Safety valve aerodynamic noise reduction method based on acoustic numerical simulation and proxy model optimization

Abstract

The invention discloses a safety valve aerodynamic noise reduction method based on acoustic numerical simulation and proxy model optimization. Belongs to the technical field of pneumatic noise control and structure optimization design of pressure system safety valves. An existing safety valve is prone to generating strong aerodynamic noise under the working condition of high-pressure relief, and a traditional analysis or experiment method is difficult to achieve accurate prediction and efficient optimization of noise under the condition of strong turbulence and multi-structure parameter coupling. A safety valve is used as a research object, a fluid calculation model under an opening working condition is established, and noise response data is obtained through aeroacoustics numerical simulation; selecting key geometric parameters of the valve port to construct a parameter design space, generating sample data by adopting a test design method, and establishing a noise prediction agent model; on the basis, an optimization algorithm is introduced to optimize structural parameters, and the noise reduction effect is verified through numerical simulation again. The method is suitable for aerodynamic noise prediction analysis and structural noise reduction optimization design of pressure relief components such as safety valves in nuclear power and other high-pressure gas pressure systems.

Description

Technical Field

[0001] This paper belongs to the field of structural optimization design for noise reduction of nuclear power safety valves, and involves a structural optimization design method for nuclear power safety valves based on a surrogate model. Background Technology

[0002] In nuclear power plant pressure systems, spring-loaded safety valves are widely used for overpressure protection due to their compact structure, sensitive response, and high reliability. When the system medium pressure reaches a set threshold, high-pressure gas is released at high speed through the valve port, forming a strong turbulent shear layer and complex vortex structure near the valve port, thus generating significant aerodynamic noise. This noise not only easily causes acoustic vibration and acoustic fatigue problems in the piping system, but may also have adverse effects on equipment safety and the health of operating personnel. Therefore, the prediction and control of safety valve pressure relief noise has always been an important research direction in pressure system design and optimization.

[0003] In existing technologies, research on noise issues in safety valves and similar valves mainly focuses on analytical methods, experimental methods, and numerical simulation methods. Some studies theoretically estimate valve noise by establishing simplified flow and acoustic analytical models. However, these methods typically rely on idealized assumptions and struggle to accurately reflect the complex noise characteristics caused by strong turbulence and multi-scale structural coupling under actual operating conditions. Other studies employ experimental testing methods, measuring noise under different operating conditions and structural parameters by building noise test benches. While this method can obtain relatively realistic data, it is costly and time-consuming under extreme operating conditions such as high pressure and high flow velocity in nuclear power plant safety valves, and it is difficult to systematically implement in a multi-parameter space, limiting its engineering applications. In recent years, with the development of computational fluid dynamics and aeroacoustics, research on safety valve noise based on numerical simulation has gradually increased. Predicting flow fields and noise using turbulence models and acoustic equations provides new means for valve noise analysis. However, directly using high-fidelity numerical models for multi-parameter structural optimization results in enormous computational demands, making it difficult to meet the needs of rapid design and engineering iteration.

[0004] Furthermore, existing research largely focuses on improving single structural parameters or local structures, lacking systematic modeling of noise variations under the combined effects of multiple key geometric parameters. Simultaneously, the contribution of different parameters to noise often relies on empirical judgment, lacking interpretable and quantifiable analytical methods, resulting in insufficient targeting and reliability of optimization designs. Therefore, although existing research has made some progress in noise prediction or local noise reduction, it remains difficult to achieve efficient prediction of safety valve noise, identification of key parameters, and rapid and reliable structural optimization under conditions of strong turbulence and multi-parameter coupling.

[0005] In summary, existing technologies suffer from several drawbacks, including difficulty in accurately predicting aerodynamic noise of safety valves under conditions of strong turbulence and coupling of multiple structural parameters, high costs of experiments and high-fidelity simulations, difficulty in quantifying and identifying the noise contribution of key parameters, and a lack of efficient and systematic multi-parameter noise reduction optimization methods. Summary of the Invention

[0006] To address the shortcomings of existing technologies, such as the difficulty in accurately predicting aerodynamic noise of safety valves under conditions of strong turbulence and multi-structural parameter coupling, high costs of experiments and high-fidelity simulations, difficulty in quantifying and identifying the noise contribution of key parameters, and the lack of efficient and systematic multi-parameter noise reduction optimization methods, the technical solution provided by this invention is as follows: A method for reducing pneumatic noise in safety valves based on acoustic numerical simulation and surrogate model optimization includes: The steps are as follows: obtain the structural and operating parameters of the safety valve under the open condition, establish a fluid calculation model of the safety valve and set the inlet and outlet boundary conditions, and output the safety valve flow model for numerical analysis. The steps include meshing the safety valve flow model, verifying mesh independence, determining the computational mesh that meets the computational accuracy requirements, and outputting the numerical discrete model for simulation. The steps are as follows: Perform aeroacoustic numerical simulation based on a numerical discrete model to obtain the unsteady flow field inside the safety valve and the corresponding far-field noise response, and output the sound pressure level data at each monitoring location. The steps include: setting up an acoustic monitoring surface and multiple monitoring points downstream of the safety valve outlet to collect and analyze noise response data and output noise characterization data for optimization. The steps include selecting preset geometric parameters of the safety valve orifice as design variables, setting their value range, constructing a parameter design space, and outputting parameter space information for experimental design. The steps involve generating multiple sets of parameter samples based on parameter space information using experimental design methods, obtaining corresponding noise characterization data for each parameter sample, forming a sample dataset, and outputting training data for modeling. The steps are as follows: constructing a surrogate model for predicting safety valve noise based on the sample dataset, selecting the surrogate model with the best prediction accuracy, and outputting the noise response mapping relationship. Based on the surrogate model, the steps involve optimizing key geometric parameters to obtain the optimal combination of structural parameters that meets the noise reduction target, and then outputting the optimized safety valve structural parameters.

[0007] Furthermore, a preferred implementation method is provided, in which the aeroacoustic numerical simulation uses the large eddy simulation method to solve the unsteady turbulent flow field inside the safety valve, and converts the flow field information into a far-field noise response based on aeroacoustic theory.

