Method for establishing grid division database of turbomachinery, computing device and application thereof

Abstract

A method for establishing a turbomachine grid division database, comprising the following steps: providing a plurality of blade profile samples, providing blade profile characteristic parameters and grid control parameters of the blade profile samples; normalizing the blade profile characteristic parameters and constructing a distance function; performing k-means clustering analysis on the blade profile samples according to the distance function, and dividing the blade profile samples into k clusters; performing clustering iteration until a clustering result meeting an iteration condition is obtained, and obtaining a turbomachine grid division database including the grid control parameters, the classification result and the clustering center of the blade profile samples. The method can establish a database for quantitatively comparing the similarity of blade profile samples, and is used for quickly dividing grids of blade profiles. The application further provides a computing device and a turbomachine grid division method.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of aero-engines, and particularly relates to a method for establishing a grid division database of a turbomachine, a computing device and application thereof. BACKGROUND

[0002] In turbomachine fluid simulation calculation, a H-O-H structured grid is generally used to generate a calculation domain grid. A suitable grid topology is first given, but the grid topology of a blade generally has many control parameters. In the grid generation process, the control parameters need to be adjusted to generate a high-quality grid. When generating a grid for different blade profiles, the control parameters also change greatly. The change of the control parameters is often related to the geometric parameters such as the inlet configuration angle and the outlet configuration angle of the blade profile. Therefore, when the blade profile is changed, the grid control parameters need to be manually adjusted to achieve high grid quality. This process is tedious and time-consuming, and increases the grid generation time of simulation calculation. Therefore, providing a turbomachine grid division database to realize rapid division of the turbomachine grid has high practical value for improving the calculation efficiency of simulation. SUMMARY

[0003] The present application aims to provide a method for establishing a turbomachine grid division database to provide a database for rapid grid division. The present application also provides a turbomachine grid division method.

[0004] In an embodiment of one aspect of the present application, a method for establishing a turbomachine grid division database is provided, which uses a k-means method to perform cluster analysis on known blade profile samples. The k-means is one of the typical algorithms of unsupervised machine learning, and is a division method for finding k clusters through iteration, so that the loss function corresponding to the clustering result is minimized. The embodiment of the present application classifies blade profiles with similar structural features into a category through k-means clustering, and quantifies the similarity between different blade profile samples through a distance function. The method specifically includes the following steps:

[0005] A plurality of blade profile samples are provided, and blade profile feature parameters and grid control parameters of the blade profile samples are provided. The blade profile feature parameters are normalized, and a distance function is constructed. According to the distance function, k-means cluster analysis is performed, the blade profile database is divided into k clusters by giving a k value. An iteration condition is set. When the blade profile feature parameters and the grid control parameters in the same cluster do not satisfy the iteration condition, the clustering is repeated until the blade profile feature parameters and the grid control parameters in the same cluster satisfy the iteration condition. A turbomachine grid division database including the blade profile feature parameters, the grid control parameters, the classification results and the cluster centers of the samples is obtained.

[0006] The method can quantitatively classify different airfoils, and establish an airfoil grid division database with different cluster centers. When a new airfoil needs to be divided into grids, the same quantification method can be used to classify the new airfoil, quickly determine the airfoil sample closest to the structural characteristics of the new airfoil from the database, and quickly realize automatic grid division based on the grid control parameters of the airfoil sample, thereby effectively improving the grid division efficiency and reducing the working hours consumed in the grid division in the simulation analysis process.

[0007] Further, in some embodiments, the airfoil feature parameters include a plurality of combinations of inlet geometric angle, outlet geometric angle, blade rotation direction, inlet twist angle, outlet twist angle, grid circle angle size, and whether it is an end-bent blade.

[0008] Further, in some embodiments, the iteration condition includes a plurality of combinations of consistent blade rotation direction, consistent end-bent blade, inlet geometric angle difference not greater than 12°, outlet geometric angle difference not greater than 12°, inlet twist angle difference not greater than 25°, and outlet twist angle difference not greater than 25°.

