Centrifugal fan impeller profile aerodynamic and acoustic optimization method and centrifugal fan

Abstract

This invention relates to the field of aeroacoustic optimization technology, and discloses a method for aerodynamic and acoustic optimization of centrifugal fan impeller profiles and a centrifugal fan. The method involves: using the tangential offset of the centrifugal fan and the inlet / outlet installation angles of the blades as design variables to parametrically model the impeller blade profile, obtaining the impeller fluid domain mesh; performing steady-state CFD solutions to obtain the convergent flow field and impeller aerodynamic performance indices, and using the convergent flow field as initial conditions to solve the pressure pulsation time series of transient CFD; calculating the far-field radiated sound pressure level at the blade passing frequency and its harmonics; and performing multi-objective optimization based on the design variables, impeller aerodynamic performance indices, and far-field radiated sound pressure level to generate the optimal impeller profile geometry file. This invention solves the technical problem in the prior art where aerodynamic optimization and acoustic optimization are independent and cannot be coordinated, ultimately outputting the Pareto optimal impeller profile with the minimum far-field radiated sound pressure level under the constraints of total pressure rise and aerodynamic efficiency.

Description

Technical Field

[0001] This invention relates to the field of aeroacoustic optimization technology, and in particular to a method for aerodynamic and acoustic optimization of centrifugal fan impeller profile and a centrifugal fan. Background Technology

[0002] The geometry of the impeller profile in a centrifugal fan directly determines the fan's aerodynamic efficiency and noise level. Existing technologies employ single-objective steady-state CFD simulations, optimizing only pressure rise or efficiency separately. This fails to simultaneously assess the impact of unsteady aerodynamic excitation generated by impeller rotation on acoustic performance. Consequently, aerodynamic optimization and noise control have long been independent design phases, making it impossible to quantitatively obtain the optimal impeller profile scheme for far-field radiated sound pressure level while ensuring aerodynamic performance. Summary of the Invention

[0003] The main objective of this invention is to provide a method for aerodynamic and acoustic optimization of centrifugal fan impeller profile and a centrifugal fan. This invention solves the technical problem that aerodynamic optimization and acoustic optimization are independent of each other and cannot be coordinated and balanced in the prior art. The final output is the Pareto optimal impeller profile with the minimum far-field radiated sound pressure level under the constraints of total pressure rise and aerodynamic efficiency.

[0004] To achieve the above objectives, the present invention provides a method for aerodynamic and acoustic optimization of centrifugal fan impeller profiles, comprising the following steps: Using the tangential offset of the centrifugal fan and the inlet and outlet installation angles of the blades as design variables, the impeller blade profile is parametrically modeled to obtain the impeller fluid domain mesh. Steady-state CFD is performed on the impeller fluid domain mesh to obtain the convergent flow field and impeller aerodynamic performance indexes. The convergent flow field is then used as the initial condition to solve the pressure fluctuation time series of transient CFD. Calculate the far-field radiated sound pressure level at the blade passage frequency and its harmonics based on the pressure pulsation time series. Based on the design variables, the impeller aerodynamic performance indicators, and the far-field radiated sound pressure level, a multi-objective optimization solution is performed to generate the optimal impeller profile geometry file.

[0005] Optionally, in a first implementation of the first aspect of the present invention, the step of using the tangential offset of the centrifugal fan and the inlet / outlet installation angle of the blades as design variables to parametrically model the impeller blade profile and obtain the impeller fluid domain mesh includes: Using the tangential offset of the centrifugal fan and the inlet and outlet installation angles of the blades as design variables, the impeller blade profile is parametrically modeled to obtain the impeller geometric model. The blade wall boundary layer region in the impeller geometry model is meshed to obtain the impeller fluid domain mesh bound to the design variables.

[0006] Optionally, in a second implementation of the first aspect of the present invention, the step of performing steady-state CFD solution on the impeller fluid domain mesh to obtain the convergent flow field and impeller aerodynamic performance indicators, and using the convergent flow field as initial conditions to solve the pressure fluctuation time series of transient CFD, includes: In the COMSOL rotating machinery fluid flow interface, the convergent flow field of the impeller fluid domain mesh is solved using the mass flow inlet boundary, pressure outlet boundary, and no-slip wall boundary as boundary conditions. The total pressure rise and shaft power are calculated based on the convergent flow field, and the aerodynamic performance index of the impeller is obtained by dividing the product of the rated volumetric flow rate and the total pressure rise by the shaft power. Using the convergent flow field as the initial condition, transient CFD solution is performed in the sliding grid mode to obtain the pressure pulsation time series.

[0007] Optionally, in a third implementation of the first aspect of the present invention, the step of using the convergent flow field as an initial condition to perform transient CFD solution in a sliding grid mode to obtain the pressure fluctuation time series includes: Using the velocity field, pressure field, and turbulent flow field in the convergent flow field as initial conditions, the transient flow field of the impeller fluid domain mesh is solved in the sliding mesh mode of the COMSOL rotating machinery fluid flow interface. Based on the transient flow field, the pressure pulsation time series of each monitoring point on the blade pressure surface and suction surface and the volute tongue region are extracted during the rotation cycle.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, the step of calculating the far-field radiated sound pressure level at the blade passage frequency and its harmonics based on the pressure pulsation time series includes: The pressure pulsation time series is input into the FW-H acoustic analog interface of the COMSOL acoustic module, and the sum of the thickness noise term and the load noise term is used as the sound source to solve the far-field sound pressure time history. A fast Fourier transform is performed on the far-field sound pressure time history to extract the corresponding far-field radiated sound pressure level at the blade passage frequency and its harmonics.