[0008] Furthermore, a preferred embodiment is provided in which the acoustic monitoring surface is set at a predetermined distance downstream of the safety valve outlet, and the monitoring points are evenly distributed along the circumference or semicircle.

[0009] Furthermore, a preferred embodiment is provided, wherein key geometric parameters include the outer diameter of the valve disc, the nozzle diameter, and the edge depth of the anti-impact disc.

[0010] Furthermore, a preferred embodiment is provided, wherein the experimental design method is the Latin hypercube experimental design method, which is used to generate uniformly distributed parameter samples within the parameter design space.

[0011] Furthermore, a preferred implementation method is provided, wherein the surrogate model is a noise prediction model constructed based on the sample dataset, and the optimal surrogate model is obtained by screening through prediction accuracy evaluation.

[0012] Based on the same inventive concept, this invention also provides a safety valve pneumatic noise reduction device based on acoustic numerical simulation and surrogate model optimization, comprising: The module obtains the structural and operating parameters of the safety valve under the open condition, establishes the fluid calculation model of the safety valve and sets the inlet and outlet boundary conditions, and outputs the safety valve flow model for numerical analysis. Mesh the safety valve flow model and verify its independence. Determine the computational mesh that meets the computational accuracy requirements and output the module of the numerical discrete model for simulation. A module that performs aeroacoustic numerical simulation based on a numerical discrete model to obtain the unsteady flow field inside the safety valve and the corresponding far-field noise response, and outputs the sound pressure level data at each monitoring location. An acoustic monitoring surface is arranged downstream of the safety valve outlet, and multiple monitoring points are set up to collect and analyze noise response and output noise characterization data for optimization. A module that selects the preset geometric parameters of the safety valve orifice as design variables and sets their value range, constructs the parameter design space, and outputs the parameter space information for experimental design. Based on parameter space information, an experimental design method is used to generate multiple sets of parameter samples, and corresponding noise characterization data is obtained for each parameter sample to form a sample dataset. A module that outputs training data for modeling is then output. A proxy model for predicting safety valve noise is constructed based on the sample dataset, and the proxy model with the best prediction accuracy is selected. The module that outputs the noise response mapping relationship is then used. Based on the surrogate model, the module optimizes and calculates key geometric parameters to obtain the optimal combination of structural parameters that meets the noise reduction target, and outputs the optimized safety valve structural parameters.

[0013] Based on the same inventive concept, the present invention also provides a computer storage medium for storing a computer program, wherein when the computer program is read by a computer, the computer executes the method described thereon.

[0014] Based on the same inventive concept, the present invention also provides a computer, including a processor and a storage medium, wherein when the processor reads a computer program stored in the storage medium, the computer executes the method described thereon.

[0015] Based on the same inventive concept, the present invention also provides a computer program product, which, when executed, implements the method described.

[0016] Compared with the prior art, the advantages of the technical solution provided by the present invention are as follows: This approach employs aeroacoustic numerical simulation based on the coupling of Large Eddy Simulation (LES) and the FW-H equations. This allows for the simultaneous characterization of transient flow structures and far-field noise radiation near the valve orifice under conditions of high-pressure venting and strong turbulence in safety valves. This capability stems from the combined application of LES turbulence modeling and FW-H acoustic solutions. Compared to common empirical formulas or simplified analytical models in existing technologies, this method avoids overly idealized assumptions about the flow field and sound sources, keeping noise prediction results within a small error range compared to experimental measurements, thus significantly improving the physical reliability of noise predictions.

[0017] By rationally arranging acoustic monitoring points and simultaneously introducing the characteristics of internal and far-field monitoring during the numerical simulation stage, the scheme can not only obtain the overall sound pressure level but also reflect the spatial directivity distribution characteristics of noise. This effect stems from the configuration of the monitoring point array and acoustic receiving surface. Unlike existing studies that only measure noise at a single location or limited angle, this feature makes the noise assessment more comprehensive and provides a reliable basis for the rational definition of subsequent optimization objectives.

[0018] A Latin hypercube experimental design is employed to construct a characteristic of the design sample space, enabling key structural parameters to achieve uniform coverage of the entire design space with a limited number of samples. This effect stems from the application of the LHS sampling strategy. Compared to traditional random sampling or single-parameter scanning methods, this feature effectively avoids sample clustering and omission issues, significantly reducing the number of simulations required while maintaining the accuracy of the surrogate model, thereby improving the overall modeling and optimization efficiency.

[0019] By introducing a surrogate model to establish the mapping relationship between key geometric parameters and noise response, this approach enables noise prediction to replace high-fidelity numerical simulation with low computational cost. This effectiveness stems from the construction and comparison of various surrogate models, with support vector regression being the preferred final prediction model. Compared to existing research that relies on repeated computation via CAA, this feature makes multi-parameter, multi-round optimization feasible in engineering applications.

[0020] By employing a multi-model accuracy evaluation and selection process to identify the optimal surrogate model, noise prediction goes beyond simply being "usable." It is quantitatively validated using metrics such as the coefficient of determination and mean squared error. This effectiveness stems from the systematic evaluation of the surrogate model's prediction accuracy within the methodology. Unlike some studies that rely solely on experience to select models, this approach ensures high reliability of the surrogate model both globally and locally, thus providing a stable computational foundation for subsequent optimization.

[0021] The introduction of SHAP interpretability analysis to quantitatively assess the noise contribution of structural parameters enables the scheme to clearly identify the degree of noise influence of different geometric parameters in different spatial locations and directions. This effect stems from the application of SHAP value analysis in the method. Compared with the existing approach that relies on experience or qualitative judgment of key parameters, this feature provides an interpretable and quantifiable basis for structural optimization, avoiding blind parameter adjustments.

[0022] The use of heuristic optimization algorithms and surrogate models as fitness functions for parameter optimization enables the scheme to achieve globally or near-globally optimal noise reduction structure design under multiple constraints. This effect stems from the application of the Grey Wolf optimization algorithm in the approach. Compared with traditional optimization methods based on local search or manual experience adjustment, this feature significantly improves optimization efficiency and result stability.

