An engine test-oriented wind tunnel flow field multi-physics coupling simulation method

Abstract

The application relates to the technical field of engine simulation test, and discloses a wind tunnel flow field multi-physical field coupling simulation method for engine test, which comprises the following steps: constructing a fluid and solid calculation domain, initializing a simulation transaction snapshot sequence; calculating a local activity index of the fluid domain, dividing the fluid domain into an active subdomain for executing a micro time step and a blunt subdomain for executing a macro time step; discretizing a simulation time axis into atomic transactions, listening to external instruction call snapshot triggering rollback and performing step length realignment; performing cross-field state projection during the atomic transactions, predicting a physical quantity load of a coupling interface and performing smooth correction; updating an interface mapping interpolation matrix based on structure displacement, calculating a fluid-solid energy residual, performing algebraic compensation on the load to drive physical field evolution; and cyclically calculating and outputting a simulation data sequence. Through subdomain step length division, snapshot rollback and energy residual compensation, the application improves the calculation efficiency and numerical stability of the multi-physical field coupling simulation under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of engine simulation and testing technology, specifically to a multi-physics coupling simulation method for wind tunnel flow fields in engine testing. Background Technology

[0002] In wind tunnel testing of aerospace engines, test specimens undergo structural deformation under the influence of high-speed airflow, resulting in a strong interaction between the fluid dynamics field and the mechanical field of the solid structure. To predict operating conditions before the test and to verify data after the test, multiphysics coupling simulation methods are typically used to numerically simulate the test process.

[0003] Current fluid-structure interaction (FSI) simulations typically employ either synchronous strongly coupled computation or asynchronous loosely coupled computation. When using synchronous strongly coupled computation, both the fluid and solid computational domains require globally uniform, small time steps to ensure computational stability. However, because the rates of change of physical states differ across different regions within the flow field, using small time steps even in regions with relatively stable airflow conditions results in numerous invalid solution steps, leading to excessively long overall simulation times and low computational efficiency.

[0004] To reduce computational load, engineering often employs asynchronous loosely coupled methods based on a fixed span, where the fluid and solid domains are solved independently with different time steps, and boundary data is exchanged at preset time nodes. However, simulating real engine testing often involves strong nonlinear disturbances such as sudden changes in external control commands or drastic shifts in shock wave positions. Under these transient conditions, a fixed asynchronous time step cannot respond promptly to high-frequency changes in physical state, leading to time delays and misalignments at the fluid-solid interface during cross-field data transfer. Combined with spatial interpolation errors at mismatched mesh interfaces, this misalignment generates virtual work at the fluid-solid interface. As the number of solution steps increases, the virtual work error at the interface accumulates, causing cross-field energy non-conservation, which in turn triggers numerical oscillations in the interface physical loads, ultimately leading to simulation divergence or loss of key transient physical characteristics. Existing simulation methods struggle to achieve an effective balance between computational efficiency and numerical stability in handling complex transient conditions. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a multiphysics coupling simulation method for wind tunnel flow fields in engine testing. This method solves the problem that existing fluid-structure interaction simulations struggle to balance computational efficiency across the entire flow domain with numerical stability at the fluid-structure interface when facing complex and transient engine operating conditions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multiphysics coupling simulation method for wind tunnel flow fields in engine testing, comprising the following steps: S1. Construct the fluid computational domain and the solid computational domain, and initialize the simulation transaction snapshot sequence; S2, the local activity index of the computational fluid computational domain, which divides it into an active subdomain that executes micro-time steps and an insensitive subdomain that executes macro-time steps; S3. Discretize the simulation time axis into atomic transactions with a period of macroscopic time step, listen for external instructions to call the simulation transaction snapshot sequence to trigger rollback, and perform step realignment to microscopic time step. S4. During atomic transactions, cross-field state projection is performed using the historical state of the active subdomain within the micro-time step to predict the physical load of the coupling interface and perform smooth correction. S5. Based on the structural displacement generated by the solid computation domain, update the interface mapping interpolation matrix, calculate the fluid-structure energy residual to perform algebraic compensation on the physical quantity load of the coupled interface, thereby driving the evolution of the physical field. S6. Repeat steps S2 to S5 until the simulation ends, and output the multiphysics coupling simulation data sequence.

[0007] Preferably, step S2 specifically includes: calculating the pressure gradient, temperature gradient, and velocity curl of each physical unit in the computational domain; normalizing these parameters and then weighted summing them to obtain a local activity index; when the local activity index is greater than a preset threshold, the corresponding physical unit is assigned to an active subdomain and a micro-time step is allocated; when the local activity index is less than or equal to the preset threshold, the corresponding physical unit is assigned to an insensitive subdomain and an amplified macro-time step is allocated. This process establishes a time step scheduling mechanism by extracting the spatial differences in the flow field characteristic gradients, thereby reducing the computational load in smooth flow regions while ensuring the computational accuracy of shock waves and areas of strong disturbance.

[0008] Preferably, step S3 specifically includes: in the execution of each atomic transaction, sampling and filtering the external instructions to calculate their signal fluctuation factor within the current macroscopic time step; when the signal fluctuation factor exceeds a preset arbitration threshold, interrupting the currently executing atomic transaction and clearing intermediate variables; according to the index pointer recorded in the simulation transaction snapshot sequence, reverting the physical state to the start time of the atomic transaction, and forcing each subdomain to uniformly adopt the microscopic time step for execution step realignment calculation. The above operations introduce a rollback and recalculation mechanism based on state backup to avoid numerical divergence caused by time scale mismatch at the point of change in operating conditions, and maintain the computational continuity of the transient change process.

[0009] Preferably, the process of performing cross-field state projection and predicting the physical loads at the coupling interface in step S4 includes: extracting historical transient features of the microscopic time step using a second-order Taylor expansion to calculate preliminary predicted values ​​of the interface loads; inputting the spatial coordinate vector, local Mach number, dimensionless thickness of the boundary layer, and surface temperature of the fluid-structure interaction interface as input features into a pre-trained deep residual network, and outputting a nonlinear correction; and superimposing the nonlinear correction onto the preliminary predicted values ​​of the interface loads to generate the predicted physical loads at the coupling interface. This process combines a time extrapolation model with a deep residual network to compensate for the time step differences caused by asynchronous solving, providing continuous dynamic boundary inputs for the solid computation domain.

[0010] Preferably, the smoothing correction process in step S4 includes: constructing a circular buffer smoothing pool in memory, aligning and caching the predicted coupling interface physical quantity loads with the measured loads fed back in real time from the fluid solution according to timestamps; dynamically determining the relaxation factor based on the variance of the residual fluctuations; and using the relaxation factor to weight and fuse the predicted coupling interface physical quantity loads with the measured loads to generate the corrected coupling interface physical quantity loads. This process eliminates the interface numerical jumps caused by the discontinuity of load data at the asynchronous step-size alternation points in the fluid-structure interaction domain.

[0011] Preferably, step S5, which updates the interface mapping interpolation matrix based on the structural displacement generated by the solid computational domain, includes: constructing a dynamic octree spatial index structure to track structural displacement, and introducing a distance threshold filtering algorithm to match the fluid computational domain grid nodes and solid computational domain structural unit nodes with topological associations; based on the spatially scattered point distribution characteristics of the matched nodes, using radial basis functions with regularized diagonal terms to construct local topological interpolation relationships, and generating the updated interface mapping interpolation matrix. This step solves the problem of spatial coordinate misalignment between the non-matching grids during the deformation of the wind tunnel flow field and the loaded structure.

[0012] Preferably, the process of calculating the fluid-structure interaction energy residual to perform algebraic compensation for the physical quantity loads at the coupling interface in step S5 includes: statistically analyzing the first time integral value of the work done by the flow field pressure at the fluid-side interface and the second time integral value of the work done by the structural node response at the structural-side interface within the current atomic transaction cycle; calculating the difference between the first and second time integral values, dividing this difference by the first time integral value with a correction deviation constant to obtain the energy residual comparison factor as the fluid-structure interaction energy residual; if the energy residual comparison factor is greater than a preset deviation threshold, activating the energy conservation constraint operator, and performing algebraic compensation on the amplitude of the physical quantity loads at the coupling interface applied to the solid computational domain through scaling and gradient compensation. This step constructs an error correction process based on the principle of energy equivalence, compensating for the virtual work calculation deviation in the fluid-structure interaction process, and maintaining the physical conservation of the coupling calculation.