[0009] Optionally, in the fifth implementation of the first aspect of the present invention, before performing multi-objective optimization based on the design variables, the impeller aerodynamic performance indicators, and the far-field radiated sound pressure level to generate the optimal impeller profile geometry file, the following method is further included: An initial sample point set is generated within the upper and lower bounds of the design variables, and the impeller aerodynamic performance index and far-field radiated sound pressure level of each sample point in the initial sample point set are calculated to construct a training sample set. Based on the training sample set, train a first surrogate model optimized for impeller aerodynamic performance indicators and a second surrogate model optimized for far-field radiated sound pressure level until the prediction accuracy of both the first and second surrogate models meets the preset conditions, thus obtaining a dual-objective surrogate model.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, the step of training a first surrogate model optimized for impeller aerodynamic performance indicators and a second surrogate model optimized for far-field radiated sound pressure level based on the training sample set, until the prediction accuracy of both the first surrogate model and the second surrogate model meets a preset condition, to obtain a dual-objective surrogate model, includes: Based on the training sample set, train a first surrogate model optimized for impeller aerodynamic performance indicators and a second surrogate model optimized for far-field radiated sound pressure level. Calculate the first prediction determination coefficient for the sample points corresponding to the first surrogate model and the second prediction determination coefficient for the sample points corresponding to the second surrogate model. Repeat this process until the first prediction determination coefficient is not lower than the first accuracy threshold and the second prediction determination coefficient is not lower than the second accuracy threshold to obtain a bi-objective surrogate model.

[0011] Optionally, in the seventh implementation of the first aspect of the present invention, the step of performing multi-objective optimization based on the design variables, the impeller aerodynamic performance indicators, and the far-field radiated sound pressure level to generate an optimal impeller profile geometry file includes: Using the aforementioned dual-objective surrogate model, multi-objective optimization solutions are performed on the design variables, the impeller aerodynamic performance indicators, and the far-field radiated sound pressure level to obtain the optimal combination of design variables; The optimal combination of design variables is substituted into the impeller fluid domain mesh for simulation, and the optimal impeller profile geometry file is exported.

[0012] Optionally, in the eighth implementation of the first aspect of the present invention, the step of employing the bi-objective surrogate model to perform multi-objective optimization on the design variables, the impeller aerodynamic performance indicators, and the far-field radiated sound pressure level to obtain the optimal combination of design variables includes: Using the bi-objective surrogate model, the design variables, the impeller aerodynamic performance index, and the far-field radiated sound pressure level are iteratively optimized and solved to obtain multiple combinations of first design variables; By filtering the multiple combinations of first design variables, a second combination of design variables is obtained; The solution with the smallest predicted far-field radiated sound pressure level in the second design variable combination is denormalized to the original design variable space to generate the optimal design variable combination.

[0013] The present invention also provides a centrifugal fan, comprising: The parametric modeling module is used to parametrically model the impeller blade profile by taking the tangential offset of the centrifugal fan and the inlet and outlet installation angles of the blades as design variables, and obtain the impeller fluid domain mesh. The simulation solution module is used to perform steady-state CFD solution on the impeller fluid domain grid to obtain the convergent flow field and impeller aerodynamic performance index, and uses the convergent flow field as the initial condition to solve the pressure pulsation time series of transient CFD. The calculation module is used to calculate the far-field radiated sound pressure level at the blade passage frequency and its harmonics based on the pressure pulsation time series. The geometry file generation module is used to perform multi-objective optimization based on the design variables, the impeller aerodynamic performance indicators and the far-field radiated sound pressure level, and generate the optimal impeller profile geometry file.

[0014] In summary, this invention directly writes the tangential offset of the blade profile midpoint control point and the inlet / outlet installation angle into the COMSOL geometric parameter table, realizing end-to-end parameter-driven geometric reconstruction, mesh generation, and simulation solution. Based on the steady-state CFD converged flow field, it switches to the sliding mesh mode to perform transient CFD solution, driving unsteady-state calculations with initial conditions that are physically closer to the periodic steady state. This effectively shortens the length of the transient transition section and completes the sequential coupling of the flow field and FW-H acoustic analogy within a single COMSOL platform. It eliminates the format conversion errors introduced by cross-software data transfer in existing technologies, resulting in higher physical consistency between the predicted results of unsteady pressure pulsation on the blade surface and far-field radiated sound pressure level. By replacing the computationally expensive complete simulation chain with an anisotropic Kriging surrogate model, and combining leave-one-out cross-validation and adaptive point addition mechanism to ensure the accuracy of the surrogate model, the large-scale optimization of NSGA-II in the bi-objective Pareto space can be completed within an engineering-acceptable timeframe. The final output is the Pareto optimal impeller profile with the minimum far-field radiated sound pressure level under the constraints of total pressure rise and aerodynamic efficiency. This solves the technical problem in the prior art where aerodynamic optimization and acoustic optimization are independent and cannot be coordinated and balanced. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the steps of a centrifugal fan impeller profile aerodynamic and acoustic optimization method in one embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the parametric modeling steps in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the CFD solution steps in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the calculation steps for far-field radiated sound pressure level in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the steps of multi-objective optimization in an embodiment of the present invention; Figure 6 This is a block diagram of the centrifugal fan structure in an embodiment of the present invention.

[0016] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Reference Figure 1 This embodiment provides a method for optimizing the aerodynamic and acoustic profile of a centrifugal fan impeller, including the following steps: S10, the tangential offset of the centrifugal fan and the inlet and outlet installation angles of the blades are used as design variables to parametrically model the impeller blade profile and obtain the impeller fluid domain mesh. S20, steady-state CFD solution is performed on the impeller fluid domain grid to obtain the convergent flow field and impeller aerodynamic performance index, and the convergent flow field is used as the initial condition to solve the pressure pulsation time series of transient CFD. S30, calculate the far-field radiated sound pressure level at the blade passing frequency and its harmonics based on the pressure pulsation time series; S40 performs multi-objective optimization based on design variables, impeller aerodynamic performance indicators, and far-field radiated sound pressure level to generate the optimal impeller profile geometry file.

[0019] In this embodiment of the invention, the tangential offset of the blade profile midpoint control point and the inlet / outlet installation angle are directly written into the COMSOL geometric parameter table, realizing full-link parameter driving of geometric reconstruction, mesh generation and simulation solution. Based on the steady-state CFD converged flow field, the transient CFD solution is executed by switching to the sliding mesh mode. The unsteady-state calculation is driven by the initial conditions that are physically closer to the periodic steady state, which effectively shortens the length of the transient transition section. The sequential coupling of the flow field and FW-H acoustic analogy is completed within the single COMSOL platform, eliminating the format conversion error introduced by cross-software data transfer in the prior art, and making the prediction results of unsteady pressure pulsation on the blade surface and far-field radiated sound pressure level have higher physical consistency. By replacing the computationally expensive complete simulation chain with an anisotropic Kriging surrogate model, and combining leave-one-out cross-validation and adaptive point addition mechanism to ensure the accuracy of the surrogate model, the large-scale optimization of NSGA-II in the bi-objective Pareto space can be completed within an engineering-acceptable timeframe. The final output is the Pareto optimal impeller profile with the minimum far-field radiated sound pressure level under the constraints of total pressure rise and aerodynamic efficiency. This solves the technical problem in the prior art where aerodynamic optimization and acoustic optimization are independent and cannot be coordinated and balanced.

