Compressor through-flow modeling and stator end bending integrated optimization method and device

Abstract

The invention provides a compressor through-flow modeling and stator end bending integrated optimization method and device. An optimization area is parameterized, a parameterized model is generated, an initial sample database is generated based on the parameterized model, and a prediction model is obtained according to training of the initial sample database. The prediction model is used for replacing fluid dynamic simulation, the optimization period is shortened, and the optimization design efficiency is greatly improved. Through-flow modeling and stator end bending optimization are subjected to collaborative optimization at the same time instead of traditional isolated optimization of a single component, the stability margin of the gas compressor is widened, and the aerodynamic performance of the gas compressor is remarkably improved. Besides, the prediction model can adapt to performance prediction of the gas compressor under different working conditions, and can be quickly expanded to full-working-condition optimization only by supplementing samples of corresponding working conditions, so that the universality and reusability of the optimization method are greatly improved.

Description

Technical Field

[0001] This application relates to the field of flow control technology, and more specifically, to an integrated optimization method and apparatus for compressor flow path design and stator end bending. Background Technology

[0002] A gas turbine is a thermal engine that uses gas as a working fluid to drive a high-speed rotating impeller. The compressor, as one of the most important core components of a gas turbine, directly determines the overall performance of the gas turbine.

[0003] Currently, compressors exhibit various unstable flow phenomena, which hinder further performance improvements. To suppress these unstable flow phenomena, engineers have conducted research on compressor flow path design and blade optimization, both of which have resulted in varying degrees of performance enhancement.

[0004] However, existing optimization schemes often exhibit flow interference at the junction of the blades and endwalls, resulting in limited effectiveness in mitigating unstable flow in the compressor and thus limiting the improvement of compressor performance. Summary of the Invention

[0005] To address the aforementioned issues, this application provides an integrated optimization method and apparatus for compressor flow path design and stator end bend. Through the integrated optimization design of flow path design and stator end bend, the stability margin of the compressor is broadened and the performance of the compressor is improved.

[0006] In a first aspect, this application provides an integrated optimization method for compressor flow path design and stator end bend, including:

[0007] Within the optimized region of the compressor's flow passage, the geometric control parameters of the flow passage design and the airfoil geometry parameters of the stator are obtained respectively, and a parameterized model is generated based on the geometric control parameters of the flow passage design and the airfoil geometry parameters of the stator. The parameterized model is subjected to structured mesh generation, and the flow field of the meshed parameterized model is numerically simulated to obtain at least two performance index parameters of the compressor corresponding to the parameterized model. Based on the parameterized model, the geometric control parameters of the flow path and the airfoil geometry parameters of the stator are sampled to generate an initial sample database; wherein, the initial sample database includes: multiple sets of sample data and at least two performance index parameters of the compressor corresponding to each set of sample data, and each set of sample data includes: the geometric control parameters of the flow path and the airfoil geometry parameters of the stator; The artificial neural network model is trained based on the initial sample database to obtain a prediction model; wherein, the prediction model is used to predict each set of input sample data and output at least two performance index parameters corresponding to the set of sample data; The prediction model is solved by multi-objective optimization so that each of the at least two performance index parameters output by the prediction model reaches its Pareto optimal solution set. In the Pareto optimal solution set, the performance index parameter with the best overall performance is selected, and the optimization region on the flow passage of the compressor is integrated and optimized based on the sample data corresponding to the selected performance index parameter.

[0008] Secondly, this application provides an integrated optimization device for compressor flow path design and stator end bending, comprising: The parameter acquisition unit is used to acquire the geometric control parameters of the flow path and the airfoil geometry parameters of the stator in the optimized region of the flow path of the compressor, and to generate a parameterized model based on the geometric control parameters of the flow path and the airfoil geometry parameters of the stator. The flow field simulation unit is used to perform structured mesh generation on the parameterized model and to perform flow field numerical simulation on the meshed parameterized model to obtain at least two performance index parameters of the compressor corresponding to the parameterized model. A database unit is used to generate an initial sample database by performing parameter sampling processing on the geometric control parameters of the flow path and the airfoil geometry parameters of the stator based on the parameterized model; wherein, the initial sample database includes: multiple sets of sample data and at least two performance index parameters of the compressor corresponding to each set of sample data, and each set of sample data includes: the geometric control parameters of the flow path and the airfoil geometry parameters of the stator; The training unit is used to train the artificial neural network model based on the initial sample database to obtain a prediction model; wherein the prediction model is used to predict each set of input sample data and output at least two performance index parameters corresponding to the set of sample data. The optimization unit is used to perform multi-objective optimization on the prediction model so that each of the at least two performance index parameters output by the prediction model reaches its Pareto optimal solution set. An integrated optimization design unit is used to select the performance index parameter with the best overall performance from the Pareto optimal solution set, and to perform integrated optimization design on the optimization area of ​​the compressor flow passage based on the sample data corresponding to the selected performance index parameter.

[0009] Thirdly, this application provides an electronic device, including a memory and a processor; The memory is used to store computer programs; The processor is configured to, when executing the computer program, implement the integrated optimization method for compressor flow path design and stator end bending as described in the first aspect.

[0010] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the integrated optimization method for compressor flow path design and stator end bending as described in the first aspect.

[0011] This application provides an integrated optimization method, apparatus, electronic device, and readable storage medium for compressor flow path design and stator end bending. It parameterizes the optimization region and generates a parameterized model, then generates an initial sample database based on the parameterized model, and trains a predictive model using this database. The predictive model replaces fluid dynamics simulation, reducing the optimization cycle and significantly improving optimization design efficiency. By simultaneously and collaboratively optimizing flow path design and stator optimization, rather than through isolated optimization of single components, it effectively solves local flow loss problems such as secondary flow and airflow separation on blade surfaces, widening the compressor's stability margin and significantly improving its aerodynamic performance. The predictive model is adaptable to performance predictions under different compressor operating conditions (such as different speeds and flow rates). With only a small number of additional operating condition samples, it can be quickly extended to full-condition optimization, greatly improving the versatility and reusability of the technical solution. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating an integrated optimization method for compressor flow path design and stator end bending provided in an embodiment of this application.

[0013] Figure 2 This is a schematic diagram of the compressor meridional view and the distribution of shape control points in the embodiments of this application.

[0014] Figure 3 This is a schematic diagram showing the relative positions of the parameterized cross sections of the stator in the embodiments of this application.

[0015] Figure 4 This is a schematic diagram of the mesh generation structure of the parametric model in the embodiments of this application.

[0016] Figure 5 This is a comparison chart of numerical simulation results and experimental values ​​in the embodiments of this application.

[0017] Figure 6 This is a two-dimensional scatter plot of the design point optimization results in the embodiments of this application.

[0018] Figure 7 This is a schematic diagram comparing the flow path and stator vane shape of the prototype and the optimal solution in the embodiments of this application.

[0019] Figure 8 This is a schematic diagram comparing the compressor characteristics of the prototype and the optimal solution in the embodiments of this application.

[0020] Figure 9 This is a schematic diagram showing the distribution of the installation angle, intake angle, and angle of attack along the blade height in the embodiments of this application.

[0021] Figure 10 This is a schematic diagram of the streamline distribution at the stator endwall in an embodiment of this application.

[0022] Figure 11 This is a schematic diagram of the absolute Mach number and entropy distribution at the stator endwall in an embodiment of this application.

[0023] Figure 12 This is a schematic diagram showing the distribution of static pressure coefficient, vorticity, and total pressure loss coefficient in the embodiments of this application.

[0024] Figure 13 This is a schematic diagram of the structure of an integrated optimization device for compressor flow path design and stator end bending provided in an embodiment of this application.

[0025] Figure 14 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0026] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, specific embodiments of this application are described in detail below with reference to the accompanying drawings. Although some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the accompanying drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0027] It should be understood that the steps described in the method embodiments of this application may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this application is not limited in this respect.

[0028] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the description below. It should be noted that the concepts of "first," "second," etc., mentioned in this application are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0029] It should be noted that the terms "one" and "more" used in this application are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0030] The names of the messages or information exchanged between multiple devices in the embodiments of this application are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0031] In related technologies, the core function of a compressor is to compress a gaseous working medium, thereby increasing its pressure and energy levels. In the compressor's structure, the flow passage enables efficient flow and compression of the gaseous working medium from inlet to outlet. It serves as the gas passage through each stage of the compressor (a unit consisting of a row of rotor blades and a row of stator blades following it), and is also the core object of compressor aerodynamic optimization.