[0009] Further, in some embodiments, the iteration condition includes that the grid node number difference in the grid control parameters is not more than 10, and the inclination angle difference at the same position in the grid control parameters is not more than 20°.

[0010] Further, in some embodiments, the normalization calculation method of the feature parameters of the airfoil is X inew = W i (X i -X imin ) / (X imax -X imin ), where X inew is the normalized data, i represents different feature parameters, X imin is the minimum value of the feature parameter i, X imax is the maximum value of the feature parameter i, and W i is a given weight value.

[0011] Further, in some embodiments, the distance function D uses Euclidean distance calculation:

[0012] where j and k represent any two airfoils.

[0013] Further, in some embodiments, the process of giving a k value is as follows: a profile coefficient SC i =(b i -a i ) / max(b i ,ai ), where b i Let a be the minimum distance from leaf-type sample i to samples in other clusters. i Let i be the average distance from the sample in this cluster; given an initial k value, perform clustering and calculate the silhouette coefficient SC for all the leaf-shaped samples. i The average value of k is SC; the k value is adjusted, the clustering is repeated and SC is calculated again until a k value that satisfies |SC-1|=min|SC-1| is obtained.

[0014] According to another aspect of the present invention, a computing device is provided, the computing device including a memory and a processor, wherein the memory stores a computing program, and when the processor executes the computing program, it is able to implement the method for establishing a turbomachinery mesh partitioning database provided in any of the foregoing embodiments.

[0015] According to another embodiment of the present invention, a method for mesh generation in turbomachinery is provided. The method uses the turbomachinery mesh generation database provided by the method for establishing a turbomachinery mesh generation database provided in any of the foregoing embodiments to perform mesh generation on airfoils. The method includes the following steps: providing a target airfoil to be meshed; extracting and normalizing the airfoil feature parameters of the target airfoil; calculating the distance function of the target airfoil to obtain a target distance function; comparing the target airfoil with the target distance function values ​​of each cluster center, and assigning the target airfoil to the cluster containing the cluster center with the smallest target distance function value as the target cluster; calculating the target distance function values ​​of the target airfoil with each airfoil in the target cluster, and assigning the mesh control parameters of the airfoil sample corresponding to the minimum value of the target distance function to the target airfoil to complete the mesh generation.

[0016] Furthermore, in some embodiments, the mesh division is generated through HOH structured mesh calculation.

[0017] The application of this turbomachinery mesh generation database fully utilizes quantified airfoil characteristic parameters to classify target airfoils and determine the airfoil sample in the database that is closest to the target airfoil. By assigning the mesh control parameters of the existing airfoil sample to the target airfoil, rapid mesh generation is achieved, which effectively reduces the workload of repeated mesh generation and improves the computational efficiency of simulation. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the establishment of a turbomachinery mesh generation database in one embodiment;

[0019] Figure 2 This is a flowchart illustrating rapid mesh generation using a turbomachinery mesh generation database in one embodiment;

[0020] Figure 3 This is a schematic diagram of the blade rotation direction in one embodiment;

[0021] Figure 4 This is a schematic diagram of the cross-sectional structure of a blade-shaped sample in one embodiment;

[0022] Figure 5 This is a schematic diagram of the topology of the leaf-shaped sample grid in one embodiment.

[0023] The purpose of the above-described drawings is to provide a detailed description of the present invention so that those skilled in the art can understand the technical concept of the invention, and not to limit the invention. For the sake of brevity, the above-described drawings only schematically depict the structures related to the technical features of the present invention, and do not depict the complete structure and all details strictly according to actual scale. Detailed Implementation

[0024] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0025] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment herein. The phrase appearing in various places in the specification does not necessarily refer to the same embodiment, nor is it limited to mutually exclusive, independent, or alternative embodiments. Those skilled in the art will understand that the embodiments herein can be combined with other embodiments without structural conflict. In the description herein, "a plurality of" means at least two.