[0020] In one example, such as Figure 2 Using the tangential offset of the centrifugal fan and the inlet / outlet installation angles of the blades as design variables, parametric modeling of the impeller blade profile is performed to obtain the impeller fluid domain mesh, including: S11, the tangential offset of the centrifugal fan and the inlet and outlet installation angles of the blades are used as design variables to parametrically model the impeller blade profile and obtain the impeller geometric model; S12, mesh the boundary layer region of the blade wall in the impeller geometry model to obtain the impeller fluid domain mesh bound to the design variables.

[0021] In this example, a basic parameter table for the impeller, front cover, rear cover, and volute flow channel is established in the geometry modeling module of COMSOL Multiphysics. The impeller outer diameter, inner diameter, axial blade height, number of blades, design volumetric flow rate, design speed, and design total pressure rise are written into the model as fixed constraint parameters, without being included in the optimization variables. Then, the centerline of a single blade, i.e., the keel line, is placed in the meridional plane unfolded coordinate system, and a cubic B-spline curve is used to describe the blade profile, so that the continuous blade surface no longer relies on manual drawing but is automatically generated by the control point positions. Six B-spline control points can be arranged along the blade chord direction, with the first and last control points fixed at the leading and trailing edges of the blade to ensure continuous connection between the blade inlet, outlet, and the reference impeller flow channel. The four middle control points are adjustable control points, and their tangential offsets, along with the blade inlet and outlet installation angles, are used as design variables to form a 6-dimensional design variable vector. ,in to This represents the offset of the four intermediate control points along the tangential direction. and These represent the blade inlet and outlet installation angles, respectively. To avoid blade self-intersection, interference between adjacent blades, or abrupt changes in inlet and outlet angles due to parameter variations, upper and lower bound constraints are set for each variable in the parameter table. For example, the tangential offset of the intermediate control point is controlled within ±10% of the total package angle, the inlet installation angle is controlled within ±10° of the nominal value, and the outlet installation angle is controlled within ±15° of the nominal value. This constraint is used as a validity check condition for parametric scanning. When any combination of variables is input, COMSOL reads the vector through the parametric scanning interface, automatically reconstructs the single-blade B-spline profile, and then expands the centerline into a blade entity based on the blade thickness distribution. Subsequently, a corresponding number of three-dimensional blades are generated through a rotation array, and Boolean merging is performed with the front and rear cover plates to form a complete impeller solid geometry. On this basis, Boolean difference operation is performed between the impeller solid and the volute flow channel solid to obtain a closed impeller fluid domain geometry, so that the geometric model is not an isolated blade model, but a fluid computation region that can be directly entered into the CFD solution process.

[0022] The fluid domain is configured as a physics-controlled mesh, with locally refined regions established near the blade pressure surface, suction surface, leading edge, trailing edge, and blade tip. These locations generate strong velocity and pressure gradients, directly impacting pressure pulsation and acoustic calculation accuracy. For the blade wall boundary layer region, a five-layer prismatic boundary layer mesh is used, with the thickness of the first layer determined based on boundary layer control relationships. Estimate the wall friction speed, and then... get The air kinematic viscosity is The calculated first-layer thickness is then incorporated into the COMSOL mesh parameters to ensure that the flow boundary layer near the blade wall can be analyzed, rather than being averaged out by the coarse mesh. Size constraints are set for the leading edge, trailing edge, and vicinity of the volute tongue to capture local unsteady excitations caused by blade rotation in transient CFD. Simultaneously, mesh quality and element transition smoothness are maintained at the interface between the rotating and stationary domains to avoid affecting the stability of the slip mesh calculation due to mesh distortion. To ensure comparability of calculation results under different combinations of design variables, a unified mesh generation template is established. This template automatically performs geometric updates, local refinement, boundary layer generation, and quality checks after each parameter change. Mesh independence is verified using coarse, medium, and fine meshes. The relative deviation of the total pressure rise between the medium and fine meshes is less than 1.5%. Once this condition is met, the medium mesh is used as the standard mesh configuration for sample points.

[0023] In one example, such as Figure 3 Steady-state CFD solutions were performed on the impeller fluid domain mesh to obtain the convergent flow field and impeller aerodynamic performance indices. The convergent flow field was then used as initial conditions to solve the transient CFD pressure fluctuation time series, including: S21, in the COMSOL rotating machinery fluid flow interface, the convergent flow field of the impeller fluid domain mesh is solved using the mass flow inlet boundary, pressure outlet boundary and no-slip wall boundary as boundary conditions; S22, based on the convergent flow field, calculate the total pressure rise and shaft power, and divide the product of the rated volumetric flow rate and the total pressure rise by the shaft power to obtain the impeller aerodynamic performance index; S23, using the convergent flow field as the initial condition, performs transient CFD solution in the sliding grid mode to obtain the pressure fluctuation time series.

[0024] In this example, based on the impeller fluid domain mesh, we directly access the "Rotating Machinery, Fluid Flow" interface of COMSOL, defining the impeller region as the rotating domain and the volute and inlet / outlet channels as the stationary domain, so that the relative motion generated by the impeller rotation can be correctly represented in the solver; SST is selected in the steady-state CFD stage. - A turbulence model was developed, and the steady flow field was solved within a multi-reference frame. The inlet was set as the mass flow rate inlet boundary, and the inlet mass flow rate could be calculated according to... The settings include the standard state values ​​for air density. The outlet is set as a pressure outlet boundary with a reference pressure of zero. The blade walls, front and rear shroud walls, and volute walls are all set as no-slip wall boundaries. The interface between the rotating and stationary domains is treated using a frozen rotor method. This allows the steady-state calculation to simultaneously reflect the energy input from the impeller rotation and the constraint effect of the volute stationary boundary on the flow field. The solution can be obtained using COMSOL's built-in Paradiso direct solver combined with fully coupled Newton iteration, reducing the normalized residuals of the momentum equation, continuity equation, and turbulence equation to a minimum. The following are the primary convergence conditions, while also ensuring that the difference between the outlet mass flow rate and the inlet mass flow rate is less than 1. As a mass conservation check condition, the steady-state result is considered as the convergent flow field of the impeller fluid domain grid only when the residual convergence and the inlet and outlet airflow balance are satisfied simultaneously.