[0032] Near the endwalls of the flow passage (the endwalls are the boundary walls of the flow passage, forming the upper and lower boundaries of the gas flow, and their function is to limit the radial range of the gas and guide the gas to flow along a predetermined path), the gas flow field structure is complex and variable, with various unstable flow phenomena such as endwall boundary layer separation, corner flow, secondary flow, and leakage flow. These unstable flow phenomena are important factors hindering the development of high-performance compressors.

[0033] To suppress unstable flow near the endwalls and achieve higher compressor load, efficiency, and stability margin, many researchers have focused on compressor flow path design and blade optimization. Historically, flow path design and blade optimization have been conducted independently, resulting in insufficient parameterization at the blade-endwall junction. Furthermore, independently designed blades often exhibit flow interference at the endwall junction, limiting the effectiveness of mitigating flow problems near the endwalls. In recent years, to overcome the limitations of existing independent designs and further suppress unstable flow near the endwalls to improve compressor performance, integrated design principles, guided by aircraft wing-body blending design, have been preliminarily explored in end-area flow control. However, there is a lack of research on integrated optimization design of flow path design and stator end bends.

[0034] To address the problem that existing technologies often involve relatively independent flow path design and blade optimization, making it difficult to effectively coordinate and control endwall flow, thus limiting the improvement of compressor stability margin and aerodynamic performance, this embodiment provides an integrated optimization method, apparatus, electronic device, and storage medium for compressor flow path design and stator end bend.

[0035] like Figure 1 As shown, the present application provides an integrated optimization method for compressor flow path design and stator end bending, including steps S100-S600: S100: Within the optimized region of the compressor's flow passage, the geometric control parameters of the flow passage design and the airfoil geometry parameters of the stator are obtained respectively, and a parameterized model is generated based on the geometric control parameters of the flow passage design and the airfoil geometry parameters of the stator.

[0036] In this step, the optimal region for the compressor's flow path is first determined. For example, the optimal region could be one or more stages of the flow path.

[0037] This application parameterizes the flow path and stator of the optimization region and constructs a parameterized model, thereby transforming the complex compressor geometry and aerodynamic characteristics into a quantifiable and controllable parameterized expression, providing a unified model carrier for subsequent simulation and optimization, and reducing the dimensionality of optimization variables.

[0038] In an optional implementation, specific steps for obtaining the geometric control parameters of the flow path are provided, including steps S101 and S102: S101: Determine the meridional curve of the flow path, perform parametric control on the meridional curve, and obtain multiple control points and the coordinate parameters corresponding to each control point.

[0039] S102: Determine the coordinate parameters of each control point as the geometric control parameters of the flow pattern.

[0040] In this optional embodiment, the meridional plane refers to the plane passing through the compressor axis in the compressor structure. The meridional curve of the flow path includes: the intersection of the meridional plane and the hub (hub flow path) and the intersection of the meridional plane and the casing (casing flow path). The hub flow path and the casing flow path are parametrically controlled, i.e., several control points are set on each hub flow path and each casing flow path. A coordinate system is established with the compressor axis as the axial direction and the perpendicular line between the control points and the compressor axis as the radial direction. The flow path shape is changed by adjusting the axial and radial coordinates of several control points. Specifically, in this optional embodiment, the coordinates (axial and radial coordinates) of the control points on each hub flow path and each casing flow path are controlled by an interpolation algorithm (such as spline curves or Bézier curves) to change the shape of the meridional curve, thereby changing the flow path shape.

[0041] As described above, in this optional embodiment, the difficulty of optimizing the geometric shape can be reduced by parametrically controlling the flow path. During parametric control, setting several control points and adjusting their coordinates using an interpolation algorithm allows for adjustment of the flow path shape while ensuring the smoothness and rationality of the modification.

[0042] In an optional implementation, specific steps for obtaining the stator's airfoil geometry parameters are provided, including steps S103 and S104: S103: Divide the stator blade into multiple sections along the blade height direction of the stator blade.

[0043] S104: Determine the inlet angle, outlet angle, and mounting angle of each of the aforementioned sections as the stator's airfoil geometry parameters.

[0044] In this step, the inlet angle is the angle between the tangent of the leading edge of the blade's mid-curve and the compressor's axial direction; the outlet angle is the angle between the tangent of the trailing edge of the blade's mid-curve and the compressor's axial direction; and the mounting angle is the angle between the chord line of the blade section and the compressor's axial direction.

[0045] In this optional embodiment, in the compressor, the stator blade height direction refers to the radial height of the stator from the side closest to the compressor shaft to the outer side furthest from the compressor shaft. Because the velocity and pressure distribution of the airflow inside the compressor varies greatly at different blade height positions, the aerodynamic characteristics of the entire blade cannot be accurately described using parameters from a single cross-section. Therefore, it is necessary to divide the stator into multiple cross-sections perpendicular to the blade height direction along the blade height direction, and select several cross-sections for optimization.

[0046] For each cross-section, the airfoil mid-curve and airfoil chord of the stator within that cross-section are first determined. The airfoil mid-curve is a curve connecting the centers of all inscribed circles within the cross-section. In this optional embodiment, a Bézier curve is used to define the airfoil mid-curve for each cross-section. By controlling several control points on the Bézier curve, the curvature and degree of bending of the airfoil mid-curve can be precisely changed, thereby enabling aerodynamic shaping and optimization of the stator within the cross-section. The airfoil chord is a straight line connecting the leading edge (LE) and trailing edge (TE) of the airfoil. Next, based on the airfoil mid-curve and airfoil chord, several airfoil geometric parameters on the stator, such as the inlet angle, outlet angle, and mounting angle, are determined.

[0047] In this design, the angle between the tangent of the blade's mid-curve at the leading edge (also known as the Bessel control line) and the compressor axis is the inlet angle; the angle between the tangent of the blade's mid-curve at the trailing edge (TE) and the compressor axis is the outlet angle; and the angle between the blade's chord and the compressor axis is the installation angle. Finally, in this optional embodiment, the installation angle is selected as the blade geometry parameter to be optimized on the stator, and the installation angle is randomly perturbed.

[0048] It should be noted that the installation angle determines the stator's guiding effect on gas flow. A larger installation angle results in a greater turning angle when the airflow passes through the stator, which can easily lead to negative angle of attack flow and cause the airflow to detach at the blade holder. This may increase the pressure ratio, but the flow loss is relatively large. Conversely, a smaller installation angle results in a gentler turning angle when the airflow passes through the stator, with less flow loss but a limited increase in pressure ratio.

[0049] In an optional implementation, specific steps are provided for generating a parameterized model based on the geometric control parameters of the flow path and the airfoil geometry parameters of the stator, including steps S105-S107: S105: Based on the geometric control parameters of the flow path design, a closed two-dimensional flow path profile is formed, and the two-dimensional flow path profile is stretched along the circumference of the compressor to obtain the flow path model.

[0050] In this step, the hub streamline and casing streamline of the compressor can be accurately fitted using geometric control parameters. Connecting the starting point (compressor inlet) and ending point (compressor outlet) of the hub streamline and casing streamline respectively forms a closed two-dimensional flow channel profile along the axial direction. This two-dimensional flow channel profile is the meridional curve of the compressor flow channel. Rotating and stretching this two-dimensional flow channel profile along the circumference of the compressor (the direction around the compressor axis) generates a hollow three-dimensional flow channel model.

[0051] S106: Interpolate the leaf geometry parameters of the stator along the leaf height direction to obtain the stator blade model.

[0052] In this step, linear interpolation and spline interpolation methods are used to interpolate and calculate the discrete cross-sectional airfoil geometric parameters along the stator blade height direction. This ensures a smooth transition of the blade profile from the hub to the casing, avoiding flow losses caused by abrupt changes in local parameters. Using the installation angle corresponding to each position after interpolation as a constraint, and combining it with the preset airfoil mid-curve, airfoil cross-sections at each blade height position are generated. Then, all airfoil cross-sections are spliced ​​and merged along the blade height direction to form a complete three-dimensional stator blade model.

[0053] S107: The parameterized model is obtained by embedding the stator blade model into the flow channel model.