[0013] Preferably, step S6, which involves repeating the iterative convergence determination process from steps S2 to S5, includes: monitoring the decrease rate of the residual norm of the fluid dynamics field and the decrease rate of the residual norm of the solid structure field in the atomic transaction calculation; dynamically assigning weights to the two decrease rates based on the load pulsation frequency of the fluid-structure interaction interface, and calculating a joint convergence factor; and dynamically adjusting the iteration depth within the current macroscopic time step based on the joint convergence factor. This determination logic takes into account the nonlinear differences between the fluid and solid domains at different calculation stages and adaptively controls the number of iterations within the macroscopic time step.

[0014] Preferably, the method further includes a step of predictive scaling of the step size based on the joint convergence factor: when the joint convergence factor fails to reach the target threshold within a preset maximum number of iterations, the current macroscopic time step is reduced by a preset ratio; the physical state of the previous convergence time step is used as a reference, and the boundary conditions are reloaded to perform a trial calculation of the physical field evolution. This mechanism ensures that when abrupt changes in the physical field exceed the time step solution range, the computation process can re-establish a numerically stable state by adjusting the time reference.

[0015] Preferably, the process of outputting the multiphysics coupling simulation data sequence in step S6 includes: after the atomic transaction completes the calculation submission, extracting the high-dimensional physical quantity data from the fluid and solid computational domains into a dedicated mirror buffer; using principal component analysis operators to perform key feature dimensionality reduction on the extracted high-dimensional physical quantity data to generate a low-dimensional structured data array; adding a global timestamp to the low-dimensional structured data array to complete time series alignment, generating and outputting the multiphysics coupling simulation data sequence. This process reduces the dimensionality of large-scale grid data, forming structured data with strict time series characteristics, which facilitates subsequent engineering comparison and analysis.

[0016] This invention provides a multiphysics coupling simulation method for wind tunnel flow fields in engine testing. It offers the following advantages: 1. This invention divides the fluid computational domain into an active subdomain executing micro-time steps and an insensitive subdomain executing macro-time steps by calculating the local activity index of the fluid computational domain. This approach uses a smaller time step in regions of drastic flow field changes to ensure computational accuracy, while using a larger time step in regions of stable flow to reduce the number of solutions. This maintains the accuracy of solving key local features of the flow field while reducing the overall computational load of multiphysics coupled simulations and improving computational efficiency.

[0017] 2. This invention discretizes the simulation time axis into atomic transactions with a macroscopic time step period. When abnormal fluctuations in external commands are detected, a rollback is triggered by calling the simulation transaction snapshot sequence, and step realignment is performed towards the microscopic time step. When faced with nonlinear disturbances such as sudden changes in control commands during engine testing, this mechanism can promptly revert the physical states on both sides of the fluid-structure interaction and uniformly reduce the time scale for recalculation, avoiding numerical divergence caused by the mismatch between the time step and the physical disturbance scale, thus improving the stability of the simulation process.

[0018] 3. Based on the cross-field state projection prediction of physical quantity loads, this invention updates the interface interpolation matrix through the displacement of the solid domain structure and performs algebraic compensation for the physical quantity loads at the coupled interface by calculating the fluid-structure interaction energy residual. This method directly corrects the virtual work errors generated during mismatched mesh interpolation and asynchronous time progression, compensates for energy calculation deviations at the fluid-structure interaction interface, maintains the physical energy conservation during cross-field data exchange, and prevents calculation failures caused by error accumulation in long-term coupled simulations. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the dynamic hierarchical partitioning process of the computational domain based on feature gradients in this invention. Figure 2 This is a schematic diagram illustrating the asynchronous scheduling and logical arbitration principle driven by atomic transactions in this invention. Figure 3 This is a schematic diagram illustrating the principle of topological sensing mapping and energy conservation closed-loop correction in this invention. Figure 4 This is a schematic diagram illustrating the adaptive iterative evolution and data sequence integration output principle of the present invention; Figure 5 A bar chart comparing the computation time of different multiphysics coupling simulation methods of this invention; Figure 6 This is a line graph showing the evolution of energy residuals at the coupling interface during the severe deformation intervals of this invention. Figure 7 This is a scatter envelope diagram showing the tracking and comparison of peak dynamic stress under extreme transient conditions according to the present invention. Detailed Implementation

[0020] Please see the appendix Figure 1 -Appendix Figure 4 This invention provides a multi-physics coupled simulation method for wind tunnel flow fields for engine testing, comprising the following steps: S1, construct the fluid computational domain and the solid computational domain, set the fluid control equations and the solid control equations, and initialize the corresponding simulation transaction snapshot sequence; S2 calculates the physical quantity gradient of physical units within the fluid computing domain in real time, generates a local activity index, and divides the fluid computing domain into an active subdomain that executes micro-time steps and an insensitive subdomain that executes macro-time steps based on the local activity index. S3 discretizes the simulation time axis into multiple atomic transactions with a macroscopic time step as the period, listens to external control command signals, and determines whether to use the simulation transaction snapshot sequence to perform transaction rollback and step realignment according to a preset threshold. S4 performs cross-field state projection prediction during atomic transaction execution, uses historical states of micro-time steps to predict the physical load of the coupling interface, and uses a buffer attenuation factor to smooth the predicted physical load of the coupling interface. S5 updates the interface mapping interpolation matrix based on the structural displacement generated by the solid computation domain, simultaneously calculates the fluid-structure energy residual of the coupled interface, and performs algebraic compensation on the physical quantity loads of the coupled interface using the energy conservation constraint operator. S6. Repeat the above steps until the logical evolution of the preset simulation time axis is completed, and output a multiphysics coupling simulation data sequence containing the flow field, structural response and coupling interface state.

[0021] In this embodiment of the invention, a fluid computational domain and a solid computational domain are first constructed using mesh generation technology. The fluid computational domain employs the conserved form of the Navier-Stokes equations as the fluid governing equations, while the solid computational domain employs the coupled equations of nonlinear structural dynamics and heat conduction as the solid governing equations. At the start of the computation, the state variables of all spatial units are archived and indexed, forming a sequence of simulation transaction snapshots. Each snapshot in the simulation transaction snapshot sequence contains the states of fluid density, velocity, total energy, and solid displacement, temperature, and other physical quantities, serving as the starting reference for subsequent atomic transaction calculations.

[0022] During the simulation, the computational engine continuously acquires the pressure gradient, temperature gradient, and velocity gradient of each physical unit in the fluid computational domain. These gradients are normalized and then weighted to obtain the local activity index. The local activity index is used to characterize the severity of transient changes in the flow field. When the local activity index is greater than a preset threshold (for example, set to 0.65-0.85), the physical unit is classified into an active subdomain, and its computation time step is set to a microscopic time step with a higher response frequency. When the local activity index is less than or equal to the preset threshold, the physical unit is classified into an insensitive subdomain, and its computation time step is set to a macroscopic time step magnified by a preset factor. In this embodiment, the preset factor is preferably in the range of 10-100.

[0023] The logical evolution of this method uses atomic transactions as the basic scheduling unit. One atomic transaction corresponds to a physical time span of one macroscopic time step. Before each atomic transaction starts, the scheduling engine records the latest index of the current simulation transaction snapshot sequence. During the execution of an atomic transaction, the amplitude changes of external control command signals are monitored in real time. If the change in the external control command signal exceeds a preset threshold (e.g., the signal amplitude mutation rate exceeds 5%), the arbitration logic will forcibly terminate the currently executing atomic transaction, clear the intermediate variables generated during that period, and revert the entire state to the transaction start time according to the index pointer recorded in the simulation transaction snapshot sequence. At the same time, the step size realignment calculation is forcibly performed by the microscopic time step.

