Method and device for calculating surge margin of centrifugal compressor based on flow field feature recognition

Abstract

The present application provides a method and device for calculating the surge margin of a centrifugal compressor based on flow field feature recognition. The method includes: obtaining the characteristic lines of the original centrifugal compressor flow and pressure, adjusting the design parameters of the original centrifugal compressor to calculate N characteristic lines, and extracting the flow field distribution cloud map to form an image set; analyzing the position of the calculation point corresponding to each image in the image set on the characteristic line, and determining the label of each image; dividing the image set with labels into a sub-feature training set and a comprehensive feature training set; inputting the images in the sub-feature training set into the centrifugal compressor flow field feature determination model to obtain a flow field feature recognition model; using the comprehensive feature training set for secondary training to obtain a comprehensive flow field feature recognition model; calculating the surge margin of the newly generated centrifugal compressor geometry. The method and device for calculating the surge margin of a centrifugal compressor based on flow field feature recognition provided by the present application can accurately and quickly calculate the margin in the aerodynamic optimization process.

Description

Technical Field

[0001] The present application relates to the technical field of aerodynamic performance analysis and surge margin prediction of centrifugal compressors, and in particular to a method and device for calculating the surge margin of a centrifugal compressor based on flow field feature recognition. Background Art

[0002] Centrifugal compressors are widely used in high-performance mechanical equipment such as aircraft engines, gas turbines, and industrial compressors. Their key role is to increase the gas pressure from low to high pressure to meet the gas pressure requirements of subsequent work processes. Under high load conditions, centrifugal compressors are prone to surge, that is, violent periodic fluctuations in gas flow, resulting in performance degradation or even equipment damage. Therefore, accurate and rapid prediction of the surge margin of centrifugal compressors has become a key issue in the design and optimization of centrifugal compressors.

[0003] The traditional method of calculating surge margin requires searching for surge boundary through numerical calculation. There are usually two ways to determine the surge boundary. One is to use the method of gradually increasing the outlet back pressure, and the calculation point gradually approaches the near-surge point until the calculation diverges to obtain the surge boundary; the other is to use the near-surge point estimation method to estimate the surge boundary by calculating and estimating the flow field of the near-surge point, design point and near-blocking point. The first method requires multiple full three-dimensional CFD (Computational Fluid Dynamics) calculations, which is relatively time-consuming; the second method can effectively reduce the number of CFD calculations, but it will lose the accuracy of margin calculations, affect the gradient generation in the multi-objective algorithm and proxy model, and thus affect the optimization results. Summary of the invention

[0004] In view of this, the present application provides a method and device for calculating the surge margin of a centrifugal compressor based on flow field feature recognition, which can accurately and quickly calculate the margin in the aerodynamic optimization process and improve the efficiency of aerodynamic optimization.

[0005] Specifically, the present application is implemented through the following technical solutions:

[0006] In a first aspect, the present application provides a method for calculating surge margin of a centrifugal compressor based on flow field feature recognition, the method comprising:

[0007] Obtaining characteristic lines of flow and pressure of the original centrifugal compressor, adjusting the design parameters of the original centrifugal compressor to calculate N characteristic lines, and extracting the flow field distribution cloud map to form an image set;

[0008] Analyze the position of the calculation point corresponding to each image in the image set on the characteristic line to determine the label of each image;

[0009] The labeled image set is divided into a sub-feature training set and a comprehensive feature training set, wherein the sub-feature training set includes shock wave intensity and shape, supersonic flow area, and flow separation area features;

[0010] Inputting the images in the sub-feature training set into the centrifugal compressor flow field feature determination model to respectively train the shock wave intensity and shape, supersonic flow area and flow separation area features to obtain a flow field feature recognition model;

[0011] Performing secondary training on the flow field feature recognition model using the comprehensive feature training set to obtain a comprehensive flow field feature recognition model;

[0012] Based on the comprehensive flow field feature recognition model, the surge margin of the newly generated centrifugal compressor geometry is calculated.

[0013] A second aspect of the present application provides a centrifugal compressor surge margin calculation device based on flow field feature recognition, the device comprising an extraction module, an analysis module, a segmentation module, a training module and a calculation module;

[0014] The extraction module is used to obtain the characteristic lines of the original centrifugal compressor flow and pressure, adjust the design parameters of the original centrifugal compressor to calculate N characteristic lines, and extract the flow field distribution cloud map to form an image set;

[0015] The analysis module is used to analyze the position of the calculation point corresponding to each image in the image set on the characteristic line to determine the label of each image;

[0016] The segmentation module is used to segment the image set with labels into a sub-feature training set and a comprehensive feature training set, wherein the sub-feature training set includes shock wave intensity and shape, supersonic flow area and flow separation area features;

[0017] The training module is used to input the images in the sub-feature training set into the centrifugal compressor flow field feature determination model to perform training on shock wave intensity and shape, supersonic flow area and flow separation area features respectively, so as to obtain a flow field feature recognition model;

[0018] The training module is further used to perform secondary training on the flow field feature recognition model using the comprehensive feature training set to obtain a comprehensive flow field feature recognition model;

[0019] The calculation module is used to calculate the surge margin of the newly generated centrifugal compressor geometry based on the comprehensive flow field feature recognition model.