[0054] In this step, the generated stator blade model is precisely placed at the designated position on the flow channel model according to the circumferential arrangement rules of the compressor (such as the number of blades and installation phase), ensuring that the hub end of the stator is in contact with the hub wall of the flow channel and the casing end is in contact with the casing wall of the flow channel, thus obtaining a parametric model. The final integrated parametric model has a flow channel shape determined by the flow geometry control parameters, and blade morphology determined by the airfoil profiles of each section of the stator. During optimization, only the geometry control parameters of the flow channel and the airfoil geometry parameters of the stator need to be adjusted, and the parametric model will automatically update the corresponding geometry, eliminating the need for remodeling. This allows minor adjustments to the parameters to be directly mapped to changes in the model morphology, avoiding repetitive operations in traditional modeling. In this optional embodiment, flow channel modeling is driven by geometry control parameters, and blade modeling is driven by airfoil geometry parameters. The two models are then integrated to form a complete parametric model of the compressor flow channel that can be used for simulation and optimization.

[0055] S200: Perform structured mesh generation on the parameterized model and numerical simulation of the flow field on the meshed parameterized model to obtain at least two performance index parameters of the compressor corresponding to the parameterized model.

[0056] In this step, a flow field numerical simulation (CFD simulation) is performed on the parameterized model to establish the correlation between the parameterized model and compressor performance parameters (such as compressor efficiency and pressure ratio). This facilitates the direct acquisition of compressor performance parameters corresponding to the modified parameterized model after subsequent modifications.

[0057] Understandably, in numerical simulation (CFD) of flow field using a parametric model, the parametric model needs to be meshed first. Approximate solutions are then performed on each mesh cell to obtain performance parameters such as velocity, pressure, and temperature within that cell. Finally, the flow field distribution of the entire parametric model is pieced together using the calculation results from all mesh cells.

[0058] In an optional implementation, specific steps are provided for structural mesh generation of the parameterized model and flow field numerical simulation of the meshed parameterized model, including steps S201-S202: S201: The parametric model is divided into multiple mesh blocks using a composite topology approach.

[0059] S202: Perform flow field numerical simulation on each of the grid blocks to obtain at least two fluid parameters for each grid block; wherein, the fluid parameters of multiple grid blocks are determined as performance index parameters of the parameterized model.

[0060] It should be noted that composite topology refers to the use of multiple structured mesh topologies, such as O-type and H-type topologies, within the same parametric model. Composite topology divides the parametric model of the compressor flow path into multiple independent mesh blocks.

[0061] Furthermore, after dividing the parametric model into multiple grid blocks, the turbulent viscosity coefficient of the parametric model can be corrected by the turbulent model to obtain the turbulent viscosity coefficient of each grid block on the parametric model, thereby obtaining the distribution of the turbulent viscosity coefficient of the parametric model. This makes it easier to obtain flow field numerical simulation results that are more consistent with the real turbulent diffusion and dissipation characteristics in the flow field numerical simulation.

[0062] After the parametric model is divided into multiple grid blocks, a CFD (Computational Fluid Dynamics) solver (such as Fluent or CFX) is used to discretize and solve each grid block in the parametric model to obtain the fluid parameters within each grid block. The fluid parameters within each grid block include, but are not limited to: flow velocity (axial, radial, and circumferential components), pressure, and temperature.

[0063] Fluid parameters from multiple grid blocks are transferred and coupled through flux conservation conditions at the grid block interfaces. The CFD solver ensures continuity of parameters such as velocity and pressure at the interfaces of adjacent grid blocks, stitching them together to form complete flow field data. The spliced ​​full flow field data is then used to calculate the compressor's performance parameters using engineering formulas, including: total pressure ratio (outlet total pressure / inlet total pressure), adiabatic efficiency (actual enthalpy rise / ideal enthalpy rise), surge margin, and total pressure loss coefficient.

[0064] S300: Based on the parameterized model, the geometric control parameters of the flow path and the airfoil geometry parameters of the stator are sampled to generate an initial sample database; wherein, the initial sample database includes: multiple sets of sample data and at least two performance index parameters of the compressor corresponding to each set of sample data, and each set of sample data includes: the geometric control parameters of the flow path and the airfoil geometry parameters of the stator.

[0065] In this step, the geometric control parameters of the flow path and the airfoil geometry parameters of the stator are sampled. Each set of sample data is input into the parameterized model to obtain the parameterized model corresponding to the current set of sample data. Based on the parameterized model and at least two corresponding compressor performance parameters obtained in step S200, the performance parameters corresponding to the parameterized model of the current set of sample data are determined. Each set of input sample data and its corresponding performance parameters are associated and stored to obtain an initial sample database. By constructing the initial sample database, sufficient datasets are provided for the training and validation of the artificial neural network model.

[0066] It should be noted that during the sampling process, multiple sets of sample points are generated within the adjustable range of sampling parameters to ensure the uniformity and representativeness of the samples. This avoids model prediction bias caused by uneven sample distribution and improves the generalization ability of subsequent artificial neural network models. Specifically, engineered experimental design methods can be used to ensure the uniformity of samples within the parameter space, such as Latin hypercube sampling (LHS), orthogonal experimental design, or uniform sampling. By using a small number of samples, the performance characteristics of a high-dimensional parameter space can be characterized, while also ensuring representativeness.

[0067] In an optional implementation, specific steps are provided for generating an initial sample database by performing parameter sampling processing on the geometric control parameters of the flow path and the airfoil geometry parameters of the stator based on the parameterized model, including steps S301-S303: S301: Sample and combine the geometric control parameters of the flow path and the airfoil geometric parameters of the stator to obtain multiple sets of sample data.

[0068] S302: Based on the parameterized model, determine at least two performance index parameters corresponding to each group of sample data.

[0069] S303: An initial sample database is constructed based on multiple sets of sample data and at least two performance index parameters of the compressor corresponding to each set of sample data.

[0070] In this optional embodiment, the geometric control parameters of the flow path and the stator airfoil geometry parameters can be sampled based on Latin hypercube sampling, orthogonal experimental design, or uniform sampling, respectively. It is understood that when sampling, the value ranges of the flow path geometric control parameters and stator airfoil geometry parameters need to be determined. The value range is based on the structural parameters of the existing design, and can be directly referenced from the value ranges in existing published literature. Alternatively, according to optimization requirements, a perturbation range can be set based on the existing value range (the perturbation range can be set between ±5% and ±30% of the baseline value). The extracted geometric control parameter samples are combined with the airfoil geometry parameter samples (or combined according to the experimental design table) to generate multiple sets of independent sample data, where each set of sample data corresponds to a unique flow path shape and stator installation angle. Each set of combined sample data is input into the parametric model to obtain the parametric model corresponding to the current set of sample data. Based on the parametric model and at least two corresponding compressor performance parameters, determine the performance parameters of the parametric model corresponding to the current group of sample data; associate and store each group of input sample data with its corresponding performance parameters to obtain an initial sample database. The initial sample database can be directly used for training artificial neural network models.

[0071] S400: The artificial neural network model is trained based on the initial sample database to obtain a prediction model; wherein the prediction model is used to predict each set of input sample data and output at least two performance index parameters corresponding to the set of sample data.

[0072] In this step, based on the initial sample database, the sample data in the initial sample database is used as the input to the artificial neural network model, and the performance index parameters in the initial sample database are used as the target output (i.e., labels) of the artificial neural network model. By back-optimizing the model parameters such as weights, a predictive model is obtained after training. The predictive model can quickly predict performance index parameters based on sample data. This step replaces time-consuming CFD simulation with a predictive model, reducing performance prediction time from hours to milliseconds, significantly improving optimization efficiency, achieving non-linear fitting between parameters and performance, and solving the problem of insufficient accuracy of traditional empirical formulas.

[0073] In an optional implementation, specific steps are provided for training an artificial neural network model based on the initial sample database to obtain a prediction model, including steps S401-S404: S401: Normalize the sample data and performance index parameters in the initial sample database.

[0074] In the normalization process, the normalized parameter results are calculated separately for each type of parameter in the sample data and performance index parameters. Then, the normalization process is performed on all parameter results to achieve the transformation of all sample data and performance index parameters into normalized form, thus avoiding interference between different units.

[0075] S402: Input the normalized sample data into the artificial neural network model to obtain the prediction index parameters.

[0076] In this step, the sample data is weighted and summed, activated, and passed layer by layer through the neurons of the artificial neural network model. Finally, normalized prediction index parameters are obtained from the output layer, which are the predicted performance index values ​​of the artificial neural network model for the sample data. It should be noted that in this embodiment, the artificial neural network model uses a convolutional neural network (CNN), which includes an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer.

