Full-time-domain fluid dynamics simulation optimization method and system

Abstract

The invention relates to the technical field of computational fluid dynamics, and discloses a full-time-domain fluid dynamics simulation optimization method and system, and the method comprises the following steps: S1, collecting flow field data through a sensor, and constructing a time-domain Navi er-Stokes fluid dynamics model; s2, inputting the model in the S1 into a machine learning prediction module; s3, calculating a pneumatic phase lag amount and generating a complex frequency domain phase response function; s4, generating a virtual damping control instruction in real time; and S5, feeding back the damping control instruction optimized in the step S4 to the CFD solver. According to the method, reliable extrapolation of the evolution trend of the flow field is realized by fusing a physical solver and a two-channel prediction mechanism of the LSTM, the simulation time step length is remarkably expanded in combination with dynamic generation of a discrete phase compensation function, and the problem of rapid increase of computing resources caused by strict limitation of the time step length is solved; therefore, the simulation period of a complex fluid system is shortened.

Description

Technical Field

[0001] The present invention relates to the technical field of computational fluid dynamics, and in particular to a full-time domain fluid dynamics simulation optimization method and system. Background Art

[0002] The full-time-domain fluid dynamics simulation optimization method and system is a cutting-edge field that deeply integrates computational fluid dynamics (CFD) and real-time control technology. It aims to completely cover the time evolution process of fluid motion through high-precision numerical simulation, and dynamically optimize system control parameters based on the simulation results. It is widely used in key fields such as aerodynamic design of aerospace vehicles, intelligent control of industrial fluid equipment, and efficient operation of energy systems. It is the core technical means to solve complex fluid problems.

[0003] Current mainstream technologies rely on iterative calculations of traditional physical models or purely data-driven prediction schemes, which have serious efficiency bottlenecks. In order to meet numerical stability conditions (such as the CFL criterion), the simulation step size needs to be sharply reduced as the grid is refined, resulting in an exponential increase in the number of calculation steps in long-time domain simulations. Summary of the Invention

[0004] In order to remedy the above shortcomings, the present invention provides a full-time domain fluid dynamics simulation optimization method and system, aiming to improve the problem that the simulation step size needs to be sharply reduced with the refinement of the grid, resulting in an exponential increase in the number of calculation steps in long-time domain simulations.

[0005] In a first aspect, the present invention provides the following technical solution, a full-time domain fluid dynamics simulation optimization method, comprising the following steps:

[0006] S1, collect flow field data through sensors and build a time-domain Navier-Stokes fluid dynamics model;

[0007] S2, input the model of S1 into the machine learning prediction module, and output the flow field evolution trend in the future time step. The prediction module integrates the dual-channel calculation results of the physical solver and the LSTM neural network;

[0008] S3, based on the flow field trend of the S2 dual-channel calculation results, calculate the aerodynamic phase lag and generate a complex frequency domain phase response function, which is discretized into a differential equation executable by a digital filter through z-transformation;

[0009] S4, dynamically configure digital filter parameters according to the differential equation of S3 and generate virtual damping control instructions in real time;

[0010] S5: Feedback the damping control instructions optimized in S4 to the CFD solver to drive the next round of flow field modeling and simulation calculations.

[0011] Preferably, in S1, the dual-channel calculation adopts a gated fusion mechanism, and the fusion weight is dynamically calculated by the Sigmoid function through the following formula:

[0012]

[0013] Among them, σ is the Sigmoid function controlled by the temperature parameter T, u LSTM is the LSTM prediction value, u NS Solve the simplified NS equations.

[0014] Preferably, in S3, the phase response function is in the form of the following formula:

[0015]

[0016] The discretization process uses the bilinear transformation method, and the parameter K p , T d ,α,τ is dynamically updated according to the flow field spectrum entropy value.

[0017] Preferably, in S4, the dynamic configuration includes:

[0018] When the vortex shedding frequency of the flow field is detected to be greater than the threshold, the system switches to the Butterworth topology and widens the cutoff frequency.

[0019] When the phase error is >5%, the group delay recalibration algorithm is triggered.

[0020] Preferably, the step S3 further includes matching the current flow field type from a pre-stored phase response template library, wherein the template library contains digitized Bode diagram data of more than 20 aircraft airfoils.

[0021] Preferably, in said S5, the next round of flow field modeling and simulation calculation is driven to adopt the adjoint gradient method to update the virtual damping parameters, and the gradient calculation balances the storage and recalculation overhead through the checkpoint technology.

[0022] Preferably, in S2, the LSTM neural network training adopts a physical constraint loss function, through the following formula:

[0023]

[0024] Among them, λ is the penalty coefficient of the conservation term, which forces the mass conservation equation to be satisfied.