[0026] One embodiment of the present invention provides a method for establishing a turbomachinery mesh database, enabling rapid mesh generation during simulation calculations. The method flow is as follows: Figure 1 As shown, the basic steps include:

[0027] Step 1: Collect blade profile samples, extract blade profile feature parameters from the samples, and provide mesh control parameters. In a preferred embodiment, the blade profile feature parameters include inlet geometry angle, outlet geometry angle, blade rotation direction, inlet twist angle, outlet twist angle, mesh fillet size, and whether it is an end-bent blade.

[0028] Step 2: Normalize the leaf shape feature parameters and construct a distance function. In a preferred embodiment, the distance function is calculated using a weight-based normalization method: X inew =W i (X i -X imin ) / (X imax -X imin ), where X inewFor the normalized data, i represents different feature parameters, X imin X is the minimum value among the characteristic parameters i. imax W is the maximum value among the characteristic parameters i. i Given weight values. Distance function. Where j and k represent any two leaf shapes.

[0029] Step 3: Perform k-means clustering analysis based on the distance function. Given a value of k, divide the leaf-shaped database into k clusters, and obtain the classification results that satisfy the iteration conditions through multiple clustering.

[0030] In a preferred embodiment, step 3.1 is further included: given an initial k value, clustering is performed and the silhouette coefficient SC is calculated. i =(b i -a i ) / max(b i ,a i ) and the average profile coefficient SC of all leaf type samples, where b i Let a be the minimum distance from leaf-type sample i to samples in other clusters. i is the average distance from i to the samples in this cluster; the value of k is adjusted until SC is closest to 1, thus obtaining a determined value of k.

[0031] In a preferred embodiment, the iteration should meet the following conditions: consistent blade rotation direction, consistency of whether the blades are end-bent, inlet geometric angle difference not exceeding 12°, outlet geometric angle difference not exceeding 12°, inlet torsion angle difference not exceeding 25°, and outlet torsion angle difference not exceeding 25°. In a preferred embodiment, the iteration should also meet the following conditions: the difference in the number of grid nodes in the grid control parameters does not exceed 10, and the difference in tilt angle at the same location in the grid control parameters does not exceed 20°.

[0032] Step 4: Obtain the impeller mesh mechanical partitioning database, which includes cluster centers, classification results, blade shape feature parameters, and mesh control parameters.

[0033] Another embodiment of the present invention provides a computing device, which can be a general-purpose computer or a special-purpose computing device. The computing device includes a memory and a processor. The memory stores a program, and when the processor executes the program, it can implement the method for establishing a turbomachinery mesh partitioning database provided in the foregoing embodiments. The program can be written in a general-purpose computer language, such as Python or Java.

[0034] Another embodiment of the present invention provides an application of a turbomachinery mesh generation database, which is used to quickly mesh the blade profile using the turbomachinery mesh generation database provided in the foregoing embodiments. The process is as follows:Figure 2 As shown, it includes:

[0035] Step 5.1: Given the target airfoil;

[0036] Step 5.2: Extract the leaf shape feature parameters of the target leaf shape;

[0037] Step 5.3: Normalize the leaf shape feature parameters and establish the target distance function;

[0038] Step 5.4: Classify the target leaf shape according to the target distance function, and take the cluster containing the cluster center with the smallest target distance function value as the target cluster;

[0039] Step 5.5: Compare the target distance function values ​​corresponding to each leaf shape sample in the target cluster, and select the leaf shape sample corresponding to the minimum value as the leaf shape with the highest similarity to the target leaf shape. Assign the mesh control parameters of this leaf shape sample to the target leaf shape and complete the mesh generation. In a preferred embodiment, this mesh generation process is generated through HOH structure and mesh calculation.