[0023] By verifying the optimization results through aeroacoustic simulation, the noise reduction effect is not merely a model prediction but is directly confirmed by high-fidelity simulation. This effect stems from the comparative analysis of the flow field and noise contour maps before and after optimization. Compared to some studies that lack a closed-loop verification mechanism, this feature ensures the effectiveness of the optimization scheme at the actual physical level.

[0024] Through the synergistic effect of the above features, this scheme achieves a significant noise reduction effect while ensuring the basic flow performance and structural rationality of the safety valve. This effect is a result of the combined effect of multiple steps, including numerical simulation, surrogate modeling, parameter interpretation and optimization decision-making, which distinguishes it from existing research paths that rely on single prediction or local optimization.

[0025] It is applicable to the aerodynamic noise prediction analysis and structural noise reduction optimization design of pressure relief components such as safety valves in nuclear power and other high-pressure gas pressure systems. Attached Figure Description

[0026] Figure 1 The computational workflow of numerical simulation; Figure 2 Verification of computational grid independence; Figure 3 Experimental test bench for measuring the noise of safety valves; Figure 4 Software platform for noise characterization testing of safety valves; Figure 5 Verification of the repeatability of experimental measurements; Figure 6 Comparison of experimental flow rate and simulated flow rate; Figure 7 Comparison of experimental and simulated noise results; Figure 8 Key parameters of the safety valve orifice; Figure 9 Distribution of sampling points in experimental design; Figure 10 The prediction accuracy of each proxy model; Figure 11 SHAP values ​​of key parameters at monitoring points along the axis; Figure 12 SHAP values ​​of key parameters corresponding to monitoring points on the acoustic directivity surface; Figure 13 Comparison of simulation cloud plots before and after optimization; Figure 14 Comparison of noise simulation results before and after optimization. Detailed Implementation

[0027] To make the advantages and benefits of the technical solution provided by the present invention clearer, the technical solution provided by the present invention will now be described in further detail with reference to the accompanying drawings, specifically: Implementation Method 1: This implementation method provides a safety valve pneumatic noise reduction method based on acoustic numerical simulation and surrogate model optimization, including: The steps are as follows: obtain the structural and operating parameters of the safety valve under the open condition, establish a fluid calculation model of the safety valve and set the inlet and outlet boundary conditions, and output the safety valve flow model for numerical analysis. The steps include meshing the safety valve flow model, verifying mesh independence, determining the computational mesh that meets the computational accuracy requirements, and outputting the numerical discrete model for simulation. The steps are as follows: Perform aeroacoustic numerical simulation based on a numerical discrete model to obtain the unsteady flow field inside the safety valve and the corresponding far-field noise response, and output the sound pressure level data at each monitoring location. The steps include: setting up an acoustic monitoring surface and multiple monitoring points downstream of the safety valve outlet to collect and analyze noise response data and output noise characterization data for optimization. The steps include selecting preset geometric parameters of the safety valve orifice as design variables, setting their value range, constructing a parameter design space, and outputting parameter space information for experimental design. The steps involve generating multiple sets of parameter samples based on parameter space information using experimental design methods, obtaining corresponding noise characterization data for each parameter sample, forming a sample dataset, and outputting training data for modeling. The steps are as follows: constructing a surrogate model for predicting safety valve noise based on the sample dataset, selecting the surrogate model with the best prediction accuracy, and outputting the noise response mapping relationship. Based on the surrogate model, the steps involve optimizing key geometric parameters to obtain the optimal combination of structural parameters that meets the noise reduction target, and then outputting the optimized safety valve structural parameters.

[0028] The aeroacoustic numerical simulation uses the large eddy simulation method to solve the unsteady turbulent flow field inside the safety valve, and converts the flow field information into a far-field noise response based on aeroacoustic theory.

[0029] The acoustic monitoring surface is set at a predetermined distance downstream of the safety valve outlet, and the monitoring points are evenly distributed along the circumference or semicircle.

[0030] Key geometric parameters include valve disc outer diameter, nozzle diameter, and anti-impact disc edge depth.

[0031] The experimental design method is the Latin hypercube experimental design method, which is used to generate uniformly distributed parameter samples within the parameter design space.

[0032] The surrogate model is a noise prediction model built based on the sample dataset, and the optimal surrogate model is selected by evaluating the prediction accuracy.

[0033] Based on the same inventive concept, this invention also provides a safety valve pneumatic noise reduction device based on acoustic numerical simulation and surrogate model optimization, comprising: The module obtains the structural and operating parameters of the safety valve under the open condition, establishes the fluid calculation model of the safety valve and sets the inlet and outlet boundary conditions, and outputs the safety valve flow model for numerical analysis. Mesh the safety valve flow model and verify its independence. Determine the computational mesh that meets the computational accuracy requirements and output the module of the numerical discrete model for simulation. A module that performs aeroacoustic numerical simulation based on a numerical discrete model to obtain the unsteady flow field inside the safety valve and the corresponding far-field noise response, and outputs the sound pressure level data at each monitoring location. An acoustic monitoring surface is arranged downstream of the safety valve outlet, and multiple monitoring points are set up to collect and analyze noise response and output noise characterization data for optimization. A module that selects the preset geometric parameters of the safety valve orifice as design variables and sets their value range, constructs the parameter design space, and outputs the parameter space information for experimental design. Based on parameter space information, an experimental design method is used to generate multiple sets of parameter samples, and corresponding noise characterization data is obtained for each parameter sample to form a sample dataset. A module that outputs training data for modeling is then output. A proxy model for predicting safety valve noise is constructed based on the sample dataset, and the proxy model with the best prediction accuracy is selected. The module that outputs the noise response mapping relationship is then used. Based on the surrogate model, the module optimizes and calculates key geometric parameters to obtain the optimal combination of structural parameters that meets the noise reduction target, and outputs the optimized safety valve structural parameters.

[0034] A computer storage medium is also provided for storing a computer program, which, when read by the computer, executes the method.