[0025] In a second aspect, the present invention provides the following technical solution: a full-time domain fluid dynamics simulation and optimization system, the system comprising:

[0026] Flow field sensor data acquisition module, supporting millisecond-level communication of OPC-UA protocol;

[0027] Hybrid computing engine, integrating real-time NS equation solver and LSTM predictor;

[0028] Phase response function library, which stores digital transfer function templates of aircraft airfoils;

[0029] Reconfigurable IIR / FIR filter array, supporting online switching of topology;

[0030] Virtual damping controller, injecting UDF parameters into CFD software through lock-free communication interface.

[0031] In a third aspect, the invention provides the following technical solution: a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned full-time-domain fluid dynamics simulation optimization method when executing the computer program.

[0032] In a fourth aspect, the present invention provides the following technical solution: a readable storage medium having a computer program stored thereon, and the computer program, when executed by a processor, implements the above-mentioned full-time-domain fluid dynamics simulation optimization method.

[0033] The present invention has the following beneficial effects:

[0034] 1. In the present invention, by integrating the dual-channel prediction mechanism of the physical solver and LSTM, reliable extrapolation of the flow field evolution trend is achieved. Combined with the dynamic generation of the discrete phase compensation function, the simulation time step is significantly extended. Compared with the existing solutions that rely on single physical model iteration or pure data-driven prediction, this solves the problem of rapid growth of computing resources caused by strictly limited time steps, thereby shortening the simulation cycle of complex fluid systems.

[0035] 2. This invention achieves closed-loop optimization of virtual damping parameters through real-time parameter adjustment of a spectral-adaptive digital filter. It automatically switches filter topology when strong transient eddy current signatures are detected, and triggers recalibration when phase deviation exceeds a limit. Compared to existing fixed-parameter hardware compensators or offline optimization solutions, this solves the control instability problem caused by phase lag in dynamic flow fields, thereby improving the stability of the fluid control system.

[0036] 3. In the present invention, the amount of real-time calculation is reduced by intelligent matching of the pre-stored airfoil phase response template library, and the storage and recalculation overhead is optimized by the accompanying gradient method of the checkpoint technology. Compared with the existing solution in the full time domain accompanying solution that requires the full storage of flow field snapshots, the defect of excessive memory usage in long time domain simulation is solved, thereby reducing the demand for high-performance computing hardware resources.

[0037] 4. In the present invention, the data-driven prediction is constrained by embedding the LSTM loss function of the mass conservation equation to achieve a deep integration of machine learning and physical mechanisms. The physical constraint strength is dynamically adjusted in the loss function. Compared with the pure data-driven model without physical constraints in the prior art, the divergence problem caused by the violation of the fluid conservation law in the prediction results is solved, thereby ensuring the physical rationality of long-time domain simulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flow chart of the full-time domain fluid dynamics simulation optimization method proposed by the present invention;

[0039] Figure 2 This is the system framework diagram of the full-time domain fluid dynamics simulation and optimization system proposed in this invention. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0041] Example 1

[0042] Reference Figure 1 In a first embodiment of the present invention, the present invention provides a full-time domain fluid dynamics simulation optimization method, comprising the following steps:

[0043] S1, collect flow field data through sensors and build a time-domain Navier-Stokes fluid dynamics model;

[0044] S2, inputs the model of S1 into the machine learning prediction module, outputs the flow field evolution trend in the future time step, and the prediction module integrates the dual-channel calculation results of the physical solver and the LSTM neural network;

[0045] S3, based on the flow field trend of the S2 dual-channel calculation results, calculate the aerodynamic phase lag and generate a complex frequency domain phase response function, which is discretized into a differential equation executable by a digital filter through z-transformation;

[0046] S4, dynamically configure digital filter parameters according to the differential equation of S3 and generate virtual damping control instructions in real time;

[0047] S5: Feedback the damping control instructions optimized in S4 to the CFD solver to drive the next round of flow field modeling and simulation calculations.

[0048] The “construction of a time-domain Navier-Stokes model” in S1 provides a physical constraint basis for subsequent predictions, avoiding purely data-driven instability;

[0049] S2 "Input machine learning prediction module" specifies that the model output must be used as the input source of the prediction module, establishing a data dependency chain;

[0050] S3 “Calculating phase lag based on predicted trends” clarifies that the generation of compensation functions must rely on the prediction results to form a causal logic;

[0051] S4 "Configure parameters according to discrete equations" requires that filter parameter adjustment must be based on the previous discretization results, excluding manual intervention;

[0052] S5 “Feedback to CFD Solver” forces a closed-loop iteration, ensuring that parameter optimization is applied to the simulation process.