[0025] The total pressure mass flow rate weighted average value of the inlet and outlet sections is extracted from the convergent flow field and calculated according to the total pressure rise relationship. The total pressure increment provided by the impeller to the airflow is obtained, and the aerodynamic torque on the impeller is extracted from the rotation domain. And combined with angular velocity relationship Calculate shaft power The impeller aerodynamic performance index is obtained by dividing the product of the rated volumetric flow rate and the total pressure rise by the shaft power. . Indicates total pressure rise, and These represent the weighted average of total pressure and mass flow rate at the outlet and inlet, respectively. Indicates impeller torque. Indicates angular velocity. Indicates the rated speed. Indicates the rated volumetric flow rate. Indicates shaft power. This indicates the aerodynamic performance of the impeller.

[0026] Instead of restarting the calculation from a zero velocity field or a uniform pressure field, the steady-state convergent velocity field, pressure field, and turbulent flow field are used as the initial conditions for transient CFD. Then, in the same COMSOL rotating machinery fluid flow interface, the calculation switches from multi-reference steady-state mode to slip mesh mode, allowing the impeller rotation domain to undergo realistic time progression relative to the volute stationary domain, and solving the unsteady flow field. During the transient solution process, pressure monitoring points are placed on the blade pressure surface, suction surface, and volute tongue region, and pressure variation data over time are continuously output according to a set time step, obtaining a pressure pulsation time series. The steady-state convergent flow field already contains a relatively reasonable velocity, pressure, and turbulence distribution within the impeller's main flow channel. Using it as the transient initial condition reduces the number of rotational cycles required to transition from the initial field to a periodic steady state, making the captured pressure pulsation data closer to the actual blade passage and dynamic-static interference excitation, without being contaminated by initial transient disturbances.

[0027] In one example, using the convergent flow field as the initial condition, a transient CFD solution is performed in the slip grid mode to obtain the pressure fluctuation time series, including: Using the velocity field, pressure field, and turbulent flow field in the convergent flow field as initial conditions, the transient flow field of the impeller fluid domain mesh is solved in the sliding mesh mode of the COMSOL rotating machinery fluid flow interface. Based on the transient flow field, the pressure pulsation time series of each monitoring point on the blade pressure surface and suction surface, as well as the monitoring point in the volute tongue region, are extracted during the rotation cycle.

[0028] In this example, the velocity, pressure, and turbulent flow fields that have converged in the steady-state CFD are transferred from the steady-state study step to the transient study step. In COMSOL's "Rotating Machinery, Fluid Flow" interface, the multi-reference frame calculation settings previously used for steady-state solutions are switched to slip mesh mode. This allows the impeller rotational domain to advance in real time relative to the stationary volute domain at its rated speed, instead of continuing to use the frozen rotor approximation. In the initial value settings of the transient study step, "Solution from previous study step" is selected, and field variables such as velocity, pressure, turbulent kinetic energy, and specific dissipation rate are mapped respectively. This allows the transient calculation to start from a near-periodic stable flow field state, thereby reducing large non-physical transition fluctuations caused by a zero initial field. Simultaneously, the inlet mass flow rate, outlet pressure, and the no-slip walls of the blades and volute are kept consistent with the steady-state stage. Slip mesh pairing is enabled at the interface between the rotating and stationary domains, so that the relative positional relationship between the blades and the tongue is recalculated in each time step. When advancing the time step, the time step can be set to the time corresponding to 1° of impeller rotation, ensuring sufficient time resolution for the blade passage process. For a typical centrifugal fan with no more than 12 blades, this setting guarantees at least 30 time steps within each blade passage cycle, thus meeting the time-domain resolution requirements for the blade passage frequency and the first three harmonics. The total transient calculation time can be set to 10 complete rotation cycles. The first 5 cycles are used to eliminate disturbances generated during the transition from the steady-state initial field to the transient periodic steady state, while the last 5 cycles serve as an effective data window. The flow field is judged to have entered a periodic steady state by ensuring that the pressure amplitude deviation between adjacent cycles at representative blade wall monitoring points is less than 2%.

[0029] After entering the data acquisition stage, 20 pressure monitoring points are evenly arranged along the chord length on the pressure and suction surfaces of each blade to ensure that the load fluctuations on the blade surface can be completely sampled. At the same time, 10 additional pressure monitoring points are arranged in the volute tongue region near the interface between the rotation and stationary domains to capture the dynamic-static interference pressure pulsations generated when the blade sweeps across the tongue. Subsequently, the wall pressure change data of each monitoring point over time is continuously recorded according to the sampling frequency of the transient time step, and only the data within the last 5 rotation cycles are extracted to form a pressure pulsation time series. Finally, the monitoring data of the blade pressure surface, suction surface, and volute tongue region are uniformly saved as .txt or .csv files as the sound source input for FW-H acoustic analogy calculation.

[0030] In one example, such as Figure 4 The far-field radiated sound pressure level at the blade passage frequency and its harmonics is calculated based on the pressure pulsation time series, including: S31, input the pressure pulsation time series into the FW-H acoustic analog interface of the COMSOL acoustic module, and use the sum of the thickness noise term and the load noise term as the sound source to solve the far-field sound pressure time history. S32 performs a fast Fourier transform on the far-field sound pressure time history and extracts the corresponding far-field radiated sound pressure level at the blade passage frequency and its harmonics.