[0024] To maintain data continuity at the fluid-structure interaction interface during normal operation of the atomic transaction, this method performs cross-field state projection prediction. Utilizing the completed historical microscopic time step, the predicted load at the current macroscopic step endpoint of the atomic transaction is calculated using a finite difference extrapolation algorithm. The predicted load serves as the dynamic boundary condition input to the solid computational domain. When the calculation reaches the macroscopic step endpoint, this method uses a buffer attenuation factor to weight and fuse the predicted load with the real-time load calculated from the current flow field, generating the final physical quantity load at the coupling interface. The buffer attenuation factor employs an exponential decay form to eliminate numerical jumps at the coupling interface caused by asynchronous step size switching.

[0025] Data interaction in physical space is achieved through an interface mapping interpolation matrix. This matrix is ​​constructed using the radial basis function method and dynamically reconstructed based on the structural displacements generated by the solid computational domain, ensuring topological consistency between the fluid computational domain elements and the solid computational domain elements in their deformed spatial positions. At the end of each atomic transaction, the energy flow balance at the coupling interface is verified. The fluid-structure interaction (FSI) energy residual between the mechanical work done by the fluid on the interface and the energy absorbed by the solid is calculated. If the FSI energy residual is not zero, the energy conservation constraint operator is activated, applying corresponding algebraic compensation increments to the physical quantity loads at the coupling interface, ensuring that the overall simulation evolution conforms to the laws of physical conservation.

[0026] The aforementioned atomic transactions constitute a closed-loop cyclical evolution process. By continuously updating the simulation transaction snapshot sequence, this method adaptively adjusts the ratio of the macroscopic time step to the microscopic time step based on the convergence speed of the fluid-structure energy residual. Each subdomain advances collaboratively according to the dynamically divided calculation step size until the simulation time axis reaches the preset endpoint. Finally, it outputs a multiphysics coupled simulation data sequence composed of the calculation results of each atomic transaction, completing the numerical simulation of the engine wind tunnel test process.

[0027] This invention provides a multiphysics coupled simulation system for wind tunnel flow fields in engine testing. The implementation methods for physical field construction and snapshot initialization are described below: Based on the perturbation characteristics of the flow field distribution caused by the complex geometry in wind tunnel tests, this method uses a preprocessing module to geometrically discretize the test environment inside the wind tunnel. The fluid computational domain covers the wind tunnel contraction section, test section, and exhaust expansion section, using unstructured tetrahedral or polyhedral meshes for space filling to capture the complex flow characteristics near the engine nozzle. The solid computational domain meshes the engine wall and internal support structure, typically using hexahedral meshes or second-order tetrahedral elements to ensure the accuracy of structural stress calculations.

[0028] Considering the accuracy of physical quantity exchange at the fluid-structure interaction interface, in this embodiment, the fluid computational domain and the solid computational domain maintain geometric consistency at the interface. A spatial correlation balance relationship is established through common node technology or non-aligned interface mapping technology to ensure that the algebraic sum of the interpolation weights is always 1 during the coordinate mapping process, thereby avoiding the generation of non-physical numerical source terms.

[0029] Within the computational domain of fluid dynamics, this method establishes fluid governing equations based on the conservation of mass, momentum, and energy. For high-speed compressible flow within a wind tunnel, the fluid governing equations adopt the conservation form of the Navier-Stokes equations as follows: ; In the formula, The vector of conserved fluid variables; It is a time variable; , , These are the spatial coordinate variables in three dimensions of a Cartesian coordinate system; , , respectively along direction, direction, The inviscid convection flux vector in the direction; , , respectively along direction, direction, The viscous dissipation flux vector in the direction.

[0030] This set of equations describes the transport laws of energy and momentum in high-speed airflow by dynamically balancing convective flux and diffusion flux, providing a mathematical basis for subsequent capture of shock waves and boundary layer disturbances.

[0031] For turbulence effects in the computational domain of fluids, those skilled in the art can select a suitable closed model based on the Reynolds average method or the large eddy simulation method. The specific implementation of this model is a well-known technique in the field and will not be elaborated here.

[0032] Within the solid-state computational domain, this method configures solid-state governing equations to describe the dynamic response and heat conduction processes of the structure. The solid-state governing equations include stress balance equations and heat diffusion equations, and their matrix form is as follows: ; In the formula, The mass matrix of this method is used to construct the structure; This represents the displacement response vector of the structural node; The time variable for solving the dynamic response of the structural field; Here is the damping matrix of the structural system; Here is the stiffness matrix of the structural system; This is the equivalent external load vector acting on the interface of the structural entity.

[0033] To ensure the stability of the numerical solution, in this embodiment, the stiffness matrix is... Perform a singularity check; if the matrix condition number exceeds a preset threshold of 10, the check will be performed. 12 Then, regularization is performed by applying weak constraint operators to prevent the displacement vector from being regularized during the solution process. Numerical divergence has occurred.

[0034] Based on the rollback requirement of atomic transactions, this method constructs a sequence of simulation transaction snapshots in memory. The simulation transaction snapshot sequence is a structured data queue with timestamp indexes, used to store the complete physical state of the computational domain at specific discrete points in time.

[0035] During initialization, this method encapsulates the density, velocity, pressure, temperature, and intermediate variables of the turbulence model for all elements in the fluid computational domain, along with the displacement, velocity, acceleration, stress elongation, and temperature distribution of each node in the solid computational domain, into a transaction snapshot and stores it at the front of the queue. Each transaction snapshot is assigned a unique incrementing index value, which serves as a logical primary key and is bound to the timing logic of the full-field simulation.

[0036] To address unsteady disturbances caused by operating condition adjustments during wind tunnel testing, the storage depth of the simulation transaction snapshot sequence is determined by the cache configuration of this method. By establishing a pointer mapping table, this method can locate state data at any index position in real time.

[0037] When external intervention signals are detected or computational divergence occurs during subsequent calculations, this method restores the computational state on the time axis by calling backup data from the simulation transaction snapshot sequence. During this process, the method performs a working condition timestamp alignment check to ensure that the fluid state and solid displacement after rollback are at the same physical moment, with the alignment error controlled within 1% of the microscopic time step.

[0038] This method receives external environmental parameters through an input module to set initial boundary conditions. For the fluid computational domain, far-field pressure, total temperature, and inlet velocity vector are set; for the solid computational domain, zero displacement constraints are set at the structural supports, and initial values ​​for the thermodynamic coupling between the fluid and solid at the physical interface are established.

[0039] As a preferred approach, the initial pressure field value is set based on the static pressure calibration value of the wind tunnel test section to reduce numerical disturbances during the simulation startup phase. After initialization, this method triggers global time-stepping logic, guiding the fluid control equations and solid control equations into a time-domain evolution loop.

[0040] Please see the appendix Figure 1 , Figure 1 This is a flowchart of dynamic hierarchical partitioning of the computational domain based on feature gradients according to an embodiment of the present invention. The implementation method of the multiphysics coupled simulation system for wind tunnel flow fields for engine testing provided by the present invention for dynamic hierarchical partitioning of the computational domain is described below: Considering the objective physical characteristics of wind tunnel flow fields, which exhibit both highly pulsating and gently evolving regions in their spatial distribution, this method employs a feature extraction module to monitor the fluid computational domain in real time to improve the accuracy of capturing key regions within limited computational resources. At the beginning of each macroscopic time step in the simulation evolution, this method initiates a dynamic hierarchical partitioning logic, utilizing the spatial non-uniformity of the flow field's physical gradient to allocate computational weights on demand.

