A method and system for dynamic particle adaptive rendering based on an autonomous and controllable GPU

Abstract

This application provides a dynamic particle adaptive rendering method and system based on an autonomous and controllable GPU. Specifically, a distributed fiber optic sensor array is deployed within a monitoring area to collect multi-band vibration waveform data of dynamic particle flows under strong electromagnetic interference. Combined with particle hydrodynamic constraint rules, a topological association between nodes in a dynamic particle knowledge graph is generated. Through a federated learning framework, multiple GPU computing nodes collaboratively update the knowledge graph and dynamically allocate particle rendering resolution based on real-time changes in the topology. Finally, based on the fluctuation range of the vibration waveform data received in real time, GPU rendering control commands are generated, triggering the GPU computing nodes to adaptively adjust the rendering granularity of particle trajectories according to the priority of energy transfer paths. The technical solution provided in this application enables efficient and accurate rendering of dynamic particle flows under strong electromagnetic interference environments, improving the visualization effect of particle flow inside mechanical cavities.

Description

Technical Field

[0001] This application relates to the field of fiber optic sensing technology, and in particular to a dynamic particle adaptive rendering method and system based on an autonomous and controllable GPU. Background Technology

[0002] In multi-user collaborative virtual simulation particle systems for scientific research and education, dynamic particle adaptive rendering technology is a key requirement. These systems simulate complex physical, chemical, or biological processes, such as fluid dynamics or molecular motion. Users need to observe the dynamic changes of particles in real time and adjust the rendering precision according to their needs to achieve efficient visual presentation with limited hardware resources. In multi-user collaborative scenarios, different users may focus on different areas or details; the system needs to dynamically allocate computing resources to achieve adaptive particle rendering while ensuring the real-time performance and stability of the rendering. Furthermore, with the widespread adoption of domestically developed and controllable GPUs, how to fully utilize their computing power has become the core of the technological implementation.

[0003] Currently, existing solutions primarily employ a GPU-accelerated hierarchical particle rendering method. This method constructs a hierarchical structure for the particle system, dividing particles into different levels of detail and dynamically adjusting the rendering precision based on the user's perspective and particle density. The system utilizes the parallel computing power of the GPU to spatially partition and hierarchically cluster particles. Simultaneously, by analyzing the user's perspective and particle motion states in real time, it dynamically allocates computing resources, prioritizing the rendering of high-precision particles in areas of user interest, while using lower-precision rendering for distant or less important areas. Furthermore, this method incorporates a GPU load balancing algorithm to ensure efficient resource allocation in multi-user collaborative scenarios.

[0004] However, hierarchical particle rendering methods still have some drawbacks. First, they rely on fixed hierarchical partitioning rules, making it difficult to adapt to complex and ever-changing particle distributions and user needs, potentially leading to oversimplification or resource waste. Second, GPU load balancing algorithms are complex to implement, and resource contention during concurrent access by multiple users can easily cause performance degradation or rendering delays. Furthermore, existing solutions lack sufficient optimization support for domestically controlled GPUs, failing to fully utilize their unique hardware characteristics and limiting further improvements in rendering efficiency. These issues may affect system performance stability and user collaborative experience. Summary of the Invention

[0005] This application provides a dynamic particle adaptive rendering method and system based on an autonomous and controllable GPU to solve the problem of limited dynamic particle adaptive capability in the prior art.

[0006] Firstly, this application provides a dynamic particle adaptive rendering method based on an autonomous and controllable GPU, including:

[0007] A distributed fiber optic sensor array is deployed within the closed monitoring area of ​​the fluid machinery to collect multi-band vibration waveform data of dynamic particle flow under strong electromagnetic interference environment.

[0008] Based on the multi-band vibration waveform data and the preset particle hydrodynamic constraint rules, the node association topology of the dynamic particle knowledge graph is generated.

[0009] The dynamic particle knowledge graph among multiple GPU computing nodes is distributed and collaboratively updated through a federated learning framework. Based on the real-time changes in the inter-node association topology obtained during the distributed collaborative update process, a dynamic allocation strategy for particle rendering resolution of each GPU computing node is determined.

[0010] Based on the particle rendering resolution dynamic allocation strategy and the vibration waveform data fluctuation range fed back in real time by the fiber optic sensing array, a GPU rendering control command is generated for the dynamic particle flow inside the mechanical cavity. The rendering control command triggers the GPU computing node to perform adaptive adjustment of the rendering granularity of the particle trajectory based on the priority of the energy transfer path in the dynamic particle knowledge graph.

[0011] Optionally, the topological association between nodes in the dynamic particle knowledge graph is divided into multiple update units, each of which contains information on changes in particle collision characteristics and the probability of energy transfer path breakage.

[0012] A data transmission channel between GPU computing nodes is established in the federated learning framework through a hybrid communication protocol. A cooperative mechanism with local constraints is adopted to synchronize only the topological fragments between nodes that have a mechanical relationship with the target GPU computing node region, thereby realizing the distributed cooperative update of the dynamic particle knowledge graph.

[0013] In the federated learning framework, a priority queue is established, and local update weights are generated based on the product of the probability of energy transfer path breakage and particle collision frequency in the update unit.

[0014] When the local update weight exceeds a preset threshold, the rendering resolution adjustment event of the corresponding GPU computing node is triggered based on the sorting of the priority queue.

[0015] Based on the real-time changes in the inter-node topology obtained during the distributed collaborative update process, and combined with the rendering resolution adjustment event and the real-time bandwidth utilization of the data transmission channel, a dynamic allocation strategy for particle rendering resolution is generated.

[0016] Optionally, in the federated learning framework, a data transmission channel between GPU computing nodes is established through a hybrid communication protocol for transmitting updated data of the dynamic particle knowledge graph;

[0017] Based on the radius of curvature of the energy transfer path within the rendering area of ​​the target GPU computing node, the boundary of the synchronization region of the spatial locality constraint is defined. The boundary of the synchronization region contains inter-node associated topological segments that share a continuous energy flux vector with the rendering area of ​​the target GPU computing node.

[0018] Through the data transmission channel, only the topological fragments associated between nodes within the boundary of the synchronization region are synchronized, thereby realizing the distributed collaborative update of the dynamic particle knowledge graph.

[0019] Optionally, based on the phase difference jump characteristics of adjacent sensing nodes in the multi-band vibration waveform data collected by the fiber optic sensing array, the abrupt interval of the radius of curvature of the energy transfer path is identified.

[0020] A vortex core energy density distribution map is generated based on the abrupt interval of the radius of curvature of the energy transfer path. The vortex core energy density distribution map includes energy focusing hotspots formed by the reflection of particle flow on the inner wall of the mechanical cavity.

[0021] In the process of delineating the boundary of the synchronous region under the spatial locality constraint, the radial diffusion front of the energy focusing hotspot in the energy density distribution map of the vortex core is used as the dynamic adjustment baseline of the synchronous region boundary.

[0022] When the rate of change of the radius of curvature of the energy transfer path exceeds the preset fluid instability threshold, a real-time re-partitioning event of the boundary of the synchronization region is triggered. The real-time re-partitioning event prioritizes the retention of the inter-node associated topology fragments that share the continuous energy flux vector with the rendering region of the target GPU computing node, and performs reverse compensation for the energy conservation error of the unsynchronized region based on the position offset of the energy focusing hotspot.

[0023] Optionally, based on the spatial distribution characteristics of particle collision frequencies in the multi-band vibration waveform data collected by the fiber optic sensing array, the probability of energy transfer path breakage with hydrodynamic coupling to the rendering area of ​​the target GPU computing node is extracted from the dynamic particle knowledge graph.

[0024] The probability of energy transfer path breakage is convolved with the particle collision frequency in the corresponding region in the time domain to generate local topology update weights.

[0025] Optionally, based on the high-frequency phase difference jump characteristics in the multi-band vibration waveform data collected by the fiber optic sensing array, the instantaneous spatial coordinates of the energy focusing hotspot are identified.

[0026] Extract the radial diffusion front trajectory of the energy focusing hotspot from the energy density distribution map of the vortex core;

[0027] Based on the positional offset of the energy focusing hotspot within a preset time window, a leading-edge expansion rate compensation factor is generated. The compensation factor is dynamically calculated by the ratio of the fluctuation amplitude of the high-frequency phase difference jump event to the thermal expansion coefficient of the inner wall of the cavity.

[0028] During the dynamic adjustment of the boundary of the synchronization region, the tangent direction of the radial diffusion front trajectory is projected onto the graph increment transmission channel of the federated learning framework.

[0029] When the energy flux vector direction offset angle is detected to exceed the preset turbulent anisotropy threshold, a local retraction operation of the synchronization region boundary is triggered, and the dynamic adjustment baseline of the synchronization region boundary is redefined based on the curvature change point distribution of the radial diffusion front trajectory.

[0030] Optionally, high-frequency phase difference jump events and low-frequency energy attenuation gradient fluctuation ranges can be screened and extracted from the fluctuation range of multi-frequency vibration waveform data fed back in real time by the fiber optic sensing array.

[0031] The selected high-frequency phase difference jump events are spatiotemporally matched with the priority of energy transfer paths in the dynamic particle knowledge graph to generate spatiotemporal matching results.

[0032] Based on the particle rendering resolution dynamic allocation strategy, and combined with the spatiotemporal matching results, a dynamic resolution decay gradient table is generated.

[0033] Based on the dynamic resolution attenuation gradient table and the spatial distribution density of high-frequency phase difference jump events acquired in real time, GPU rendering control instructions for the dynamic particle flow inside the mechanical cavity are generated.

[0034] Secondly, this application provides a dynamic particle adaptive rendering system based on an independently controllable GPU, comprising:

[0035] The acquisition module deploys a distributed fiber optic sensor array within the closed monitoring area of ​​the fluid machinery to acquire multi-band vibration waveform data of dynamic particle flow under strong electromagnetic interference environment through the fiber optic sensor array.

[0036] The generation module generates the node association topology of the dynamic particle knowledge graph based on the multi-band vibration waveform data and the preset particle hydrodynamic constraint rules.

[0037] The computing module performs distributed collaborative updates of the dynamic particle knowledge graph among multiple GPU computing nodes through a federated learning framework, and determines the dynamic allocation strategy of particle rendering resolution for each GPU computing node based on the real-time changes in the inter-node association topology obtained during the distributed collaborative update process.

[0038] The adjustment module generates GPU rendering control instructions for the dynamic particle flow inside the mechanical cavity based on the particle rendering resolution dynamic allocation strategy and the vibration waveform data fluctuation range fed back by the fiber optic sensing array in real time. The rendering control instructions trigger the GPU computing node to perform adaptive adjustment of the rendering granularity of the particle trajectory based on the priority of the energy transfer path in the dynamic particle knowledge graph.

[0039] Thirdly, embodiments of this application provide a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement a dynamic particle adaptive rendering method based on an autonomous and controllable GPU as described in the first aspect above.