[0020] The method and device for calculating the surge margin of a centrifugal compressor based on flow field feature recognition provided by the present application obtain the characteristic lines of the original centrifugal compressor flow and pressure and adjust the design parameters to calculate multiple characteristic lines, and use the flow field feature recognition model to estimate the surge margin, thereby avoiding a large number of tedious CFD calculations for each sample to accurately search for the surge boundary. Only the original and N characteristic lines after adjusting the design parameters need to be calculated, and the model prediction can be used, which reduces the amount of calculation and improves the calculation efficiency; the surge margin of the newly generated centrifugal compressor geometry is calculated based on the comprehensive flow field feature recognition model. Once the model training is completed, for the new compressor geometry, only the design point flow calculation needs to be performed to obtain the flow characteristic parameters and input them into the model, and the surge margin can be quickly obtained. There is no need to perform complex iterative calculations like traditional methods, which saves a lot of time. In addition, by training the shock wave intensity and shape, supersonic flow area and flow separation area characteristics respectively, and using the comprehensive feature training set for secondary training, the model can comprehensively and accurately identify the flow field feature combination pattern. Compared with the traditional single feature or limited point estimation method (such as the near-surge point estimation method), it can more accurately reflect the relationship between the actual flow field state inside the compressor and surge, thereby improving the accuracy of surge margin calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A flow chart of a first embodiment of a method for calculating a surge margin of a centrifugal compressor based on flow field feature recognition provided by the present application;

[0022] Figure 2 A schematic diagram of characteristic lines shown in this application;

[0023] Figure 3 The flow field distribution cloud diagram shown in this application;

[0024] Figure 4 A schematic diagram of an image with a label shown in the present application;

[0025] Figure 5 This is a schematic diagram of the centrifugal compressor flow field similarity principle shown in this application;

[0026] Figure 6 This is a structural schematic diagram of a second embodiment of a centrifugal compressor surge margin calculation device based on flow field feature recognition provided in the present application. DETAILED DESCRIPTION

[0027] Here, exemplary embodiments will be described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0028] The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items.

[0029] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0030] Specific embodiments are given below to introduce the technical solution of the present application in detail.

[0031] Embodiment 1:

[0032] Figure 1 This is a flow chart of the first embodiment of the method for calculating the surge margin of a centrifugal compressor based on flow field feature recognition provided by this application. Figure 1 , the method provided in this embodiment may include:

[0033] S101, obtaining characteristic lines of flow rate and pressure of an original centrifugal compressor, adjusting design parameters of the original centrifugal compressor to calculate N characteristic lines, and extracting a flow field distribution cloud map to form an image set.

[0034] It should be noted that there are many types of characteristic lines. Specifically, in this embodiment, it can be a flow-pressure ratio characteristic line, wherein the characteristic line intuitively shows the boosting ability of the compressor under different flow conditions, and is the core indicator for evaluating the performance of the compressor. In addition, it should be noted that the horizontal coordinate of the characteristic line is the flow rate, and the vertical coordinate is the pressure ratio. The design point on it represents the design condition of the compressor, at which the efficiency is the highest; the surge boundary is the lower limit flow rate for stable operation of the compressor. If the flow rate is lower than this value, the internal flow field of the compressor will become unstable, causing surge; the blocking point corresponds to the upper limit flow rate. If the flow rate is too high, the airflow will be blocked and the efficiency will drop sharply. The design parameters include blade geometry parameters, impeller diameter and wheel width, diffuser geometry parameters, etc., wherein the blade geometry parameters may include the number of blades, blade inlet installation angle, blade outlet angle, etc., and the diffuser geometry parameters may include the diffuser expansion angle and length, etc. It should be noted that the characteristic line of the original centrifugal compressor flow and pressure ratio is obtained, including:

[0035] (1) Edit calculation files based on full 3D calculation tools and determine design points and boundary conditions.

[0036] The flow characteristics of the centrifugal compressor are simulated by editing the calculation file through the full three-dimensional CFD calculation tool. Editing the calculation file includes setting the geometric parameters of the centrifugal compressor, the physical properties of the working fluid, and simulation parameters such as boundary conditions. Among them, the initial condition of the simulation is usually selected at the design point, which is the ideal working state when the centrifugal compressor is designed. Specifically, it is the working point of the centrifugal compressor under rated conditions, that is, under a specific flow rate and pressure ratio, the centrifugal compressor can operate stably and efficiently. The full three-dimensional CFD calculation tool performs flow calculations under the design point conditions to ensure that the flow field distribution is reasonable under this state, and there is no stall or surge.

[0037] (2) Starting from the design point, the back pressure is increased according to the preset step size, the changes in flow rate and pressure ratio are recorded, and a full three-dimensional calculation is performed after each change in back pressure.

[0038] The calculation results of the design point usually include parameters such as flow rate, pressure ratio, temperature, etc. Starting from the design point, the back pressure value is gradually increased according to the preset step size. The size of the preset step size should be set according to the performance range of the compressor and the simulation accuracy requirements to ensure that the process of the centrifugal compressor gradually approaching the surge boundary can be captured. For example, the preset step size can be 10kPa. Among them, back pressure refers to the reverse pressure that the airflow is subjected to after passing through the outlet of the centrifugal compressor. During the centrifugal compressor test, by gradually increasing the back pressure, the performance changes of the centrifugal compressor from normal working state to approaching the surge boundary can be observed.