[0041] S201, this method extracts the transient flow field characteristics of each physical element in the fluid computational domain. To comprehensively characterize the drastic evolution of the flow field, the extracted parameters include pressure gradient, temperature gradient, and curl of the velocity vector. These parameters directly reflect the shock wave intensity and turbulent dissipation level near the engine nozzle. In this embodiment, the calculation of each gradient uses the central difference operator based on the least squares method, which has good geometric adaptability in unstructured meshes. For any physical element, the quantification of its comprehensive characteristic intensity is achieved by calculating the local activity index, the mathematical expression of which is as follows: ; In the formula, For grid encryption indicator factors; , , These are the weighting coefficients for the pressure gradient, temperature gradient, and vorticity, respectively. Let be the magnitude of the pressure gradient vector of the target element in the flow field; This serves as the reference pressure baseline value for the global computational domain. To prevent the denominator of division from approaching zero, a minimal regularization constant is used; Let be the magnitude of the temperature gradient vector of the target element in the flow field; This serves as the reference temperature base value for the global computational domain. Let be the magnitude of the velocity curl vector of the target element in the flow field; This serves as the reference vorticity baseline value for the global computational domain.

[0042] To ensure the robustness of the algorithm under extreme vacuum or cryogenic conditions, a correction operator is introduced into the denominator term in this embodiment. Its value is 10 -10 This is used to avoid the risk of numerical overflow caused by the denominator approaching zero. Weighting coefficients , , The values ​​of all are in the range [0,1], and their sum is 1; As a preferred approach, when the simulation target tends to observe the shock wave structure, increasing... The weighting; when focusing on the combustion thermal response, the weighting should be increased accordingly. The percentage.

[0043] S202, this method classifies the spatial distribution attributes of the fluid computational domain based on the local activity index. This method presets a classification threshold, which is determined based on the statistical standard deviation of historical wind tunnel test data. In this embodiment, the preferred value range of the classification threshold is 0.65 to 0.85. When the local activity index of a physical unit is greater than the classification threshold, the unit is marked as a member of the active subdomain.

[0044] Active subdomains are typically distributed in the shock wave trapping zone, boundary layer separation points, and fuel injection mixing zone. If the local activity index is less than or equal to the classification threshold, the unit is marked as a member of the insensitive subdomain, mainly covering the straight flow region near the wind tunnel wall and the far-field uniform flow region. This judgment logic, based on multi-dimensional physical characteristics weighted rather than a single extreme value, can more accurately locate the unsteady core region in the flow field.

[0045] S203, for the partitioned heterogeneous subdomains, this method performs asynchronous mapping of time steps. Physical units in the active subdomains are assigned microscopic time steps to capture high-frequency pressure wave fluctuations; physical units in the insensitive subdomains are assigned macroscopic time steps.

[0046] The macroscopic time step is set to an integer multiple of the microscopic time step, and this multiple is dynamically given by the computational efficiency evaluation model. Based on the consistency requirement of energy transfer, this method introduces a spatial smoothing operator in the transition layer region across subdomains to ensure that the step size transition between adjacent active and insensitive units is gradient-like, avoiding numerical reflection caused by abrupt changes in time step at the subdomain boundary.

[0047] S204, This method performs dynamic refreshing of the subdomain topology. Based on the migration characteristics of the flow field structure, such as the movement of the shock wave position with the controlled command, this method performs a rescan of the full-field local activity index every preset number of atomic transaction cycles. In this embodiment, the preset number is 5-10 cycles.

[0048] In this embodiment, the method utilizes the current flow field prediction trend to pre-activate insensitive units that will enter the high feature gradient region in the next stage. Through this feedforward adjustment mechanism, it is possible to ensure that the adjustment of the subdomain boundary precedes the diffusion of physical fluctuations, thereby maintaining the continuity and integrity of the simulated physical process.

[0049] Based on the above partitioning results, this method transforms the fluid computational domain from a unified time evolution model into a hierarchical computational model with spatiotemporal heterogeneity. This approach significantly reduces computational redundancy in the far-field smooth flow region while ensuring simulation accuracy in core areas such as the engine nozzle. After completing the subdomain partitioning, this method feeds back the step size attribute of each element to the solution control module, entering the next stage of multi-field coupled computation.

[0050] Please see the appendix Figure 2 , Figure 2 This is a schematic diagram of asynchronous scheduling and logical arbitration driven by atomic transactions according to an embodiment of the present invention. The implementation method of the multiphysics coupling simulation system for wind tunnel flow fields for engine testing provided by the present invention, specifically for asynchronous scheduling and logical arbitration, is described below: To address the timescale mismatch between external high-frequency control commands and the large time lag in numerical simulation during wind tunnel testing, this method introduces an atomic transaction processing mechanism. An atomic transaction refers to a set of physical evolution operations performed in the fluid and solid computational domains within a macroscopic time step. It possesses logical indivisibility and aims to ensure the logical integrity of the numerical computation process and the consistency of data states, guaranteeing that before the end of the cycle, the physical states of all computational domains are either fully committed or completely rolled back.

[0051] S301, this method establishes a real-time monitoring and preprocessing mechanism for external control signals. Control commands in wind tunnel tests include engine throttle valve opening signals, nozzle geometry adjustment parameters, and wind tunnel total temperature and total pressure commands. These signals are injected into the simulation system in real time via a high-speed serial bus or industrial Ethernet interface.

[0052] Within the execution cycle of each atomic transaction, the frequency synchronization module discretizes and samples the external signal, calculating the rate of change of the signal within the current macroscopic time step. As a preferred approach, to eliminate high-frequency electromagnetic interference noise, this method employs a moving average filter to smooth the control signal, thereby providing a realistic physical disturbance input for subsequent logic arbitration.

[0053] Based on the causal relationship between the disturbance amount and the simulation convergence, S302, this method performs logic arbitration based on the signal disturbance intensity. The logic arbitration aims to assess whether the severity of the external operating condition will lead to simulation distortion. This method quantifies the external disturbance by calculating the signal fluctuation factor, the formula of which is as follows: ; In the formula, The global state change perception index; The total number of key physical state variables participating in the dynamic evaluation of this method; For the index number of the physical state variable; For the first A physical state variable at a microscopic moment The predicted value; For the first At the current moment, the physical state variables are... The measured values; For the first The maximum amplitude boundary of a physical state variable within its historical evolution cycle; This is the current macroscopic trial calculation step size; This is a small numerical offset introduced to prevent denominator anomalies; For the first The sensitivity evaluation weight coefficients corresponding to each physical state variable are set according to the degree of influence of the physical quantity on the flow field structure, and the values ​​range from [0,1].

[0054] Considering that the denominator term may cause numerical calculation anomalies under static operating conditions, a correction factor is introduced in this embodiment. Its value is 10 -8 This is used to avoid the risk of division by zero.

[0055] This formula evaluates the comprehensive interference of command changes on the stability of the method by normalizing and weighting the average of multi-dimensional parameters. Based on the above calculation results, this method sets a preset arbitration threshold, which is dynamically adjusted according to the residual convergence speed of the previous cycle. As a preferred implementation, the initial benchmark value of the preset arbitration threshold is set between 0.15 and 0.25. If the comprehensive signal fluctuation factor is less than or equal to the preset arbitration threshold, this method determines that the current operating condition is stable and allows the simulation state vector to step to the next macroscopic time step.

[0056] S303, Asynchronous driving and state commit logic for atomic transactions. During the execution of the macroscopic step size, the active subdomain performs multiple high-frequency calculations using the microscopic time step size, while the insensitive subdomain runs a complete macroscopic step size under the maintenance of the load under the interface projection. To confirm the physical authenticity of the asynchronous calculation results, this method initiates state consistency verification after all physical units within the atomic transaction have completed the cumulative calculation of the preset step size.

[0057] In this embodiment, the verification is based on the displacement continuity and force balance residual at the fluid-structure interface. When the interface load deviation is less than a preset tolerance threshold (in this embodiment, the preset tolerance threshold is set to 5%), the method performs an atomic transaction commit operation, updates the current physical field data to the simulation transaction snapshot sequence, and releases the relevant memory. If the interface load deviation is greater than or equal to the preset tolerance threshold, it is determined that the current asynchronous calculation has caused physical state distortion, the commit is rejected, and the rollback logic of S304 is triggered.