[0035] A computer is also provided, including a processor and a storage medium, wherein the computer executes the method when the processor reads a computer program stored in the storage medium.

[0036] A computer program product is also provided, which, when executed, implements the method described.

[0037] Implementation Method Two: This implementation method is a further detailed description of the technical solution provided in Implementation Method One, specifically: Modeling the safety valve and determining the noise analysis conditions are crucial steps in providing the foundational model and boundary conditions for subsequent numerical simulations and optimizations. Specifically, based on the actual structural form of the safety valve, a three-dimensional fluid computational model consistent with the real valve is established. Its axisymmetric characteristics are utilized to construct a semi-symmetric three-dimensional computational domain to reduce computational costs. Simultaneously, the fluid flow region inside the valve is extracted, and the outlet section is appropriately extended to ensure sufficient gas flow development within the computational domain. Building upon this, the inlet pressure, outlet pressure, and medium parameters under the open state of the safety valve are set to determine the pressure relief conditions, providing a unified input for subsequent flow and noise calculations.

[0038] The fluid computational domain is meshed and its independence is verified. This step is used to obtain a numerical discretization model that balances computational accuracy and efficiency. Specifically, a structured hexahedral mesh is used to discretize the fluid domain, with local refinement at the valve orifice and in regions with strong flow gradients. By progressively refining the mesh and comparing the valve outlet flow rate and acoustic power indices, the convergence of key physical quantities with varying mesh count is determined, thereby identifying the final mesh scheme that meets the accuracy requirements. This mesh model serves as the unified input for subsequent flow and acoustic simulations.

[0039] Aeroacoustic numerical simulations are conducted based on a defined grid model to obtain the flow characteristics and noise response of the safety valve under target operating conditions. Specifically, the large eddy simulation method is used to solve the unsteady turbulent flow field inside the safety valve. Large-scale vortices are calculated directly, while small-scale vortices are approximated using a subgrid-scale model. After obtaining stable transient flow data, aeroacoustic theory is introduced to calculate the flow-induced sound sources, and the sound source information is converted into far-field sound pressure response using an acoustic propagation model, thereby obtaining the time-domain and frequency-domain noise data for each monitoring location.

[0040] While completing the numerical simulation, acoustic monitoring points are strategically arranged to acquire noise characterization data. This step is used to form the target output required for subsequent optimization. Specifically, a circular or semi-circular acoustic receiving array is arranged at a specified distance downstream of the safety valve outlet, so that the monitoring points are evenly distributed in the circumference to comprehensively reflect the spatial directivity characteristics of the noise. At the same time, monitoring points are set up inside the valve or near the outlet to capture the acoustic characteristics of the noise source area. Finally, the sound pressure level data corresponding to each monitoring point is output as the basic indicator for noise evaluation and optimization.

[0041] After obtaining the noise results under a single structural condition, the key structural parameters of the safety valve are selected and a design space is constructed. This step is used to clarify the design variables for subsequent parameter optimization. Specifically, based on the structural characteristics of the safety valve, the outer diameter of the valve disc, the nozzle diameter, and the edge depth of the anti-impact disc are selected as key geometric parameters affecting noise, and reasonable value ranges are set for each parameter to form a multi-dimensional parameter design space, providing input for experimental design and sample generation.

[0042] The Latin hypercube experimental design method is used to generate multi-parameter sample points, which is used to cover the entire design space with a limited number of computations. Specifically, the value range of each structural parameter is divided into several equally probable intervals, and random combinations are performed within each interval to generate a uniformly distributed set of sample parameters. For each set of sample parameters, the aforementioned modeling, mesh generation, and aeroacoustic numerical simulation process are repeated to output the corresponding sound pressure level results, thereby forming a training dataset and a validation dataset containing input parameters and noise response.

[0043] A surrogate model for the noise response of a safety valve is constructed based on sample data. This step is used to replace high-fidelity numerical simulation with low computational cost. Specifically, various surrogate modeling methods are used to fit the mapping relationship between structural parameters and noise response. The prediction accuracy of different surrogate models is evaluated by error index. Finally, the surrogate model with the highest prediction accuracy and optimal stability is selected as the noise prediction model. The output of this model is used for subsequent parameter sensitivity analysis and optimization calculations.

[0044] After the surrogate model is established, a contribution analysis is performed on the key structural parameters. This step is used to clarify the degree of influence of each parameter on the noise. Specifically, interpretability analysis methods are used to analyze the noise prediction results at different monitoring points and in different spatial directions, quantify the contribution ratio of each structural parameter in noise formation, thereby identifying the key parameters with the greatest impact on noise and providing a basis for optimization.

[0045] Based on a clear understanding of the influence of parameters, a noise reduction optimization model for safety valves is constructed, and parameter optimization is performed. This step is used to obtain the optimal structural scheme that satisfies the constraints. Specifically, minimizing the sound pressure level at each monitoring point within the study area is taken as the optimization objective, and the range of structural parameter values ​​is taken as the constraint. The established surrogate model is used as the fitness function, and a heuristic optimization algorithm is introduced to perform a global search for parameter combinations, outputting the optimal or near-optimal combination of structural parameters that satisfies the noise reduction objective.

[0046] The optimized safety valve structure is then validated and analyzed to confirm the authenticity and reliability of the optimization effect. Specifically, a numerical model of the safety valve is re-established based on the optimized structural parameters, and the aeroacoustic numerical simulation process is repeated. The flow field characteristics, pressure distribution, and changes in sound pressure level at each monitoring point are compared before and after optimization. This verifies that the noise level is significantly reduced while ensuring the basic flow performance of the safety valve, thus completing the closed loop of the entire safety valve aerodynamic noise reduction method.