[0077] S403: Determine the loss value between the predicted index parameters and the normalized performance index parameters using a loss function.

[0078] S404: If the loss value is greater than or equal to a preset threshold, the weights and biases of the artificial neural network model are updated using the gradient descent algorithm, and the normalized sample data is input into the updated artificial neural network model to obtain new prediction index parameters and a redefined loss value, until the redefined loss value is less than the preset threshold.

[0079] As we can understand, the loss function is a mathematical function that quantifies the difference between the prediction metric parameters and the performance metric parameters. It calculates the loss value (e.g., mean squared error or mean absolute error) between the prediction metric parameters and the performance metric parameters. This loss value provides an evaluation metric for optimizing the artificial neural network model. A larger loss value indicates a greater deviation between the artificial neural network model's prediction and the actual situation, requiring further optimization; a smaller loss value indicates higher prediction accuracy. After each update of the weights and biases of the artificial neural network model, all normalized sample data are re-inputted into the updated model, and forward propagation is performed to calculate the new loss value. This process continues until the prediction accuracy of the artificial neural network model meets the prediction requirements, at which point iterative training stops, and the artificial neural network model at this point becomes the final usable prediction model.

[0080] By iteratively optimizing the model, the prediction bias of the artificial neural network is continuously reduced, allowing the model to fully learn the nonlinear mapping between "parameters" and "performance" in the sample database, ultimately achieving the required accuracy for prediction. The trained prediction model can output performance metrics for any input parameters within milliseconds, completely replacing time-consuming CFD simulations and providing efficient support for subsequent multi-objective optimization.

[0081] S500: Perform multi-objective optimization on the prediction model to achieve a Pareto optimal solution set for each of the at least two performance index parameters output by the prediction model.

[0082] In this step, multiple performance indicators of the compressor (such as maximizing efficiency, maximizing pressure ratio, and maximizing surge margin) are interdependent. For example, increasing the pressure ratio may lead to a decrease in efficiency, while expanding the surge margin may sacrifice peak efficiency. Using these multiple performance parameters as optimization objectives, the prediction model is optimized using a multi-objective approach, resulting in a Pareto optimal solution set where each objective is balanced. The essence of a Pareto optimal solution is that no single solution can improve a particular performance indicator without decreasing at least one of the other performance indicators. The set of all Pareto optimal solutions constitutes the Pareto optimal solution set. By using the Pareto optimal solution set, the limitations of single-objective optimization are overcome, taking into account the overall performance of the compressor, avoiding situations where one performance indicator is optimal while other indicators deteriorate, and providing multiple optimal solutions for selection.

[0083] In an optional implementation, specific steps are provided for multi-objective optimization of the prediction model to ensure that each of the at least two performance index parameters output by the prediction model reaches a Pareto optimal solution set, including steps S501-S503: S501: Select multiple sets of sample data from the initial sample database as original samples; wherein, the performance index parameters corresponding to each set of sample data are the original performance parameters.

[0084] In this step, multiple sets of sample data with performance indicators within the acceptable engineering range (e.g., efficiency ≥ 0.88, pressure ratio ≥ 1.8, surge margin ≥ 15%) are selected from the existing initial sample database and used as the basis for cross-combination. This not only eliminates invalid sample data from the initial sample database (such as parameter combinations with severe airflow separation or extremely poor performance), but also reduces the computational load of cross-combination, avoids invalid sample data interfering with the selection of the optimal solution, and ensures that the original samples themselves have a certain performance foundation, making it easier to generate high-quality design samples after cross-combination. The selected samples are called the original samples, and their corresponding CFD simulation performance parameters are called the original performance parameters.

[0085] It should be noted that each set of sample data includes: flow path geometric control parameters and stationary vane geometry parameters. Therefore, the original sample, composed of the sample data, also includes flow path geometric control parameters and stationary vane geometry parameters.

[0086] S502: The original samples are cross-combined to obtain multiple sets of design samples.

[0087] In this step, all geometric control parameters and airfoil geometric parameters in the original sample data are fully paired to ensure complete coverage of the parameter space. During this full pairing, it is necessary to first determine the sample set J of geometric control parameters and the sample set Q of airfoil geometric parameters contained in the original samples. Each geometric control parameter in sample set J is then paired with each airfoil geometric parameter in sample set Q to obtain multiple sets of design samples.

[0088] It should be noted that the cross-referenced design samples must meet geometric feasibility requirements (e.g., no flow channel interference, blade installation angle within a reasonable aerodynamic range); otherwise, they must be marked as invalid samples. Each set of design samples is input into the prediction model, and at least two performance index parameters output by the prediction model are determined as design performance parameters. By cross-combining the original samples, new design samples not present in the initial sample library are generated, enabling the discovery of high-quality, unsampled regions in the parameter space and improving the diversity of optimization schemes.

[0089] Each set of design samples is input into the trained prediction model, which outputs at least two performance index parameters. These at least two performance index parameters are then determined as the design performance parameters. By using the prediction model to replace CFD simulation to evaluate the performance of the design samples, the optimization cycle is shortened from "days" to "hours," significantly improving efficiency.

[0090] S503: Select target performance index parameters that meet the performance requirements from the original performance parameters and the design performance parameters, and determine the original sample or design sample corresponding to the target performance index parameters as the optimal solution set; wherein, the performance requirement is: each performance index parameter among at least two performance index parameters reaches its Pareto optimal solution.

[0091] All original and design performance parameters are integrated to form a dataset containing sample parameter combinations and multiple performance indicators. In this dataset, all samples that are not dominated by any other samples (dominance means that two samples are mutually superior or inferior, and no two samples are always superior to or equal to the other; for example, each set of sample data includes geometric control parameters and airfoil geometry parameters. For two sample data X and Y, if the geometric control parameters and airfoil geometry parameters of sample data X are both superior to or equal to the geometric control parameters and airfoil geometry parameters of sample data Y, then sample data X is superior to or equal to sample data Y; if sample data X and sample data Y are mutually superior or inferior in both geometric control parameters and airfoil geometry parameters, then sample data X and sample data Y are not mutually dominant) have corresponding performance parameters that are the target performance indicator parameters. These original or design samples together constitute the final optimal solution set, which is the Pareto optimal solution set.

[0092] S600: Select the performance index parameter with the best overall performance from the Pareto optimal solution set, and perform integrated optimization design on the optimization region of the compressor flow passage based on the sample data corresponding to the selected performance index parameter.

[0093] In this step, based on engineering requirements (such as prioritizing efficiency under a certain operating condition or stability under another), the performance index parameters with the best overall performance are selected from the Pareto solution set. Alternatively, multi-objective performance can be transformed into a single comprehensive evaluation value, thereby achieving quantitative ranking and selecting the optimal performance index parameters. Sample data is obtained by back-correlation of the performance index parameters. Based on the sample data, the geometry of the compressor flow passage is corrected to achieve precise optimization of the compressor design, ultimately improving its aerodynamic performance or operational stability.

[0094] In an optional implementation, specific steps are provided for selecting the performance index parameter with the best overall performance from the Pareto optimal solution set, including steps S601-S602: S601: Calculate the comprehensive performance evaluation value corresponding to each solution unit in the Pareto optimal solution set using a weighted summation method.

[0095] S602: Determine the performance index parameter corresponding to the solution unit with the largest comprehensive performance evaluation value as the performance index parameter with the best comprehensive performance; wherein, each solution unit is a set of original samples or design samples in the Pareto optimal solution set.

[0096] In this optional implementation, when calculating the comprehensive performance evaluation value, at least two performance indicators (such as adiabatic efficiency, total pressure ratio, and surge margin) output by the prediction model are selected as the basic dimensions for comprehensive evaluation. Because the dimensions and numerical ranges of different performance indicators vary significantly (e.g., efficiency ranges from 0.8 to 0.95, and pressure ratio ranges from 1.5 to 3.0), all performance indicators need to be normalized first to eliminate dimensions. Based on actual engineering requirements, corresponding weight coefficients are assigned to each normalized performance indicator. For each solution unit, the normalized performance indicators are weighted and summed according to their weights to obtain the comprehensive performance evaluation value. By comparing the comprehensive performance evaluation values ​​of all solution units in the Pareto optimal solution set, the solution unit with the largest comprehensive performance evaluation value is selected as the optimal solution in terms of overall performance. In this embodiment, by converting performance indicators into quantitative comprehensive performance evaluation values, the decision-making process becomes objective and reproducible. In addition, by adjusting the weighting coefficients, different engineering design preferences can be quickly adapted. For example, if efficiency is prioritized, the weighting coefficient corresponding to the efficiency value increases; or if stability is prioritized, the weighting coefficient corresponding to the pressure ratio value increases.