[0058] Based on the self-healing requirement of the simulation logic, in S304, this method executes the transaction rollback and reconstruction logic triggered by instruction mutation. If the comprehensive signal fluctuation factor is greater than the preset arbitration threshold, or if the interface load deviation is detected to be greater than or equal to the preset tolerance threshold (i.e., internal state distortion) in S303, it is determined that the current simulation environment has been destroyed, such as when the engine experiences emergency throttling or when the wind tunnel experiences shock wave oscillation.

[0059] At this point, this method immediately interrupts all currently executing asynchronous computation processes and retrieves the state data of the previous commit point in the simulation transaction snapshot sequence via a pointer reset instruction. Based on the above rollback operation, in order to capture the non-steady details of the mutation process, this method forcibly reconstructs the current scheduling strategy, adjusts the precision of the replayed time period division to micro-time steps, and disables asynchronous acceleration logic across the entire field.

[0060] This rollback and recalculation mechanism ensures that when the simulation process encounters physical nonlinear transitions, it can still capture transient evolutions using the smallest physical feature resolution scale, thus avoiding numerical oscillations at signal abrupt changes.

[0061] Based on the aforementioned logical arbitration and asynchronous scheduling, this method achieves precise compatibility with extreme unstable conditions while maintaining computational efficiency for most stable operating conditions. This dynamically switching transaction processing architecture effectively resolves the mapping conflict problem between multi-source control instructions and computational domains with different step sizes. After completing the execution of the logical arbitration branch, this method passes the final determined time step topology and state vector to the cross-field state projection module.

[0062] The multiphysics coupled simulation system for wind tunnel flow fields for engine testing provided by this invention is described below for its implementation method of cross-field state projection and smoothing correction: Considering the significant differences between the fluid computational domain and the solid computational domain in terms of solution frequency and physical quantity evolution rate, this method introduces a cross-field state projection mechanism in order to eliminate the time gap between heterogeneous domains and achieve physical-logical alignment.

[0063] S401, this method constructs a projection model of interface physical quantities based on spatiotemporal correlation. Because the high-speed flow loads in wind tunnel tests have extremely strong unsteady characteristics, maintaining a constant load within a macroscopic step size will lead to step-type numerical errors in the solid domain response. Therefore, this method extracts historical transient characteristics at the interface in the flow field and uses a time extrapolation model to compensate for step size mismatch.

[0064] In this embodiment, a second-order Taylor expansion is used to predict the physical quantities at the interface, and its mathematical expression is as follows: ; In the formula, The physical point and time at which the interface is located during dynamic load prediction; To predict the fluid-solid interface at microscopic time The physical load vector; For the interface at the currently known time The steady-state load reference value; This represents the microstep offset relative to the current moment; The first-order time rate of change of the interface load; Let be the second-order time rate of change of the interface load.

[0065] This formula provides a continuous dynamic boundary reference value for the solid domain by extracting the inertial trend of the flow field evolution. Its technical purpose is to maintain the evolution consistency of the interface energy density under asynchronous step size.

[0066] To capture the highly nonlinear mapping of interface loads under complex shock wave disturbances, S402, this method performs interface state enhancement prediction based on a deep residual network. The deep residual regression network integrated in this method consists of an input layer with 8-dimensional feature inputs, four fully connected hidden layers each containing 64 neurons activated by a linear rectified function, and an output layer introducing a skip connection structure.

[0067] The input layer data items include the spatial coordinate vector of the interface node (3D), the local Mach number (1D), the dimensionless thickness of the boundary layer (1D), the near-wall vorticity intensity (1D), the interface static pressure (1D), and the interface surface temperature (1D). These parameters were chosen because they directly determine the fluctuation frequency and amplitude of the wall pressure distribution.

[0068] After the input data is preprocessed by standard deviation normalization, it passes through each hidden layer in sequence and is then fused with the linear transformation term of the input features in the output layer using residuals.

[0069] During the training phase, this method collects historical wind tunnel calibration data as samples and uses convergent solutions from high-fidelity large eddy simulations as labels. The loss function adopts the following modified weighted mean square error function: ; In the formula, This represents the total loss function value during model training. The total number of training samples participating in network weight updates; The index number of the data sequence input for the model training samples; For the first The prior importance weight coefficients corresponding to each training sample; For the residual network model targeting the first The interface load prediction vector output by each sample; For the first The high-fidelity real label vector corresponding to each sample; The squared L2 norm of all connection weights within the network model; This is the regularization coefficient, usually taken as 10. -5 To suppress overfitting, the model can output interface load predictions after nonlinear correction. This method superimposes the nonlinear correction output by the residual network onto the initial prediction value of the Taylor expansion in S401 as the final candidate load after comprehensive enhancement, thus effectively making up for the insufficient prediction accuracy of Taylor expansion in strong discontinuity regions.

[0070] Based on the numerical mismatch that may exist between the predicted values ​​and the measured values ​​of the equations S403 performs physical quantity smoothing correction based on a buffer pool mechanism. This method constructs a circular buffer smoothing pool in memory, aligning and caching the predicted sequence with the real-time sequences fed back from each computational domain by timestamp. To prevent numerical oscillations induced by abrupt changes in interface loads, this method introduces a relaxation factor for smoothing iterations: ; In the formula, This refers to the corrected load term ultimately applied to the structural boundary; The candidate payloads are obtained after enhanced predictive fusion; This is the precise calculated value fed back by the fluid solver at the end of this step; It is a relaxation factor. The value of is dynamically determined by calculating the variance of the residuals, and its range is [0.1, 0.95]. When the prediction difference is large, this method reduces . To smooth the signal trajectory. In this embodiment, if the calculation involves inverting a matrix, this method uses singular value decomposition instead of direct inversion to ensure the robustness of matrix operations during momentum exchange.

[0071] S404, this method performs topology alignment and displacement feedback at the computational domain interface. After each round of projection prediction, this method calls the geometric mapping operator to integrate the pressure of the fluid mesh nodes into the equivalent force of the solid finite element nodes. As a preferred approach, to monitor the stability of fluid-structure interaction, this method performs a comprehensive judgment based on the bidirectional residuals of displacement and load.

[0072] If the deformation exceeds 5% of the element feature length, this method initiates local remeshing logic. Through the coordinated processing of prediction, enhancement, and correction described above, this method achieves high alignment of cross-field data under asynchronous step sizes, providing accurate boundary support for efficient stepping of atomic transactions.

[0073] Please see the appendix Figure 3 , Figure 3 This is a schematic diagram illustrating the principle of topology-sensing mapping and energy conservation closed-loop correction according to an embodiment of the present invention. The implementation method of the wind tunnel flow field multiphysics coupling simulation system for engine testing provided by the present invention, specifically for topology-sensing mapping and energy conservation correction, is described below: Based on the mismatched mesh characteristics of the engine nozzle and the airframe structure in the wind tunnel flow field, this method introduces a topology-sensing mapping and energy conservation closed-loop correction mechanism to address the spatial misalignment between the fluid mesh and the structural mesh at the interface and the resulting energy transfer imbalance. This mechanism aims to ensure that the work done at the interface and heat transfer are physically conserved on a numerical scale during cross-field data exchange.

[0074] S501, this method performs a topology-aware mesh search based on spatial bounding box filtering. Since the fluid domain typically uses dense tetrahedral meshes to capture shock waves, while the structural domain mostly uses hexahedral elements, the node distribution at the interface does not overlap. This method first constructs a dynamic octree spatial index structure to track the geometric deformation of the engine wall caused by stress in real time, and then finds the projected footprint of each fluid interface node on the structural mesh surface.

[0075] In this embodiment, the method employs an asymmetric projection operator, utilizing the intersection of the fluid node normal direction and the envelope surface of the structural element to determine the corresponding weights. As a preferred approach, the search process introduces distance threshold filtering, performing topological association only on node pairs whose Euclidean distance is less than 1.5 times the average feature size of the fluid mesh, thereby eliminating physically irrelevant far-field interference.