[0040] Fourthly, embodiments of this application provide a computer storage medium storing a computer program, which, when executed by a computer, implements a dynamic particle adaptive rendering method based on an autonomous and controllable GPU as described in the first aspect.

[0041] In this embodiment, a distributed fiber optic sensor array is deployed within a closed monitoring area of ​​fluid machinery to collect multi-band vibration waveform data of dynamic particle flow under strong electromagnetic interference. Based on the multi-band vibration waveform data and preset particle fluid dynamics constraint rules, a node association topology of a dynamic particle knowledge graph is generated. A federated learning framework is used to perform distributed collaborative updates of the dynamic particle knowledge graph among multiple GPU computing nodes. A dynamic particle rendering resolution allocation strategy for each GPU computing node is determined based on the real-time changes in the node association topology. Based on the dynamic particle rendering resolution allocation strategy and the fluctuation range of the vibration waveform data fed back in real-time by the fiber optic sensor array, GPU rendering control instructions for the dynamic particle flow inside the mechanical cavity are generated. These rendering control instructions trigger the GPU computing nodes to adaptively adjust the rendering granularity of particle trajectories based on the priority of energy transfer paths in the dynamic particle knowledge graph.

[0042] The technical solution of this application has the following beneficial effects:

[0043] This application presents a dynamic particle adaptive rendering method based on independently controllable GPUs. It collects multi-band vibration waveform data under strong electromagnetic interference environments through a distributed deployment of fiber optic sensor arrays, effectively achieving high-precision monitoring of dynamic particle flows. Based on the vibration waveform data and particle hydrodynamic constraint rules, it generates the inter-node association topology of a dynamic particle knowledge graph, providing structured physical model support for particle flow behavior. A federated learning framework is used to perform distributed collaborative updates of multiple GPU computing nodes, and a dynamic allocation strategy for particle rendering resolution is determined based on the real-time changes in the inter-node association topology, ensuring intelligent optimization and load balancing of multi-GPU computing resources. Finally, GPU rendering control instructions are generated by combining the fluctuation range of the vibration waveform data, and the rendering granularity of particle trajectories is adaptively adjusted based on the priority of energy transfer paths. This significantly improves the real-time performance and accuracy of dynamic particle rendering while reducing redundant consumption of computing resources, achieving efficient and stable rendering in complex environments.

[0044] Furthermore, by dynamically calculating the spatiotemporal increment unit and local topology update weight, precise capture of particle collisions and changes in energy transfer paths is achieved. The synchronization mechanism based on spatial locality constraints and the graph increment transmission channel significantly reduce the communication overhead between GPU computing nodes and improve data transmission efficiency. Through the intelligent triggering of the topology evolution priority queue and the rendering resolution reallocation event, the dynamic optimization allocation of GPU computing resources is ensured, thereby achieving high-precision and high-efficiency particle rendering in complex fluid environments, while reducing system latency and resource consumption, and meeting the high-performance rendering requirements of dynamic particle flows in closed mechanical cavities.

[0045] These or other aspects of this application will become more apparent from the description of the following embodiments. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart of a dynamic particle adaptive rendering method based on an autonomous and controllable GPU provided in this application is shown;

[0048] Figure 2 A schematic diagram of the structure of a dynamic particle adaptive rendering system based on an autonomous and controllable GPU provided in this application is shown.

[0049] Figure 3A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0050] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0051] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0052] This project aims to develop a multi-user collaborative virtual simulation particle system based on an independently controllable GPU. It collects multi-band vibration waveform data of dynamic particle flows through a distributed fiber optic sensor array and constructs a dynamic particle knowledge graph by combining particle hydrodynamic constraint rules. A federated learning framework is used to achieve distributed collaborative updates and dynamic allocation of rendering resolution across multiple GPU computing nodes. Finally, GPU rendering control commands are generated for the dynamic particle flows inside mechanical cavities, enabling adaptive adjustment of particle trajectory rendering granularity. This system can efficiently and accurately render complex dynamic particle flows in real-time under strong electromagnetic interference environments, meeting the high-performance requirements of multi-user collaborative virtual simulation in scientific research and education.

[0053] 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 some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] Figure 1 A flowchart of a dynamic particle adaptive rendering method based on an autonomous and controllable GPU is provided in this application embodiment, as follows: Figure 1 As shown, the method includes:

[0055] 101. A distributed fiber optic sensor array is deployed within the closed monitoring area of ​​the fluid machinery to collect multi-band vibration waveform data of dynamic particle flow under strong electromagnetic interference environment through the fiber optic sensor array.

[0056] In this step, the fiber optic sensor array is a distributed sensor network composed of multiple optical fibers. Its core principle is to use the optical properties of optical fibers (such as interference, scattering, or reflection) to detect physical changes in the environment. Within the enclosed monitoring area of ​​the fluid machinery, the fiber optic sensor array is deployed to acquire multi-band vibration waveform data of dynamic particle flows under strong electromagnetic interference.

[0057] Dynamic particle flow refers to the flow of particles or fluids within fluid machinery, whose motion state changes due to factors such as flow velocity, pressure, and turbulence.

[0058] Multi-band vibration waveform data refers to vibration signals in different frequency ranges collected by fiber optic sensing arrays, including low-frequency, medium-frequency, and high-frequency vibrations. These data can comprehensively reflect the motion characteristics of dynamic particle flows.

[0059] In this embodiment, firstly, within the enclosed monitoring area of ​​the fluid machinery, fiber optic sensor arrays are deployed in a distributed manner to ensure coverage of the entire monitoring area. The fiber optic sensor arrays monitor the vibration waveforms of the dynamic particle flow inside the machinery in real time using the optical properties of optical fibers (such as Brillouin scattering or Rayleigh scattering). Secondly, due to the electromagnetic interference resistance, high temperature resistance, and corrosion resistance of fiber optic sensors, they can operate stably in strong electromagnetic environments. The acquired multi-frequency vibration waveform data includes low-frequency (such as mechanical vibration or water flow impact), mid-frequency (such as turbulent vibration), and high-frequency (such as bubble bursting or microparticle collisions) vibration signals. Finally, these data are transmitted via optical fiber to a data processing center, where they undergo signal amplification, filtering, and digitization to form a multi-frequency vibration waveform dataset available for subsequent analysis.

[0060] Inside the turbine of a large hydroelectric power station, a distributed fiber optic sensor array was installed. Due to the strong electromagnetic interference and high-temperature, high-pressure environment inside the turbine, traditional sensors cannot operate stably. The fiber optic sensor array successfully acquired multi-band vibration waveform data of the dynamic particle flow inside the turbine, including low-frequency water flow impact vibration (0-100Hz), mid-frequency turbulent vibration (100-1000Hz), and high-frequency bubble bursting vibration (1000-10000Hz). This data provides a solid foundation for subsequent analysis, helping engineers better understand the fluid dynamics characteristics inside the turbine.

[0061] 102. Based on the multi-band vibration waveform data and the preset particle hydrodynamic constraint rules, generate the node association topology of the dynamic particle knowledge graph;

[0062] In this step, the dynamic particle knowledge graph is a graphical representation based on multi-band vibration waveform data and particle hydrodynamic constraint rules, used to describe the topological relationships between nodes (such as particles, fluid regions, or energy transfer paths) in a dynamic particle flow.

[0063] Particle fluid dynamics constraint rules are rules set according to fluid dynamics principles (such as the continuity equation, momentum conservation equation, and energy conservation equation) and are used to guide the generation of knowledge graphs.

[0064] The nodes of a knowledge graph represent key elements in a dynamic particle flow (such as fluid regions or particle swarms), and the edges represent the interaction relationships between these elements.

[0065] In this embodiment, the acquired multi-band vibration waveform data is first preprocessed, including filtering, denoising, and feature extraction. Then, based on particle fluid dynamics constraint rules, the behavior patterns of the dynamic particle flow are analyzed. For example, the mass conservation characteristics of the fluid are analyzed using the continuity equation, the momentum transfer characteristics using the momentum conservation equation, and the energy distribution characteristics using the energy conservation equation. Next, graph theory algorithms are used to represent the nodes and their relationships in the dynamic particle flow as nodes and edges of a knowledge graph. For example, nodes can represent fluid regions or particle swarms, and edges can represent energy transfer paths or momentum exchange relationships. Finally, the topological relationships between nodes in the dynamic particle knowledge graph are generated, clearly demonstrating the structure and interaction relationships of the dynamic particle flow.

[0066] Inside the turbine, based on collected multi-frequency vibration waveform data, the system analyzes the dynamic behavior of the water flow using particle fluid dynamics constraint rules. For example, low-frequency vibration data is used to identify the overall flow direction, mid-frequency vibration data is used to identify turbulent regions, and high-frequency vibration data is used to identify bubble distribution. A graph neural network is used to generate the associated topology of the dynamic particle flow inside the turbine, clearly demonstrating the interaction between water flow, turbulence, and bubbles. For example, a node in the knowledge graph represents the fluid region of the turbine blade, and edges represent the energy transfer paths between the water flow and the blade.

[0067] 103. A distributed collaborative update of the dynamic particle knowledge graph among multiple GPU computing nodes is performed through a federated learning framework, and a dynamic allocation strategy for particle rendering resolution of each GPU computing node is determined based on the real-time changes in the inter-node association topology obtained during the distributed collaborative update process.

[0068] In this step, the federated learning framework is a distributed machine learning approach that allows multiple GPU computing nodes to collaboratively update the model without sharing the original data.

[0069] Distributed collaborative updates of dynamic particle knowledge graphs refer to the process by which multiple GPU computing nodes locally update the knowledge graph based on local data and then globally integrate the updates through a federated learning framework.

[0070] The particle rendering resolution dynamic allocation strategy allocates different rendering resolutions to each GPU computing node based on the real-time changes in the topology between nodes, in order to optimize computing resources. For example, high rendering resolutions are allocated to regions with high dynamic changes, and low rendering resolutions are allocated to regions with low dynamic changes.

[0071] In this embodiment, firstly, each GPU computing node locally updates the dynamic particle knowledge graph based on local data, including adjusting the node association topology and optimizing energy transfer paths. For example, one GPU computing node is responsible for updating the association topology of the turbine blade region, while another GPU computing node is responsible for updating the association topology of the water flow region. Then, through a federated learning framework, the local update results of each node are integrated to generate a global dynamic particle knowledge graph. For example, weighted averaging or gradient aggregation methods are used to fuse the update results of each node into a global knowledge graph. Next, based on the real-time changes in the association topology between nodes, the system allocates different particle rendering resolutions to each GPU computing node. For example, high rendering resolutions are allocated to turbulent regions with high dynamic changes, and low rendering resolutions are allocated to stable regions with low dynamic changes. Finally, a dynamic allocation strategy for particle rendering resolution of each GPU computing node is determined, optimizing the use of computing resources.