[0053] In S1, dual-channel calculation adopts gated fusion mechanism, and the fusion weight is dynamically calculated by Sigmoid function through the following formula:

[0054]

[0055] Among them, σ is the Sigmoid function controlled by the temperature parameter T, u LSTM is the LSTM prediction value, u NS Solve the simplified NS equations.

[0056] Dynamic weight calculation: The output value of the Sigmoid function changes with the temperature parameter $T$, realizing automatic weight distribution between the physical model and LSTM prediction;

[0057] Technical necessity: The temperature parameter T is related to the turbulence intensity of the flow field, ensuring that the physical model is prioritized when turbulence is high, and when T→0, σ(T)→0.

[0058] In S3, the phase response function is in the form of the following formula:

[0059]

[0060] The discretization process uses the bilinear transformation method, and the parameter K p ,T d ,α,τ is dynamically updated according to the flow field spectrum entropy value.

[0061] Function structure: e -τs The term explicitly characterizes the phase delay, K p ,T d Control gain and derivative strength;

[0062] Bilinear transformation: Converts continuous transfer functions into differential equations to ensure the executable nature of digital filters;

[0063] Parameter dynamic update: K p ,T d ,α,τ are adjusted according to the flow field spectrum characteristics to adapt to non-steady state conditions.

[0064] In S4, dynamic configuration includes:

[0065] When the vortex shedding frequency of the flow field is detected to be greater than the threshold, the system switches to the Butterworth topology and widens the cutoff frequency.

[0066] When the phase error is >5%, the group delay recalibration algorithm is triggered.

[0067] Switching conditions: The vortex shedding frequency exceeds the threshold, indicating a strong transient flow. The flat passband characteristics of the Butterworth topology can maintain phase linearity.

[0068] Error threshold: 5% is the critical point of flow field control accuracy. Recalibration is required when the error exceeds the tolerance to avoid error accumulation.

[0069] Action clarity: This limits the two parameter adjustment operations to specific scenarios.

[0070] S3 also includes matching the current flow field type from a pre-stored phase response template library, which contains digitized Bode diagram data for more than 20 aircraft airfoils.

[0071] The template library is used to provide verified aircraft airfoil phase response data to reduce the amount of real-time calculations.

[0072] Matching logic: Index the corresponding template according to the flow field type (such as airfoil number, Mach number) to ensure the effectiveness of compensation.

[0073] In S5, the adjoint gradient method is used to update the virtual damping parameters to drive the next round of flow field modeling and simulation calculations. The gradient calculation balances the storage and recalculation overhead through the checkpoint technology.

[0074] Checkpoint technology: selectively stores flow field snapshots on the timeline and restarts from the most recent checkpoint during recalculation, reducing memory usage;

[0075] Balancing mechanism: The storage interval is dynamically set based on computing resources to avoid efficiency losses caused by fixed intervals.

[0076] In S2, the LSTM neural network training uses the physical constraint loss function, which is expressed as follows:

[0077]

[0078] Among them, λ is the penalty coefficient of the conservation term, which forces the mass conservation equation to be satisfied.

[0079] Conservation penalty: The Σ term enforces the mass conservation equation to prevent the prediction results from violating the laws of physics.

[0080] Adjustable coefficient: λ controls the strength of the physical constraint and can be adjusted according to the confidence level of the data (e.g., increase λ when there is little experimental data).

[0081] Example 2:

[0082] Reference Figure 2 In a second embodiment of the present invention, the present invention provides a full-time domain fluid dynamics simulation optimization system, the system comprising:

[0083] Flow field sensor data acquisition module, supporting millisecond-level communication of OPC-UA protocol;

[0084] Hybrid computing engine, integrating real-time NS equation solver and LSTM predictor;

[0085] Phase response function library, which stores digital transfer function templates of aircraft airfoils;

[0086] Reconfigurable IIR / FIR filter array, supporting online switching of topology;

[0087] Virtual damping controller, injecting UDF parameters into CFD software through lock-free communication interface.

[0088] Flow field sensor data acquisition module: The OPC-UA protocol ensures industrial sensor compatibility, and millisecond-level communication meets real-time requirements.

[0089] Hybrid computing engine: Integrates high-precision CFD solvers, reduced-order models, and data-driven algorithms. Through adaptive grid and dynamic time step technology, it achieves full-time multi-scale fluid simulation from millisecond transients to hourly evolution. The physical solver and LSTM hardware operate in parallel and exchange data through memory sharing.