[0097] As described above, the integrated optimization method for compressor flow path design and stator end bend provided in this application parameterizes the optimization region and generates a parameterized model. An initial sample database is then generated based on the parameterized model, and a prediction model is trained using this database. The prediction model replaces fluid dynamics simulation, reducing the optimization cycle and significantly improving optimization design efficiency. By simultaneously and collaboratively optimizing the flow path design and stator end bend, rather than through isolated optimization of a single component, the method effectively solves local flow loss problems such as secondary flow at the endwall and airflow separation on the blade surface, widening the compressor's stability margin and significantly improving its aerodynamic performance. The prediction model is adaptable to performance predictions under different compressor operating conditions (such as different speeds and flow rates). Only a small number of operating condition samples are needed to quickly extend the optimization to all operating conditions, greatly improving the versatility and reusability of the technical solution.

[0098] The present application will now be described in conjunction with specific application embodiments, and the specific implementation methods are as follows: 1. Taking a certain type of axial flow compressor as the research object, conduct parametric modeling. For example... Figure 2The diagram shows the meridional view of the compressor and the distribution of its shape control points. Gas enters through inlet 1, flows through the channel between hub line 2 and casing line 6, passes through rotor 3 and stator 4, and exits through outlet 5. In the compressor's axial (Z) and radial (R) coordinate systems, B-spline curves are used to parametrically control the casing endwall line and hub endwall line, respectively. Each curve has 30 control points, and the radial coordinates (R) of these control points are adjusted. i This is used to change the shape of the end wall. The coordinates of the control points for the casing line are S... i =(Z i ,R i The coordinates of the hub line control point are H. i =(Z i ,R i In this embodiment, the coordinates of the control points are used as the geometric control parameters for the flow path design.

[0099] 2. Parameterization of stators. For example... Figure 3 The diagram shows the relative positions of the parametric sections of the stator. Nine stator sections are defined along the blade height direction using linear interpolation. cj ( j =1~9). Subscript j The section number represents each section. c 1 to section c The relative positions of leaf heights of 9 are 0%. h 25% h 50% h 75% h 80% h 85% h 90% h 95% h and 100% h .in, h Let S be the stator's blade height. Project the stator along the radial (Y) axis of the coordinate system. The area occupied by the projection on the casing corresponds to the casing control point S. 19 -S 26 between.

[0100] Each section cj The airfoil's mid-curve is defined using a Bézier curve, where PS represents the pressure surface and SS represents the suction surface. β K This represents the airflow inlet angle. The tangents to the airfoil's mid-curve at the leading edge (LE) and the tangents to the airfoil's mid-curve at the trailing edge (TE) constitute the Bezier control lines of the airfoil within the cross-section. The airfoil geometry parameters within the cross-section include the inlet angle. β j Exit angle α j and installation angle γ j The angle between the tangent of the blade's arc at the leading edge and the compressor axis is the inlet angle. β j The angle between the tangent of the blade's arc at the trailing edge (TE) and the compressor axis is the outlet angle. α j Leaf-shaped chord c The angle between the compressor axis and the compressor axis is the installation angle. γ j Angle of attack i j It is the airflow inlet angle β K With installation angle γ j The angle difference between them, the angle of attack i j This is used to reflect the angle of incidence when gas enters the blade channel. In this embodiment, the blade height is 75%. h Up to 100% h The installation angle of the cross section within the range is used as the airfoil geometry parameter of the stator for co-optimization with the flow path design.

[0101] 3. Generate a parametric model based on geometric control parameters and airfoil geometric parameters. Perform structured mesh generation on the parametric model, such as... Figure 4 As shown, the periodic single-channel computational domain on the parameterized model is divided into O4H (three structured grid topologies: O-type, 4-zone, and H-type) structured grids. H-type grids are used at the inlet and outlet, while O-type grids are used on the blade surfaces. Grid blocks are periodically and perfectly matched. A butterfly-shaped grid is used to control the rotor blade tip clearance, and the grid is further refined. The circumferential transition zone of the blades is solved iteratively using the 4th-order Runge-Kutta method and the finite volume method with a central difference scheme. Multigrid technology, local time steps, and implicit residual generalization are employed to accelerate convergence. The SA turbulence model is used for numerical solutions. The total temperature and total pressure distribution along the blade height under actual experimental conditions are selected as the inlet boundary conditions. An axial inlet mode is selected, and the outlet back pressure is continuously increased to reduce the compressor flow rate until the calculation diverges. The last point of convergence is selected as the minimum flow rate point.

[0102] It should be noted that in the aerodynamic design of compressors, the minimum flow rate point (also known as the near-stall point) is the lower limit of the flow rate for stable compressor operation. When the gas flow rate is below the minimum flow rate point, the compressor will enter a stall or surge state, resulting in severe fluctuations in the compressor's flow field and intensified blade vibration. In this embodiment, when performing Computational Fluid Dynamics (CFD) simulation on the parametric model, the minimum flow rate point can be quickly located without the need for full calculations for all low flow rate conditions.

[0103] Furthermore, when performing CFD simulations on the parametric model, the CFD simulation needs to be validated. First, the performance of the existing compressor needs to be tested using a compressor performance test bench to obtain experimental data. Second, the compressor is parametrically modeled to obtain the target parametric model. Finally, CFD simulation methods are used to numerically simulate the flow field of the target parametric model, obtaining simulation results. By comparing the simulation results with the experimental data, the accuracy and reliability of the CFD simulation method are ensured, providing a reliable computational basis for subsequent optimization.

[0104] like Figure 5 The comparison graph shown here is between the numerical simulation results and the experimental values. Figure 5 (Blue CFD) and experimental data ( Figure 5 The TESTs in the two languages ​​basically overlap, among which, η It is the isentropic efficiency, and π is the total pressure ratio. n cor This is the converted speed of the compressor. At 100% n cor 90% n cor and 70% n cor The maximum total pressure ratio errors were 1.28%, 1.63%, and 1.49%, respectively, and the maximum efficiency errors were 2.61%, 2.44%, and 2.33%, respectively. The flow range of the numerical simulation was smaller than that of the experimental results. Since the errors were all less than 3%, the conclusions obtained by the fluid dynamics method were considered usable and could support the subsequent application of the integrated flow path design and stator design method.

[0105] 4. In this embodiment, isentropic efficiency and total pressure ratio are selected as the optimized operating points. The optimization objective is to significantly improve the compressor's isentropic efficiency and total pressure ratio while maintaining or slightly increasing the flow rate. The optimization design process includes: constructing a database containing 300 initial samples based on Latin hypercube sampling technology; building a nonlinear mapping relationship, i.e., a prediction model, between the geometric control parameters of the endwalls, the airfoil geometry parameters of the stator blades, and the compressor performance index parameters (efficiency, pressure ratio) through an artificial neural network training platform; and using a multi-objective genetic algorithm (such as NSGA-II) to perform multi-objective optimization on the prediction model to find the Pareto optimal solution set that can simultaneously improve efficiency and pressure ratio, and selecting the design scheme with the best overall performance from it.

[0106] During the optimization process, eight geometric control parameters for the end wall were set. i =19~26) Control parameters for free movement along the span, stator airfoil geometry parameters are set to 5 ( j=5~9) Control parameters that can be freely transformed along the positive angle, specifically the range of control parameter variation shown in Table 1.

[0107] Table 1

[0108] Based on 300 original samples in the database, 100 designed samples were generated through multiple iterative design solutions, such as... Figure 6 As shown in the figure. To simultaneously consider compressor efficiency and overall pressure ratio, and to maximize compressor performance, the optimized design sample indicated by the red box in the figure is selected as the optimization scheme (Opt.). A comparison before and after optimization is provided. Figure 7 As shown, Ori. represents the prototype, i.e., the original solution. Opt. represents the optimization, i.e., the optimized solution. Figure 7 It can be seen that at 75% h ~100% h The blade height range, the optimized stator, with the blade height increasing, the blade installation angle increases. Endwall control point S i ( i The radial coordinates (=19~26) fluctuated (partially increasing, partially decreasing). A comparison of the characteristic curves before and after optimization is shown below. Figure 8 As shown, Figure 8 NSP represents the near-stall point. Figure 8 It can be seen that the optimized scheme (Opt.) significantly improves both efficiency across the entire operating range and the total pressure ratio near stall conditions. Specifically, peak efficiency is improved by 2.43%; the total pressure ratio near stall is improved by 1.31%, and efficiency is improved by 3.64%. This demonstrates that the integrated optimization design of the flow path and stator effectively improves the overall performance of the compressor, with the most significant performance improvement observed, especially near stall conditions.