[0072] In the turbine monitoring system, multiple GPU computing nodes are responsible for updating the dynamic particle knowledge graph for different regions. Through a federated learning framework, the update results from each node are integrated into a global knowledge graph. For example, when the water flow velocity inside the turbine suddenly increases, the system detects the change in the topological relationships between nodes and dynamically adjusts the rendering resolution of the GPU computing nodes: increasing the rendering resolution for turbulent regions (e.g., 1024x1024 pixels) and decreasing the rendering resolution for stable regions (e.g., 512x512 pixels), thereby optimizing the use of computing resources.

[0073] 104. Based on the particle rendering resolution dynamic allocation strategy and the vibration waveform data fluctuation range fed back by the fiber optic sensing array in real time, a GPU rendering control instruction for the dynamic particle flow inside the mechanical cavity is generated. The rendering control instruction triggers the GPU computing node to perform adaptive adjustment of the rendering granularity of the particle trajectory based on the priority of the energy transfer path in the dynamic particle knowledge graph.

[0074] In this step, the GPU rendering control command is generated based on the particle rendering resolution dynamic allocation strategy and the vibration waveform data fluctuation range fed back in real time by the fiber optic sensor array. It is used to control the GPU computing node to render the dynamic particle flow.

[0075] The priority of energy transfer paths refers to the degree of importance of different paths in the knowledge graph to the overall dynamic particle flow. Adaptive adjustment of rendering granularity refers to dynamically adjusting the rendering precision of particle trajectories based on priority. For example, high-priority energy transfer paths are rendered with high precision, while low-priority paths are rendered with low precision.

[0076] In this embodiment, firstly, the state of the current dynamic particle flow is determined based on the fluctuation range of the vibration waveform data fed back in real time by the fiber optic sensing array. Then, based on a dynamic allocation strategy for particle rendering resolution, specific rendering control instructions are generated for each GPU computing node. These instructions include the number of particles to be rendered, the resolution, and the rendering order. For example, particle trajectories in turbulent regions are rendered with high precision to capture details, while particle trajectories in stable regions are rendered with low precision to save computing resources. Finally, the GPU computing node adaptively adjusts the rendering granularity of the particle trajectories according to the priority of energy transfer paths in the dynamic particle knowledge graph. For example, high-priority energy transfer paths are rendered with high precision, while low-priority paths are rendered with low precision.

[0077] In the turbine monitoring system, when a sudden increase in water flow velocity is detected, the system generates GPU rendering control instructions based on the fluctuation range of real-time vibration waveform data. For example, high-precision rendering (e.g., rendering 10,000 particles per frame) is performed on particle trajectories in turbulent regions to capture details, while low-precision rendering (e.g., rendering 1,000 particles per frame) is performed on particle trajectories in stable regions to save computing resources. Ultimately, the GPU computing nodes complete the efficient rendering of the dynamic particle flow inside the turbine based on the priority of energy transfer paths, providing engineers with clear visualization results.

[0078] In summary, steps 101 to 104 successfully acquired multi-band vibration waveform data of dynamic particle flow under strong electromagnetic interference by distributing fiber optic sensor arrays within the closed monitoring area of ​​the fluid machinery. Based on this data and particle fluid dynamics constraints, a topological association between nodes in the dynamic particle knowledge graph was generated, clearly demonstrating the structure and interaction relationships of the dynamic particle flow. Through a federated learning framework, multiple GPU computing nodes collaboratively updated the knowledge graph and dynamically allocated particle rendering resolution based on the real-time changes in the topological association between nodes, optimizing the use of computing resources. Finally, based on the rendering resolution strategy and the fluctuation range of the real-time vibration waveform data, GPU rendering control commands were generated, achieving efficient rendering of the dynamic particle flow inside the mechanical cavity. This scheme significantly improves the accuracy and efficiency of dynamic particle flow monitoring and rendering under strong electromagnetic interference, providing strong support for intelligent monitoring and fault diagnosis of fluid machinery.

[0079] To address the challenges of uneven resource allocation, high communication overhead, and the difficulty in balancing rendering accuracy and efficiency across multiple GPU computing nodes in dynamic particle flow simulations, a distributed collaborative update system based on a federated learning framework was developed. This system achieves efficient multi-GPU collaboration and significantly improved rendering performance through intelligent allocation of computing resources, optimized communication mechanisms, and dynamic adjustment of rendering resolution, providing high-performance support for complex dynamic particle flow simulations. In some embodiments, step 103 involves performing distributed collaborative updates of the dynamic particle knowledge graph among multiple GPU computing nodes using a federated learning framework, and determining a dynamic allocation strategy for particle rendering resolution for each GPU computing node based on the real-time changes in the inter-node association topology obtained during the distributed collaborative update process. This includes:

[0080] 201. The topological association between nodes in the dynamic particle knowledge graph is divided into spatiotemporal increment units, wherein the spatiotemporal increment units include particle collision mode mutation characteristics and energy transfer path breakage probability.

[0081] In step 201, the dynamic particle knowledge graph is a graph model used to describe the interactions and energy transfer relationships between particles. Nodes represent particles, and edges represent the relationships between particles. The spatiotemporal increment unit is the smallest unit that divides the graph in both time and space dimensions, used to describe the particle behavior and energy transfer characteristics of a local region. The particle collision mode mutation feature describes the characteristics of significant changes in particle collision behavior, such as abrupt changes in collision frequency or angle. The energy transfer path breakage probability describes the likelihood of an interruption in the energy transfer path between particles, calculated based on the distance between particles and the interaction strength.

[0082] In this embodiment, the dynamic particle knowledge graph is first spatiotemporally segmented into multiple spatiotemporal increment units. Each spatiotemporal increment unit describes the particle behavior in a local region by analyzing the changing characteristics of particle collision modes (such as abrupt changes in collision frequency and angle) and the probability of energy transfer path breakage (calculated based on inter-particle distance and interaction strength). This segmentation method allows focus on the dynamic changes in local regions, providing a foundation for subsequent incremental processing and rendering optimization. Furthermore, the segmentation of spatiotemporal increment units also considers the continuity of the time dimension, ensuring that the evolution of particle behavior can be captured over time, thereby providing more accurate data support for subsequent real-time updates and rendering.

[0083] 202. Establish an incremental transmission channel for the knowledge graph between GPU computing nodes in the federated learning framework through a hybrid communication protocol. Adopt a particle subset collaboration mechanism with spatial locality constraints to synchronize only the associated topological fragments between nodes that have a hydrodynamic coupling relationship with the rendering area of ​​the target GPU computing node, thereby realizing the distributed collaborative update of the dynamic particle knowledge graph.

[0084] In step 202, the hybrid communication protocol is a protocol that combines multiple communication technologies (such as TCP / IP and RDMA) for efficient data transmission. The map increment transmission channel is a communication channel used to transmit spatiotemporal increment units between GPU computing nodes. Spatial locality constraints are constraints based on physical spatial proximity, used to filter subsets of particles that need synchronization. Fluid dynamic coupling describes the mutual influence relationships between particles under fluid dynamics.

[0085] In this embodiment, firstly, an incremental graph transmission channel is established between GPU computing nodes within the federated learning framework using a hybrid communication protocol. Next, a particle subset synchronization mechanism with spatial locality constraints is employed to filter out associated topological fragments between nodes that have a hydrodynamic coupling relationship with the rendering region of the target GPU computing node, and only these fragments are synchronized. This mechanism reduces unnecessary data transmission, improves communication efficiency, and ensures the accuracy of the rendering region. Furthermore, the selection of the hybrid communication protocol considers dynamic changes in the network environment, enabling dynamic adjustment of the transmission strategy based on real-time network bandwidth and latency, further optimizing communication performance.

[0086] 203. In the federated learning framework, establish a topological evolution priority queue for the dynamic particle knowledge graph, and generate local topological update weights based on the product of the probability of energy transfer path breakage and particle collision frequency in the spatiotemporal increment unit.

[0087] In step 203, the topology evolution priority queue is a queue structure used to manage the priority of local topology updates in the dynamic particle knowledge graph. The local topology update weight is an index generated by multiplying the probability of energy transfer path breakage by the particle collision frequency, used to measure the significance of local topological changes. The probability of energy transfer path breakage refers to the probability that a particle energy transfer path breaks in a spatiotemporal increment unit, used to measure the stability of the topological structure. The particle collision frequency refers to the frequency at which particles collide in a spatiotemporal increment unit, used to measure the intensity of particle activity. The local topology update weight is a weight generated based on the product of the probability of energy transfer path breakage and the particle collision frequency, used to guide the updating of the local topology.

[0088] In this embodiment, a topology evolution priority queue for a dynamic particle knowledge graph is first established within the federated learning framework. The establishment of the topology evolution priority queue typically employs a priority sorting algorithm, for example, by using local topology update weights as the priority sorting criterion. Next, local topology update weights are generated based on the product of the energy transfer path breakage probability and the particle collision frequency in the spatiotemporal increment unit. The calculation of local topology update weights typically employs a weighted product algorithm, for example, by multiplying the energy transfer path breakage probability by the particle collision frequency to generate weight values. Through this step, the system can scientifically quantify the priority of local topology updates, providing a basis for subsequent rendering optimization.

[0089] 204. When the local topology update weight exceeds the preset fluid turbulence threshold, the rendering resolution reallocation event of the corresponding GPU computing node is activated based on the sorting of the topology evolution priority queue.

[0090] In step 204, the fluid turbulence threshold is a preset threshold used to determine whether a rendering resolution reassignment event needs to be triggered. The fluid turbulence threshold refers to a preset threshold used to determine the intensity of fluid turbulence and to trigger the rendering resolution reassignment event. The topology evolution priority queue refers to a priority queue generated based on local topology update weights, used to guide the order of topology updates. A GPU compute node refers to a GPU compute unit used to execute rendering tasks and to implement rendering optimization. The rendering resolution reassignment event refers to an event that reallocates the rendering resolution based on local topology update weights, used to optimize rendering quality.

[0091] In this embodiment, it is first determined whether the local topology update weight exceeds a preset fluid turbulence threshold. The fluid turbulence threshold is typically set based on fluid dynamics models or experimental data, for example, by analyzing the impact of fluid turbulence on rendering quality to set a reasonable threshold. When the local topology update weight exceeds the fluid turbulence threshold, a rendering resolution reallocation event for the corresponding GPU computing node is activated based on the sorting of the topology evolution priority queue. The rendering resolution reallocation event typically employs a dynamic resolution adjustment algorithm, for example, by increasing the rendering resolution of high-priority areas and decreasing the rendering resolution of low-priority areas to optimize rendering quality and system performance. Through this step, the system can optimize the rendering resolution in real time, ensuring maximum rendering quality and system performance.

[0092] 205. Based on the rendering resolution reallocation event and the real-time bandwidth occupancy rate of the map incremental transmission channel, generate a dynamic particle rendering resolution allocation strategy for the sealed mechanical cavity.

[0093] In step 205, the rendering resolution reallocation event is the event that triggers the GPU compute node to adjust the rendering resolution. The real-time bandwidth utilization of the map incremental transmission channel is an indicator describing the current bandwidth usage of the communication channel. A sealed mechanical cavity refers to a closed physical space used to simulate or analyze particle behavior.

