision is triggered, the current octree node is cut off to split it into sub-mesh, and the polynomial parameters are updated again until the local error is completely converged.

[0026] S3. Environment Configuration and Initial Seed Particle Multidirectional Growth: Based on the physical properties of the fluid or solid region, the corresponding geometric normal pattern is determined. Its main function is to guide the source point positioning strategy of the seed particles: When it is a normal pattern generated based on a free liquid surface, the set center of the liquid surface is obtained and pushed inward along the normal of the liquid surface by a particle diameter distance, and the point calculated in this way is used as the source point; when it is a normal pattern generated based on a geometry, the center of the triangular facets on the surface of the geometry is extracted and pushed inward along the opposite direction of the normal of the facet by a preset specific distance to perform connected domain detection. The point that successfully falls into the connected region inside the geometry is used as the source point; then the reachable source points obtained by the above positioning calculation based on different normal patterns are used as seed particles, and the seed particles are activated in three-dimensional space.

[0027] After completing the SDF construction, prepare a cuboid bounding box; the system acquires the SDF data of different regions in the geometric normal mode as constraint boundaries, including free liquid surface environment, static solid, and moving solid; after activating the seed particle, in the pre-allocated spatial grid memory, the parent particle node can expand outward to generate child node particles in up to 32 specified expansion directions, and perform coordinate translation based on the set particle radius distance to generate a coarse candidate particle cluster with translation coordinates.

[0028] S4. Spatial Pruning and Precise Positioning: A high-precision directed distance field is used to perform spatial penetration detection on the coarse candidate particle cluster, and overlapping removal is performed using one-dimensional space-filling curve encoding technology to obtain a high-precision initial particle model of the nuclear reactor without stacking or wall penetration. Specifically, spatial pruning and precise positioning include: SDF wall penetration removal: the coordinates of the coarse candidate particle cluster are substituted into the directed distance field generated in step S2 to query the SDF scalar value. When the SDF scalar value is negative or less than the preset boundary tolerance, it is determined to be an out-of-bounds particle and removed; GPU parallel deduplication test: the coordinate system of the coarse candidate particle cluster in three-dimensional space is discretized, the Morton code of each particle is calculated, and a one-dimensional space-filling curve is constructed; by sorting and comparing adjacent Morton codes, stacked sub-particles within the same Morton code index are identified and removed, achieving self-crowding removal of paternal and non-paternal particles under parallel acceleration.

[0029] Since spherical expansion inevitably leads to particle stacking and boundary crossing, this invention employs the following rigorous cascade elimination mechanism for cleaning: First, SDF wall penetration elimination is performed, quickly inputting the coordinate vectors of all coarse candidate particles into the SDF field obtained in step S2 to query the SDF scalar value, eliminating particles whose SDF calculation results are in the invalid domain, and applying liquid surface shear constraints; Second, an overlap culling test is performed, using Morton code to construct a one-dimensional space-filling curve (SFC), interleaving and encoding the discrete coordinates of three-dimensional spatial particles into a one-dimensional long integer: (4) In the formula: This represents the one-dimensional Morton code value obtained after dimensionality reduction of the particle. The bit width of the binary code representing integer coordinates; Representing the integer coordinates of the particles respectively X , Y , Z In binary representation m The value of the bit.

[0030] Based on this, the Morton code array is sorted in parallel using the GPU underlying architecture. In the sorted array, the Morton codes of sub-particles that interfere or stack with each other in space will be extremely similar or the same. Based on this, the system can quickly complete the self-crowding elimination of particles of the same parent and non-same parent, and complete the precise positioning and retention of particles.