[0090] Phase response function library: The digitized template is stored as a Bode diagram array, supporting fast interpolation calls.

[0091] Reconfigurable IIR / FIR filter array: Reconfigurable means it supports online switching of order / type (such as switching from 4th-order IIR to 8th-order FIR). Based on adaptive Kalman filtering, wavelet transform and multi-sensor fusion algorithms, it realizes real-time denoising, feature extraction and missing data reconstruction of sensor data, providing high-quality input for simulation.

[0092] Virtual damping controller: A lock-free communication interface avoids CFD solution blocking, and UDF parameter injection enables software and hardware collaboration.

[0093] Example 3

[0094] The third embodiment of the present invention is based on the same inventive concept. The present invention proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the full-time domain fluid dynamics simulation optimization method of the above embodiment are implemented.

[0095] Example 4

[0096] The fourth embodiment of the present invention is based on the same inventive concept. The present invention proposes a computer device, the terminal includes: a processor, a memory; the processor and the memory communicate with each other; the memory is used to store instructions; the processor is used to execute the instructions in the memory and execute the full-time domain fluid dynamics simulation optimization method of the above embodiment.

[0097] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0098] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent replacements for some of the technical features therein. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. Full-time domain fluid dynamics simulation optimization method, characterized in that: The following steps are involved: S1, collect flow field data through sensors and build a time-domain Navier-Stokes fluid dynamics model; S2, input the model of S1 into the machine learning prediction module, and output the flow field evolution trend in the future time step. The prediction module integrates the dual-channel calculation results of the physical solver and the LSTM neural network; S3, based on the flow field trend of the S2 dual-channel calculation results, calculate the aerodynamic phase lag and generate a complex frequency domain phase response function, which is discretized into a differential equation executable by a digital filter through z-transformation; S4, dynamically configure digital filter parameters according to the differential equation of S3 and generate virtual damping control instructions in real time; S5: Feedback the damping control instructions optimized in S4 to the CFD solver to drive the next round of flow field modeling and simulation calculations.

2. The full-time domain fluid dynamics simulation optimization method according to claim 1, characterized in that: In S1, the dual-channel calculation adopts a gated fusion mechanism, and the fusion weight is dynamically calculated by the Sigmoid function using the following formula: Among them, σ is the Sigmoid function controlled by the temperature parameter T, u LSTM is the LSTM prediction value, u NS Solve the simplified NS equations.

3. The full-time domain fluid dynamics simulation optimization method according to claim 1, characterized in that: In S3, the phase response function is in the form of the following formula: The discretization process uses the bilinear transformation method, and the parameter K p , T d ,α,τ is dynamically updated according to the flow field spectrum entropy value.

4. The full-time domain fluid dynamics simulation optimization method according to claim 1, characterized in that: In S4, the dynamic configuration includes: When the vortex shedding frequency of the flow field is detected to be greater than the threshold, the system switches to the Butterworth topology and widens the cutoff frequency. When the phase error is >5%, the group delay recalibration algorithm is triggered.

5. The full-time domain fluid dynamics simulation optimization method according to claim 1, characterized in that: The S3 further includes matching the current flow field type from a pre-stored phase response template library, wherein the template library contains digitized Bode diagram data of more than 20 aircraft airfoils.

6. The full-time domain fluid dynamics simulation optimization method according to claim 1, characterized in that: In S5, the next round of flow field modeling and simulation calculation is driven to update the virtual damping parameters using the adjoint gradient method, and the gradient calculation balances the storage and recalculation overhead through the checkpoint technology.

7. The full-time domain fluid dynamics simulation optimization method according to claim 1, characterized in that: In S2, the LSTM neural network training adopts the physical constraint loss function, which is expressed by the following formula: Among them, λ is the penalty coefficient of the conservation term, which forces the mass conservation equation to be satisfied.

8. Full-time domain fluid dynamics simulation and optimization system, characterized by: The full-time-domain fluid dynamics simulation optimization method according to any one of claims 1 to 7, wherein the system comprises: Flow field sensor data acquisition module, supporting millisecond-level communication of OPC-UA protocol; Hybrid computing engine, integrating real-time NS equation solver and LSTM predictor; Phase response function library, which stores digital transfer function templates of aircraft airfoils; Reconfigurable IIR / FIR filter array, supporting online switching of topology; Virtual damping controller, injecting UDF parameters into CFD software through lock-free communication interface.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the full-time-domain fluid dynamics simulation optimization method according to any one of claims 1 to 7 is implemented.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the full-time-domain fluid dynamics simulation optimization method according to any one of claims 1 to 7 is implemented.

Images