[0094] In this embodiment, firstly, the particle rendering resolution of each region within the sealed mechanical cavity is dynamically adjusted based on the rendering resolution reallocation event and the real-time bandwidth occupancy of the map increment transmission channel. When communication bandwidth is sufficient, the rendering resolution of highly dynamic regions is increased; when bandwidth is limited, priority is given to ensuring the rendering accuracy of critical regions. Finally, this dynamic allocation strategy optimizes resource utilization while ensuring rendering quality. Furthermore, the dynamic allocation strategy also considers the differences in physical characteristics of different regions, allocating higher rendering resolutions to highly dynamic regions, thereby further improving the accuracy of the simulation results.

[0095] Here is a specific example:

[0096] In a fluid dynamics simulation scenario within an aero-engine combustion chamber, the dynamic particle knowledge graph is first segmented into multiple spatiotemporal incremental units, and the particle collision patterns and energy transfer paths within each unit are analyzed. For example, in the high-temperature and high-pressure region of the combustion chamber, the particle collision frequency increases significantly, and the probability of energy transfer path breakage is also high; these regions are classified as highly dynamically changing spatiotemporal incremental units. Next, a graph incremental transmission channel is established between GPU computing nodes using a hybrid communication protocol, and a particle subset synchronization mechanism with spatial locality constraints is employed to synchronize only the associated topological fragments between nodes that have a fluid dynamic coupling relationship with the target GPU computing node's rendering area. For example, in the flame propagation region of the combustion chamber, only the particle subset related to flame propagation is synchronized, reducing unnecessary data transmission. Then, a topology evolution priority queue is established, generating local topology update weights based on the product of the energy transfer path breakage probability and the particle collision frequency. For example, in the flame propagation region of the combustion chamber, due to the high particle collision frequency and energy transfer path breakage probability, the local topology update weight exceeds a preset fluid turbulence threshold, triggering a rendering resolution reallocation event. Finally, based on the rendering resolution reallocation event and the real-time bandwidth utilization of the incremental transmission channel, the particle rendering resolution of each region within the combustion chamber is dynamically adjusted. For example, the rendering resolution is increased in the flame propagation region to ensure accurate simulation of the flame shape and propagation process, while the rendering resolution is appropriately reduced in other regions to optimize resource utilization.

[0097] In summary, steps 201 to 205 achieve high efficiency and accuracy in simulating particle behavior within a closed mechanical cavity, while optimizing resource utilization and communication efficiency, significantly improving data transmission efficiency and rendering area accuracy. This provides reliable technical support for simulating complex physical scenarios. Furthermore, this method considers the continuity of time and the differences in physical characteristics across different regions, further enhancing the accuracy and real-time performance of the simulation results, thus providing reliable technical support for simulating complex physical scenarios.

[0098] To address the issues of low data transmission efficiency, high latency in emergency incremental unit transmission, and unreasonable bandwidth resource allocation among multi-GPU computing nodes, a graph incremental transmission system based on a hybrid communication protocol was developed. This system achieves high-priority transmission of emergency incremental units, intelligent allocation of bandwidth resources, and a significant improvement in data transmission efficiency, providing reliable support for efficient collaboration among multi-GPU computing nodes. In some embodiments, step 203, which establishes a graph incremental transmission channel between GPU computing nodes within the federated learning framework using a hybrid communication protocol, employs a spatial locality-constrained particle subset synchronization mechanism to synchronize only the associated topological fragments between nodes that have a hydrodynamic coupling relationship with the target GPU computing node's rendering region, including:

[0099] 301. In the federated learning framework, a data transmission channel between GPU computing nodes is established through a hybrid communication protocol for transmitting updated data of the dynamic particle knowledge graph;

[0100] In step 301, the federated learning framework refers to a framework that supports distributed computing and collaborative learning, used to update and optimize the dynamic particle knowledge graph. The hybrid communication protocol refers to a protocol that combines multiple communication technologies for efficient data transmission. GPU computing nodes refer to GPU computing units used to perform rendering tasks, used for rendering optimization. Data transmission channels refer to channels used to transmit data between GPU computing nodes, used to support distributed collaborative updates. The dynamic particle knowledge graph refers to a knowledge graph describing the behavior and relationships of particles in spatiotemporal increment units, used to guide rendering optimization. Updated data refers to the update information of the dynamic particle knowledge graph, used to synchronize and optimize the knowledge graph.

[0101] In this embodiment, a data transmission channel between GPU computing nodes is first established within the federated learning framework using a hybrid communication protocol. This hybrid protocol typically combines high-speed network communication with low-latency communication technologies, such as Ethernet and Infin iBand, to achieve efficient data transmission. Next, the updated data of the dynamic particle knowledge graph is transmitted through this data transmission channel. The data transmission process typically employs data compression and chunked transmission techniques, such as compressing the updated data and transmitting it in chunks to improve transmission efficiency and reliability. Through this step, the system can efficiently establish a data transmission channel, providing support for subsequent distributed collaborative updates.

[0102] 302. Based on the radius of curvature of the energy transfer path within the rendering area of ​​the target GPU computing node, the boundary of the synchronization region of the spatial locality constraint is defined, wherein the boundary of the synchronization region contains inter-node associated topological segments that share a continuous energy flux vector with the rendering area of ​​the target GPU computing node.

[0103] In step 302, the target GPU computing node refers to the GPU computing unit currently executing the rendering task, used to determine the synchronization region boundary. The rendering region refers to the area rendered by the target GPU computing node, used to delineate the synchronization region boundary. The radius of curvature of the energy transfer path refers to the radius of curvature of the energy transfer path, used to measure the degree of curvature of the energy transfer path. Spatial locality constraint refers to the constraint conditions based on spatial locality partitioning, used to determine the synchronization region boundary. The synchronization region boundary refers to the boundary partitioned according to spatial locality constraints, used to limit the synchronization range. The continuous energy flux vector refers to the continuous energy flux vector in the energy transfer path, used to determine the inter-node associated topological segments. The inter-node associated topological segments refer to the topological segments that share the continuous energy flux vector with the rendering region of the target GPU computing node, used for synchronizing and updating the knowledge graph.

[0104] In this embodiment, the synchronization region boundary for spatial locality constraints is first delineated based on the radius of curvature of the energy transfer path within the target GPU computing node's rendering area. The delineation of the synchronization region boundary typically employs a radius of curvature threshold method; for example, by setting a radius of curvature threshold, regions with radii less than the threshold are designated as synchronization region boundaries. Next, the system identifies inter-node associated topological segments within the synchronization region boundary that share continuous energy flux vectors with the target GPU computing node's rendering area. The determination of inter-node associated topological segments typically uses an energy flux vector matching algorithm; for example, by matching continuous energy flux vectors, associated topological segments are identified. Through this step, the system can scientifically delineate the synchronization region boundary, providing a basis for subsequent distributed collaborative updates.

[0105] 303. Through the data transmission channel, only the topological fragments associated between nodes within the boundary of the synchronization region are synchronized to achieve distributed collaborative updating of the dynamic particle knowledge graph.

[0106] In step 303, the data transmission channel refers to the channel used to transmit data between GPU computing nodes to support distributed collaborative updates. The synchronization region boundary refers to the boundary defined according to spatial locality constraints to limit the synchronization range. The inter-node associated topology fragment refers to a topology fragment that shares a continuous energy flux vector with the rendering region of the target GPU computing node, used for synchronizing and updating the knowledge graph. Distributed collaborative updates refer to the collaborative updates of the dynamic particle knowledge graph achieved through distributed computing, used to optimize the knowledge graph.

[0107] In this embodiment, firstly, the topological fragments associated between nodes within the region boundary are synchronized only through the data transmission channel. The synchronization process typically employs incremental synchronization techniques, such as transmitting only updated data to reduce the amount of synchronized data. Next, distributed collaborative updates of the dynamic particle knowledge graph are implemented. Distributed collaborative updates typically employ consensus protocols or distributed locking mechanisms, such as ensuring the consistency of data across nodes to achieve collaborative updates. Through this step, the system can efficiently achieve distributed collaborative updates of the dynamic particle knowledge graph, ensuring the accuracy and real-time performance of the knowledge graph.

[0108] Here is a specific example:

[0109] In the internal fluid monitoring system of a turbine engine, the system first establishes a data transmission channel between GPU computing nodes within a federated learning framework using a hybrid communication protocol. For example, TCP / IP and RDMA technologies are employed to ensure efficient transmission of updated data from the dynamic particle knowledge graph between GPU computing nodes, such as real-time transmission of energy transfer path changes in high-frequency turbulence regions to the target GPU node. Next, the system delineates synchronization region boundaries based on the radius of curvature of the energy transfer paths within the target GPU node's rendering area. For instance, when the radius of curvature is less than 5 mm, the target GPU node's rendering area and its adjacent areas are designated as synchronization region boundaries. These boundaries contain inter-node associated topological fragments that share continuous energy flux vectors with the target GPU node's rendering area, such as energy transfer paths between high-frequency turbulence and low-frequency eddy current regions. Subsequently, the system transmits only the inter-node associated topological fragments within the synchronization region boundaries through the data transmission channel. For example, only energy transfer path change data in high-frequency turbulence regions is transmitted, reducing unnecessary data exchange and enabling distributed collaborative updates of the dynamic particle knowledge graph. Through this mechanism, the system can accurately monitor the internal fluid dynamics of a turbine engine, providing reliable data support for mechanical performance optimization.

[0110] In summary, by dynamically adjusting the bandwidth allocation ratio of the incremental transmission channel in steps 301 to 304, the transmission needs of high-priority data are prioritized in emergency situations, while ensuring the basic transmission needs of non-urgent data are met, further optimizing resource utilization and communication efficiency. Ultimately, this method significantly improves the overall performance and reliability of complex physical scenario simulations while ensuring the accuracy and real-time nature of the simulation results.

[0111] To address the issues of inaccurate synchronization region boundary delineation and accumulated energy conservation errors caused by changes in energy transfer paths within the rendering area of ​​multi-GPU computing nodes, a dynamic adjustment system for synchronization region boundaries based on the radius of curvature of the energy transfer path was developed. This system achieves accurate delineation of synchronization region boundaries, real-time tracking of energy transfer paths, and effective control of energy conservation errors, providing a reliable guarantee for efficient collaborative rendering of multi-GPU computing nodes. In some embodiments, step 303 involves establishing a topology evolution priority queue for a dynamic particle knowledge graph within the federated learning framework, generating local topology update weights based on the product of the energy transfer path breakage probability and particle collision frequency in the spatiotemporal increment unit, and when the local topology update weights exceed a preset fluid turbulence threshold, activating a rendering resolution reallocation event for the corresponding GPU computing node based on the order of the topology evolution priority queue, including:

[0112] 401. Based on the phase difference jump characteristics of adjacent sensing nodes in the multi-band vibration waveform data collected by the fiber optic sensing array, identify the abrupt interval of the radius of curvature of the energy transfer path.