[0047] in, Figure 1This diagram illustrates the overall workflow of aeroacoustic numerical simulation and noise analysis for a safety valve. The left side shows the fluid flow path of the safety valve under open conditions. High-pressure gas enters the valve body through the inlet, is released at high speed through the valve port area, and then flows out along the outlet pipe. The length of the outlet pipe is set to a multiple of the pipe diameter to ensure sufficient flow development. The middle section illustrates the meshing process of the fluid computational domain. A denser hexahedral structured mesh is used for the valve port and near-wall region to improve the analytical accuracy for strong turbulence and shear layers. The right side shows the arrangement of the acoustic monitoring surfaces. Circular or semi-circular acoustic receiving surfaces are set at a certain distance downstream of the valve outlet. Monitoring points are evenly distributed along the circumference, and the spatial distances between each receiving point and the valve outlet and pipe wall are marked to obtain noise radiation information in different directions. The bottom section shows the flow field analysis and noise analysis results, including velocity contour maps, pressure contour maps, and a polar coordinate noise directivity distribution map, intuitively reflecting the correspondence between the flow structure and noise radiation characteristics during the safety valve depressurization process.

[0048] Figure 4 This is a schematic diagram of the structural components of a safety valve noise testing platform. On the left side of the diagram are the pressure tank and the safety valve body. High-pressure gas is stored in the pressure tank and enters the safety valve for pressure relief through connecting pipelines and valve devices. The safety valve outlet is connected to the piping system, on which pressure and flow sensors are arranged to collect inlet pressure and flow data in real time. Multiple sound pressure sensors are arranged above and around the perimeter of the diagram. These sensors are fixed to a semi-circular or arc-shaped frame by brackets, forming an acoustic receiving array used to measure the noise level of the safety valve in different directions. On the right side is the data acquisition and control system, including a data acquisition box, a host computer, and related interface devices. Various sensor signals are transmitted to the host computer through the acquisition system for synchronous recording and processing, thereby achieving unified acquisition and analysis of parameters such as noise, pressure, and flow, ensuring the integrity and repeatability of experimental data.

[0049] Figure 8 This diagram illustrates the key structural parameters and geometric definitions of a safety valve. The left side of the diagram shows a cross-sectional view of the overall safety valve structure, clearly illustrating the positional relationships of key components such as the valve disc, nozzle, and anti-impact disc within the valve body. The valve disc is located above the nozzle and controls the valve's opening and closing, while the anti-impact disc is positioned below the valve disc to improve flow conditions during high-speed discharge. The middle section provides a separate enlarged view of the valve disc and nozzle structures to highlight their geometry and assembly relationship. The right side shows the annotation of key geometric parameters, clearly defining the valve disc outer diameter D1, nozzle diameter D2, and anti-impact disc edge depth H1 using dimension lines. These parameters serve as design variables for noise optimization in this design; their values ​​directly affect the flow structure and turbulence intensity near the valve orifice, thus significantly influencing the aerodynamic noise characteristics of the safety valve.

[0050] Implementation Method 3, in conjunction with Appendix Figure 1-14 This embodiment describes the technical solution provided above in further detail through specific examples. Specifically: The purpose of this embodiment is to address the problems of excessive pressure relief noise, difficulty in prediction, and high optimization costs in safety valves. It proposes a rapid and accurate optimization method for safety valve noise optimization based on a surrogate model. To achieve the above objective, this embodiment adopts the following technical solution: A design method for noise reduction optimization of safety valves based on a surrogate model is presented, using safety valves as the research object. For the opening condition of the safety valve, finite element simulation analysis is used to reasonably set noise monitoring points to obtain the valve noise. Three structural parameters are selected as design variables for optimization based on the structural characteristics. Sufficient training sample points are selected using the Latin hypercube experimental design method. The sound pressure level at the monitoring points is used as the optimization objective, and the range of structural parameters is used as the constraint condition to establish a mapping relationship between key structural parameters and noise. Combined with a heuristic algorithm, the optimal solution is found, obtaining the minimum noise structure while ensuring the reasonableness of the safety valve structure. Finally, finite element simulation is performed on the optimized structure to evaluate the optimization results and verify the effectiveness of the proposed method.

[0051] For the aforementioned finite element simulation, large eddy simulation (LES) and the FW-H equations are mainly used for prediction. The LES employs spatial filtering techniques to separate large-scale and small-scale eddies; large-scale eddies are solved directly, while small-scale eddies are approximated using a subgrid-scale model (SGS). Its governing equations can be expressed as: (1) (2) (3) In the above equation: and Represents the subgrid-scale velocity components. t For time; Represents subgrid-scale pressure, where xi is the spatial coordinate. ρ For fluid density, v For kinematic viscosity, Let be the subgrid-scale turbulent stress. The subgrid-scale stress tensor characterizes the influence of unsolved small-scale motions on the solved large-scale flow field. The solved strain rate tensor is denoted as . The definition is as follows: (4) Subgrid-scale turbulent stress The relationship between the strain rate tensor and the solved strain rate tensor can be expressed as: (5) In the formula, Represents the Kronecker function. This is the sub-grid-scale turbulent viscosity. This turbulent viscosity term is typically modeled using the Smagorinsky–Lilly model, expressed as follows: (6) In the formula It is an empirical constant, and Indicates the filter width.

[0052] The aforementioned FW-H equation is essentially a non-homogeneous wave equation, which can be derived using the continuity equation and the Navier-Stokes equations. The FW-H equation can be expressed as: (7) This indicates the sound pressure fluctuation at the disturbed location; This represents the fluid density in an undisturbed field. It is the density of the disturbed flow field; c It is the reference speed for the speed of sound; f Represents the wall function. H ( f ) is the Heaviside function, specifically represented by equation (8); δ ( f ) represents the Dirac function, specifically expressed as equation (9); u ij Indicates speed; P ij They represent the stress tensor, respectively. T ij It is the Lighthill tensor, as shown in equation (10); in the FW-H equation, the left side describes the propagation of far-field waves, and the right side contains the sound source terms. The three terms on the right side of the equation are the monopole sound source terms, which are the sound sources caused by surface acceleration, the dipole sound sources caused by surface pulse pressure, and the quadrupole sound sources generated by fluid turbulence.

[0053] (8) (9) (10) in, This represents the Kronecker function.

[0054] The Latin Hypercube Design of Experiments (LHS) addresses the need for reliable and representative training data to construct high-fidelity surrogate models, ensuring the data uniformly covers the entire design space and enhancing the effectiveness of efficient experimental design. LHS is an advanced sampling technique applicable to multidimensional spaces, employing a stratified sampling strategy to achieve comprehensive and uniform coverage of the parameter domain. This method can systematically explore the global distribution characteristics of input variables and their sensitivity to the output response.