[0039] As an optional embodiment, the preset step size varies according to the distance between the design point and the surge boundary, for example, in direct proportion, the closer the distance, the smaller the step size, so that the boundary condition can be calculated with a more accurate scale. Identify the distance between the design point and the surge boundary; calculate the preset step size according to the distance, specifically by multiplying the previous step size by the adjustment coefficient, the adjustment coefficient is a direct proportion change of the distance, for example, a linear change or a quadratic change.

[0040] It should be noted that increasing the back pressure will lead to a decrease in flow rate and increase the instability in the flow field, which is a typical feature of gradually approaching surge. After each adjustment of the back pressure, the full three-dimensional calculation tool is used to recalculate the flow field under the current conditions to capture the flow rate, pressure ratio and other changing characteristics of the flow field. During the calculation process, as the back pressure increases, the flow inside the centrifugal compressor will gradually become more complex, and shock waves, flow separation and other phenomena may occur. After each calculation, the new flow rate and pressure ratio are recorded.

[0041] (3) When the calculation result is not convergent, obtain the surge boundary information.

[0042] Repeat the process of increasing the back pressure and performing full three-dimensional calculations until the calculation results show non-convergence characteristics, that is, the compressor becomes unstable or surges. At this point, it can be considered that the surge boundary has been reached and the relevant information is recorded. Specifically, parameters such as flow, pressure ratio, and back pressure can be recorded as the surge boundary point of the centrifugal compressor. The surge boundary point represents the maximum back pressure that the compressor can reach under the current operating conditions. When the back pressure exceeds this point, the compressor will enter a surge state and can no longer operate stably.

[0043] (4) reducing the back pressure according to a preset step size, reversely searching for the characteristics of the partial blocking point, and obtaining a characteristic line of the original centrifugal compressor based on the design point, surge boundary information and the partial blocking point characteristics; the characteristic line is used to record the shape of the characteristic line, each operating point and surge boundary information.

[0044] It should be noted that the characteristic line is a smooth curve. As the flow rate decreases, the pressure ratio gradually increases, and the slope of the curve is negative and relatively stable. Specifically, after the surge boundary is determined, the operation sequence is changed to reduce the back pressure, that is, reverse search. The back pressure is gradually reduced with the same preset step size, and changes from a higher back pressure to a lower back pressure. In this process, as the back pressure gradually decreases, the flow rate will gradually increase. When the back pressure drops to a certain critical point, the flow field characteristics begin to change, which may be manifested as a decrease in local flow velocity or slow flow in the flow channel. This characteristic point is called a partial blocking point. The partial blocking point is a point that indicates that the flow field characteristics gradually recover from the blocked state, reflecting the lower limit of the flow-pressure relationship of the centrifugal compressor. When the operating conditions are close to the partial blocking point, the flow in the compressor gradually returns to normal and the flow field tends to stabilize. By determining the partial blocking point, it can be presented together with the design point and the surge boundary point on the performance curve to obtain the characteristic line of the original centrifugal compressor.

[0045] Figure 2 The characteristic line diagram shown in this application is shown in FIG. Figure 2 The characteristic line is used to record the characteristic line shape, various operating points and surge boundary information (pressure ratio and flow rate).

[0046] It should also be noted that the design parameters of the original centrifugal compressor are adjusted to calculate N characteristic lines, including:

[0047] (1) Adjust the design parameters of the original centrifugal compressor to obtain N different centrifugal compressor models.

[0048] According to actual needs, multiple different centrifugal compressor models are created by modifying the design parameters of the centrifugal compressor. Common design parameters include the geometry of the blades (number of blades, blade height, blade shape, etc.), the leading and trailing edge shapes of the blades, intake conditions, etc. Each adjustment of the design parameters will generate a new centrifugal compressor model. Through a series of adjustments, N models with different geometric and performance characteristics can be obtained.

[0049] (2) For each centrifugal compressor model, the back pressure is increased according to the preset step size, and the changes in flow rate and pressure ratio are recorded until the surge boundary is approached.

[0050] It should be noted that the main purpose of this process is to determine the working performance of the centrifugal compressor under different back pressure conditions, especially to analyze the behavior close to the surge boundary. The back pressure range generally starts from a lower back pressure (such as the normal working back pressure) and gradually increases to the area close to the surge boundary. In this process, a maximum back pressure value is usually set, and the calculation results are used to determine whether it is close to the surge boundary.

[0051] After each back pressure adjustment, the corresponding flow rate and pressure ratio need to be recorded. Specifically, when the back pressure increases, the flow rate usually decreases, especially when approaching the surge boundary, the flow rate will drop sharply; the pressure ratio usually shows a downward trend when the back pressure increases, especially when approaching surge, the pressure ratio may change sharply.

[0052] (3) Based on the recorded data, N characteristic lines corresponding to N different centrifugal compressor models are calculated.