[0113] In step 401, the fiber optic sensing array refers to an array composed of multiple fiber optic sensors used to acquire multi-band vibration waveform data. Multi-band vibration waveform data refers to the vibration waveform data acquired by the fiber optic sensing array, containing vibration information across multiple frequency bands. Adjacent sensing nodes refer to adjacent sensor nodes in the fiber optic sensing array, used to detect the phase difference of the vibration waveform data. Phase difference jump characteristics refer to the phase difference jump characteristics of the vibration waveform data between adjacent sensing nodes, used to identify abrupt changes in the radius of curvature of the energy transfer path. The radius of curvature of the energy transfer path refers to the radius of curvature of the energy transfer path, used to measure the degree of bending of the energy transfer path. Abrupt change interval refers to the interval where the radius of curvature of the energy transfer path changes abruptly, used to guide the generation of the vortex core energy density distribution map.

[0114] In this embodiment, the phase difference jump characteristics of adjacent sensing nodes are first extracted based on multi-band vibration waveform data acquired by an optical fiber sensor array. The extraction of phase difference jump characteristics typically employs signal processing techniques, such as analyzing the phase changes of the vibration waveform data using Fourier transform or wavelet transform. Next, based on the phase difference jump characteristics, abrupt intervals in the radius of curvature of the energy transfer path are identified. Identification of abrupt intervals typically employs threshold detection or abrupt point detection algorithms, for example, by setting a phase difference jump threshold. Through this step, the system can scientifically identify abrupt intervals in the radius of curvature of the energy transfer path, providing a basis for the subsequent generation of the vortex core energy density distribution map.

[0115] 402. Generate a vortex core energy density distribution map based on the abrupt change interval of the radius of curvature of the energy transfer path, wherein the vortex core energy density distribution map includes energy focusing hotspots formed by the reflection of particle flow on the inner wall of the mechanical cavity;

[0116] In step 402, the abrupt change interval of the energy transfer path curvature radius refers to the interval where the energy transfer path curvature radius changes abruptly, used to guide the generation of the vortex core energy density distribution map. The vortex core energy density distribution map is a graphic depicting the energy density distribution of the vortex core, used to guide the dynamic adjustment of the synchronization region boundary. The particle flow refers to the particle flow within the mechanical cavity, used to form energy focusing hotspots. The inner wall of the mechanical cavity refers to the inner wall of the mechanical cavity, used to reflect the particle flow. The energy focusing hotspot is the energy focusing area formed by the reflection of the particle flow on the inner wall of the mechanical cavity, used to guide the dynamic adjustment of the synchronization region boundary.

[0117] In this embodiment, a vortex core energy density distribution map is first generated based on the abrupt change in the radius of curvature of the energy transfer path. The generation of this map typically employs energy density calculation and visualization techniques, such as calculating the energy density of the vortex core region and generating the distribution map. Next, energy focusing hotspots formed by the reflection of particle flow on the inner wall of the mechanical cavity are identified within the vortex core energy density distribution map. Identification of these hotspots typically utilizes peak detection or clustering analysis algorithms, such as identifying hotspot regions by detecting peak energy density values. Through this step, the system can scientifically generate a vortex core energy density distribution map, providing a basis for subsequent dynamic adjustment of the synchronization region boundary.

[0118] 403. In the process of dividing the boundary of the synchronous region under the spatial locality constraint, the radial diffusion front of the energy focusing hotspot in the energy density distribution map of the vortex core is used as the dynamic adjustment baseline of the boundary of the synchronous region.

[0119] In step 403, the vortex core energy density distribution map refers to a graphic depicting the energy density distribution of the vortex core, used to guide the dynamic adjustment of the synchronization region boundary. The energy focusing hotspot refers to the energy focusing region formed by the reflection of particle flow on the inner wall of the mechanical cavity, used to guide the dynamic adjustment of the synchronization region boundary. The radial diffusion front refers to the radial diffusion front of the energy focusing hotspot, used as the baseline for the dynamic adjustment of the synchronization region boundary. The synchronization region boundary refers to the boundary defined according to spatial locality constraints, used to define the synchronization range. The dynamic adjustment baseline refers to the baseline used for dynamically adjusting the synchronization region boundary, used to guide the boundary adjustment.

[0120] In this embodiment, firstly, during the boundary delineation of the synchronization region under spatial locality constraints, the radial diffusion front of the energy focusing hotspot in the vortex core energy density distribution map is used as the dynamically adjusted baseline of the synchronization region boundary. The determination of the dynamically adjusted baseline typically employs a front detection or boundary tracking algorithm, for example, by detecting the diffusion front of the energy focusing hotspot. Next, the synchronization region boundary is dynamically adjusted based on the dynamically adjusted baseline. The adjustment of the synchronization region boundary typically employs a boundary expansion or contraction algorithm, for example, by expanding or contracting the boundary to adapt to changes in the energy focusing hotspot. Through this step, the system can scientifically adjust the synchronization region boundary, ensuring the rationality and real-time nature of the synchronization range.

[0121] 404. When the rate of change of the radius of curvature of the energy transfer path exceeds the preset fluid instability threshold, a real-time re-partitioning event of the boundary of the synchronization region is triggered. The real-time re-partitioning event preferentially retains the inter-node associated topology fragments that share a continuous energy flux vector with the rendering region of the target GPU computing node.

[0122] In step 404, the rate of change of the radius of curvature of the energy transfer path refers to the rate at which the radius of curvature of the energy transfer path changes, used to determine the fluid instability state. The fluid instability threshold is a preset threshold used to determine the fluid instability state, used to trigger a real-time re-division event of the synchronization region boundary. The synchronization region boundary refers to the boundary defined according to spatial locality constraints, used to limit the synchronization range. The real-time re-division event refers to the event of real-time re-division of the synchronization region boundary based on the rate of change of the radius of curvature of the energy transfer path, used to optimize the synchronization range. The target GPU computing node refers to the GPU computing unit currently executing the rendering task, used to determine the synchronization region boundary. The continuous energy flux vector refers to the continuous energy flux vector in the energy transfer path, used to determine the inter-node associated topological segments. The inter-node associated topological segments refer to the topological segments that share the continuous energy flux vector with the rendering region of the target GPU computing node, used to synchronize and update the knowledge graph.

[0123] In this embodiment, the rate of change of the radius of curvature of the energy transfer path is first detected to determine whether it exceeds a preset fluid instability threshold. The fluid instability threshold is typically set based on fluid dynamics models or experimental data, for example, by analyzing the impact of fluid instability on the synchronization range and setting a reasonable threshold. When the rate of change of the radius of curvature of the energy transfer path exceeds the fluid instability threshold, a real-time re-partitioning event of the synchronization region boundary is triggered. This real-time re-partitioning event typically employs a dynamic re-partitioning algorithm, such as re-partitioning the synchronization region boundary to adapt to changes in the energy transfer path. Next, in the real-time re-partitioning event, topological fragments associated with nodes that share a continuous energy flux vector with the target GPU computing node's rendering region are preferentially retained. The retention of these topological fragments typically employs topology matching or energy flux vector matching algorithms, such as matching continuous energy flux vectors to retain associated topological fragments. Through this step, the system can optimize the synchronization region boundary in real time, ensuring the rationality and real-time performance of the synchronization range.

[0124] Here is a specific example:

[0125] In the internal fluid monitoring system of a turbine engine, the system first identifies abrupt changes in the radius of curvature of the energy transfer path based on the phase difference jump characteristics of adjacent sensing nodes in multi-band vibration waveform data acquired by a fiber optic sensing array. For example, when the phase difference between adjacent sensing nodes suddenly increases by more than 30 degrees, the system identifies the abrupt change in the radius of curvature in the high-frequency turbulent region. Next, the system generates a vortex core energy density distribution map based on these abrupt changes in the radius of curvature of the energy transfer path. For example, the map shows energy focusing hotspots formed by the reflection of particle flow on the inner wall of the mechanical cavity; these hotspots are typically located at the edge of the turbine blades. Subsequently, during the boundary delineation of the synchronous region under spatial locality constraints, the system uses the radial diffusion front of the energy focusing hotspots in the vortex core energy density distribution map as the dynamically adjusted baseline for the synchronous region boundary. For example, the synchronous region boundary is defined by extending 5 mm outward from the hotspot. Finally, when the rate of change of the radius of curvature of the energy transfer path exceeds a preset fluid instability threshold, such as a rate of change exceeding 2 mm per second, the system triggers a real-time re-delineation event for the synchronous region boundary. For example, priority is given to preserving inter-node topological fragments that share continuous energy flux vectors with the target GPU computing node's rendering region, such as energy transfer paths in high-frequency turbulent regions. Through this mechanism, the system can accurately monitor the internal fluid dynamics of a turbine engine, providing reliable data support for mechanical performance optimization.

[0126] In summary, steps 401 to 404 achieve high efficiency and accuracy in simulating particle behavior within a closed mechanical cavity, while optimizing resource utilization and communication efficiency, providing reliable technical support for simulating complex physical scenarios. By combining multi-band vibration waveform data from an optical fiber sensor array with a dynamic particle knowledge graph, the probability of energy transfer path breakage can be accurately quantified, providing a reliable basis for local topology updates. Furthermore, by introducing multi-band spectral analysis, adaptive convolution kernels, multi-scale spatial sorting algorithms, and dynamic phase correction mechanisms, the robustness and adaptability of the method are further improved, significantly enhancing the overall performance and reliability of simulating complex physical scenarios.

[0127] To address the issues of untimely local topology updates and unreasonable rendering resolution allocation among GPU computing nodes in dynamic particle knowledge graphs, a rendering resolution reallocation system based on a topology evolution priority queue was developed. This system achieves precise triggering of local topology updates, dynamic optimization of rendering resolution, and high synchronization of system timing, providing a reliable guarantee for efficient collaborative rendering across multiple GPU computing nodes. In some embodiments, step 203, which involves establishing a priority queue within the federated learning framework and generating local update weights based on the product of the energy transfer path breakage probability and particle collision frequency in the update unit, includes:

[0128] 501. Based on the spatial distribution characteristics of particle collision frequencies in the multi-band vibration waveform data collected by the fiber optic sensing array, extract the probability of energy transfer path breakage that exists in the rendering area of ​​the target GPU computing node in the dynamic particle knowledge graph.

[0129] In step 501, the fiber optic sensing array refers to an array composed of multiple fiber optic sensors used to acquire multi-band vibration waveform data. Multi-band vibration waveform data refers to the vibration waveform data acquired by the fiber optic sensing array, containing vibration information across multiple frequency bands. Particle collision frequency refers to the frequency at which particles collide within a spatiotemporal increment unit, used to measure the intensity of particle activity. Spatial distribution characteristics refer to the spatial distribution characteristics of particle collision frequencies, used to extract the probability of energy transfer path breakage. Dynamic particle knowledge graph refers to a knowledge graph describing the behavior and relationships of particles within a spatiotemporal increment unit, used to guide rendering optimization. Target GPU computing node refers to the GPU computing unit currently performing the rendering task, used to determine the rendering region. Rendering region refers to the region rendered by the target GPU computing node, used to extract the probability of energy transfer path breakage. Fluid dynamic coupling refers to the fluid dynamic interaction between the rendering region of the target GPU computing node and its surrounding regions, used to determine the probability of energy transfer path breakage. The probability of energy transfer path breakage refers to the probability that the energy transfer path breaks within a spatiotemporal increment unit, used to measure the stability of the topology.