[0109] like Figure 9 The diagram shows the distribution of the installation angle, inlet angle, and angle of attack along the blade height near the stall point. Figure 9 The black dot in the middle represents the mounting angle of the stator in the original design (Ori.). γ j Intake angle β K and attack angle i j Distribution pattern along leaf height. The red box represents the stator installation angle in the optimized scheme (Opt.). γ j Intake angle β K and attack angle i j Distribution pattern along leaf height. Combined with the angle of attack calculation formula.i j =β K -γ i (in, β The airflow inlet angle, γ i (Based on the blade installation angle), the angle of attack can be further derived. i j Differences along the leaf height direction: Within the 0–55% blade height range, the installation angles of the original design (Ori.) and the optimized design (Opt.) are basically the same, with no significant difference in the inlet angle, and their angles of attack are also basically the same. Within the 55%–75% blade height range, the installation angle of the original design (Ori.) is larger than that of the optimized design (Opt.), and their inlet angles are similar, resulting in the original design (Ori.) having a slightly larger angle of attack than the optimized design (Opt.). However, within the 75%–100% blade height range, the installation angle of the original design (Ori.) is smaller than that of the optimized design (Opt.), and there is still no significant difference in the inlet angle, ultimately resulting in the original design (Ori.) having a larger angle of attack than the optimized design (Opt.).

[0110] like Figure 10 The diagram shows the streamline distribution at the stator endwall for the original scheme (Ori.) and the optimized scheme (Opt.). For the original scheme (Ori.)... Figure 10 In section a), the airflow close to the endwall splits upon impacting node N1 on the pressure side of the blade's leading edge: one stream flows downstream, while the other bypasses the blade's leading edge and enters the adjacent stator channel. The upstream flow interacts with the airflow exiting from point N1, forming a saddle point S in the flow field. A leading-edge horseshoe vortex is then generated behind saddle point S. Driven by both the axial reverse pressure gradient and the circumferential pressure gradient, this horseshoe vortex continuously absorbs low-energy fluid around the endwall, eventually forming a large-area corner separation at the stator blade root. Significantly affected by the corner separation, a large-scale counterflow appears near the stator suction surface. After interacting with the mainstream, the counterflow forms a separation line SL, whose starting position is located at node N2.

[0111] After integrated optimization design, the endwall flow field structure of the stator in the optimized scheme (Opt.) was significantly reconstructed. Figure 10(b) Thanks to the optimization scheme (Opt.), the reduced angle of attack of the airflow in the 75%–100% blade height range weakens the generation dynamics of horseshoe vortices, significantly suppressing them at the casing endwall. Simultaneously, the optimized design causes the casing endwall at the stator inlet to contract inwards into the channel, resulting in a smaller channel cross-sectional area. According to the principle of flow conservation in fluid mechanics, the airflow velocity at this location significantly increases, allowing high-speed airflow to directly inject into the separation region at the blade root corner, compensating for the insufficient kinetic energy of the low-energy fluid, breaking the backflow circulation of the separated flow, and further suppressing flow separation. Furthermore, the concentrated shedding vortex (CSV) shifts significantly backwards, and the counterflow phenomenon on the stator suction surface is also significantly improved with flow field optimization, with a substantial reduction in the range and intensity of counterflow, directly causing the starting node N2 of the separation line SL to move towards the casing endwall.

[0112] like Figure 11 The diagram shows the absolute Mach number and entropy distribution (entropy represents entropy, J / (kg K) is joules per kilogram Kelvin) at the stator endwall for the original scheme (Ori.) and the optimized scheme (Opt.).

[0113] It should be noted that, c This indicates the axial chord length of the blade. x % c Indicates the distance from the leading edge of the blade (LE). x The position of the section along the axial chord length.

[0114] exist Figure 11 In the absolute Mach number distribution of the original scheme (Ori.) in (a), it can be seen that from 20% c Up to 80% c A large area of ​​dark blue (low Mach number) reverse flow region appeared near the blade root endwall (represented by the 0 m / s isosurface of the axial velocity), causing the stator to be severely affected by blade root corner separation, resulting in a significant airflow blockage effect. Simultaneously, low-energy fluid accumulated in the corner region and flowed in the opposite direction, forming significant flow blockage and hindering the passage of the main flow, thus greatly reducing the effective flow area in this region. In the entropy distribution of the original scheme (Ori.), from 40%... c To 100% c The large reddish-yellow area in the reversal zone (high entropy value) indicates that severe irreversible energy loss occurred during the separation and reversal process due to friction, vortexing, and dissipation. This loss not only reduces the compressor efficiency but also further exacerbates flow instability.

[0115] exist Figure 11 In the absolute Mach number distribution of the optimal scheme (Opt.) in (b), from 20%c To 100% c The significantly reduced area of ​​the deep blue backflow zone on the root endwall indicates that lowering the blade angle of attack in the 75%–100% blade height range reduced excessive deflection of the airflow at the root, suppressing corner separation. The change in Mach number color in the root region from deep blue to yellowish-green indicates a significant increase in root airflow velocity and effective mitigation of the blockage effect. In the entropy distribution of the optimized scheme (Opt.), the area of ​​the red-yellow high-entropy region at the root significantly decreased from 20%c to 80%c, indicating that energy loss caused by corner separation was effectively controlled and flow field efficiency was improved. However, it should be noted that a significant low-velocity backflow zone appeared at the casing recess.

[0116] like Figure 12 The original scheme (Ori.) and the optimized scheme (Opt.) shown have different static pressure coefficients near the stator endwall. C p vorticity and total pressure loss coefficient ω Distribution map.

[0117] It should be noted that, c This indicates the axial chord length of the blade. x % c Indicates the distance from the leading edge of the blade (LE). x The position of the section along the axial chord length.

[0118] exist Figure 12 In the original scheme (Ori.), the airflow needs to undergo a deceleration and pressurization process as it passes through the stator channel. The static pressure coefficient of the blade surface... C p The chord length along the axis starts from 0%. c To 100% c The pressure shows a gradual increasing trend, but near the endwall, due to the influence of the backflow zone, the pressure increase is significantly insufficient. At 100% c In the trailing edge region, pressure growth even stalls, because the significant energy loss and airflow blockage in the reversing zone hinder the normal deceleration and pressurization process. Near the endwall casing, from 40%... c To 100% c A large bright green area appears within the axial range, indicating a high level of vorticity. This is caused by strong secondary vortices induced by corner separation and backflow, reflecting a high degree of flow instability. Near the endwall, at 80%... c ~100% c In the trailing edge region, a large area of ​​red and yellow appears, and the total pressure loss coefficient... ω A value close to 1.2 to 1.3 indicates that the accumulation and dissipation of low-energy fluid in the reverse flow region has resulted in severe energy loss, directly leading to the deterioration of aerodynamic performance.

[0119] exist Figure 12 In the optimized scheme (Opt.) of (b), the static pressure coefficient of the blade surface along the entire axial chord length is... C p The static pressure coefficient is generally higher than that of the original scheme (Ori.). C p Especially in 60% c ~100% c In the main flow region, the pressure increase is more uniform and greater, indicating a more efficient deceleration and pressurization process, and improved flow capacity. Near the endwall casing, the bright green high-vorticity region almost disappears, with only a small amount of blue-green area appearing in localized locations. The vorticity level drops significantly, indicating effective suppression of corner separation and secondary flow, and a significant improvement in flow stability. Near the endwall, at 80%... c ~100% c In the trailing edge region, the area of ​​the red and yellow high-loss zone is significantly reduced, and the total pressure loss coefficient... ω The decrease from 1.3 to below 0.9 indicates that energy loss has been effectively controlled. This is the core reason for the significant improvement in isentropic efficiency of the optimized scheme.