[0053] By gradually increasing the back pressure and recording the changes in flow rate and pressure ratio, characteristic lines of N different centrifugal compressors can be drawn. These characteristic lines reflect the relationship between the flow rate and pressure ratio of the centrifugal compressor under different back pressure conditions.

[0054] It should be noted that after obtaining N characteristic lines, flow field distribution cloud maps can be extracted based on the N characteristic lines to form an image set. Specifically, extracting the flow field distribution cloud maps to form an image set includes: analyzing the characteristic lines to extract the flow field information at the first leaf height, the second leaf height and the third leaf height; based on the flow field information, obtaining flow field distribution cloud maps at different geometries, different back pressures and different leaf heights to form an image set.

[0055] The characteristic line usually records the working points from the design point to the surge boundary, covering different back pressures, flow rates, and pressure ratios. Each working point corresponds to the flow field conditions at different positions inside the compressor. It should be noted that the blade height refers to the different cross-sectional positions on the centrifugal compressor blades. For example, the first blade height, the second blade height, and the third blade height can be 90%, 50%, and 10% of the blade height cross-section, respectively. At different blade heights, due to the influence of the airflow, the distribution of the flow field will change. As the back pressure increases, when the centrifugal compressor works at the surge boundary, the relative Mach number in the blade channel will gradually increase, and the flow will gradually separate. By extracting the flow field information at different blade heights of N different centrifugal compressor models, the flow field distribution cloud maps at different geometries, different back pressures, and different blade heights can be obtained. Among them, these cloud maps can intuitively show the distribution of airflow inside the centrifugal compressor, and reflect the airflow characteristics under different back pressure conditions. Specifically, such as Figure 3 As shown, the flow field distribution cloud diagram represents different flow characteristics through areas of different colors. For example, high-speed areas are represented by red, and low-pressure areas are represented by blue.

[0056] S102: Analyze the position of the calculation point corresponding to each image in the image set on the characteristic line to determine the label of each image.

[0057] It should be noted that the position of the calculation point corresponding to each image in the image set on the characteristic line is analyzed to determine the label of each image, including:

[0058] (1) Classify the position of the calculation point of each image on the characteristic line and identify the flow state corresponding to each position.

[0059] It should be noted that the purpose of this process is to determine the flow state represented by the working point (i.e., calculation point) of the centrifugal compressor based on its position on the characteristic line. The classification and identification of the flow state will help further compressor performance analysis and optimization. In the characteristic line diagram, each calculation point corresponds to a specific flow and pressure ratio combination, and each calculation point represents the operating state of the centrifugal compressor under a certain back pressure and flow condition.

[0060] (2) Assigning a label to each image according to the flow state category; the label includes at least a design point, a near-blocking point, and a near-breathing point.

[0061] It should be noted that the flow state generally includes a stable state, a blocked state, and a surge state, which can be represented by the design point, the near-blocking point, and the near-surge point, respectively. The design point is located at the starting position of the characteristic line and is the best operating condition for the compressor. It usually has the highest efficiency and stability. At the design point, the flow is stable and the compressor works in the optimal state; the near-blocking point is the operating condition close to the compressor where flow blockage occurs. At the near-blocking point, the flow may begin to become unstable, vortices may form, and performance may be reduced. The near-blocking point usually appears on the right side of the characteristic line, indicating an area with small flow and high pressure. When approaching the blocking point, there may be some flow loss and increased noise; the near-surge point is the operating condition close to the surge boundary. When the working point of the centrifugal compressor approaches the surge boundary, airflow oscillation and flow instability may occur. The surge point usually appears at the end or left side of the characteristic line, indicating a state with large flow and high back pressure. The surge boundary is the limit operating point of the compressor. Exceeding this point may cause compressor failure or serious performance degradation.

[0062] (3) All images with assigned labels are classified and stored to form a set of images with labels.

[0063] Each image with a flow state marked is recorded and stored in an image collection. These images can be classified and stored according to the label for subsequent use. Figure 4 For a schematic diagram of an image with a label shown in this application, please refer to Figure 4 , we can get the position information of each label in the graph.

[0064] S103, dividing the labeled image set into a sub-feature training set and a comprehensive feature training set, wherein the sub-feature training set includes shock wave intensity and shape, supersonic flow area, and flow separation area features.

[0065] It should be noted that the labeled image set is divided into sub-feature training sets, focusing on the shock wave intensity and shape, supersonic flow area and flow separation area features, and these key flow field features can be learned separately in the future. Among them, the change of shock wave intensity and shape may directly affect the stability and energy transfer of the airflow, the size and position of the supersonic flow area are closely related to the working state of the centrifugal compressor, and the flow separation area is one of the important factors causing surge.

[0066] Specifically, dividing the labeled image set into the sub-feature training set and the comprehensive feature training set includes: dividing the labeled image set into the sub-feature training set and the comprehensive feature training set according to a first ratio and a second ratio at random proportions; wherein the first ratio is greater than the second ratio (for example, the first ratio is 80%, and the second ratio is 20%). The random division method can ensure that each subset is representative.

[0067] S104, inputting the images in the sub-feature training set into the centrifugal compressor flow field feature determination model to perform training on shock wave intensity and shape, supersonic flow region and flow separation region features respectively, to obtain a flow field feature recognition model.