[0130] In this embodiment, the spatial distribution characteristics of particle collision frequencies are first analyzed based on multi-band vibration waveform data collected by an optical fiber sensor array. The spatial distribution characteristics of particle collision frequencies are typically analyzed using statistical methods, such as quantifying the particle collision frequency by calculating the number of particle collisions per unit volume. Next, the probability of energy transfer path breakage due to hydrodynamic coupling with the target GPU computing node's rendering region is extracted from the dynamic particle knowledge graph. The extraction of energy transfer path breakage probability typically employs machine learning or regression analysis algorithms, such as calculating the breakage probability by analyzing the relationship between particle collision frequencies and energy transfer path breakage. Through this step, the system can scientifically extract the probability of energy transfer path breakage, providing a basis for subsequent local topology update weight generation.

[0131] 502. Perform a temporal convolution operation between the probability of energy transfer path breakage and the particle collision frequency in the corresponding region to generate local topology update weights.

[0132] In step 502, the energy transfer path breakage probability refers to the probability that an energy transfer path breaks in the spatiotemporal increment unit, used to measure the stability of the topology. The particle collision frequency refers to the frequency at which particles collide in the spatiotemporal increment unit, used to measure the intensity of particle activity. Temporal convolution operation refers to performing convolution operations on the signal in the time domain, used to generate local topology update weights. The local topology update weights are weights generated based on the temporal convolution operation of the energy transfer path breakage probability and the particle collision frequency, used to guide the update of the local topology.

[0133] In this embodiment, the probability of energy transfer path breakage is first convolved with the particle collision frequency in the corresponding region in the temporal domain. Temporal convolution typically employs discrete convolution or fast Fourier transform techniques. For example, by convolving the probability of energy transfer path breakage with the particle collision frequency, local topology update weights are generated. The generation of these local topology update weights typically uses a weighted convolution algorithm, such as combining the convolution result with weight coefficients to generate update weights. Through this step, the system can scientifically generate local topology update weights, providing a basis for subsequent topology updates.

[0134] Here is a specific example:

[0135] In the internal fluid monitoring system of a turbine engine, the system first extracts the probability of energy transfer path breakage that exists in the fluid dynamic coupling region of the target GPU computing node's rendering area based on the spatial distribution characteristics of particle collision frequencies in multi-band vibration waveform data collected by a fiber optic sensor array, using a dynamic particle knowledge graph. For example, the system analyzes the particle collision frequencies in high-frequency turbulent regions and finds that their spatial distribution characteristics are highly correlated with the probability of energy transfer path breakage; when the collision frequency exceeds 1000 times per second, the breakage probability increases significantly. Next, the system performs a temporal convolution operation between the energy transfer path breakage probability and the corresponding particle collision frequencies to generate local topology update weights. For example, through convolution, the system calculates a local topology update weight of 0.8 for high-frequency turbulent regions and 0.3 for low-frequency eddy regions. Through this mechanism, the system can accurately identify potential breakage risks in energy transfer paths and provide reliable data support for the rendering strategy of GPU computing nodes, ensuring real-time monitoring and optimization of the internal fluid dynamics of the turbine engine.

[0136] In summary, steps 501 to 503 achieve high efficiency and accuracy in simulating particle behavior within a closed mechanical cavity, while optimizing resource utilization and communication efficiency, providing reliable technical support for simulating complex physical scenarios. By combining multi-band vibration waveform data from an optical fiber sensor array with a dynamic particle knowledge graph, the probability of energy transfer path breakage can be accurately quantified, providing a reliable basis for local topology updates. Furthermore, by introducing multi-band spectral analysis, adaptive convolution kernels, multi-scale spatial sorting algorithms, and dynamic phase correction mechanisms, the robustness and adaptability of the method are further improved, significantly enhancing the overall performance and reliability of simulating complex physical scenarios.

[0137] To address the boundary offset and energy flux vector direction inaccuracy caused by dynamic changes in energy focusing hotspots during the boundary delineation of synchronization regions, a dynamic adjustment system for synchronization region boundaries based on a radial diffusion front was developed. This system achieves precise dynamic adjustment of the synchronization region boundary, real-time tracking of energy transfer paths, and effective control of boundary offset, providing stable support for efficient collaborative rendering across multiple GPU computing nodes. In some embodiments, step 404, during the boundary delineation of the synchronization region under spatial locality constraints, uses the radial diffusion front of the energy focusing hotspot in the vortex core energy density distribution map as the dynamic adjustment baseline for the synchronization region boundary, including:

[0138] 601. Based on the phase difference jump characteristics of the high-frequency band in the multi-band vibration waveform data collected by the fiber optic sensing array, identify the instantaneous spatial coordinates of the energy focusing hotspot;

[0139] In step 601, the fiber optic sensor array is a device used to acquire multi-band vibration waveform data, capable of real-time monitoring of vibration within a sealed mechanical cavity. The high-frequency phase difference jump characteristic describes abrupt changes in phase difference within the high-frequency vibration waveform data, typically related to the formation or dissipation of energy focusing hotspots. An energy focusing hotspot refers to a region within a sealed mechanical cavity where energy is highly concentrated. Instantaneous spatial coordinates describe the spatial location of the energy focusing hotspot at a given moment.

[0140] In this embodiment, firstly, based on multi-band vibration waveform data collected by an optical fiber sensor array, the high-frequency phase difference jump characteristics are analyzed. This process can accurately capture the positional changes of energy focusing hotspots, providing data support for subsequent energy diffusion analysis and synchronization area adjustment. Next, the instantaneous spatial coordinates of the energy focusing hotspots are identified through the spatiotemporal distribution characteristics of the phase difference jumps. Furthermore, by introducing spectral analysis of the multi-band vibration waveform data, the accuracy of energy focusing hotspot identification is further improved, ensuring that subtle changes in energy concentration can be captured in complex fluid dynamics scenarios.

[0141] 602. Extract the radial diffusion front trajectory of the energy focusing hotspot from the energy density distribution map of the vortex core;

[0142] In step 602, the vortex core energy density distribution map is a map describing the energy density distribution in the vortex core region within the sealed mechanical cavity. The radial diffusion front trajectory describes the outward diffusion path of the energy focusing hotspot.

[0143] In this embodiment, firstly, the radial diffusion front trajectory of the energy focusing hotspot is extracted from the vortex core energy density distribution map. This process determines the diffusion direction and path of the energy focusing hotspot by analyzing the gradient change of energy density, providing a basis for subsequent front expansion rate compensation and synchronization region adjustment. Furthermore, by introducing a multi-scale energy density analysis method, the trajectory extraction strategy can be dynamically adjusted according to the diffusion characteristics of the energy focusing hotspot, further improving the accuracy of the diffusion front trajectory.

[0144] 603. Based on the positional offset of the energy focusing hotspot within a preset time window, a leading-edge expansion rate compensation factor is generated. The compensation factor is dynamically calculated by the ratio of the fluctuation amplitude of the high-frequency phase difference jump event to the thermal expansion coefficient of the inner wall of the cavity.

[0145] In step 603, the preset time window is the time range used to analyze the changes in the location of the energy focusing hotspot. The leading edge expansion rate compensation factor is a dynamic factor used to correct the expansion rate of the energy diffusion leading edge. The fluctuation amplitude of the high-frequency phase difference jump event is an indicator describing the intensity of the phase difference jump. The thermal expansion coefficient of the cavity inner wall is a physical parameter describing the degree of expansion of the cavity inner wall material under temperature changes.

[0146] In this embodiment, firstly, a frontal expansion rate compensation factor is generated based on the positional offset of the energy focusing hotspot within a preset time window. Then, the compensation factor is dynamically calculated using the ratio of the fluctuation amplitude of the high-frequency phase difference jump event to the thermal expansion coefficient of the cavity inner wall. This process dynamically corrects the expansion rate of the energy diffusion front based on the dynamic changes of the energy focusing hotspot and the physical properties of the cavity inner wall, improving the accuracy of diffusion analysis. Furthermore, by introducing an adaptive compensation algorithm, the calculation parameters of the compensation factor can be dynamically adjusted according to the diffusion characteristics of the energy focusing hotspot, further improving compensation accuracy.

[0147] 604. During the dynamic adjustment of the boundary of the synchronization region, the tangent direction of the radial diffusion front trajectory is projected onto the graph increment transmission channel of the federated learning framework.

[0148] In step 604, the synchronization region boundary is a region defined based on spatial locality constraints, used to determine the topological segments of inter-node associations that need to be synchronized. The tangent direction of the radial diffusion front trajectory is a vector describing the instantaneous propagation direction of the energy diffusion front. The map increment transmission channel is a communication channel used to transmit spatiotemporal increment units between GPU computing nodes.

[0149] In this embodiment, firstly, during the dynamic adjustment of the synchronization region boundary, the tangent direction of the radial diffusion front trajectory is projected onto the graph increment transmission channel of the federated learning framework. This process dynamically adjusts the position and range of the synchronization region boundary according to the propagation direction of the energy diffusion front, ensuring that the synchronization region remains consistent with the energy diffusion front, thus improving the efficiency and accuracy of data synchronization. Furthermore, by introducing a multi-scale projection algorithm, the projection strategy can be dynamically adjusted according to different scales of the energy diffusion front, further optimizing the adjustment efficiency of the synchronization region boundary.

[0150] 605. When the energy flux vector direction offset angle is detected to exceed the preset turbulent anisotropy threshold, a local retraction operation of the synchronization region boundary is triggered, and the dynamic adjustment baseline of the synchronization region boundary is redefined based on the curvature change point distribution of the radial diffusion front trajectory.

[0151] In step 605, the energy flux vector direction offset angle is an indicator describing the angle of offset between the energy transfer direction and a preset direction. The turbulence anisotropy threshold is a preset threshold used to determine whether a significant offset has occurred in the energy transfer direction. The local retraction operation of the synchronization region boundary refers to the operation of shrinking the synchronization region boundary inward when an offset in the energy transfer direction is detected. The curvature abrupt change point distribution describes the spatial distribution of points where the curvature of the radial diffusion front trajectory abruptly changes.

