Aero-engine fan outlet flow field multi-parameter intelligent reconstruction method under small sample condition

Abstract

The invention discloses an aero-engine fan outlet flow field multi-parameter intelligent reconstruction method under a small sample condition. The method comprises the following steps: acquiring a three-dimensional distribution characteristic data set of static temperature, static pressure, total temperature and total pressure of a fan outlet flow field under multiple working conditions based on a fan component three-dimensional numerical simulation model; constructing a double-path input intelligent reconstruction model composed of a measurement point up-sampling branch and a section feature branch, and training and optimizing the model by adopting a loss function fusing data constraint and physical mechanism constraint; and inputting the trained model by utilizing actually acquired sparse measuring point temperature and pressure information and fan working condition characteristics, and realizing high-precision reconstruction of multiple parameters of the flow field of the outlet section of the fan. According to the method, dependence on the number of sensors is remarkably reduced, interference to airflow is reduced, and the method has good generalization ability and robustness to measuring point noise and installation errors and is suitable for the field of aero-engine testing and control.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of flow field static temperature, static pressure, total temperature and total pressure reconstruction, in particular to a small sample condition under the fan outlet flow field multi-parameter intelligent reconstruction method of an aero-engine, and can be specifically applied to the field of aero-engine testing. BACKGROUND

[0002] As a highly complex thermodynamic system, the performance and reliability of an aero-engine depend on the accurate perception of internal flow field parameters. Among them, the temperature and pressure distribution of the fan outlet flow field cross-section is the key to representing the working state and stability of the compression system. In the traditional fan outlet temperature and pressure signal measurement, single-point or a few sensors can only provide representative values or average values of the cross-section signal, causing large errors in the sampling accuracy of the flow field parameters. Due to factors such as blade rotation, rotor-stator interference and wall effect, the actual flow field has complex three-dimensional non-uniform distribution characteristics. Accurate reconstruction of the entire cross-section "field" signal of the fan outlet of the aero-engine, rather than isolated "point" signal measurement, can enable the control system to better understand the real working state of the engine and realize the transition from average value-based conservative control to flow field structure-based active control. Specifically, high-resolution flow field pressure reconstruction can capture the pre-choke and pre-stall signs in real time, providing valuable early warning time for the control system and improving the engine's distortion resistance and flight safety. Intelligent flow field temperature reconstruction can identify the quality of fan outlet air temperature and provide reference for combustion chamber wall cooling efficiency, fuel atomization and evaporation, etc., reducing the generation of hot spots at the combustion chamber outlet. In addition, in the field of intelligent engines and digital twins, rich field data can provide important verification dimensions and information depth for engine model calibration, state reconstruction, fault diagnosis and residual life prediction, and is a prerequisite for engine autonomous intelligence and adaptive decision-making.

[0003] The fine and real-time reconstruction of fan outlet flow field temperature and pressure far exceeds the traditional single-point monitoring category, and its importance runs through the whole life cycle of the engine, such as advanced control, stability boundary expansion, life estimation and health management, and is the cornerstone of the development of high-performance, high-reliability and intelligent aero-engines. At present, scholars at home and abroad are still in the initial stage of research on the reconstruction of key cross-section flow fields of aero-engines, and there is no multi-parameter reconstruction method for fan outlet flow field cross-section. The small sample condition under the fan outlet flow field multi-parameter intelligent reconstruction method of an aero-engine proposed in the present application can obtain an intelligent flow field reconstruction model based on ground condition data, realize intelligent reconstruction of temperature and pressure of any cross-section of the fan outlet flow field under ground and high-altitude conditions, greatly reduce the dependence on the number of sensor measuring points, reduce the disturbance to the airflow field, and have robustness to measuring point noise and installation errors. SUMMARY

[0004] The technical problem to be solved by the present application is to overcome the defects of the prior art, based on the urgent need for multi-parameter reconstruction of the fan outlet flow field of an aero-engine, to provide a multi-parameter intelligent reconstruction method for the fan outlet flow field of an aero-engine under small sample conditions, so as to solve the problems of high sensor dependence, large flow field interference, low reconstruction precision, weak model generalization ability and the like in the prior art, and can be used for high-precision reconstruction of the static temperature, static pressure, total temperature and total pressure of the fan outlet flow field of an aero-engine.