[0068] Specifically, the images in the sub-feature training set are input into the centrifugal compressor flow field feature determination model for training respectively for shock wave intensity and shape, supersonic flow area and flow separation area features, to obtain a flow field feature recognition model, including:

[0069] (1) The centrifugal compressor flow field feature determination model extracts shock wave intensity and shape, supersonic flow area and flow separation area features according to the sub-feature training set.

[0070] It should be noted that for the centrifugal compressor flow field feature determination model, when processing the images in the sub-feature training set, the convolution layer of the convolutional neural network (CNN) can be used to extract the shock wave intensity and shape, supersonic flow area and flow separation area features. The convolution kernel slides on the image and captures local features through convolution operations with the image pixel values. For example, for the shock wave intensity and shape features, the convolution kernel design may be sensitive to the areas in the image that reflect the drastic changes in the relative Mach number, because the relative Mach number changes greatly at the shock wave. The convolution operation can highlight the features of these areas, thereby extracting the shock wave intensity (which can be represented by the intensity of the feature response) and shape (such as determining the shape by edge information captured by convolution kernels of different directions and scales). For the supersonic flow area features, the convolution kernel is designed based on the relatively high relative Mach number characteristics of the supersonic flow area. When the convolution kernel slides on the image, it responds strongly to the area exceeding a certain relative Mach number threshold, thereby effectively identifying and extracting the supersonic flow area from the entire flow field image. For the flow separation area features, the convolution kernel focuses on the areas with relatively low Mach numbers and low speeds in the image. These areas correspond to low-energy fluid clusters formed by flow separation. The boundaries, sizes, distribution and other features of these areas are identified through convolution operations.

[0071] (2) Performing dimensionality reduction processing on the extracted features, and connecting the feature vectors after dimensionality reduction to establish a global relationship between all features.

[0072] After extracting the original shock wave intensity and shape, supersonic flow area and flow separation area features, the amount of data is usually large and there may be some redundant information. Dimensionality reduction processing aims to reduce the dimension of the data while retaining key feature information. Commonly used dimensionality reduction methods include pooling operations (such as maximum pooling or average pooling). In terms of shock wave intensity and shape features, the pooling layer downsamples the feature map output by the convolution layer. For example, maximum pooling selects the maximum value in each small area as the new feature value, which can reduce the amount of data while retaining the peak value of shock wave intensity and key information of shape. For the supersonic flow area and flow separation area features, the pooling operation is similar. It simplifies the feature representation of these areas, highlights the main features, reduces the interference caused by subtle changes in the image, and makes the flow field feature determination model have a certain invariance to the position changes of the features, thereby enhancing the robustness of the model.

[0073] The purpose of connecting the reduced-dimensional shock wave intensity and shape, supersonic flow region, and flow separation region feature vectors is to establish a global relationship between all features. Through this connection, the flow field feature determination model can consider the interaction between these features as a whole. For example, in the fully connected layer, the elements of each feature vector are combined according to certain weights and biases. In this way, the flow field feature determination model can learn how changes in shock wave intensity affect the development of the flow separation region, as well as the mutual correlation between the supersonic flow region and the shock wave and flow separation region. The establishment of this global relationship enables the flow field feature determination model to comprehensively consider the synergistic effect of multiple features when processing new flow field distribution cloud maps.

[0074] (3) Based on the global relationship, when a flow field distribution cloud map is input, different flow field feature combination patterns are identified to obtain a flow field feature recognition model.

[0075] It should be noted that when a new flow field distribution cloud map is input into the flow field feature determination model, the flow field feature determination model identifies different flow field feature combination patterns based on the established global relationship. Since the flow field feature determination model has learned the relationship between various feature combination patterns and related information such as surge margin during the training process, it can quickly determine the feature combination of the current flow field based on the features extracted from the input image. For example, if the flow field feature determination model detects a combination pattern of shock waves of a specific intensity and shape, a certain range of supersonic flow areas, and corresponding flow separation area features, it can match it with the previously learned pattern to accurately identify the state of the current flow field and form a flow field feature recognition model.

[0076] S105, using the comprehensive feature training set to perform secondary training on the flow field feature recognition model to obtain a comprehensive flow field feature recognition model.

[0077] It should be noted that the flow field feature recognition model is trained twice using the comprehensive feature training set to obtain a comprehensive flow field feature recognition model, including: inputting the comprehensive feature training set into the flow field feature recognition model to obtain the interactive relationship between the shock wave intensity and shape, the supersonic flow area and the flow separation area characteristics; adjusting the flow field feature recognition model parameters based on the interactive relationship to obtain a comprehensive flow field feature recognition model.

[0078] When the comprehensive feature training set is input into the flow field feature recognition model that has been trained with sub-features, the flow field feature recognition model will process the image data according to its existing structure and parameters. In this process, the flow field feature recognition model uses components such as convolutional layers, pooling layers, and fully connected layers to re-extract the shock wave intensity and shape, supersonic flow area, and flow separation area features in the image. Unlike sub-feature training, the flow field feature recognition model pays more attention to the interaction between these features. For example, the flow field feature recognition model may find that when the shock wave intensity is within a certain range, the shape and size of the supersonic flow area will have a specific correlation with the development of the flow separation area, and this correlation shows a certain pattern in different images.