[0031] In this example, the time series of pressure pulsations at monitoring points on the blade pressure surface, suction surface, and volute tongue region are used as inputs for the acoustic solution. An "Aeroacoustics" or FW-H acoustic analogy calculation interface is established in the COMSOL acoustic module. A mapping relationship is established between the unsteady wall pressure data obtained from transient CFD and the corresponding monitoring point geometric positions, wall outward normal vectors, and impeller rotational angular velocities, allowing the pressure pulsations obtained from the flow field calculation to be directly converted into acoustic sources. Equivalent sound source surfaces near the blade wall and tongue are defined in the acoustic model, and the far-field sound pressure contribution in the FW-H equations is limited to the sum of the thickness noise term and the load noise term. For subsonic centrifugal fan operating conditions where the blade tip Mach number is less than 0.3 and the flow field does not exhibit large-scale separation, backflow, or strong turbulent jets, the quadrupole volume noise term can be neglected by estimating at a low Mach number scale, while retaining the thickness noise term and the load noise term. During the prototype verification stage, this simplification can be verified through sound pressure level tests under typical operating conditions or full-item acoustic calculations. The load noise input can be obtained by multiplying the pressure pulsation on the blade surface by the external normal vector, i.e. The thickness noise input can be obtained by calculating the wall normal velocity from the impeller rotational angular velocity and the monitoring point position vector, i.e. ,in Indicates the first The location vector of each monitoring point This represents the time series of pressure pulsations at the monitoring point. This represents the external normal vector at the monitoring point. Indicates the angular velocity of the impeller. Indicates the normal pressure per unit area. This represents the normal velocity along the wall. After defining the sound source, at a distance from the impeller center... A virtual sound pressure sensor array is arranged at the position, and 12 sensors are evenly arranged around the circumference in the plane directly above the impeller axis. This ensures that the far-field sound pressure calculation results can reflect the circumferential radiation differences and remain comparable to conventional acoustic measurement distances. The far-field sound pressure time history at each virtual sensor is obtained by integrating the solutions of each sound source surface through the FW-H interface.

[0032] A Fast Fourier Transform (FFT) was performed on the far-field sound pressure time history, with the time window limited to the last five rotational cycles of data after the aforementioned transient CFD had entered periodic stability. A Hanning window was also applied to reduce spectral leakage caused by truncation. At this point, the frequency resolution... and the frequency of blade passage and its first three harmonics , The sound pressure amplitude of each sensor is extracted. To unify multiple sensors and multiple harmonic frequencies into a single acoustic index suitable for multi-objective optimization, the total radiated effective sound pressure is calculated. Then convert it to sound pressure level. The reference sound pressure .

[0033] In one example, before generating the optimal impeller profile geometry file by performing multi-objective optimization based on design variables, impeller aerodynamic performance indicators, and far-field radiated sound pressure level, the following steps are also included: An initial sample point set is generated within the upper and lower bounds of the design variables, and the impeller aerodynamic performance index and far-field radiated sound pressure level of each sample point in the initial sample point set are calculated to construct a training sample set. The first surrogate model, which optimizes the aerodynamic performance of the impeller, and the second surrogate model, which optimizes the far-field radiated sound pressure level, are trained based on the training sample set until the prediction accuracy of both the first and second surrogate models meets the preset conditions, thus obtaining a dual-objective surrogate model.

[0034] In this example, the upper and lower bounds of the intermediate control point tangential offset, blade inlet installation angle, and blade outlet installation angle are used as sampling boundaries. Each variable is constrained within a geometrically feasible range, ensuring that the sampling points cover the main areas where the blade profile may change without generating invalid geometry such as blade self-intersection, interference between adjacent blades, or excessive deviations in inlet and outlet angles. An initial sample point set is generated using optimized Latin hypercube sampling within the design variable upper and lower bound matrices, ensuring that each variable is uniformly layered within its allowable range. The dispersion of samples in the design space is improved by maximizing the minimum Euclidean distance between samples. Since the design space dimension is... And adopt the empirical rules of the Kriging proxy model. The lower limit of the sample size is determined, therefore the minimum sample size is Meanwhile, to preserve redundancy for surrogate model training and local predictions near the Pareto front, the initial sample size can be maintained at [value missing]. After obtaining 80 sample points, the design variable combination for each sample point was sequentially written into the COMSOL parametric geometric model. This automatically completed impeller profile reconstruction, impeller entity updating, fluid domain Boolean operations, and standard mesh generation. Then, steady-state CFD, transient CFD, and FW-H acoustic solution chains were sequentially executed for each sample point. The steady-state CFD was used to output the impeller aerodynamic performance indicators corresponding to that sample point; the transient CFD was used to obtain the pressure pulsation time series; and the FW-H acoustic analogy and fast Fourier transform were used to extract the far-field radiated sound pressure level. To avoid incomparability of target values ​​between samples due to differences in mesh accuracy, boundary conditions, or solution settings, the same set of mesh generation templates, inlet mass flow rates, outlet pressures, no-slip wall conditions, transient time windows, and acoustic sensor arrangements were fixed. Samples that failed to solve, had unqualified mesh quality, or whose residuals did not meet the convergence conditions were marked and resampled. After completing the calculations for all samples, the design variables, impeller aerodynamic performance indicators, and far-field radiated sound pressure levels for each sample point are written into the training sample set to form the training data structure. ,in Indicates the first A combination of design variables, This indicates the impeller aerodynamic performance index corresponding to this combination. This indicates the far-field radiated sound pressure level corresponding to this combination.

[0035] When training the dual-objective surrogate model, the 6-dimensional design variables in the training sample set are linearly normalized to map tangential offsets and installation angles with different dimensions and value ranges to the [0,1] interval, thereby reducing the impact of variable scale differences on the covariance calculation of the surrogate model. The normalized design variables are used as input, and the impeller aerodynamic performance index is used as output to train the first surrogate model. The same input variable, far-field radiated sound pressure level, is used as output to train the second surrogate model, allowing the two surrogate models to approximate the nonlinear mapping relationship between "profile parameters—aerodynamic target" and "profile parameters—acoustic target," respectively. The surrogate model can adopt the Kriging model, and the first and second surrogate models use the same training samples but are modeled independently, thus avoiding the forced merging of aerodynamic and acoustic targets into a single weighted index. After training, the prediction determination coefficients of the two models are calculated using leave-one-out cross-validation, and whether the prediction determination coefficient is not lower than 0.90 is used as the criterion for judging whether the accuracy meets the preset conditions. If the prediction accuracy of either the first or second surrogate model fails to reach the threshold, additional sample points are added in areas with high prediction uncertainty or drastic target changes. The addition method is to add 10 samples each time by using the maximum expectation improvement criterion, and then re-execute geometric reconstruction, CFD solution and acoustic solution to expand the training sample set until both models meet the accuracy requirements, thus obtaining a dual-objective surrogate model.