[0120] As described above, compared to the original scheme (Ori.), the optimized scheme (Opt.) significantly improves efficiency across the entire operating range and the total pressure ratio near stall conditions. Specifically, peak efficiency is improved by 2.43%, and the total pressure ratio near stall is improved by 1.31%, resulting in a 3.64% improvement in overall efficiency.

[0121] Within the 75%–100% blade height range, the increased stator installation angle in the optimized scheme (Opt.) leads to a relative decrease in the airflow angle of attack in this region. This not only weakens the horseshoe vortex intensity at the stator leading edge but also, combined with the inward contraction of the optimized flow path, effectively suppresses corner separation and blockage effects by increasing the airflow velocity near the endwall and injecting kinetic energy into the corner region. This, in turn, reduces the backflow region, high-entropy region, and total pressure loss, ultimately achieving an improvement in compressor peak efficiency and near-stall pressure ratio and efficiency.

[0122] like Figure 13 As shown in the embodiment of this application, a compressor design optimization device 1300 includes: The parameter acquisition unit 1310 is used to acquire the geometric control parameters of the flow path and the airfoil geometry parameters of the stator in the optimized region of the flow path of the compressor, and generate a parameterized model based on the geometric control parameters of the flow path and the airfoil geometry parameters of the stator. The flow field simulation unit 1320 is used to perform structured meshing of the parameterized model and to perform flow field numerical simulation on the meshed parameterized model to obtain at least two performance index parameters of the compressor corresponding to the parameterized model. Database unit 1330 is used to generate an initial sample database by performing parameter sampling processing on the geometric control parameters of the flow path and the airfoil geometry parameters of the stator based on the parameterized model; wherein, the initial sample database includes: multiple sets of sample data and at least two performance index parameters of the compressor corresponding to each set of sample data, and each set of sample data includes: the geometric control parameters of the flow path and the airfoil geometry parameters of the stator; Training unit 1340 is used to train an artificial neural network model based on the initial sample database to obtain a prediction model; wherein, the prediction model is used to predict each set of input sample data and output at least two performance index parameters corresponding to the set of sample data; The optimization unit 1350 is used to perform multi-objective optimization on the prediction model so that each of the at least two performance index parameters output by the prediction model reaches its Pareto optimal solution set. The integrated optimization design 1360 is used to select the performance index parameter with the best overall performance from the Pareto optimal solution set, and to perform integrated optimization design on the optimization area of ​​the compressor flow passage based on the sample data corresponding to the selected performance index parameter.

[0123] In an alternative embodiment, the parameter acquisition unit 1310 further includes: The flow path module is used to determine the meridional curve of the flow path shape, perform parameterized control on the meridional curve to obtain multiple control points and coordinate parameters corresponding to each control point, and determine the coordinate parameters of each control point as the geometric control parameters of the flow path shape.

[0124] The blade module is used to divide the stator into multiple sections along the blade height direction, and to determine the inlet angle, outlet angle, and mounting angle of each section as the stator's airfoil geometric parameters; wherein, the inlet angle is the angle between the tangent of the leading edge of the airfoil's mid-curve and the compressor axis; the outlet angle is the angle between the tangent of the trailing edge of the airfoil's mid-curve and the compressor axis; and the mounting angle is the angle between the airfoil chord of the section and the compressor axis.

[0125] The model module is used to form a closed two-dimensional flow channel profile based on the geometric control parameters of the flow path shape, stretch the two-dimensional flow channel profile along the compressor circumference to obtain the flow channel model; interpolate the blade geometry parameters of the stator along the blade height direction to obtain the stator blade model; and embed the stator blade model into the flow channel model to obtain the parameterized model.

[0126] In an alternative embodiment, the flow field simulation unit 1320 further includes: The simulation module is used to divide the parameterized model into multiple grid blocks using a composite topology approach; to perform flow field numerical simulation on each grid block to obtain at least two fluid parameters for each grid block; wherein, the fluid parameters of the multiple grid blocks are determined as performance index parameters of the parameterized model.

[0127] In an alternative embodiment, the database unit 1330 further includes: The data module is used to sample and combine the geometric control parameters of the flow path and the airfoil geometric parameters of the stator to obtain multiple sets of sample data; based on the parameterized model, at least two performance index parameters corresponding to each set of sample data are determined; and an initial sample database is constructed based on the multiple sets of sample data and the at least two performance index parameters of the compressor corresponding to each set of sample data.

[0128] In an alternative embodiment, the training unit 1340 further includes: The training module is used to normalize the sample data and performance index parameters in the initial sample database; input the normalized sample data into the artificial neural network model to obtain the predicted index parameters; determine the loss value between the predicted index parameters and the normalized performance index parameters through a loss function; if the loss value is greater than a preset threshold, update the weights and biases of the artificial neural network model through the gradient descent algorithm, and input the normalized sample data into the updated artificial neural network model to obtain new predicted index parameters and a redefined loss value, until the redefined loss value is less than the preset threshold.

[0129] In an alternative embodiment, the optimization unit 1350 further includes: The set module is used to select multiple sets of sample data as original samples from the initial sample database; wherein the performance index parameter corresponding to each set of sample data is the original performance parameter; the original samples are cross-combined to obtain multiple sets of design samples, and each set of design samples is input into the prediction model to determine at least two performance index parameters output by the prediction model as design performance parameters; target performance index parameters that meet the performance requirements are selected from the original performance parameters and the design performance parameters, and the original samples or design samples corresponding to the target performance index parameters are determined as the optimal solution set; wherein the performance requirement is: each of the at least two performance index parameters reaches its Pareto optimal solution.

[0130] In one alternative embodiment, the integrated optimization design 1360 further includes: The optimization module is used to calculate the comprehensive performance evaluation value corresponding to each solution unit in the Pareto optimal solution set using a weighted summation method; and to determine the performance index parameter corresponding to the solution unit with the largest comprehensive performance evaluation value as the performance index parameter with the best comprehensive performance; wherein, each solution unit is a set of original samples or design samples in the Pareto optimal solution set.

[0131] like Figure 14 As shown in the embodiment of this application, an electronic device 1400 includes a memory 1420 and a processor 1410; the memory 1420 is used to store a computer program; the processor 1410 is used to implement the model reasoning method as described above when the computer program is executed.

[0132] Alternatively, an electronic device 1400 includes a memory 1420 and a processor 1410 coupled to the memory 1420; the memory 1420 is configured to store a computer program; and the processor 1410 is configured to perform the following operations when the computer program is executed: Within the optimized region of the compressor's flow passage, the geometric control parameters of the flow passage design and the airfoil geometry parameters of the stator are obtained respectively, and a parameterized model is generated based on the geometric control parameters of the flow passage design and the airfoil geometry parameters of the stator. The parameterized model is subjected to structured mesh generation, and the flow field of the meshed parameterized model is numerically simulated to obtain at least two performance index parameters of the compressor corresponding to the parameterized model. Based on the parameterized model, the geometric control parameters of the flow path and the airfoil geometry parameters of the stator are sampled to generate an initial sample database; wherein, the initial sample database includes: multiple sets of sample data and at least two performance index parameters of the compressor corresponding to each set of sample data, and each set of sample data includes: the geometric control parameters of the flow path and the airfoil geometry parameters of the stator; The artificial neural network model is trained based on the initial sample database to obtain a prediction model; wherein, the prediction model is used to predict each set of input sample data and output at least two performance index parameters corresponding to the set of sample data; The prediction model is solved by multi-objective optimization so that each of the at least two performance index parameters output by the prediction model reaches its Pareto optimal solution set. In the Pareto optimal solution set, the performance index parameter with the best overall performance is selected, and the optimization region on the flow passage of the compressor is integrated and optimized based on the sample data corresponding to the selected performance index parameter.

[0133] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the integrated optimization method for compressor flow path design and stator end bending as described above.