[0040] In a preferred embodiment, the process of establishing a turbomachinery mesh generation database and using the database for mesh generation is as follows:

[0041] First, collect data on n airfoil profiles, including airfoil characteristic parameters and grid control parameters for compressors, fans, and turbines, as shown in Table 1.

[0042]

[0043] Table 1 Leaf type data

[0044] Among them, the leaf shape characteristic parameter a i This represents the leaf shape characteristic parameters of the i-th leaf shape, combined with... Figure 4 The subscripts 1-7 represent the inlet geometry angle α1 (6), outlet geometry angle α2 (7), blade rotation direction, inlet twist angle, outlet twist angle, grid fillet size, and whether it is an end-bend blade, respectively. The inlet twist angle is the difference between the inlet profile angle at the hub and the inlet profile angle at the tip; the outlet twist angle is the difference between the inlet profile angle at the hub and the outlet profile angle at the tip; the blade rotation direction is as follows: Figure 3 As shown, the helical direction characteristic parameter is defined as 1 for blade type a and 0 for blade type b; the same assignment method is used for whether it is an end-bend blade. Mesh control parameter p i This represents the mesh control parameters for the i-th blade shape, with subscripts 1-10 indicating the parameters for that blade shape in different situations. Figure 5 The different control parameters p1-p in the leaf-shaped sample mesh topology shown 10 .

[0045] Normalize the feature parameters of the leaf shape and construct a weighted distance function: X inew =W i (X i -X imin ) / (X imax -X imin ), where X inew For the normalized data, i takes values ​​from 1 to 7, representing the different characteristic parameters mentioned above (inlet geometry angle, outlet geometry angle, blade rotation direction, inlet twist angle, outlet twist angle, mesh fillet size, and whether it is an end-bend blade), X imin X is the minimum value among the characteristic parameters i. imax W is the maximum value among the characteristic parameters i. i The given weight values ​​are shown in Table 2.

[0046]

[0047] Table 2 Range of Feature Values

[0048] The distance D between samples of different leaf shapes is represented by Euclidean distance:

[0049]

[0050] Where j and k represent any two leaf-shaped samples.

[0051] Based on the leaf shape feature parameters and the distance function, k-means distance analysis was performed to divide the leaf shape data into k clusters. The specific parameters of the leaf shape sample features are shown in Table 3.

[0052]

[0053]

[0054] Table 3. Leaf type sample characteristic parameters and classification results

[0055] Table 3 contains a total of 36 blade type samples. First, based on blade rotation direction, the samples can be clearly divided into two categories: moving blades and stationary blades. Moving blades have a rotation direction of 0, and stationary blades have a rotation direction of 1. Among the stationary blades with a rotation direction of 1, observing the inlet and outlet construction angles, samples 1 and 2 have inlet and outlet construction angles near 0, significantly different from other stationary blades, and can be classified into one category; samples 3 and 4 have inlet construction angles close to 0, while outlet construction angles are near 120, significantly different from other stationary blades, and can be classified into one category; samples 14–16 have relatively small inlet and outlet construction angles, and can be classified into one category; among the remaining stationary blade samples, samples 5–10 are classified into one category based on a corner radius of 0; samples 17–21 all have a corner radius of 3 and are end-curved blades, and are classified into one category; samples 11–13 and 22 are not end-curved blades and have corner radii that are not 0, and are classified into another category. Similarly, the blades with a rotation direction of 0 are divided into two categories: samples 23-29 have a corner radius of 0, while samples 31-36 have a corner radius that is not 0. Based on the above analysis, the samples can be roughly divided into 8 categories, i.e., initial k=8. First, a clustering is performed, and the silhouette coefficient SC is defined. i =(b i -a i ) / max(b i ,a i ), where b i Let a be the minimum distance from leaf-type sample i to samples in other clusters. i Given the average distance from i to the samples in this cluster, calculate the average silhouette coefficient SC for all samples.