[0036] In one example, a first surrogate model optimized for impeller aerodynamic performance indicators and a second surrogate model optimized for far-field radiated sound pressure level are trained based on a training sample set until the prediction accuracy of both the first and second surrogate models meets preset conditions, resulting in a bi-objective surrogate model, including: A first surrogate model optimized for impeller aerodynamic performance indicators and a second surrogate model optimized for far-field radiated sound pressure level were trained based on the training sample set. Calculate the first prediction determination coefficient for the sample points corresponding to the first surrogate model and the second prediction determination coefficient for the sample points corresponding to the second surrogate model. Repeat this process until the first prediction determination coefficient is not lower than the first accuracy threshold and the second prediction determination coefficient is not lower than the second accuracy threshold to obtain the bi-objective surrogate model.

[0037] In this example, the training sample set is used as a unified data source, and the 6-dimensional design variables in each sample point are used as model inputs. The impeller aerodynamic performance index is used as the first output target, and the far-field radiated sound pressure level is used as the second output target. Before training, linear normalization is performed on the design variables to map the tangential offset and the inlet / outlet installation angle, although they have different dimensions and ranges, to the same dimensionless interval. This avoids one type of variable having excessive weight in the surrogate model training due to its larger numerical scale. Two independent Kriging surrogate models with identical input samples are established, with the first surrogate model denoted as... The second surrogate model is used to fit the nonlinear relationship between design variables and impeller aerodynamic performance indicators. This method is used to fit the nonlinear relationship between design variables and far-field radiated sound pressure level, thus preserving the individual variation patterns of aerodynamic and acoustic targets, rather than merging them into a single weighted score. During training, an anisotropic Matérn 5 / 2 kernel function can be used to establish the spatial correlation between samples, and the correlation length hyperparameters for each dimension can be determined by maximizing the logarithmic marginal likelihood function. This allows the model to identify which design variables are more sensitive to aerodynamic performance and which are more sensitive to noise indicators.

[0038] After the initial training of the first and second proxy models, the prediction determination coefficients are calculated using leave-one-out cross-validation. This involves temporarily removing one sample from the training set each time, training the model with the remaining samples, predicting the target value of the removed sample, and then summing the prediction errors of all samples. ,in Indicates the first The true target value of each sample Indicates the removal of the first The predicted value obtained after 10 samples This represents the mean of the target sample value. This formula is used to measure the predictive ability of the surrogate model on samples not included in the current training. The prediction determination coefficient of the first surrogate model is denoted as... The prediction determination coefficient of the second proxy model is denoted as... And both the first precision threshold and the second precision threshold can be set to 0.90. and When the aerodynamic and acoustic surrogate models are both explained, it indicates that they can explain more than 90% of the variance of the objective function and can be used as fast evaluators for multi-objective optimization. If any prediction determination coefficient is lower than the corresponding threshold, it indicates that the current sample set does not adequately cover the nonlinear changes of the objective function. Based on the prediction uncertainty or maximum expectation improvement criterion, new design variable samples are added to the region with large errors. Parametric modeling, steady-state CFD, transient CFD, and FW-H acoustic calculations are then re-executed on the new samples. The obtained real aerodynamic performance indicators and far-field radiated sound pressure levels are added to the training sample set. Then, the two surrogate models are re-normalized, retrained, and recalculated. and This process is repeated until both prediction determination coefficients simultaneously meet the threshold requirements. The resulting bi-objective surrogate model is a callable model group composed of the first surrogate model and the second surrogate model. Its input is the same set of normalized design variables, and its outputs correspond to the predicted values ​​of turbine aerodynamic performance and far-field radiated sound pressure level, respectively.

[0039] In one example, such as Figure 5 Based on design variables, impeller aerodynamic performance indicators, and far-field radiated sound pressure level, a multi-objective optimization solution is performed to generate the optimal impeller profile geometry file, including: S41 uses a dual-objective surrogate model to perform multi-objective optimization solutions on design variables, impeller aerodynamic performance indicators and far-field radiated sound pressure level, and obtains the optimal combination of design variables. S42, substitute the optimal combination of design variables into the impeller fluid domain mesh for simulation, and export the optimal impeller profile geometry file.

[0040] In this example, a bi-objective surrogate model is integrated into a multi-objective optimization program, and the normalized 6-dimensional design variables are used as the search objects. Each candidate individual corresponds to a set of blade intermediate control point tangential offset, blade inlet installation angle, and blade outlet installation angle. The first surrogate model quickly predicts the impeller aerodynamic performance index of the candidate profile, and the second surrogate model quickly predicts the far-field radiated sound pressure level of the candidate profile. The common optimization direction is to maintain or improve aerodynamic performance and reduce far-field radiated sound pressure level, thereby avoiding the simple weighting of aerodynamic and acoustic objectives into a single objective, which would obscure the trade-off relationship. The NSGA-II multi-objective genetic algorithm is run in the [0,1] normalized design variable space. The initial population can be 100, and the maximum number of generations can be set to 200. Selection, crossover, and mutation are all completed through surrogate model evaluation, which transforms the objective function calculation that originally required a large number of complete COMSOL simulations into a fast surrogate model query. During the iteration process, individuals that exceed the upper and lower bounds of the variables are directly truncated to the legal boundary, and the convergence and distribution of the Pareto solution set are continuously maintained through non-dominated sorting and crowding distance. When the change rate of the Pareto front hypervolume index is less than 0.1% for 20 consecutive generations, the optimization search can be considered to be stable. If the condition is not met even after reaching the maximum number of generations, the last generation of non-dominated solutions is used as the candidate solution set. Based on the candidate solution set, the solution is selected in conjunction with engineering constraints. For example, the solution that satisfies the condition that the total pressure rise is not lower than the design total pressure rise and the aerodynamic efficiency decrease does not exceed the allowable tolerance is retained. Among the solutions that satisfy the constraints, the design variable combination with the smallest predicted value of far-field radiated sound pressure level is selected. Since this combination is still in the normalized variable space, it is denormalized back to the original design variable space to obtain the optimal design variable combination that can be directly written into the COMSOL parameter table.