[0031] This invention innovatively introduces a "pseudo-normal vector" mechanism, which perfectly handles the cusp singularity problem at nuclear reactor geometric components by weighting and accumulating the normals of adjacent surfaces. This fundamentally avoids the SDF positive and negative field scalar errors caused by rapid changes in normal angles and the inaccuracy of "wall-penetrating" judgments in subsequent physics calculations, greatly reducing the spatial memory burden and ensuring local accuracy. This invention employs "P / H adaptive and dynamic spatial downsampling fitting" technology, eliminating the need for a global high-density grid. Instead, it uses high-order polynomial fitting (P-adaptive) to extract distance field parameters and performs local octree splitting (H-adaptive) on error-boundary regions. Without increasing the memory burden of redundant grids, it achieves high-precision distance resolution for complex nuclear reactor boundaries, enabling efficient GPU parallel culling and defect-free modeling. After multi-directional growth of seed particles, this invention combines SDF physical constraints with Morton codes. Morton code is a one-dimensional space filling curve technique. It can reduce the three-dimensional proximity relationship to a one-dimensional array. By utilizing the high-speed sorting performance of the GPU, it can quickly lock and eliminate all redundant particles that are stacked together, ensuring the spatial smoothness of the generated particles and avoiding the distortion of the flow field and pressure field at the initial moment of CFD calculation.

[0032] Based on the above inventive concept, this invention also proposes a particle generation system for pretreatment of complex reactor geometries, comprising: The acquisition module is used to acquire the initial geometric mesh data of the nuclear reactor fluids and solids, establish the bounding box, and calculate the pseudo normal vectors of the vertices and edges of the bounding box geometric model. The directed distance field construction module is used to construct a directed distance field that includes polynomial order adaptive fitting and geometric topological morphological fission features, using the bounding box as the global search space and the pseudo-normal vector as the criterion for determining whether the spatial point is inside or outside the geometric model. The particle swarm generation module is used to determine the geometric normal pattern based on the physical properties of the fluid or solid region of the nuclear reactor, locate the activation source point of the seed particle based on the pattern, activate the seed particle in the spatial region with a positive scalar value in the directed distance field, take the activated seed particle as the parent node, determine the outward growth direction based on the gradient direction of the directed distance field, and grow outward along the preset growth direction to generate a coarse candidate particle cluster. The generation module is used to substitute the coordinates of each particle in the coarse candidate particle cluster into the constructed directed distance field, and remove out-of-bounds particles with negative directed distance values; the coordinate system of the remaining particles after removal is discretized in three-dimensional space, and the Morton code of each remaining particle is calculated; by sorting the Morton codes and comparing adjacent ones, stacked particles in the same Morton code index are identified and removed, and the final initial particle distribution of the nuclear reactor is obtained.

[0033] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A particle generation method for pretreatment of particles with complex reactor geometry, characterized in that, Includes the following steps: Acquire the initial geometric mesh data of the fluid and solid surfaces in the nuclear reactor, establish a bounding box, and calculate the pseudo normal vectors of the vertices and edges of the bounding box geometric model; Using the bounding box as the global search space and the pseudo normal vector as the criterion for determining whether a spatial point is located inside or outside the geometric model, a directed distance field containing polynomial order adaptive fitting and geometric topological morphological fission features is constructed using a dynamic spatial downsampling fitting strategy. The geometric normal pattern is determined based on the physical properties of the fluid or solid region of the nuclear reactor. The activation source point of the seed particle is located based on the pattern, and the seed particle is activated in the spatial region with a positive scalar value in the directed distance field. Using the activated seed particle as the parent node, the outward growth direction is determined according to the gradient direction of the directed distance field, and the particle grows outward along the preset growth direction to generate a coarse candidate particle cluster. Substitute the coordinates of each particle in the coarse candidate particle cluster into the constructed directed distance field, and remove out-of-bounds particles with negative directed distance values; discretize the coordinate system of the remaining particles in three-dimensional space, and calculate the Morton code of each remaining particle; by sorting and comparing the Morton codes, identify and remove stacked particles within the same Morton code index, and obtain the final initial particle distribution of the nuclear reactor.