[0056] The clustering model is trained by increasing or decreasing the value of k, and the model's performance is evaluated based on the SC value. When SC is closest to 1, i.e., |SC-1| = min|SC-1|, the clustering model is considered to have reached a good state. The value of k is adjusted, and finally min|SC-1| is obtained when k = 13. Therefore, the given value of k in k-means clustering is 13.

[0057] Analyzing the clustering results, k-means clustering tends to converge to local optima, requiring multiple clustering iterations to determine if the results meet expectations. The iterative criteria for blade shape characteristic parameters include: criteria for judging similarity in blade shape features: 1) Blades of the same category have the same rotation direction; 2) Blades of the same category are either rounded or without rounded corners; 3) Blades of the same category have consistent end-bending blade characteristics; 4) The inlet angle deviation of blades of the same category is no greater than 12°; 5) The outlet angle deviation of blades of the same category is no greater than 12°; 6) The inlet twist angle deviation of blades of the same category is no greater than 25°; 7) The outlet twist angle deviation of blades of the same category is no greater than 25°.

[0058] Simultaneously, the similarity of the mesh control parameters of the leaf shape samples should also be judged. The mesh control parameters of each leaf shape sample are shown in Table 4, where H1, H2, H3, H4, BS1, BP1, BS2, BP2, and O1 represent the number of mesh nodes at different positions, and Hin, Hout, Sin, and Sout represent the tilt angles at different positions. The iterative criteria for the mesh control parameters are: 1) the deviation of the number of mesh nodes at the same position for leaf shapes of the same category does not exceed 10; 2) the deviation of the tilt angle at the same position for leaf shapes of the same category does not exceed 20°.

[0059]

[0060]

[0061] Table 4. Control parameters for leaf-shaped sample grids

[0062] When the feature parameters and grid control parameters of leaf-type samples in the same cluster all satisfy the above iterative criteria, the clustering result is considered to meet the requirements, and the final clustering result is shown in Table 3. It can be considered that the distance D can accurately reflect the similarity between two leaf types at this point. The cluster centers in each cluster are shown in Table 5.

[0063]

[0064] Table 5 Cluster Centers

[0065] Thus, a turbomachinery mesh partitioning database is obtained, including blade shape feature parameters, mesh control parameters, classification results, and cluster centers of blade shape samples.

[0066] Next, the database is used to perform mesh generation on a target airfoil. The characteristic parameters of the target airfoil are shown in Table 6.

[0067]

[0068] Table 6 Target Leaf Shape Characteristic Parameters

[0069] The leaf shape characteristic parameters of the target leaf shape are normalized, a target distance function is established, and the target distance function values ​​between the target leaf shape and each cluster center are calculated, as shown in Table 7.

[0070]

[0071] Table 7 Distance between target leaf shape and cluster center

[0072] Therefore, it can be determined that the target leaf shape is the closest to cluster 12, and the target leaf shape is classified into cluster 12.

[0073] Next, the distances of each leaf shape sample in cluster 12 to the target leaf shape are compared. The distance of sample 23 is determined to be 0.1346, and the distance of sample 24 is determined to be 0.0025. Sample 24 is the leaf shape sample most similar to the target leaf shape. The mesh control parameters of sample 24 are assigned to the target sample, and the HOH structured mesh is calculated and generated.

[0074] The above process can be programmed using general-purpose computer languages ​​such as Python and implemented on a general-purpose computer. The time required for rapid blade mesh generation using a turbomachinery mesh generation database is one to two orders of magnitude lower than that for manual mesh control parameter adjustment, significantly improving the simulation efficiency of turbomachinery. In a preferred embodiment, generating mesh control parameters for 10 rows of blades using the turbomachinery mesh generation database takes 1 minute, while manually adjusting the mesh control parameters for 10 rows of blades takes 100 minutes.