[0055] Unlike traditional random sampling, Latin hypercube sampling divides each parameter dimension into equally probable intervals and precisely selects one sample within each interval. This design ensures a uniform distribution of samples throughout the multidimensional space while effectively avoiding problems such as local sample clustering or insufficient sampling. Therefore, Latin hypercube sampling significantly improves the efficiency of parameter space exploration, and is particularly suitable for sensitivity analysis and surrogate model construction of high-dimensional, nonlinear systems. Based on these advantages, this study uses Latin hypercube sampling to sample the design space, laying a reliable data foundation for subsequent surrogate model construction and parameter optimization. The mathematical expression is as follows: Let... Given an input random vector, where each component is independent, the cumulative distribution function is: LHS generates N samples. The process is described as follows: (1) Generate two random matrices and All sizes ;in Each column yes A random arrangement. elements And all elements are independent.

[0056] (2) No. One sample in The values ​​in each dimension are: (11) in For the first The inverse cumulative distribution function of the variables.

[0057] It was determined which subinterval the point fell into; It is a random offset within this sub-interval For the surrogate model, since different physical phenomena often require customized mathematical models to achieve accurate approximation, this implementation method selects the surrogate model technology with the best adaptability for the noise prediction problem of safety valves. Five different surrogate models were selected and compared: Kriging model (KRG), radial basis function model (RBF), support vector regression model (SVR), multinomial response surface model (PRS), and artificial neural network model (ANN). The kernel functions and key configuration parameters used in each model are summarized in Table 1.

[0058]

[0059] The fitting performance of the surrogate model is evaluated using prediction accuracy metrics, which can be categorized into global accuracy metrics and local accuracy metrics based on the evaluation scope. Global accuracy metrics reflect the model's overall fitting ability across the entire design space, including mean squared error (MSE), root mean square error (RMSE), and coefficient of determination (R²). Smaller MSE and RMSE values ​​indicate higher global prediction accuracy and smaller errors. The coefficient of determination, ranging from [0,1], reflects the proportion of data variance explained by the model; a value closer to 1 indicates stronger explanatory power. In contrast, local accuracy metrics focus on prediction bias within specific regions. A commonly used metric is the maximum absolute percentage error (MAPE). A smaller MAPE value indicates higher predictive reliability in the local region. The mathematical expressions for these metrics are summarized in Table 2.

[0060]

[0061] The heuristic algorithm described is the Grey Wolf Algorithm (GWO).

[0062] Given the axisymmetric nature of the safety valve, this implementation employs a semi-symmetric three-dimensional (1 / 2 three-dimensional) computational model to reduce simulation costs and improve optimization efficiency. The detailed workflow of the numerical simulation is as follows: Figure 1 As shown.

[0063] The specific simulation procedure is as follows: (1) First, the fluid flow domain of the safety valve is extracted. To ensure sufficient flow development, the outlet section is extended to 5 times the pipe diameter. The inlet pressure is set to 0.1 MPa, the outlet pressure is defined as 0 MPa, and the medium is an ideal gas. (2) This domain is discretized using a structured hexahedral mesh. Mesh independence analysis was performed based on the maximum acoustic power level (APLvo) at the valve port and the mass flow rate at the outlet. The results are shown in Table 3 and Figure 2Finally, a grid size of 1 mm was selected for the valve orifice. (3) A circular acoustic receiver array with a radius of 1 meter was set up 1 meter downstream of the valve outlet. The monitoring points were evenly distributed along the perimeter, starting from the horizontal monitoring point 1 on the right and continuing clockwise at 10° intervals to fully capture the directivity of the radiated noise. In addition, in order to ensure signal stability and improve simulation reliability, a monitoring point was set at the nozzle outlet to track the internal noise source. (4) CAA simulation was performed on the safety valve. The simulation duration was set to 0.3 seconds and the time step was 1×10. -4 seconds, convergence residual is 1×10 -5 (5) Extract the acoustic signal and compare it with experimental measurements to evaluate the reliability of the noise prediction.

[0064]

[0065] To accurately measure the sound pressure performance of safety valves, a noise test bench was designed according to GB / T 17213. The overall experimental setup is as follows: Figure 3 As shown, a semi-circular acoustic array with a radius of 1 meter is used, centered 1 meter downstream of the valve outlet. Sound level sensors are installed at 30° intervals along the semicircle to capture noise emissions from the safety valve from multiple directions.

[0066] To facilitate data acquisition and analysis, a dedicated safety valve data acquisition system was developed based on the LabVIEW platform, such as... Figure 4 As shown in the figure. This system supports real-time display and recording of key parameters such as noise, pressure, and flow rate during the test. The specifications of the sensors, data acquisition cards, and other key equipment used in the experiment are summarized in Table 4.

[0067]

[0068] To evaluate the repeatability of the experiment, three independent tests were conducted under the same operating conditions. The results show that the flow rate and noise measurements obtained from the test bench exhibited small errors in multiple tests under the same inlet pressure, confirming the reliability of the experimental setup in characterizing the safety valve's performance. Relevant data are as follows: Figure 5 As shown.

[0069] Based on the flow field structure of the test rig, the model was reconstructed and simulation calculations were performed. The experimental flow rates and simulation results were then compared and analyzed. Researchers analyzed the simulated flow field and compared the calculated mass flow rates with experimental data to verify the accuracy of the simulation results. The flow rate comparison results are as follows: Figure 6 As shown in the figure, the comparison between the simulated flow data and the experimentally measured data shows that the maximum error between the two is less than 3%. This result confirms that the turbulence model based on Large Eddy Simulation (LES) has high reliability in the flow field simulation of slide valve pressure reducing valves (safety valves).