[0041] The optimal design variable combination is re-substituted into the parameterized impeller fluid domain model. The inverse-normalized tangential offset and inlet / outlet installation angles are written into the COMSOL geometry parameter table, triggering B-spline blade profile reconstruction, blade solid rotation array, front and rear cover plate merging, and Boolean difference operation between the impeller solid and the volute flow channel, thus regenerating the optimal impeller geometry. The fluid domain is then re-divided using the standard mesh template that has passed the mesh independence verification. After completing the geometry and mesh update, a complete high-fidelity verification simulation is performed, i.e., the actual aerodynamic performance and actual far-field radiated sound pressure level of the optimal scheme are recalculated in the order of steady-state CFD, transient CFD, and FW-H acoustic analogy. The simulation results are then compared with the prediction results of the surrogate model for error checking. If the error meets the engineering allowable range, the design variable combination is confirmed to be effective, and the optimal impeller profile geometry file is output through COMSOL's geometry export function. The file format can be selected as STEP, Parasolid, IGES, or STL according to the needs of manufacturing, 3D modeling, or subsequent verification, while retaining the corresponding parameter table, mesh file, and simulation result file.

[0042] In one example, a bi-objective surrogate model is used to perform multi-objective optimization on design variables, impeller aerodynamic performance indicators, and far-field radiated sound pressure level, obtaining the optimal combination of design variables, including: A dual-objective surrogate model was used to iteratively optimize and solve the design variables, impeller aerodynamic performance indicators and far-field radiated sound pressure level, resulting in multiple combinations of first design variables. By filtering multiple combinations of first design variables, a second combination of design variables is obtained; The solution that minimizes the predicted far-field radiated sound pressure level in the second design variable combination is inversely normalized to the original design variable space to generate the optimal design variable combination.

[0043] In this example, a bi-objective surrogate model is used as the objective function evaluator for the NSGA-II multi-objective genetic algorithm. An initial population is generated in the normalized 6-dimensional design variable space, so that each individual corresponds to a set of blade intermediate control point tangential offset, blade inlet installation angle, and blade outlet installation angle. In each generation iteration, the design variables corresponding to the individual are input into the first surrogate model and the second surrogate model. The first surrogate model predicts the impeller aerodynamic performance index, and the second surrogate model predicts the far-field radiated sound pressure level. According to the principle of "keeping or improving aerodynamic performance as much as possible and reducing far-field radiated sound pressure level as much as possible", non-dominated sorting is performed so that the solution that is not simultaneously superior to other individuals in terms of aerodynamic and acoustic performance is retained. The dispersion of the solution set on the Pareto front is maintained by the crowding distance. For example, the population size can be 100, the maximum number of generations can be 200, the selection operator is a binary tournament selection, the crossover operator is a simulated binary crossover with a crossover probability of 0.9, the mutation operator is a polynomial mutation with a mutation probability of 1 / 6, and variables that exceed the [0,1] normalization boundary after mutation or crossover are directly truncated to the boundary value, thus ensuring that all candidate curves are within the feasible design space. As the iteration progresses, the algorithm continuously uses a bi-objective surrogate model to quickly evaluate a large number of candidate schemes without having to repeat a complete COMSOL simulation for each scheme. When the rate of change of the Pareto front hypervolume index is less than 0.1% for 20 consecutive generations, the iterative optimization can be considered to have converged. If the convergence condition is not met after 200 generations, the non-dominated solution in the last generation is used as a combination of multiple first design variables.

[0044] Multiple combinations of first design variables are screened. During the screening, schemes with predicted total pressure rise lower than the design total pressure rise are eliminated, as are schemes with aerodynamic efficiency lower than the allowable lower limit of the reference impeller. For example, the efficiency constraint is that the absolute value of efficiency decreases by no more than 3 percentage points. Therefore, only combinations of first design variables that simultaneously meet the total pressure rise constraint and the efficiency constraint can enter the set of second design variable combinations.

[0045] In the second combination of design variables, the predicted far-field radiated sound pressure levels are compared, and the solution with the smallest predicted sound pressure level is selected as the final candidate solution. Since the final candidate solution is still located in the normalized design variable space, each variable is denormalized from the [0,1] interval back to the original design variable space according to the same upper and lower bounds used when training the surrogate model, so that it becomes the true intermediate control point tangential offset, blade inlet installation angle, and blade outlet installation angle again. The generated optimal combination of design variables contains complete 6-dimensional original parameter values ​​and can be directly written into the COMSOL geometry parameter table for driving impeller profile reconstruction, fluid domain mesh update, and high-fidelity simulation verification.

[0046] Reference Figure 6 This embodiment provides a centrifugal fan, including: The parametric modeling module 61 is used to parametrically model the impeller blade profile by taking the tangential offset of the centrifugal fan and the inlet and outlet installation angles of the blades as design variables, and to obtain the impeller fluid domain mesh. The simulation solution module 62 is used to perform steady-state CFD solution on the impeller fluid domain mesh to obtain the convergent flow field and impeller aerodynamic performance index, and uses the convergent flow field as the initial condition to solve the pressure pulsation time series of transient CFD. Calculation module 63 is used to calculate the far-field radiated sound pressure level at the blade passage frequency and its harmonics based on the pressure pulsation time series. The geometry file generation module 64 is used to perform multi-objective optimization based on design variables, impeller aerodynamic performance indicators and far-field radiated sound pressure level, and generate the optimal impeller profile geometry file.

[0047] In this embodiment, the specific implementation of each unit in the above device embodiment is described in the above method embodiment, and will not be repeated here.

[0048] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0049] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for optimizing the aerodynamic and acoustic profile of a centrifugal fan impeller, characterized in that, include: Using the tangential offset of the centrifugal fan and the inlet and outlet installation angles of the blades as design variables, the impeller blade profile is parametrically modeled to obtain the impeller fluid domain mesh. Steady-state CFD is performed on the impeller fluid domain mesh to obtain the convergent flow field and impeller aerodynamic performance indexes. The convergent flow field is then used as the initial condition to solve the pressure fluctuation time series of transient CFD. Calculate the far-field radiated sound pressure level at the blade passage frequency and its harmonics based on the pressure pulsation time series. Based on the design variables, the impeller aerodynamic performance indicators, and the far-field radiated sound pressure level, a multi-objective optimization solution is performed to generate the optimal impeller profile geometry file.

2. The method for optimizing the aerodynamic and acoustic profile of a centrifugal fan impeller according to claim 1, characterized in that, The process involves using the tangential offset of the centrifugal fan and the inlet / outlet installation angles of the blades as design variables to parametrically model the impeller blade profile, resulting in an impeller fluid domain mesh, including: Using the tangential offset of the centrifugal fan and the inlet and outlet installation angles of the blades as design variables, the impeller blade profile is parametrically modeled to obtain the impeller geometric model. The blade wall boundary layer region in the impeller geometry model is meshed to obtain the impeller fluid domain mesh bound to the design variables.