[0076] To improve the accuracy of the interpolation process and avoid local numerical singularities, this method constructs an interpolation matrix based on radial basis functions and utilizes weighted combinations of spatially scattered points to achieve smooth transfer of physical quantities. In step S502, this method performs physical quantity interpolation mapping at the fluid-structure interface. By introducing global support parameters, this method maps the pressure and temperature properties of fluid nodes to structural nodes.

[0077] When constructing a system of linear equations to solve for the interpolation weights, considering that coplanar support points or excessively close distances can lead to singularities in the mapping matrix, this method introduces a regularization term on the diagonal of the matrix.

[0078] The load mapping process is as follows: ; In the formula, The equivalent scalar field values ​​mapped and assigned to the target structural nodes; This represents the number of fluid reference nodes located within the support domain that participate in local topology interpolation. The topology traversal index of the fluid reference node for the flow from the mesh node to the structure; For the first A mapping radial basis function centered on a fluid source node; For the target structure node in the mapping space and the first Euclidean distance between fluid source nodes; To solve the inverse of a system of linear equations The interpolation contribution weight of each fluid node; To ensure that the spatial interpolation calculations possess Galilean translation invariance for first-order polynomials; This is the three-dimensional absolute coordinate vector of the target structure node during interpolation calculation.

[0079] This formula uses polynomial terms to eliminate mapping bias, ensuring the continuity of load distribution during mesh deformation. In this embodiment, the method checks the value of the determinant of the mapping matrix; if this value is less than a preset minimum value of 10... -12 If the regularization coefficient is increased, the numerical stability will be improved.

[0080] Based on the conservation laws of physics and mechanics, the virtual work applied by the fluid domain to the interface should be equal to the virtual work received by the structural domain. However, in the mathematical mapping process of non-matching meshes, discrete errors can lead to unequal virtual work, thereby causing artificial dissipation or surge in system energy.

[0081] S503, a closed-loop correction based on the principle of energy equivalence. To eliminate this error, this method introduces an energy feedback correction loop, which statistically analyzes the total interfacial pressure work on the fluid side and the structural deformation work on the structure side within one atomic transaction cycle, and calculates the energy residual comparison factor between the two. Its mathematical expression is as follows: ; In the formula, To evaluate the energy residual alignment factor in the cross-field mesh mapping process; This refers to the physical region at the mismatched interface between the fluid and the structure. This represents the environmental pressure vector on the target mesh node in the fluid domain. This represents the spatial velocity vector of the interface mesh node corresponding to the fluid domain. The integral evolution of the time variable for the work done by the physical quantity at the interface; The discrete area of ​​the spatial infinitesimal element of the fluid side interface; This is the equivalent contact load vector applied to the structural domain nodes after mapping and transformation; This is the displacement response velocity vector of the deformation node corresponding to the structural domain; The discrete area of ​​the spatial infinitesimal element of the structural side interface; To avoid the abnormality of the energy ratio denominator approaching zero under zero-power conditions such as cold start (exhaust cold rotation). Correction deviation constant.

[0082] Considering that the total work done on the fluid side tends to zero at the moment of wind tunnel startup or shutdown, a correction term with a value of 10 is introduced into the denominator of the formula. -6 This factor is used to mitigate the risk of the denominator approaching zero in division operations. It quantifies cross-field energy transfer deviation, providing a basis for subsequent closed-loop compensation.

[0083] S504, this method performs load amplitude redistribution correction. The output result is determined based on a weighted logic of the energy residual comparison factor and the physical field residual trend. When the energy residual comparison factor is greater than a preset deviation threshold (in this embodiment, the threshold is set to 0.01), it is determined that there is energy leakage in the current mapping.

[0084] This method initiates a closed-loop correction procedure, using global scaling operators and local gradient compensation operators to fine-tune the magnitude of the load vectors at the structural nodes. This step, by forcibly matching the power consumption between the two fields, ensures that coupling this method does not generate non-physical numerical gains within the current macroscopic step size, thereby maintaining computational stability under high-frequency oscillations.

[0085] Based on the aforementioned topology-aware mapping and closed-loop correction, this method effectively solves the numerical truncation error caused by complex geometries and mismatched meshes. This energy-conservation-based feedback mechanism not only compensates for the accuracy deficiencies of simple spatial interpolation but also provides a conservation guarantee for the long-term stable operation of atomic transactions in strongly coupled environments. After completing the closed-loop correction, this method loads the final consistency load into subsequent computation modules.

[0086] Please see the appendix Figure 4 , Figure 4This is a schematic diagram illustrating the adaptive iterative evolution and data sequence integration output principle according to an embodiment of the present invention. The implementation method for convergence evolution and result output of the wind tunnel flow field multiphysics coupling simulation system for engine testing provided by the present invention is described below: To address the numerical instability caused by continuous changes in operating conditions during wind tunnel testing, and to ensure the reliability of the simulation results over long periods of operation, this method introduces an adaptive iterative evolution and data sequence integration output mechanism. This mechanism dynamically adjusts the computational weights and step size redundancy to ensure that the evolution of the physical field always remains within the convergence range, and achieves the integrated transformation of high-dimensional simulation data into a structured sequence.

[0087] S601, this method implements an adaptive iterative strategy based on residual evolution trajectories. During the convergence iteration of each atomic transaction, this method monitors the decreasing ratio of the flow field residual and the structural displacement residual in real time. Considering the differences in the nonlinear intensity of fluid-structure interaction under different operating conditions, a single convergence criterion is insufficient to adapt to full-condition simulation. In this embodiment, this method constructs a joint convergence evaluation index that integrates multiple criteria, dynamically adjusting the iteration depth within the current macroscopic step size through residual volatility. Its adaptive adjustment logic follows the following criteria: ; In the formula, To dynamically quantify the joint convergence state factor of atomic transaction processes; represents the convergence weight coefficient of the fluid dynamics field in the multidimensional evaluation system; To evaluate the count variable of the iterative convergence substep within the current macroscopic time step; For the fluid dynamics field in the first The residual boundary norm during the next iteration; Let be the residual reference norm of the fluid dynamics field at the current macroscopic step starting point; represents the convergence weight coefficient of the solid structure field in the multidimensional evaluation system; For the solid structure field in the first The residual boundary norm during the next iteration; Let be the residual reference norm of the solid structure field at the current macroscopic step starting point; To prevent division overflow caused by perfect matching of initial residuals, a minimum stability constant is set to 10. 10 This is used to avoid the risk of division overflow caused by an initial residual of zero. In this embodiment, the weighting coefficients are dynamically allocated based on the load pulsation frequency of the fluid-structure interface, typically satisfying... + =1.

[0088] When a shock wave is detected sweeping over a large area, this method automatically increases... This strengthens the convergence constraints of the flow field. The technical purpose of this formula is to provide a logical basis for the allocation of computational resources by quantifying the convergence state of the two fields.

[0089] Based on the continuity requirement of simulation calculations, S602, this method performs predictive scaling and self-healing of the time step. If the joint convergence factor does not reach the target threshold within a preset maximum number of iterations (in this embodiment, this threshold is preferably set to 10),... -4 If the current step size fails to capture drastic physical transients, this method will call the step size scaling operator to reduce the current macroscopic time step size that has failed to converge, according to the golden ratio.

[0090] As a preferred approach, this method introduces a rollback mechanism, reverting the state to the previous successfully converged time step and reloading the boundary conditions. This predictive scaling ensures that the simulation frequency automatically matches the evolution rate when physical quantities undergo abrupt changes, thereby maintaining the self-consistency of the numerical computation.

[0091] To address the storage bottleneck caused by tens of millions of grid nodes in multiphysics simulations, this method (S603) constructs a data sequence integration and storage architecture based on high-performance parallel files. After each physics step converges and is successfully submitted, this method enters the data integration and output phase.

[0092] This method utilizes asynchronous I / O streaming technology to allocate a dedicated mirror buffer in memory. Whenever an atomic transaction completes a state update, the data management module packages and encapsulates core physical quantities such as flow field pressure distribution, spatial velocity vector, structural temperature gradient, and interface displacement. To ensure time alignment of multi-source data, this method adds a global high-precision timestamp to each data packet, ensuring that the transient characteristics of the fluid-structure interaction interface are perfectly matched in time.