[0152] In this embodiment, firstly, when the detected energy flux vector direction offset angle exceeds a preset turbulent anisotropy threshold, a local retraction operation of the synchronization region boundary is triggered. Next, based on the distribution of curvature abrupt change points in the radial diffusion front trajectory, the baseline of the synchronization region boundary is redefined. This process can promptly adjust the synchronization region boundary when a significant shift occurs in the energy transfer direction, ensuring that the synchronization region remains consistent with the dynamic changes of the energy diffusion front, further improving the accuracy and real-time performance of data synchronization. Furthermore, by introducing a dynamic baseline adjustment algorithm, the position and shape of the baseline can be dynamically adjusted according to changes in the distribution of curvature abrupt change points, further optimizing the adjustment effect of the synchronization region boundary.

[0153] The following is a specific example:

[0154] In a fluid dynamics simulation scenario within an aero-engine combustion chamber, the high-frequency phase difference jump characteristics are first analyzed based on multi-band vibration waveform data acquired by a fiber optic sensor array. For example, in the high-temperature, high-pressure region of the combustion chamber, significant phase difference jumps occur in the high-frequency band due to the formation of energy focusing hotspots. Then, during the dynamic adjustment of the synchronization region boundary, the tangent direction of the radial diffusion front trajectory is projected onto the graph increment transmission channel of the federated learning framework. For example, in the flame propagation region, the synchronization region boundary is dynamically adjusted according to the propagation direction of the energy diffusion front. Finally, when the energy flux vector direction offset angle exceeds a preset turbulence anisotropy threshold, a local retraction operation of the synchronization region boundary is triggered, and the baseline of the synchronization region boundary is redefined based on the curvature abrupt change point distribution of the radial diffusion front trajectory. For example, in the flame propagation region, due to a significant shift in the energy transfer direction, the synchronization region boundary contracts inward, and the baseline is redefined based on the curvature abrupt change point distribution.

[0155] In summary, steps 601 to 605 achieve accurate identification and dynamic tracking of energy focusing hotspots and diffusion fronts within a sealed mechanical cavity, while optimizing the adjustment efficiency and accuracy of the synchronization region boundary, providing reliable technical support for the simulation of complex fluid dynamics scenarios. By projecting the tangent direction of the radial diffusion front trajectory onto the incremental transmission channel of the spectrum, the synchronization region boundary can be dynamically adjusted to ensure its consistency with the energy diffusion front. By detecting the energy flux vector direction offset angle and triggering a local retraction operation of the synchronization region boundary, the synchronization region boundary can be adjusted in a timely manner, further improving the accuracy and real-time performance of data synchronization. Ultimately, this method significantly improves the overall performance and reliability of complex fluid dynamics scenario simulations while maintaining simulation effectiveness.

[0156] To address the issues of inaccurate resolution allocation and accumulated particle trajectory rendering errors during dynamic particle flow rendering, a particle trajectory adaptive adjustment system based on GPU rendering control commands was developed. This system achieves precise control of particle trajectory rendering granularity, dynamic optimization of resolution allocation, and effective compensation for rendering errors, providing reliable support for high-performance rendering of complex dynamic particle flows. In some embodiments, step 105, based on the particle rendering resolution dynamic allocation strategy and the fluctuation range of vibration waveform data fed back in real time by the fiber optic sensing array, generates GPU rendering control commands for the dynamic particle flow inside the mechanical cavity. These rendering control commands trigger the GPU computing node to perform adaptive adjustment of the particle trajectory rendering granularity based on the priority of energy transfer paths in the dynamic particle knowledge graph, including:

[0157] 701. Filter and extract high-frequency phase difference jump events and low-frequency energy attenuation gradient fluctuation ranges from the multi-frequency vibration waveform data fed back in real time by the fiber optic sensing array.

[0158] In step 701, the fiber optic sensing array is a device used to acquire multi-band vibration waveform data, capable of real-time monitoring of vibration within a sealed mechanical cavity. High-frequency phase difference jump events describe abrupt changes in phase difference within high-frequency vibration waveform data, typically associated with the formation or dissipation of energy focusing hotspots. Low-frequency energy attenuation gradient fluctuation ranges describe fluctuations in the energy attenuation gradient within low-frequency vibration waveform data, typically associated with breaks or weakening of the energy transfer path.

[0159] In this embodiment, firstly, high-frequency phase difference jump events and low-frequency energy attenuation gradient fluctuation ranges are screened and extracted from the multi-band vibration waveform data fed back in real time by the fiber optic sensing array. This process, by analyzing the spectral characteristics of the vibration waveform data, can accurately capture the dynamic changes of energy focusing hotspots and energy transfer paths, providing data support for subsequent spatiotemporal matching and resolution adjustment. Furthermore, by introducing an adaptive filtering algorithm, the screening and extraction strategy can be adjusted according to the dynamic characteristics of the vibration waveform data, further improving the accuracy and efficiency of data extraction.

[0160] 702. Perform spatiotemporal matching between the selected high-frequency phase difference jump events and the priority of energy transfer paths in the dynamic particle knowledge graph to generate spatiotemporal matching results;

[0161] In step 702, the dynamic particle knowledge graph is a graphical model used to describe the interactions and energy transfer relationships between particles. The priority of energy transfer paths is an index generated based on the probability of path breakage and particle collision frequency, used to measure the importance of energy transfer paths. Spatiotemporal matching is the process of matching high-frequency phase difference jump events with energy transfer paths in the temporal and spatial dimensions.

[0162] In this embodiment, firstly, the selected high-frequency phase difference jump events are spatiotemporally matched with the priority of energy transfer paths in the dynamic particle knowledge graph. This process analyzes the spatiotemporal distribution characteristics of the high-frequency phase difference jump events to determine their correspondence with energy transfer paths, generating spatiotemporal matching results. These results provide a reliable basis for subsequent resolution adjustment and particle trajectory compensation. Furthermore, by introducing a multi-scale matching algorithm, the matching strategy can be dynamically adjusted according to different scales of the high-frequency phase difference jump events, further improving matching accuracy.

[0163] 703. Based on the particle rendering resolution dynamic allocation strategy, and combined with the spatiotemporal matching results, generate a dynamic resolution decay gradient table.

[0164] In step 703, the particle rendering resolution dynamic allocation strategy refers to a strategy that dynamically allocates resolution based on particle motion state and rendering requirements, used to optimize rendering quality. The spatiotemporal matching result refers to the matching result between particle motion state and rendering requirements in time and space, used to generate a dynamic resolution decay gradient table. The dynamic resolution decay gradient table is a table describing the resolution decay gradient in the spatiotemporal dimension, used to guide the generation of GPU rendering control instructions.

[0165] In this embodiment, a dynamic resolution attenuation gradient table is first generated based on a particle rendering resolution dynamic allocation strategy and spatiotemporal matching results. The generation of the dynamic resolution attenuation gradient table typically employs gradient calculation and interpolation techniques, such as calculating the resolution attenuation gradient in the spatiotemporal dimension and generating the gradient table. The gradient table generation process usually uses adaptive algorithms, such as dynamically adjusting the attenuation gradient according to particle motion states and rendering requirements. Through this step, the system can scientifically generate a dynamic resolution attenuation gradient table, providing a basis for subsequent GPU rendering control command generation.

[0166] 704. Based on the dynamic resolution attenuation gradient table and the spatial distribution density of high-frequency phase difference jump events acquired in real time, generate GPU rendering control instructions for the dynamic particle flow inside the mechanical cavity.

[0167] In step 704, the dynamic resolution decay gradient table refers to a table describing the resolution decay gradient in the spatiotemporal dimension, used to guide the generation of GPU rendering control instructions. High-frequency phase difference jump events refer to phase difference jump events in high-frequency vibration waveform data, used to measure the intensity of particle activity. Spatial distribution density refers to the spatial distribution density of high-frequency phase difference jump events, used to guide the generation of GPU rendering control instructions. GPU rendering control instructions are instructions used to control the GPU to render dynamic particle flows, used to optimize rendering quality.

[0168] In this embodiment, GPU rendering control instructions for the dynamic particle flow inside the mechanical cavity are first generated based on the dynamic resolution attenuation gradient table and the spatial distribution density of high-frequency phase difference jump events acquired in real time. The generation of GPU rendering control instructions typically employs instruction generation algorithms, such as combining the dynamic resolution attenuation gradient table with the spatial distribution density of high-frequency phase difference jump events to generate rendering instructions. The generation process of rendering instructions typically employs optimization algorithms, such as optimizing instruction generation based on rendering requirements and system performance. Through this step, the system can efficiently generate GPU rendering control instructions, ensuring the rendering quality of the dynamic particle flow and system performance.

[0169] Here is a specific example:

[0170] In the internal fluid monitoring system of a turbine engine, the system first filters high-frequency phase difference jump events and low-frequency energy decay gradient fluctuation ranges from the multi-band vibration waveform data fed back in real time by the fiber optic sensor array. For example, high-frequency phase difference jump events are concentrated at the turbine blade edges, while low-frequency energy decay gradient fluctuation ranges appear at the cavity center. Next, the system performs spatiotemporal matching of the high-frequency phase difference jump events with the priority of energy transfer paths in a dynamic particle knowledge graph, generating matching results. For example, high-priority energy transfer paths are matched to high-frequency turbulence regions, and low-priority energy transfer paths are matched to low-priority eddy regions. Subsequently, the system generates a dynamic resolution decay gradient table based on the particle rendering resolution dynamic allocation strategy and the spatiotemporal matching results. For example, the resolution decay gradient in the high-frequency turbulence region decreases by 10% per second, and in the low-frequency eddy region by 5% per second. Finally, the system generates GPU rendering control commands based on the dynamic resolution decay gradient table and the spatial distribution density of the real-time high-frequency phase difference jump events. For example, when the density of phase difference jump events in the high-frequency turbulence region exceeds 5 times per second, the system triggers the GPU computing node to increase the particle rendering resolution in that region, ensuring monitoring accuracy and efficiency. Through this mechanism, the system can accurately capture the fluid dynamics inside the turbine engine, providing reliable support for optimizing mechanical performance.

[0171] In summary, steps 701 to 704 achieve high-precision rendering and dynamic adjustment of particle trajectories within a sealed mechanical cavity, while optimizing rendering efficiency and simulation accuracy, providing reliable technical support for simulating complex fluid dynamics scenarios. By combining multi-band vibration waveform data from a fiber optic sensor array with high-frequency phase difference jump events, the dynamic changes in energy focusing hotspots and energy transfer paths can be accurately captured. By spatiotemporally matching high-frequency phase difference jump events with the priority of energy transfer paths, reliable spatiotemporal matching results can be generated. By generating a dynamic resolution attenuation gradient table and a particle trajectory compensation factor, the rendering accuracy of particle trajectories can be dynamically adjusted. By integrating the dynamic resolution attenuation gradient table and the particle trajectory compensation factor into the GPU rendering control instructions, the GPU computing nodes can be triggered to execute particle trajectory rendering and adaptive adjustment according to the rendering granularity corresponding to the priority of energy transfer paths. Ultimately, this method significantly improves the overall performance and reliability of complex fluid dynamics scenario simulations while ensuring simulation effectiveness.