[0079] By processing a large number of images in the comprehensive feature training set, the flow field feature recognition model can deeply explore the interactive relationship between the shock wave intensity and shape, the supersonic flow area and the flow separation area features. After obtaining the interactive relationship between the features, the flow field feature recognition model calculates the error between the predicted result and the actual label based on the label of the image in the comprehensive feature training set. Based on this error, the back propagation algorithm is used to adjust the parameters of the flow field feature recognition model. For example, if the flow field feature recognition model predicts that the surge boundary distance corresponding to a certain image is significantly different from the actual label, the back propagation algorithm will calculate the gradient of the error for each parameter of the flow field feature recognition model (including the convolution kernel weight of the convolution layer, the parameters of the pooling layer, and the weight and bias of the fully connected layer). Based on this gradient information, the flow field feature recognition model will adjust the parameters to obtain a comprehensive flow field feature recognition model.

[0080] It should be noted that the comprehensive flow field feature recognition model obtained after adjusting the parameters based on the interactive relationship can more accurately reflect the complex relationship between multiple key features in the centrifugal compressor flow field. It is no longer limited to the recognition of a single feature and the understanding of a simple combination, but can more accurately judge the relationship between the flow field state and surge based on these complex interactive relationships. When faced with a new centrifugal compressor geometry or flow field image under operating conditions, the comprehensive flow field feature recognition model can more effectively utilize the interactive information between these features, improve the accuracy of surge margin calculation, and provide a more reliable basis for the performance evaluation and optimization of the centrifugal compressor.

[0081] S106. Calculate the surge margin of the newly generated centrifugal compressor geometry based on the comprehensive flow field feature recognition model.

[0082] Specifically, based on the comprehensive flow field feature recognition model, the surge margin of the newly generated compressor geometry is calculated, including:

[0083] (1) Performing design point flow calculation on the newly generated centrifugal compressor geometry to obtain flow characteristic parameters of the centrifugal compressor geometry.

[0084] It should be noted that the process of design point flow calculation is usually based on computational fluid dynamics (CFD) methods. First, determine the design point back pressure, which is set based on the normal operating range and design conditions of the compressor. Then, set the initial conditions such as the inlet total temperature, total pressure, and airflow angle, and use the design point back pressure as the boundary condition. By solving the basic equations of fluid mechanics, the flow field parameters of the newly generated compressor geometry at the design point can be obtained. These flow characteristic parameters include the distribution of velocity, pressure, temperature, etc. in the flow field, as well as the design point efficiency, etc.

[0085] (2) Inputting the flow characteristic parameters into the comprehensive flow field feature recognition model to identify the flow field feature combination pattern of the centrifugal compressor geometry.

[0086] The information formed by the flow characteristic parameters of the newly generated compressor geometry is input into the comprehensive flow field feature recognition model. The comprehensive flow field feature recognition model processes the input data according to the knowledge learned during the training process. Through structures such as convolutional layers, pooling layers, and fully connected layers, the comprehensive flow field feature recognition model extracts characteristic information such as shock wave intensity and shape, supersonic flow area, and flow separation area from the flow characteristic parameters, and identifies the combination pattern of these features. For example, the comprehensive flow field feature recognition model can determine the intensity and shape characteristics of the shock wave in the current flow field, the range and shape of the supersonic flow area, and the position and size of the flow separation area, and then determine the combination pattern between them, such as whether there is a situation where high-intensity shock waves and large-area flow separation areas appear at the same time.

[0087] (3) The identified flow field feature combination pattern is compared with the operating point characteristics on the characteristic line of the original centrifugal compressor to determine the current flow state of the newly generated centrifugal compressor geometry.

[0088] The identified flow field feature combination pattern of the newly generated compressor geometry is compared with the operating point characteristics on the characteristic line of the original centrifugal compressor. The characteristic line of the original centrifugal compressor contains information on operating points such as the design point, surge boundary, and partial blockage point, each of which has a specific combination of flow field features. By comparison, it can be determined which operating point on the original characteristic line is most similar or close to the flow state of the newly generated compressor geometry at the design point. For example, if the shock wave intensity is strong and the flow separation area is large in the flow field feature combination pattern of the newly generated compressor geometry, which is similar to the characteristics of the operating point close to the surge point on the original characteristic line, then it can be preliminarily judged that the newly generated compressor geometry may be close to the surge state at the design point; if it is consistent with the characteristics of the operating point near the design point, it indicates that it is near the normal design condition. This comparison is based on the principle of centrifugal compressor flow field similarity, that is, compressors with different geometric structures should have similar flow field characteristics and performance under similar flow conditions. Specifically, Figure 5 As shown, the trends of the centrifugal compressor pressure ratio characteristic lines are similar.

[0089] (4) Based on the current flow state and the calculation results of the comprehensive flow field feature identification model, calculate and output the surge margin of the newly generated centrifugal compressor geometry.