[0070] To improve the clarity of the experimental data, this implementation method performed symmetry processing on the noise data collected by the semi-circular monitoring point array, and then compared the processed data with the simulation results. The comparison of the noise results between the experiment and the simulation is as follows: Figure 7 As shown. Analysis of the simulated noise data indicates that the acoustic directivity of the slide valve-type pressure reducing valve (safety valve) exhibits significant dipole characteristics. At the 90° and 270° monitoring positions, the sound pressure level (SPL) reaches its minimum, and both values ​​are below the industrial noise limit of 85 dB. Since this SPL level meets the relevant specifications, it is not discussed further in this embodiment. For regions outside the angle ranges [80°, 100°] and [260°, 280°], the maximum deviation between the simulation results and experimental data is 3.678 dB, corresponding to a relative error of 3.96%. These results confirm that the aeroacoustic (CAA) method has high predictive reliability for the noise radiation characteristics of safety valves.

[0071] The core idea of ​​surrogate models is to approximate complex physical or engineering phenomena through simplified mathematical representations, thereby significantly improving computational efficiency. Constructing such models requires clearly defining the key input parameters and the target physical quantity to be predicted. For slide valve pressure reducing valves (safety valves), the main source of noise generation is the turbulence intensity near the valve orifice, and the key parameters of the valve orifice significantly affect the sound pressure level (SPL) at a specific monitoring point. Therefore, this implementation aims to establish a mapping relationship between the key geometric parameters of the valve orifice and the sound pressure level at a specified location. The key parameters considered in the study are as follows: Figure 8 As shown: D 1 represents the outer diameter of the valve disc; D 2 represents the valve nozzle diameter; H 1 represents the edge depth of the shock-absorbing disc.

[0072] The value ranges of key parameters are shown in Table 5. A total of 21 sample points and their corresponding experimental results were selected to form the training dataset for the surrogate model; additionally, 7 independent sample points were generated as the validation dataset. The spatial distribution of the training and validation samples is shown in Table 5. Figure 9 As shown, the samples uniformly cover the entire design space, providing reliable data support for achieving high-precision model construction and robustness verification.

[0073]

[0074] Based on sample data obtained through Latin hypercube sampling (LHS), this implementation method generates high-fidelity noise simulation data through aeroacoustic (CAA) simulation calculations. Subsequently, these simulation results are used to train various surrogate models. The quantitative comparison results of the accuracy of each surrogate model are as follows: Figure 10As shown: The prediction results of the Kriging model (KRG) are shown below. Figure 10 (a) The results of the polynomial response surface model (PRS) are shown in [reference needed]. Figure 10 (b) The results of the radial basis function (RBF) model are shown in Figure 10 (c) The results of the Support Vector Regression (SVR) model are shown in [reference needed]. Figure 10 (d) The results of the Artificial Neural Network (ANN) model are shown in [the original text]. Figure 10 (e).

[0075] The prediction accuracy of each surrogate model is summarized in Figure 10 (f) This implementation method employs multiple metrics for validation to evaluate the performance of different models. Among these, the coefficient of determination (R²) of the Support Vector Regression (SVR) model is used. 2 The model achieved a maximum value of 0.9851, indicating excellent data interpretability. Furthermore, the model's mean square error (MSE), root mean square error (RMSE), and maximum absolute percentage error (MAPE) were all minimized, demonstrating extremely small prediction bias and confirming its superior performance in predicting noise levels in slide valve-type pressure reducing valves (safety valves). The deviation between the SVR model's predicted noise value and the measured value did not exceed 5 dB, further validating the high reliability of predictions based on the SVR model.

[0076] To better achieve the optimization goals, analyzing the impact of key parameters on the pneumatic noise of the safety valve is crucial. SHAP analysis can meet the interpretability requirements of the surrogate model, and compared with Sobol sensitivity analysis, this method has lower computational costs and more intuitive results. Therefore, this implementation uses SHAP analysis to assess the degree of influence of each key parameter on the noise level—the higher the contribution value of a parameter, the greater its impact on the noise output.

[0077] First, to investigate whether the contribution of each parameter to noise is related to the distance from the monitoring point, SHAP analysis was conducted at three monitoring points on the downstream axis of the valve outlet. The results showed that at all three monitoring points, the contribution rate of the anti-impact disc edge depth (H1) was consistently the highest, followed by the valve disc outer diameter (D1), while the valve nozzle diameter (D2) had the lowest influence. Specifically, the contribution rate of H1 at monitoring point 1 was 84.05%, decreasing to 59.82% at monitoring point 2, and reaching 58.28% at monitoring point 3. In summary, H1 is the parameter with the greatest impact on the safety valve noise; its contribution rate decreases with increasing distance from the valve outlet and tends to stabilize at a location approximately 1 meter downstream of the valve outlet. Detailed analysis results are shown in Figure 13: the results for monitoring point 1 are shown in Figure 13. Figure 11 (a), the results of monitoring point 2 are shown in [reference]. Figure 11(b), the results corresponding to monitoring point 3 are shown in Figure 11 (c).

[0078] Secondly, to investigate whether the contribution of each parameter to noise changes with angular position, this embodiment conducted SHAP analysis on monitoring points distributed on the acoustic directivity surface. The results show that the edge depth of the shock-absorbing disc (H1) remains the parameter with the highest noise contribution; furthermore, compared to other directions, H1's contribution is higher in the horizontal and vertical directions relative to the valve axis. Detailed results can be found in [link to detailed results]. Figure 12 .

[0079] The optimization objective of this implementation method is to minimize the sound pressure level (SPL) at all monitoring points within the study area. These monitoring points are located 1 m downstream of the valve and 1 m from the pipe wall. To achieve this objective, an optimization framework was developed, using equation (17) as the optimization equation and a support vector regression (SVR) model as the fitness function. Through optimization of the three key parameters, the optimal noise reduction design scheme for the safety valve was finally obtained.

[0080] (17) Given that the Grey Wolf Optimization Algorithm (GWO) possesses excellent global exploration and local exploitation capabilities, as well as advantages such as fast convergence speed and high solution accuracy, this implementation method uses this algorithm to optimize the key parameters corresponding to noise reduction at each monitoring point. The optimized values ​​of the key parameters are listed in Table 6.