3. The method for optimizing the aerodynamic and acoustic profile of a centrifugal fan impeller according to claim 1, characterized in that, The steady-state CFD solution of the impeller fluid domain mesh is used to obtain the convergent flow field and impeller aerodynamic performance indicators. The convergent flow field is then used as the initial condition to solve the pressure fluctuation time series of the transient CFD, including: In the COMSOL rotating machinery fluid flow interface, the convergent flow field of the impeller fluid domain mesh is solved using the mass flow inlet boundary, pressure outlet boundary, and no-slip wall boundary as boundary conditions. The total pressure rise and shaft power are calculated based on the convergent flow field, and the aerodynamic performance index of the impeller is obtained by dividing the product of the rated volumetric flow rate and the total pressure rise by the shaft power. Using the convergent flow field as the initial condition, transient CFD solution is performed in the sliding grid mode to obtain the pressure pulsation time series.

4. The method for optimizing the aerodynamic and acoustic profile of a centrifugal fan impeller according to claim 3, characterized in that, The step of using the convergent flow field as initial conditions and performing transient CFD solution in slip grid mode to obtain the pressure fluctuation time series includes: Using the velocity field, pressure field, and turbulent flow field in the convergent flow field as initial conditions, the transient flow field of the impeller fluid domain mesh is solved in the sliding mesh mode of the COMSOL rotating machinery fluid flow interface. Based on the transient flow field, the pressure pulsation time series of each monitoring point on the blade pressure surface and suction surface and the volute tongue region are extracted during the rotation cycle.

5. The method for optimizing the aerodynamic and acoustic profile of a centrifugal fan impeller according to claim 1, characterized in that, The calculation of the far-field radiated sound pressure level at the blade passage frequency and its harmonics based on the pressure pulsation time series includes: The pressure pulsation time series is input into the FW-H acoustic analog interface of the COMSOL acoustic module, and the sum of the thickness noise term and the load noise term is used as the sound source to solve the far-field sound pressure time history. A fast Fourier transform is performed on the far-field sound pressure time history to extract the corresponding far-field radiated sound pressure level at the blade passage frequency and its harmonics.

6. The method for optimizing the aerodynamic and acoustic profile of a centrifugal fan impeller according to claim 5, characterized in that, Before generating the optimal impeller profile geometry file by performing multi-objective optimization based on the design variables, the impeller aerodynamic performance indicators, and the far-field radiated sound pressure level, the process also includes: An initial sample point set is generated within the upper and lower bounds of the design variables, and the impeller aerodynamic performance index and far-field radiated sound pressure level of each sample point in the initial sample point set are calculated to construct a training sample set. Based on the training sample set, train a first surrogate model optimized for impeller aerodynamic performance indicators and a second surrogate model optimized for far-field radiated sound pressure level until the prediction accuracy of both the first and second surrogate models meets the preset conditions, thus obtaining a dual-objective surrogate model.

7. The method for optimizing the aerodynamic and acoustic profile of a centrifugal fan impeller according to claim 6, characterized in that, The process involves training a first surrogate model optimized for impeller aerodynamic performance indicators and a second surrogate model optimized for far-field radiated sound pressure level based on the training sample set, until the prediction accuracy of both the first and second surrogate models meets preset conditions, resulting in a dual-objective surrogate model, including: Based on the training sample set, train a first surrogate model optimized for impeller aerodynamic performance indicators and a second surrogate model optimized for far-field radiated sound pressure level. Calculate the first prediction determination coefficient for the sample points corresponding to the first surrogate model and the second prediction determination coefficient for the sample points corresponding to the second surrogate model. Repeat this process until the first prediction determination coefficient is not lower than the first accuracy threshold and the second prediction determination coefficient is not lower than the second accuracy threshold to obtain a bi-objective surrogate model.

8. The method for optimizing the aerodynamic and acoustic profile of a centrifugal fan impeller according to claim 7, characterized in that, The process of performing multi-objective optimization based on the design variables, the impeller aerodynamic performance indicators, and the far-field radiated sound pressure level to generate the optimal impeller profile geometry file includes: Using the aforementioned dual-objective surrogate model, multi-objective optimization solutions are performed on the design variables, the impeller aerodynamic performance indicators, and the far-field radiated sound pressure level to obtain the optimal combination of design variables; The optimal combination of design variables is substituted into the impeller fluid domain mesh for simulation, and the optimal impeller profile geometry file is exported.

9. The method for optimizing the aerodynamic and acoustic profile of a centrifugal fan impeller according to claim 8, characterized in that, The method employs the bi-objective surrogate model to perform multi-objective optimization on the design variables, the impeller aerodynamic performance indicators, and the far-field radiated sound pressure level, obtaining the optimal combination of design variables, including: Using the bi-objective surrogate model, the design variables, the impeller aerodynamic performance index, and the far-field radiated sound pressure level are iteratively optimized and solved to obtain multiple combinations of first design variables; By filtering the multiple combinations of first design variables, a second combination of design variables is obtained; The solution with the smallest predicted far-field radiated sound pressure level in the second design variable combination is denormalized to the original design variable space to generate the optimal design variable combination.

10. A centrifugal fan, characterized in that, The steps for implementing the centrifugal fan impeller profile aerodynamic and acoustic optimization method according to any one of claims 1 to 9 include: The parametric modeling module is used to parametrically model the impeller blade profile by taking the tangential offset of the centrifugal fan and the inlet and outlet installation angles of the blades as design variables, and obtain the impeller fluid domain mesh. The simulation solution module is used to perform steady-state CFD solution on the impeller fluid domain grid to obtain the convergent flow field and impeller aerodynamic performance index, and uses the convergent flow field as the initial condition to solve the pressure pulsation time series of transient CFD. The calculation module is used to calculate the far-field radiated sound pressure level at the blade passage frequency and its harmonics based on the pressure pulsation time series. The geometry file generation module is used to perform multi-objective optimization based on the design variables, the impeller aerodynamic performance indicators and the far-field radiated sound pressure level, and generate the optimal impeller profile geometry file.

Images