[0075] The purpose of the above embodiments is to provide a further detailed description of the present invention in conjunction with the accompanying drawings, so that those skilled in the art can understand the technical concept of the present invention. Within the scope of the present invention, optimization or equivalent substitution of the method steps involved, as well as combination of implementation methods in different embodiments without causing a conflict of principles, all fall within the protection scope of the present invention.

Claims

1. A method for establishing a mesh generation database for turbomachinery, characterized in that, Includes the following steps: Multiple blade profile samples are provided, along with blade profile feature parameters and mesh control parameters for each sample. The blade profile feature parameters include multiple combinations of inlet geometry angle, outlet geometry angle, blade rotation direction, inlet twist angle, outlet twist angle, mesh fillet size, and whether the blade is end-bent. The leaf shape feature parameters are normalized to construct a distance function; Based on the distance function, k-means clustering analysis is performed, and given a value of k, the leaf-shaped database is divided into k clusters; Set an iteration condition: if the leaf shape feature parameter and the grid control parameter in the same cluster do not meet the iteration condition, repeat the clustering until the leaf shape feature parameter and the grid control parameter in the same cluster meet the iteration condition. The iteration conditions include multiple combinations of the following: consistent blade rotation direction, consistent end-bend blades, inlet geometric angle difference not exceeding 12°, outlet geometric angle difference not exceeding 12°, inlet torsion angle difference not exceeding 25°, and outlet torsion angle difference not exceeding 25°; the difference in the number of grid nodes in the grid control parameters does not exceed 10, and the difference in tilt angle at the same position in the grid control parameters does not exceed 20°; A turbomachinery mesh partitioning database is obtained, including blade feature parameters, mesh control parameters, classification results, and cluster centers of blade samples.

2. The method for establishing a turbomachinery mesh database according to claim 1, characterized in that, The normalized calculation method for the characteristic parameters of the extracted leaf shape is as follows: X inew =W i (X i -X imin ) / (X imax -X imin ), where X inew For the normalized data, i represents different feature parameters, X imin X is the minimum value among the characteristic parameters i. imax W is the maximum value among the characteristic parameters i. i For the given weight values.

3. The method for establishing a turbomachinery mesh database according to claim 2, characterized in that, The distance function D is calculated using Euclidean distance: Where j and k represent any two leaf shapes.

4. The method for establishing a turbomachinery mesh database according to claim 1, characterized in that, The process of giving a given value of k includes the following steps: Establish the profile coefficient SC i =(b i -a i ) / max(b i ,a i ), where b i Let a be the minimum distance from leaf-type sample i to samples in other clusters. i The average distance from i to the samples in this cluster; Given an initial k value, cluster the samples and calculate the silhouette coefficient SC for all the leaf shape samples. i The average value of SC; Adjust the value of k, cluster again and calculate SC, until a value of k that satisfies |SC-1|=min|SC-1| is obtained.

5. A computing device, comprising a memory and a processor, characterized in that, The memory stores a calculation program, which, when executed by the processor, enables the implementation of the method for establishing a turbomachinery mesh partitioning database as described in any one of claims 1 to 4.

6. A method for mesh generation in turbomachinery, characterized in that, A turbomachinery mesh database is provided, wherein the turbomachinery mesh database is obtained using the turbomachinery mesh data establishment method as described in any one of claims 1 to 4; Provide a target leaf shape to be meshed, extract the leaf shape feature parameters of the target leaf shape and normalize them, and calculate the distance function of the target leaf shape to obtain the target distance function; The target airfoil is compared with the target distance function values ​​of each cluster center in the turbomachinery mesh partitioning database, and the target airfoil is assigned to the cluster containing the cluster center with the smallest target distance function value as the target cluster; Calculate the target distance function value between the target leaf shape and each leaf shape in the target cluster, and assign the grid control parameter of the leaf shape sample corresponding to the minimum value of the target distance function to the target leaf shape to complete the grid division.

7. The impeller machinery mesh generation method according to claim 6, characterized in that, The mesh division is generated through HOH structured mesh calculation.

Images