2. The particle generation method for pretreatment of complex reactor geometries according to claim 1, characterized in that, The pseudo-normal vectors of the vertices and edges of the bounding box geometry model include vertex pseudo-normals and edge pseudo-normals; the vertex pseudo-normal is obtained by summing the normals of all adjacent triangular faces of the vertex according to the weighted summation of the interior angles of each face at the vertex; the edge pseudo-normal is obtained by summing the unit normals of two adjacent faces sharing the edge.

3. The particle generation method for pretreatment of complex reactor geometries according to claim 1, characterized in that, After acquiring the initial geometric mesh data of the nuclear reactor fluid and solid, an accelerated search topology is established using an octree and AABB bounding box tree structure; the bounding box is extended outward by 8 times the radius of the target particle based on the original volume mesh bounding box.

4. The particle generation method for pretreatment of complex reactor geometry particles according to claim 3, characterized in that, The polynomial-order adaptive fitting generates Gaussian integration points within the octree unit to construct a spatial distance matrix, and obtains the fitted distance value by solving the coefficients of the higher-order polynomial; the specific process is as follows: ； in, Represents coordinate point ( x , y , z The scalar value of the directed distance field obtained by fitting at point ); Indicates the highest order of the polynomial. The coordinates are respectively in x , y , z The exponent in the direction; This represents the spatial distance coefficient.

5. The particle generation method for pretreatment of complex reactor geometries according to claim 4, characterized in that, The geometric topology fission performs error verification on the polynomial fitting distance value. When the local error exceeds the preset threshold, the current octree unit is triggered to split into sub-mesh, and polynomial fitting is recursively performed until the error converges.

6. The particle generation method for pretreatment of complex reactor geometries according to claim 1, characterized in that, The coarse candidate particle cluster is generated by expanding the parent node outward in a specified direction to generate child node particles in a pre-allocated spatial grid memory. The child node particles are translated based on a set particle radius distance to generate a coarse candidate particle cluster with translation coordinates.

7. The particle generation method for pretreatment of complex reactor geometry particles according to claim 1, characterized in that, The removal of out-of-bounds particles with negative directed distance values ​​specifically means that when a particle's directed distance value is negative or less than a preset boundary tolerance, it is determined to be an out-of-bounds particle and removed.

8. The particle generation method for pretreatment of complex reactor geometry particles according to claim 1, characterized in that, The process of discretizing the coordinates of the remaining particles in three-dimensional space and calculating the Morton code for each remaining particle is specifically expressed as follows: ； in, This represents the one-dimensional Morton code value obtained after dimensionality reduction of the particle. The bit width of the binary code representing integer coordinates; Representing the integer coordinates of the particles respectively X , Y , Z In binary representation m The value of the bit.

9. A particle generation system for pretreatment of particles with complex reactor geometry, characterized in that, include: The acquisition module is used to acquire the initial geometric mesh data of the nuclear reactor fluids and solids and to establish the bounding box; And calculate the pseudo normal vectors of the vertices and edges of the bounding box geometry model; The directed distance field construction module is used to construct a directed distance field that includes polynomial order adaptive fitting and geometric topological morphological fission features by using the bounding box as the global search space, the pseudo normal vector as the sign determination criterion for whether the spatial point is located inside or outside the geometric model, and adopting a dynamic space downsampling fitting strategy. The particle swarm generation module is used to determine the geometric normal pattern based on the physical properties of the fluid or solid region of the nuclear reactor, locate the activation source point of the seed particle based on the pattern, and activate the seed particle in a spatial region with a positive scalar value in the directed distance field. Using the activated seed particle as the parent node, the outward growth direction is determined according to the gradient direction of the directed distance field, and the particle grows outward along the preset growth direction to generate a coarse candidate particle cluster. The generation module is used to substitute the coordinates of each particle in the coarse candidate particle cluster into the constructed directed distance field and remove out-of-bounds particles with negative directed distance values. The remaining particles after elimination are discretized in a three-dimensional coordinate system, and the Morton code of each remaining particle is calculated. By sorting the Morton codes and comparing adjacent ones, stacked particles within the same Morton code index are identified and eliminated, thus obtaining the final initial particle distribution of the nuclear reactor.

Images