[0134] Alternatively, a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the following operations: Within the optimized region of the compressor's flow passage, the geometric control parameters of the flow passage design and the airfoil geometry parameters of the stator are obtained respectively, and a parameterized model is generated based on the geometric control parameters of the flow passage design and the airfoil geometry parameters of the stator. The parameterized model is subjected to structured mesh generation, and the flow field of the meshed parameterized model is numerically simulated to obtain at least two performance index parameters of the compressor corresponding to the parameterized model. Based on the parameterized model, the geometric control parameters of the flow path and the airfoil geometry parameters of the stator are sampled to generate an initial sample database; wherein, the initial sample database includes: multiple sets of sample data and at least two performance index parameters of the compressor corresponding to each set of sample data, and each set of sample data includes: the geometric control parameters of the flow path and the airfoil geometry parameters of the stator; The artificial neural network model is trained based on the initial sample database to obtain a prediction model; wherein, the prediction model is used to predict each set of input sample data and output at least two performance index parameters corresponding to the set of sample data; The prediction model is solved by multi-objective optimization so that each of the at least two performance index parameters output by the prediction model reaches its Pareto optimal solution set. In the Pareto optimal solution set, the performance index parameter with the best overall performance is selected, and the optimization region on the flow passage of the compressor is integrated and optimized based on the sample data corresponding to the selected performance index parameter.

[0135] Electronic device 1400, which can serve as a server or client in this application, is described below as an example of hardware devices applicable to various aspects of this application. Electronic device 1400 is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device 1400 can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0136] Electronic device 1400 includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) or a computer program loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The computing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0137] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc. In this application, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application according to actual needs. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units can be implemented in hardware or as software functional units.

[0138] Although embodiments of this application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for integrated optimization of compressor flow path design and stator end bend, characterized in that, include: Within the optimized region of the compressor's flow passage, the geometric control parameters of the flow passage design and the airfoil geometry parameters of the stator are obtained respectively, and a parameterized model is generated based on the geometric control parameters of the flow passage design and the airfoil geometry parameters of the stator. The parameterized model is subjected to structured mesh generation, and the flow field of the meshed parameterized model is numerically simulated to obtain at least two performance index parameters of the compressor corresponding to the parameterized model. Based on the parameterized model, the geometric control parameters of the flow path and the airfoil geometry parameters of the stator are sampled to generate an initial sample database; wherein, the initial sample database includes: multiple sets of sample data and at least two performance index parameters of the compressor corresponding to each set of sample data, and each set of sample data includes: the geometric control parameters of the flow path and the airfoil geometry parameters of the stator; The artificial neural network model is trained based on the initial sample database to obtain a prediction model; wherein, the prediction model is used to predict each set of input sample data and output at least two performance index parameters corresponding to the set of sample data; The prediction model is solved by multi-objective optimization so that each of the at least two performance index parameters output by the prediction model reaches its Pareto optimal solution set. In the Pareto optimal solution set, the performance index parameter with the best overall performance is selected, and the optimization region on the flow passage of the compressor is integrated and optimized based on the sample data corresponding to the selected performance index parameter.

2. The integrated optimization method for compressor flow path design and stator end bend according to claim 1, characterized in that, Obtain the geometric control parameters for the flow path design, including: The meridional curve of the flow pattern is determined, and the meridional curve is parametrically controlled to obtain multiple control points and the coordinate parameters corresponding to each control point; The coordinate parameters of each control point are determined as the geometric control parameters of the flow path design.

3. The integrated optimization method for compressor flow path design and stator end bend according to claim 1, characterized in that, Obtain the airfoil geometry parameters of the stator, including: The stator is divided into multiple sections along the blade height direction, and the inlet angle, outlet angle and installation angle of each section are determined as the blade geometry parameters of the stator. Wherein, the inlet angle is the angle between the tangent of the leading edge of the blade's mid-curve and the compressor's axial direction; the outlet angle is the angle between the tangent of the trailing edge of the blade's mid-curve and the compressor's axial direction; and the mounting angle is the angle between the chord line of the blade section and the compressor's axial direction.

4. The integrated optimization method for compressor flow path design and stator end bend according to claim 1, characterized in that, The generation of a parameterized model based on the geometric control parameters of the flow path design and the airfoil geometry parameters of the stator includes: Based on the geometric control parameters of the flow path design, a closed two-dimensional flow path profile is formed, and the flow path model is obtained by stretching the two-dimensional flow path profile along the circumference of the compressor. The blade geometry parameters of the stator are interpolated and transitioned along the blade height direction to obtain the stator blade model; The parameterized model is obtained by embedding the stator blade model into the flow channel model.

5. The integrated optimization method for compressor flow path design and stator end bend according to claim 1, characterized in that, The process of structurally meshing the parameterized model and performing flow field numerical simulation on the meshed parameterized model includes: The parameterized model is divided into multiple grid blocks using a composite topology approach; Numerical flow field simulation is performed on each of the grid blocks to obtain at least two fluid parameters for each grid block; Among them, the fluid parameters of multiple grid blocks are determined as the performance index parameters of the parameterized model.

6. The integrated optimization method for compressor flow path design and stator end bend according to claim 1, characterized in that, The step of generating an initial sample database by sampling the geometric control parameters of the flow path and the airfoil geometry parameters of the stator based on the parameterized model includes: The geometric control parameters of the flow path and the airfoil geometry parameters of the stator are sampled and combined to obtain multiple sets of sample data; Based on the parameterized model, at least two performance index parameters are determined for each group of sample data. An initial sample database is constructed based on multiple sets of sample data and at least two performance parameters of the compressor corresponding to each set of sample data.

7. The integrated optimization method for compressor flow path design and stator end bend according to claim 1, characterized in that, The process of training the artificial neural network model based on the initial sample database to obtain the prediction model includes: The sample data and performance index parameters in the initial sample database are normalized. The normalized sample data is input into the artificial neural network model to obtain the prediction index parameters; The loss value between the predicted index parameters and the normalized performance index parameters is determined by the loss function. If the loss value is greater than a preset threshold, the weights and biases of the artificial neural network model are updated using the gradient descent algorithm, and the normalized sample data is input into the updated artificial neural network model to obtain new prediction index parameters and a redefined loss value, until the redefined loss value is less than the preset threshold.

8. The integrated optimization method for compressor flow path design and stator end bend according to claim 1, characterized in that, The step of performing multi-objective optimization on the prediction model to achieve a Pareto optimal solution set for each of the at least two performance index parameters output by the prediction model, including: Multiple sets of sample data are selected from the initial sample database as original samples; wherein, the performance index parameters corresponding to each set of sample data are the original performance parameters; The original samples are cross-combined to obtain multiple sets of design samples. Each set of design samples is input into the prediction model, and at least two performance index parameters output by the prediction model are determined as design performance parameters. Select target performance index parameters that meet the performance requirements from the original performance parameters and the design performance parameters, and determine the original sample or design sample corresponding to the target performance index parameters as the optimal solution set. The performance requirement is that each of at least two performance index parameters reaches its Pareto optimal solution.

9. The integrated optimization method for compressor flow path design and stator end bend according to claim 1, characterized in that, The performance index parameters that best overall performance are selected from the Pareto optimal solution set include: The comprehensive performance evaluation value corresponding to each solution unit in the Pareto optimal solution set is calculated using a weighted summation method. The performance index parameter corresponding to the solution unit with the largest comprehensive performance evaluation value is determined as the performance index parameter with the best comprehensive performance; wherein, each solution unit is a set of original samples or design samples in the Pareto optimal solution set.

10. An integrated optimization device for compressor flow path design and stator end bending, characterized in that, include: The parameter acquisition unit is used to acquire the geometric control parameters of the flow path and the airfoil geometry parameters of the stator in the optimized region of the flow path of the compressor, and to generate a parameterized model based on the geometric control parameters of the flow path and the airfoil geometry parameters of the stator. The flow field simulation unit is used to perform structured mesh generation on the parameterized model and to perform flow field numerical simulation on the meshed parameterized model to obtain at least two performance index parameters of the compressor corresponding to the parameterized model. A database unit is used to generate an initial sample database by performing parameter sampling processing on the geometric control parameters of the flow path and the airfoil geometry parameters of the stator based on the parameterized model; wherein, the initial sample database includes: multiple sets of sample data and at least two performance index parameters of the compressor corresponding to each set of sample data, and each set of sample data includes: the geometric control parameters of the flow path and the airfoil geometry parameters of the stator; The training unit is used to train the artificial neural network model based on the initial sample database to obtain a prediction model; wherein the prediction model is used to predict each set of input sample data and output at least two performance index parameters corresponding to the set of sample data. The optimization unit is used to perform multi-objective optimization on the prediction model so that each of the at least two performance index parameters output by the prediction model reaches its Pareto optimal solution set. An integrated optimization design unit is used to select the performance index parameter with the best overall performance from the Pareto optimal solution set, and to perform integrated optimization design on the optimization area of ​​the compressor flow passage based on the sample data corresponding to the selected performance index parameter.

Images