[0081]

[0082] To verify the optimization results, aeroacoustic simulation analysis was conducted on the optimized safety valve design. Figure 13 The flow field characteristics and pressure contour maps before and after optimization were compared. Simulation results show that, under the same operating conditions of inlet pressure 0.1 MPa and outlet pressure 0 MPa, the structure optimized by the Grey Wolf Optimization (GWO) algorithm has a maximum flow velocity of 263.25 m / s, which is 5.39 m / s lower than the original design. The flow velocity contour map before optimization is shown below. Figure 13 As shown in (a), the optimized result is as follows: Figure 13 As shown in (b); similarly, the pressure contour map before optimization is shown in [image / data]. Figure 13 As shown in (c), the corresponding optimization results are as follows: Figure 13 As shown in (d).

[0083] Figure 14 The comparison of sound pressure levels (SPL) at the monitoring points before and after optimization is presented. The results show that the average sound pressure level decreased by 6.685 dB across the entire study area. These findings confirm that the valve port structure optimized by GWO can effectively reduce the noise of the safety valve.

[0084] The above description of several specific embodiments further details the technical solution provided by the present invention in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the above-described specific embodiments are not intended to limit the present invention. Any reasonable modifications and improvements to the present invention, combinations of embodiments, and equivalent substitutions based on the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for reducing pneumatic noise in safety valves based on acoustic numerical simulation and surrogate model optimization, characterized in that, include: The steps are as follows: obtain the structural and operating parameters of the safety valve under the open condition, establish a fluid calculation model of the safety valve and set the inlet and outlet boundary conditions, and output the safety valve flow model for numerical analysis. The steps include meshing the safety valve flow model, verifying mesh independence, determining the computational mesh that meets the computational accuracy requirements, and outputting the numerical discrete model for simulation. The steps are as follows: Perform aeroacoustic numerical simulation based on a numerical discrete model to obtain the unsteady flow field inside the safety valve and the corresponding far-field noise response, and output the sound pressure level data at each monitoring location. The steps include: setting up an acoustic monitoring surface and multiple monitoring points downstream of the safety valve outlet to collect and analyze noise response data and output noise characterization data for optimization. The steps include selecting preset geometric parameters of the safety valve orifice as design variables, setting their value range, constructing a parameter design space, and outputting parameter space information for experimental design. The steps involve generating multiple sets of parameter samples based on parameter space information using experimental design methods, obtaining corresponding noise characterization data for each parameter sample, forming a sample dataset, and outputting training data for modeling. The steps are as follows: constructing a surrogate model for predicting safety valve noise based on the sample dataset, selecting the surrogate model with the best prediction accuracy, and outputting the noise response mapping relationship. Based on the surrogate model, the steps involve optimizing key geometric parameters to obtain the optimal combination of structural parameters that meets the noise reduction target, and then outputting the optimized safety valve structural parameters.

2. The method for reducing pneumatic noise of a safety valve based on acoustic numerical simulation and surrogate model optimization according to claim 1, characterized in that, The aeroacoustic numerical simulation uses the large eddy simulation method to solve the unsteady turbulent flow field inside the safety valve, and converts the flow field information into a far-field noise response based on aeroacoustic theory.

3. The method for reducing pneumatic noise of a safety valve based on acoustic numerical simulation and surrogate model optimization according to claim 1, characterized in that, The acoustic monitoring surface is set at a predetermined distance downstream of the safety valve outlet, and the monitoring points are evenly distributed along the circumference or semicircle.

4. The method for reducing pneumatic noise of a safety valve based on acoustic numerical simulation and surrogate model optimization according to claim 1, characterized in that, Key geometric parameters include valve disc outer diameter, nozzle diameter, and anti-impact disc edge depth.

5. The method for reducing pneumatic noise of a safety valve based on acoustic numerical simulation and surrogate model optimization according to claim 1, characterized in that, The experimental design method is the Latin hypercube experimental design method, which is used to generate uniformly distributed parameter samples within the parameter design space.

6. The method for reducing pneumatic noise of a safety valve based on acoustic numerical simulation and surrogate model optimization according to claim 1, characterized in that, The surrogate model is a noise prediction model built based on the sample dataset, and the optimal surrogate model is selected by evaluating the prediction accuracy.

7. A pneumatic noise reduction device for a safety valve based on acoustic numerical simulation and surrogate model optimization, characterized in that, include: The module obtains the structural and operating parameters of the safety valve under the open condition, establishes the fluid calculation model of the safety valve and sets the inlet and outlet boundary conditions, and outputs the safety valve flow model for numerical analysis. Mesh the safety valve flow model and verify its independence. Determine the computational mesh that meets the computational accuracy requirements and output the module of the numerical discrete model for simulation. A module that performs aeroacoustic numerical simulation based on a numerical discrete model to obtain the unsteady flow field inside the safety valve and the corresponding far-field noise response, and outputs the sound pressure level data at each monitoring location. An acoustic monitoring surface is arranged downstream of the safety valve outlet, and multiple monitoring points are set up to collect and analyze noise response and output noise characterization data for optimization. A module that selects the preset geometric parameters of the safety valve orifice as design variables and sets their value range, constructs the parameter design space, and outputs the parameter space information for experimental design. Based on parameter space information, an experimental design method is used to generate multiple sets of parameter samples, and corresponding noise characterization data is obtained for each parameter sample to form a sample dataset. A module that outputs training data for modeling is then output. A proxy model for predicting safety valve noise is constructed based on the sample dataset, and the proxy model with the best prediction accuracy is selected. The module that outputs the noise response mapping relationship is then used. Based on the surrogate model, the module optimizes and calculates key geometric parameters to obtain the optimal combination of structural parameters that meets the noise reduction target, and outputs the optimized safety valve structural parameters.

8. A computer storage medium for storing computer programs, characterized in that, When the computer program is read by the computer, the computer executes the method of claim 1.

9. A computer, comprising a processor and a storage medium, characterized in that, When the processor reads the computer program stored in the storage medium, the computer executes the method of claim 1.

10. A computer program product, as a computer program, is characterized by: When the computer program is executed, it implements the method of claim 1.

Images