[0090] The surge margin is calculated based on the current flow state of the newly generated compressor geometry and the calculation results of the comprehensive flow field feature recognition model. Specifically, the information output by the comprehensive flow field feature recognition model may include the relative position of the current flow field state from the surge boundary, etc. Then, the surge margin calculation formula is used. For the surge margin calculation formula, please refer to the description of the relevant technology and will not be repeated here.

[0091] The method for calculating the surge margin of a centrifugal compressor based on flow field feature recognition provided in the present application obtains the characteristic lines of the original centrifugal compressor flow and pressure and calculates multiple characteristic lines by adjusting the design parameters, and uses the flow field feature recognition model to estimate the surge margin, thereby avoiding a large number of tedious CFD calculations for each sample to accurately search for the surge boundary. Only the original and N characteristic lines after adjusting the design parameters need to be calculated, and the model prediction can be used, which reduces the amount of calculation and improves the calculation efficiency; the surge margin of the newly generated centrifugal compressor geometry is calculated based on the comprehensive flow field feature recognition model. Once the model training is completed, for the new compressor geometry, only the design point flow calculation needs to be performed to obtain the flow characteristic parameters and input them into the model, and the surge margin can be quickly obtained. There is no need for complex iterative calculations like traditional methods, which saves a lot of time. In addition, by training the shock wave intensity and shape, supersonic flow area and flow separation area characteristics respectively, and using the comprehensive feature training set for secondary training, the model can comprehensively and accurately identify the flow field feature combination pattern. Compared with the traditional single feature or limited point estimation method (such as the near-surge point estimation method), it can more accurately reflect the relationship between the actual flow field state inside the compressor and surge, thereby improving the accuracy of surge margin calculation. In addition, by adjusting the design parameters of the original centrifugal compressor to calculate multiple characteristic lines, and extracting the flow field distribution cloud map at different geometries, back pressures and blade heights, this method can adapt to centrifugal compressors with different geometric structures and various operating conditions. Whether it is a newly designed compressor or an improvement of an existing type, this method can be used to calculate the surge margin. It has strong versatility and the whole process is easy to operate.

[0092] Corresponding to the aforementioned embodiment of a method for calculating a surge margin of a centrifugal compressor based on flow field feature recognition, the present application also provides an embodiment of a device for calculating a surge margin of a centrifugal compressor based on flow field feature recognition.

[0093] Embodiment 2:

[0094] Figure 6 This is a schematic diagram of the structure of the second embodiment of the centrifugal compressor surge margin calculation device based on flow field feature recognition provided by this application. Figure 6 , the device provided in this embodiment includes an extraction module 610, an analysis module 620, a segmentation module 630, a training module 640 and a calculation module 650;

[0095] The extraction module 610 is used to obtain the characteristic lines of the flow rate and pressure of the original centrifugal compressor, adjust the design parameters of the original centrifugal compressor to calculate N characteristic lines, and extract the flow field distribution cloud map to form an image set;

[0096] The analysis module 620 is used to analyze the position of the calculation point corresponding to each image in the image set on the characteristic line to determine the label of each image;

[0097] The segmentation module 630 is used to segment the labeled image set into a sub-feature training set and a comprehensive feature training set, wherein the sub-feature training set includes shock wave intensity and shape, supersonic flow area, and flow separation area features;

[0098] The training module 640 is used to input the images in the sub-feature training set into the centrifugal compressor flow field feature determination model to perform training on shock wave intensity and shape, supersonic flow area and flow separation area features, so as to obtain a flow field feature recognition model;

[0099] The training module 640 is further used to perform secondary training on the flow field feature recognition model using the comprehensive feature training set to obtain a comprehensive flow field feature recognition model;

[0100] The calculation module 650 is used to calculate the surge margin of the newly generated centrifugal compressor geometry based on the comprehensive flow field feature recognition model.

[0101] The device of this embodiment can be used to perform Figure 1 The steps, specific implementation principles and implementation processes of the method embodiment shown are similar and will not be repeated here.

[0102] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0103] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0104] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for calculating surge margin of a centrifugal compressor based on flow field feature recognition, characterized in that: The method comprises: Obtaining characteristic lines of flow and pressure of the original centrifugal compressor, adjusting the design parameters of the original centrifugal compressor to calculate N characteristic lines, and extracting the flow field distribution cloud map to form an image set; Analyze the position of the calculation point corresponding to each image in the image set on the characteristic line to determine the label of each image; The labeled image set is divided into a sub-feature training set and a comprehensive feature training set, wherein the sub-feature training set includes shock wave intensity and shape, supersonic flow area, and flow separation area features; Inputting the images in the sub-feature training set into the centrifugal compressor flow field feature determination model to respectively train the shock wave intensity and shape, supersonic flow area and flow separation area features to obtain a flow field feature recognition model; Performing secondary training on the flow field feature recognition model using the comprehensive feature training set to obtain a comprehensive flow field feature recognition model; Calculating the surge margin of the newly generated centrifugal compressor geometry based on the comprehensive flow field feature recognition model; The step of extracting the flow field distribution cloud map to form an image set includes: Analyzing the characteristic line to extract flow field information at the first blade height, the second blade height and the third blade height; Based on the flow field information, flow field distribution cloud images at different geometries, different back pressures and different leaf heights are obtained to form an image set.