[0093] Considering that the directly output raw data contains a large amount of redundancy and is not conducive to direct comparison with experimental measurement points, S604 executes the integrity verification and formatted output of the method's state. This method performs the final data feature extraction, extracting key features from the high-dimensional space using principal component analysis operators. The logic for generating the data sequence is as follows: ; In the formula, This is a low-dimensional structured data sequence generated after performing multi-source field mapping; This is the state mapping matrix generated based on principal component extraction; The original computational domain physical feature vector is fed back by the fluid solver; The physical feature vector of the solid component is fed back by the solid finite element solver; To extract coupled physical feature vectors at the fluid-structure mapping interface; The tensor transpose operator is used to encapsulate multi-domain non-homogeneous data. A data position calibration compensation vector is established for hardware sensing measurement bias.

[0094] In this embodiment, if a rank deficiency occurs during the calculation of the mapping matrix, this method uses singular value decomposition to replace the standard inversion operation to eliminate the numerical bias caused by matrix singularity. The technical purpose of this step is to condense large-scale field distribution data into a time series that can characterize the evolution of the core performance of this method, thereby improving the efficiency of subsequent experimental comparison and analysis.

[0095] Based on the aforementioned adaptive iterative evolution and data integration output, this method achieves closed-loop management from dynamic numerical evolution to structured data storage. This feedback control and feature mapping technology based on residual trajectories enables the method provided by this invention to meet the technological requirements of long-duration stability simulation in engine wind tunnel testing. By verifying the integrity of the output sequence, it provides reliable data source support for engine performance evaluation.

[0096] To fully demonstrate the feasibility, efficiency, and accuracy of the wind tunnel flow field multiphysics coupling simulation system for engine testing provided by this invention in practical engineering applications, the following detailed explanation of the system evolution process and final comparison results is based on a specific application scenario of a wind tunnel transient test of a certain type of supersonic ramjet engine at Mach 3.0.

[0097] Execution flow of specific application examples: In a hypersonic wind tunnel with a total pressure of 1.5 MPa and an incoming Mach number of 3.0, a ramjet engine was subjected to long-duration wind blowing. During the mid-run (with a set physical time of 0.25 seconds), an external high-frequency command for emergency throttling and secondary ignition was suddenly applied. Due to the accompanying violent shock wave movement, boundary layer separation, and high-temperature thermoelastic warping of structural components, this operation is a typical problem involving strong nonlinearity and multiple time scales.

[0098] The specific execution flow of this invention is as follows: 1. In the preprocessing stage, the wind tunnel test section and the engine body are geometrically discretized. The fluid computational domain (covering the air intake, combustion chamber and exhaust nozzle) is divided into approximately 15 million unstructured tetrahedral meshes to accurately capture the complex shock wave system; the solid computational domain of the engine casing and supporting components is divided into approximately 2 million high-order hexahedral meshes.

[0099] After completing the mesh initialization, under the stable blowing and cold working condition of Mach number 3.0, the global pressure field, temperature field and the initial stress state of the structure are packaged into the 0th simulation transaction snapshot and stored in the distributed cache for later use.

[0100] 2. As the flow field advances to the inner channel section, shock waves begin to converge at the inlet lip. The feature extraction module monitors the pressure and temperature gradients across the entire field in real time. At this point, the local activity index of the nozzle convergence section and the inlet lip region surges beyond a preset threshold, immediately marking these regions as active sub-domains and assigning them microsecond-level micro-time steps; while the flat airflow region of the wind tunnel's outer bypass duct is classified as an insensitive sub-domain, executing millisecond-level macro-time steps.

[0101] 3. When the physical evolution approaches 0.25 seconds, a strong disturbance command is injected into the external control bus, causing the throttle valve to open fully instantaneously and triggering secondary ignition. Since the rate of change of this control signal within a single macroscopic time step exceeds the logic arbitration threshold of this invention, if the current asynchronous macroscopic step size is continued for calculation, it will cause non-physical divergence in the expansion segment.

[0102] At this point, the self-healing mechanism is activated, immediately halting and destroying any uncommitted computational memory, and precisely retrieving the last calibrated simulation snapshot from 0.25 seconds prior. After completing the state rollback, it forcibly switches the step size of all regions to the microscopic level for recalculation, perfectly capturing the strong expansion wave generated at the moment of ignition.

[0103] 4. Under the intense thermal flux impact after ignition, the computational steps of the fluid and solid still overlap. To provide real-time boundary conditions for the solid simulation engine, the built-in deep residual network receives features such as spatial coordinates, transient Mach number, and boundary layer thickness at the fluid-solid interface, and quickly outputs the predicted value of the thermal shock load after nonlinear fitting. This predicted load enters the annular buffer smoothing pool, and is weighted and fused with the accurate value fed back by the solver using a relaxation factor, successfully eliminating the step oscillation of interface pressure caused by independent solving of heterogeneous domains.

[0104] 5. Under long-term thermal aerodynamic loads, the thin-walled structure of the engine exhaust nozzle undergoes thermoelastic deformation, leading to spatial misalignment of the fluid-solid interface mesh. An octree-based spatial bounding box search is employed to re-identify the mapping target point. At this point, an energy residual comparison factor of 0.8% due to discrete interpolation is detected (i.e., there is a physical leakage between fluid work and solid absorption work).

[0105] The conservation constraint operator is immediately triggered, performing inverse algebraic compensation scaling on the pressure increment mapped to the structural cell, forcibly smoothing the fluid-structure virtual work difference within the macroscopic step size to near zero. This mechanism eliminates the total energy dissipation or explosion that is very likely to occur in long-term asynchronous simulations.

[0106] 6. After surviving the ignition mutation period, the joint convergence factor decreases, and the macroscopic time step calculation is dynamically restored. Finally, for the tens of millions of grid node data, principal component analysis is used to perform feature dimensionality reduction, encapsulating the core node features under stress into a time-series structure array containing pressure, dynamic strain, and temperature, accurately outputting and ending the entire simulation cycle.

[0107] To quantitatively demonstrate the technical superiority of this invention, the simulation system provided by this invention is compared and verified with existing technologies. The comparison scheme is set as follows: Method A (traditional strongly coupled scheme): Represents the industry's highest precision benchmark. It uses extremely small global micro time steps for strongly synchronous iterative calculations across the entire domain, without asynchronous space partitioning.

[0108] Method B (traditional loosely coupled asynchronous scheme): adopts asynchronous solution with a fixed proportion of the region, but lacks transaction rollback, AI load prediction and energy conservation forced closed-loop correction mechanism.

[0109] Method C (the present invention): A novel system solution that fully implements dynamic subdomain asynchronous scheduling, snapshot-triggered rollback, AI-enhanced projection, and energy closed-loop correction.

[0110] By simulating a real physical experiment for 0.5 seconds under the same cluster computing power, and taking the average value of multiple tests, the following comprehensive evaluation data table of system simulation performance was obtained (Table 1): Table 1: Comprehensive Comparison of Performance Indicators of Different Multiphysics Coupling Simulation Methods

[0111] Based on Table 1 and the attached figures, the verification results are analyzed in detail as follows: 1. Please refer to the appendix. Figure 5 Time consumption analysis shows that Method A, due to its globally minimal step size, generates a massive amount of invalid computation, resulting in a time consumption of up to 115.5 hours. This invention (Method C), by accurately determining the local activity index, locks 80% of the computing power in the shock wave and boundary layer core region, with a total time consumption of only 28.4 hours. Compared to Method A, the computational efficiency is improved by 75.4%, even surpassing Method B, which simply pursues speed.