[0172] Figure 2 This application provides a schematic diagram of the structure of a dynamic particle adaptive rendering system based on an autonomous and controllable GPU, as shown in the embodiment. Figure 2 As shown, the system includes:

[0173] The acquisition module 21 is deployed in a distributed manner within the closed monitoring area of ​​the fluid machinery using an optical fiber sensor array. The optical fiber sensor array is used to acquire multi-band vibration waveform data of dynamic particle flow under strong electromagnetic interference environment.

[0174] The generation module 22 generates the node association topology of the dynamic particle knowledge graph based on the multi-band vibration waveform data and the preset particle hydrodynamic constraint rules.

[0175] The computing module 23 performs distributed collaborative updates of the dynamic particle knowledge graph among multiple GPU computing nodes through a federated learning framework, and determines the dynamic allocation strategy of particle rendering resolution for each GPU computing node based on the real-time changes in the inter-node association topology obtained during the distributed collaborative update process.

[0176] The adjustment module 24 generates GPU rendering control instructions for the dynamic particle flow inside the mechanical cavity based on the particle rendering resolution dynamic allocation strategy and the vibration waveform data fluctuation range fed back by the fiber optic sensing array in real time. The rendering control instructions trigger the GPU computing node to perform adaptive adjustment of the rendering granularity of the particle trajectory based on the priority of the energy transfer path in the dynamic particle knowledge graph.

[0177] Figure 2 The aforementioned dynamic particle adaptive rendering system based on an autonomous and controllable GPU can execute... Figure 1 The implementation principle and technical effects of the dynamic particle adaptive rendering method based on an autonomous and controllable GPU described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit performs operations in the dynamic particle adaptive rendering system based on an autonomous and controllable GPU in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0178] In one possible design, Figure 2 The illustrated embodiment of a dynamic particle adaptive rendering system based on an autonomous and controllable GPU can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0179] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0180] The processing component 32 is used for the above Figure 1 The embodiment describes a dynamic particle adaptive rendering method based on an autonomous and controllable GPU.

[0181] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0182] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0183] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0184] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0185] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0186] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0187] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is a dynamic particle adaptive rendering method based on an autonomous and controllable GPU.

[0188] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0189] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0190] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0191] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A dynamic particle adaptive rendering method based on an autonomous and controllable GPU, characterized in that, include: A distributed fiber optic sensor array is deployed within the closed monitoring area of ​​the fluid machinery to collect multi-band vibration waveform data of dynamic particle flow under strong electromagnetic interference environment. Based on the multi-band vibration waveform data and the preset particle hydrodynamic constraint rules, the node association topology of the dynamic particle knowledge graph is generated. The dynamic particle knowledge graph among multiple GPU computing nodes is distributed and collaboratively updated through a federated learning framework. Based on the real-time changes in the inter-node association topology obtained during the distributed collaborative update process, a dynamic allocation strategy for particle rendering resolution of each GPU computing node is determined. Based on the particle rendering resolution dynamic allocation strategy and the vibration waveform data fluctuation range fed back by the fiber optic sensing array in real time, a GPU rendering control instruction for the dynamic particle flow inside the mechanical cavity is generated. The rendering control instruction triggers the GPU computing node to perform adaptive adjustment of the rendering granularity of the particle trajectory based on the priority of the energy transfer path in the dynamic particle knowledge graph. Specifically, a distributed collaborative update of the dynamic particle knowledge graph among multiple GPU computing nodes is performed using a federated learning framework. Based on the real-time changes in the inter-node association topology obtained during the distributed collaborative update process, a dynamic allocation strategy for particle rendering resolution of each GPU computing node is determined, including: The topological association between nodes in the dynamic particle knowledge graph is divided into multiple update units, each of which contains information on changes in particle collision characteristics and the probability of energy transfer path breakage. A data transmission channel between GPU computing nodes is established in the federated learning framework through a hybrid communication protocol. A cooperative mechanism with local constraints is adopted to synchronize only the topological fragments between nodes that have a mechanical relationship with the target GPU computing node region, thereby realizing the distributed cooperative update of the dynamic particle knowledge graph. In the federated learning framework, a priority queue is established, and local update weights are generated based on the product of the probability of energy transfer path breakage and particle collision frequency in the update unit. When the local update weight exceeds a preset threshold, the rendering resolution adjustment event of the corresponding GPU computing node is triggered based on the sorting of the priority queue. Based on the real-time changes in the inter-node topology obtained during the distributed collaborative update process, and combined with the rendering resolution adjustment event and the real-time bandwidth utilization of the data transmission channel, a dynamic allocation strategy for particle rendering resolution is generated.

2. The method according to claim 1, characterized in that, A data transmission channel between GPU computing nodes is established within the federated learning framework using a hybrid communication protocol. A locally constrained collaborative mechanism is employed to synchronize only topological fragments between nodes that have a mechanical relationship with the target GPU computing node's rendering region, thereby achieving distributed collaborative updates of the dynamic particle knowledge graph, including: In the federated learning framework, a data transmission channel between GPU computing nodes is established through a hybrid communication protocol to transmit updated data of the dynamic particle knowledge graph. Based on the radius of curvature of the energy transfer path within the rendering area of ​​the target GPU computing node, the boundary of the synchronization region with spatial locality constraints is defined. The boundary of the synchronization region contains inter-node associated topological segments that share a continuous energy flux vector with the rendering area of ​​the target GPU computing node. Through the data transmission channel, only the topological fragments associated between nodes within the boundary of the synchronization region are synchronized, thereby realizing the distributed collaborative update of the dynamic particle knowledge graph.

3. The method according to claim 2, characterized in that, Based on the radius of curvature of the energy transfer path within the rendering region of the target GPU computing node, the synchronization region boundary of the spatial locality constraint is defined. The synchronization region boundary contains inter-node associated topological segments that share a continuous energy flux vector with the rendering region of the target GPU computing node, including: Based on the phase difference jump characteristics of adjacent sensing nodes in the multi-band vibration waveform data collected by the fiber optic sensing array, the abrupt interval of the radius of curvature of the energy transfer path is identified. A vortex core energy density distribution map is generated based on the abrupt interval of the radius of curvature of the energy transfer path. The vortex core energy density distribution map includes energy focusing hotspots formed by the reflection of particle flow on the inner wall of the mechanical cavity. In the process of delineating the boundary of the synchronous region under the spatial locality constraint, the radial diffusion front of the energy focusing hotspot in the energy density distribution map of the vortex core is used as the dynamic adjustment baseline of the synchronous region boundary. When the rate of change of the radius of curvature of the energy transfer path exceeds the preset fluid instability threshold, a real-time re-partitioning event of the boundary of the synchronization region is triggered. The real-time re-partitioning event prioritizes the preservation of inter-node associated topology fragments that share continuous energy flux vectors with the rendering region of the target GPU computing node.

4. The method according to claim 1, characterized in that, In the federated learning framework, a priority queue is established, and local update weights are generated based on the product of the energy transfer path breakage probability and the particle collision frequency in the update unit, including: Based on the spatial distribution characteristics of particle collision frequencies in the multi-band vibration waveform data collected by the fiber optic sensing array, the probability of energy transfer path breakage with hydrodynamic coupling to the rendering area of ​​the target GPU computing node is extracted from the dynamic particle knowledge graph. The probability of energy transfer path breakage is convolved with the particle collision frequency in the corresponding region in the time domain to generate local topology update weights.

5. The method according to claim 3, characterized in that, In the process of delineating the boundary of the synchronization region under the spatial locality constraint, the radial diffusion front of the energy focusing hotspot in the vortex core energy density distribution map is used as the dynamically adjusted baseline of the synchronization region boundary, including: Based on the high-frequency phase difference jump characteristics in the multi-band vibration waveform data collected by the fiber optic sensing array, the instantaneous spatial coordinates of the energy focusing hotspot are identified. Extract the radial diffusion front trajectory of the energy focusing hotspot from the energy density distribution map of the vortex core; Based on the positional offset of the energy focusing hotspot within a preset time window, a leading-edge expansion rate compensation factor is generated. The compensation factor is dynamically calculated by the ratio of the fluctuation amplitude of the high-frequency phase difference jump event to the thermal expansion coefficient of the inner wall of the cavity. During the dynamic adjustment of the boundary of the synchronization region, the tangent direction of the radial diffusion front trajectory is projected onto the graph increment transmission channel of the federated learning framework. When the offset angle of the energy focusing hotspot vector direction is detected to exceed the preset turbulent anisotropy threshold, a local retraction operation of the synchronization region boundary is triggered, and the dynamic adjustment baseline of the synchronization region boundary is redefined based on the curvature change point distribution of the radial diffusion front trajectory.

6. The method according to claim 1, characterized in that, Based on the particle rendering resolution dynamic allocation strategy and the vibration waveform data fluctuation range fed back in real time by the fiber optic sensing array, GPU rendering control instructions for the dynamic particle flow inside the mechanical cavity are generated, including: The high-frequency phase difference jump event and the low-frequency energy attenuation gradient fluctuation range are filtered and extracted from the fluctuation range of the multi-frequency vibration waveform data fed back in real time from the fiber optic sensing array. The selected high-frequency phase difference jump events are spatiotemporally matched with the priority of energy transfer paths in the dynamic particle knowledge graph to generate spatiotemporal matching results. Based on the particle rendering resolution dynamic allocation strategy, and combined with the spatiotemporal matching results, a dynamic resolution decay gradient table is generated. Based on the dynamic resolution attenuation gradient table and the spatial distribution density of high-frequency phase difference jump events acquired in real time, GPU rendering control instructions for dynamic particle flow inside the mechanical cavity are generated.

7. A dynamic particle adaptive rendering system based on an autonomous and controllable GPU, used to execute the dynamic particle adaptive rendering method based on an autonomous and controllable GPU as described in any one of claims 1 to 6, characterized in that, include: The acquisition module deploys a distributed fiber optic sensor array within the closed monitoring area of ​​the fluid machinery to acquire multi-band vibration waveform data of dynamic particle flow under strong electromagnetic interference environment through the fiber optic sensor array. The generation module generates the node association topology of the dynamic particle knowledge graph based on the multi-band vibration waveform data and the preset particle hydrodynamic constraint rules. The computing module performs distributed collaborative updates of the dynamic particle knowledge graph among multiple GPU computing nodes through a federated learning framework, and determines the dynamic allocation strategy of particle rendering resolution for each GPU computing node based on the real-time changes in the inter-node association topology obtained during the distributed collaborative update process. The adjustment module generates GPU rendering control instructions for the dynamic particle flow inside the mechanical cavity based on the particle rendering resolution dynamic allocation strategy and the vibration waveform data fluctuation range fed back by the fiber optic sensing array in real time. The rendering control instructions trigger the GPU computing node to perform adaptive adjustment of the rendering granularity of the particle trajectory based on the priority of the energy transfer path in the dynamic particle knowledge graph.

8. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the dynamic particle adaptive rendering method based on an autonomous and controllable GPU as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements a dynamic particle adaptive rendering method based on an autonomous and controllable GPU as described in any one of claims 1 to 6.

Images