2. The method according to claim 1, characterized in that: The calculating of the surge margin of the newly generated compressor geometry based on the comprehensive flow field feature recognition model includes: Performing design point flow calculation on the newly generated centrifugal compressor geometry to obtain flow characteristic parameters of the centrifugal compressor geometry; Inputting the flow characteristic parameters into the comprehensive flow field feature recognition model to identify the flow field feature combination pattern of the centrifugal compressor geometry; The identified flow field feature combination pattern is compared with the operating point characteristics on the characteristic line of the original centrifugal compressor to determine the current flow state of the newly generated centrifugal compressor geometry; Based on the current flow state and the calculation results of the comprehensive flow field feature recognition model, the surge margin of the newly generated centrifugal compressor geometry is calculated and output.

3. The method according to claim 1, characterized in that The image in the sub-feature training set is input into the centrifugal compressor flow field feature determination model for training the shock wave intensity and shape, supersonic flow area and flow separation area features to obtain a flow field feature recognition model, including: The centrifugal compressor flow field feature determination model extracts shock wave intensity and shape, supersonic flow area and flow separation area features according to the sub-feature training set; Performing dimensionality reduction processing on the extracted features, and connecting the feature vectors after dimensionality reduction to establish a global relationship between all features; Based on the global relationship, when the flow field distribution cloud map is input, different flow field feature combination patterns are identified to obtain a flow field feature recognition model.

4. The method according to claim 1, characterized in that The method of performing secondary training on the flow field feature recognition model using the comprehensive feature training set to obtain a comprehensive flow field feature recognition model comprises: Inputting the comprehensive feature training set into the flow field feature recognition model to obtain the interactive relationship between the shock wave intensity and shape, the supersonic flow area and the flow separation area features; The flow field feature recognition model parameters are adjusted based on the interactive relationship to obtain a comprehensive flow field feature recognition model.

5. The method according to claim 1, characterized in that The analyzing the position of the calculation point corresponding to each image in the image set on the characteristic line to determine the label of each image includes: Classify the position of the calculation point of each image on the characteristic line and identify the flow state corresponding to each position; Assign a label to each image according to the flow state category; the label includes at least a design point, a near-blocking point, and a near-breathing point; All images with assigned labels are classified and stored to form a set of images with labels.

6. The method according to claim 1, characterized in that The step of dividing the labeled image set into a sub-feature training set and a comprehensive feature training set comprises: The labeled image set is divided into a sub-feature training set and a comprehensive feature training set according to a first ratio and a second ratio at a random ratio; wherein the first ratio is greater than the second ratio.

7. The method according to claim 1, characterized in that The method of obtaining the characteristic line of the original centrifugal compressor flow rate and pressure includes: Edit calculation files based on full 3D calculation tools to determine design points and boundary conditions; Starting from the design point, the back pressure is increased according to the preset step size, the changes in flow rate and pressure ratio are recorded, and a full three-dimensional calculation is performed after each change in back pressure; When the calculation result is non-convergent, the surge boundary information is obtained; The back pressure is reduced according to a preset step size, the blocking point characteristics are reversely searched, and the characteristic line of the original centrifugal compressor is obtained based on the design point, surge boundary information and the blocking point characteristics; the characteristic line is used to record the characteristic line shape, each working point and surge boundary information.

8. The method according to claim 1, characterized in that The adjusting the design parameters of the original centrifugal compressor to calculate N characteristic lines includes: Adjust the design parameters of the original centrifugal compressor to obtain N different centrifugal compressor models; For each centrifugal compressor model, the back pressure is increased according to the preset step size, and the changes in flow rate and pressure ratio are recorded until the surge boundary is approached; N characteristic lines corresponding to N different centrifugal compressor models are calculated based on the recorded data.

9. A centrifugal compressor surge margin calculation device based on flow field feature recognition, characterized in that: The device comprises an extraction module, an analysis module, a segmentation module, a training module and a calculation module; The extraction module is used to obtain the characteristic lines of the original centrifugal compressor flow and pressure, adjust the design parameters of the original centrifugal compressor to calculate N characteristic lines, and extract the flow field distribution cloud map to form an image set; The analysis module is used to analyze the position of the calculation point corresponding to each image in the image set on the characteristic line to determine the label of each image; The segmentation module is used to segment the image set with labels into a sub-feature training set and a comprehensive feature training set, wherein the sub-feature training set includes shock wave intensity and shape, supersonic flow area and flow separation area features; The training module is used to input the images in the sub-feature training set into the centrifugal compressor flow field feature determination model to perform training on shock wave intensity and shape, supersonic flow area and flow separation area features respectively, so as to obtain a flow field feature recognition model; The training module is further used to perform secondary training on the flow field feature recognition model using the comprehensive feature training set to obtain a comprehensive flow field feature recognition model; The calculation module is used to calculate the surge margin of the newly generated centrifugal compressor geometry based on the comprehensive flow field feature recognition model; The step of extracting the flow field distribution cloud map to form an image set includes: Analyzing the characteristic line to extract flow field information at the first blade height, the second blade height and the third blade height; Based on the flow field information, flow field distribution cloud images at different geometries, different back pressures and different leaf heights are obtained to form an image set.

Images