[0112] 2. Please refer to the appendix. Figure 6 In the 0.2-0.3 second shock wave sweeping period of intense deformation, Method B suffers from algebraic losses due to numerous mismatched mesh interpolation, causing the energy residual alignment factor to diverge to -4.50%, resulting in numerical chatter caused by the absorption of spurious energy in the solid domain. In contrast, this invention (Method C) performs rigorous energy closed-loop correction, ensuring that its residual polygonal line always closely wraps around the baseline, and the cumulative energy error over the entire time period is strictly controlled within an extremely narrow tolerance band of 0.02%, eliminating the non-physical divergence caused by high-frequency fluid-structure oscillations.

[0113] 3. Please refer to the appendix. Figure 7 By comparing simulation and actual wind tunnel sensor data, it was found that when the control command abruptly changed at 0.25 seconds to trigger the shock wave, Method B completely missed the microsecond-level transient thermal stress spikes due to its long step size (fit accuracy was only 82.0%). This invention (Method C), however, thanks to its snapshot rollback mechanism and AI state prediction, accurately captured the microsecond-level shock ripples at the moment of command abrupt change. Its output timing characteristics showed a fit of up to 98.6% with the actual target test data.

[0114] The above embodiments and experimental data fully demonstrate that the present invention achieves near-equivalent absolute simulation accuracy under harsh conditions where the time consumption is only about one-quarter of that of the traditional benchmark scheme. It overcomes the engineering pain points of existing technologies where ensuring accuracy requires excessive time, while increasing speed easily leads to divergence and loss of high-frequency characteristics, providing highly reliable technical support for digital twin testing of aero-engines in complex operating environments.

Claims

1. A multiphysics coupling simulation method for wind tunnel flow fields in engine testing, characterized in that, Includes the following steps: S1. Construct the fluid computational domain and the solid computational domain, and initialize the simulation transaction snapshot sequence; S2. Calculate the local activity index of the fluid computation domain and divide it into an active subdomain that executes micro-time steps and an insensitive subdomain that executes macro-time steps. S3. Discretize the simulation time axis into atomic transactions with the macroscopic time step as the period, listen for external instructions to call the simulation transaction snapshot sequence to trigger rollback, and perform step realignment to the microscopic time step; S4. During the atomic transaction, cross-field state projection is performed using the historical state of the active subdomain within the microscopic time step to predict the physical quantity load of the coupling interface and perform smooth correction. S5. Based on the structural displacement generated by the solid computation domain, update the interface mapping interpolation matrix, calculate the fluid-structure energy residual to perform algebraic compensation on the physical quantity load of the coupled interface, thereby driving the evolution of the physical field. S6. Repeat steps S2 to S5 until the simulation ends, and output the multiphysics coupling simulation data sequence.

2. The multiphysics coupling simulation method for wind tunnel flow fields for engine testing according to claim 1, characterized in that, Step S2 specifically includes: The pressure gradient, temperature gradient, and velocity curl of each physical unit in the fluid computational domain are calculated, and the local activity index is obtained by weighted synthesis after normalization of the above parameters. When the local activity index is greater than the preset partitioning threshold, the corresponding physical unit is assigned to the active subdomain and the microscopic time step is allocated. When the local activity index is less than or equal to the preset partitioning threshold, the corresponding physical unit is assigned to the insensitive subdomain and the amplified macroscopic time step is allocated.

3. The multiphysics coupling simulation method for wind tunnel flow fields for engine testing according to claim 1, characterized in that, Step S3 specifically includes: In each of the atomic transactions, the external instructions are sampled and filtered to calculate their signal fluctuation factor within the current macroscopic time step. When the signal fluctuation factor exceeds the preset arbitration threshold, the currently executing atomic transaction is interrupted and intermediate variables are cleared. Based on the index pointer recorded in the simulation transaction snapshot sequence, the physical state is rolled back to the start time of the atomic transaction, and each subdomain is forced to uniformly use the micro time step to perform the step realignment calculation.

4. The multiphysics coupling simulation method for wind tunnel flow fields for engine testing according to claim 1, characterized in that, The process of performing cross-field state projection and predicting the physical loads at the coupling interface in step S4 includes: The historical transient features of the micro-time step are extracted using a second-order Taylor expansion, and the preliminary predicted value of the interface load is calculated. The spatial coordinate vector of the fluid-structure interaction interface, the local Mach number, the dimensionless thickness of the boundary layer, and the surface temperature are used as input features and fed into a pre-trained deep residual network to output a nonlinear correction. The nonlinear correction is superimposed on the preliminary predicted value of the interface load to generate the predicted physical load of the coupled interface.

5. The multiphysics coupling simulation method for wind tunnel flow fields for engine testing according to claim 4, characterized in that, The smoothing correction process in step S4 includes: A circular buffer smoothing pool is constructed in memory to align and cache the predicted physical loads of the coupling interface with the measured loads fed back in real time from the fluid solution according to timestamps. The relaxation factor is determined based on the variance dynamics of the residual fluctuation; The predicted physical load of the coupling interface is weighted and fused with the measured load using the relaxation factor to generate the corrected physical load of the coupling interface.

6. The multiphysics coupling simulation method for wind tunnel flow fields for engine testing according to claim 1, characterized in that, The process of updating the interface mapping interpolation matrix based on the structural displacement generated by the solid computational domain in step S5 includes: A dynamic octree spatial index structure is constructed to track the displacement of the structure, and a distance threshold filtering algorithm is introduced to match the fluid computing domain grid nodes and solid computing domain structural unit nodes with topological associations. Based on the spatially scattered point distribution characteristics of the matching nodes, a local topological interpolation relationship is constructed using a radial basis function with a regularized diagonal term to generate the updated interface mapping interpolation matrix.

7. The multiphysics coupling simulation method for wind tunnel flow fields for engine testing according to claim 1, characterized in that, The process of calculating the fluid-structure energy residual in step S5 to perform algebraic compensation for the physical quantity loads at the coupling interface includes: The first time integral value of the work done by the flow field pressure at the fluid-side interface and the second time integral value of the work done by the structural node response at the structural-side interface are calculated within the current atomic transaction cycle. Calculate the difference between the first time integral value and the second time integral value, and divide the difference by the first time integral value with a correction deviation constant to obtain the energy residual comparison factor as the fluid-structure energy residual; If the energy residual comparison factor is greater than the preset deviation threshold, the energy conservation constraint operator is activated, and the amplitude of the physical quantity load applied to the coupling interface of the solid computation domain is algebraically compensated through scaling and gradient compensation.

8. The multiphysics coupling simulation method for wind tunnel flow fields for engine testing according to claim 1, characterized in that, The iterative convergence determination process of repeating steps S2 to S5 in step S6 includes: Monitor the rate of decrease of the reference norm of the hydrodynamic field residuals and the rate of decrease of the reference norm of the solid structure field residuals in atomic transaction calculations; The joint convergence factor is obtained by dynamically assigning weights to the two decreasing ratios based on the load pulsation frequency of the fluid-structure interaction interface. The iteration depth within the current macroscopic time step is dynamically adjusted based on the joint convergence factor.

9. The multiphysics coupling simulation method for wind tunnel flow fields for engine testing according to claim 8, characterized in that, The iterative convergence determination process of repeating steps S2 to S5 in step S6 also includes a step of predictive scaling of the step size based on the joint convergence factor: When the joint convergence factor fails to reach the target threshold within the preset maximum number of iterations, the current macroscopic time step is reduced by a preset ratio. Using the physical state of the previous convergence time step as a baseline, the boundary conditions are reloaded to perform a trial calculation of the physical field evolution.

10. The multiphysics coupling simulation method for wind tunnel flow fields for engine testing according to claim 1, characterized in that, The process of outputting the multiphysics coupling simulation data sequence in step S6 includes: After the atomic transaction completes the calculation and is submitted, the high-dimensional physical quantity data in the fluid computing domain and the solid computing domain are extracted to a dedicated mirror buffer; Principal component analysis operators are used to reduce the dimensionality of key features in the extracted high-dimensional physical quantity data, generating a low-dimensional structured data array. A global timestamp is applied to the low-dimensional structured data array to complete the time sequence alignment, and the multiphysics coupling simulation data sequence is generated and output.

Images