[0005] The present application adopts the following technical solutions:

[0006] A multi-parameter intelligent reconstruction method for the fan outlet flow field of an aero-engine under small sample conditions, characterized in that it comprises the following steps:

[0007] Step 1), based on the three-dimensional numerical simulation model of the fan component of an aero-engine, simulating and collecting three-dimensional distribution characteristic data sets of the static temperature, static pressure, total temperature and total pressure of the fan outlet flow field under multiple operating conditions;

[0008] Step 2), constructing a multi-parameter intelligent reconstruction model architecture for the fan outlet flow field of an aero-engine, designing a loss function, and training and optimizing the flow field multi-parameter intelligent reconstruction model;

[0009] Step 3), using the temperature and pressure information of small sample points to reconstruct the static temperature, static pressure, total temperature and total pressure of the outlet flow field.

[0010] Further, the specific steps of step 1) are as follows:

[0011] Step 1.1), for the three-dimensional numerical simulation model of the fan component of an aero-engine, carry out grid division and grid independence verification work to obtain high-quality structured grid;

[0012] Step 1.2), select operating condition points of the fan component under different inlet total temperature, inlet total pressure, flow rate and rotational speed, set the three-dimensional numerical simulation flow model, boundary conditions and calculation parameters of the fan component;

[0013] Step 1.3), obtain the performance parameters of the fan component and the cross-sectional static temperature, static pressure, total temperature and total pressure distribution of the outlet flow field;

[0014] Step 1.4), repeat steps 1.2) to 1.3) until the flow field multi-parameter three-dimensional distribution characteristic data set covering multiple operating condition points under different inlet total temperature, total pressure, equivalent rotational speed and equivalent flow rate is obtained;

[0015] Step 1.5), normalize the three-dimensional distribution characteristic data of the flow field static temperature, static pressure, total temperature and total pressure, and reduce the dimension of the single-channel flow field distribution.

[0016] Further, the specific steps of step 2) are as follows:

[0017] Step 2.1) The intelligent reconstruction model architecture of the multi-parameter flow field at the aero-engine fan outlet consists of two input branches: the measurement point upsampling branch and the cross-sectional feature branch. The latter branch is merged to output the multi-parameter distribution of the flow field in a single-channel cross-section.

[0018] Step 2.2) The loss function of the multi-parameter intelligent reconstruction model of the air engine fan outlet flow field consists of two parts: data constraints and mechanism constraints. The data constraints minimize the numerical error between the model-reconstructed flow field and the CFD simulation flow field. The mechanism constraints minimize the numerical error between the static temperature and the total temperature, the static pressure and the total pressure at the same point in the model-reconstructed flow field, and the Mach number 1 calculated from the static temperature and the total temperature and the Mach number 2 calculated from the static pressure and the total pressure.

[0019] Step 2.3) Initialize the network structure parameters of the multi-parameter intelligent reconstruction model of the flow field, design the temperature and pressure multi-parameter flow field reconstruction accuracy evaluation index, and train, test and optimize the multi-parameter intelligent reconstruction model of the flow field based on the dataset obtained in Step 1).

[0020] Furthermore, the specific steps of step 2.1) are as follows:

[0021] Step 2.1.1), the input of the sampling branch at the measurement point is the static temperature, static pressure, total temperature, and total pressure signals of the probe measurement point. The branch structure includes modules such as interpolation, transposed convolution, convolution, pooling, dropout, batch normalization, and fully connected.

[0022] Step 2.1.2), the input for the cross-sectional feature branch is the fan's reduced speed, pressure ratio, and cross-sectional position, and the branch structure includes a fully connected module;

[0023] Step 2.1.3), after the sampling branch and the cross-sectional feature branch at the measuring point are merged, the single-channel cross-sectional static temperature, static pressure, total temperature, and total pressure intelligent reconstructed flow field is output through the fully connected module.

[0024] Furthermore, the specific steps of step 2.3) are as follows:

[0025] Step 2.3.1), based on experience and trial and error, select the branch structure, number of layers, and number of nodes of the multi-parameter intelligent reconstruction model network for the flow field;

[0026] Step 2.3.2) divide the dataset into a ground condition training set and a ground-high-altitude condition test set, and use the Adam optimizer to train in batches at a specified learning rate;

[0027] Step 2.3.3) evaluates the relative error, root mean square error, peak signal-to-noise ratio, and structural similarity index of different cross sections in the reconstructed static temperature, static pressure, total temperature, and total pressure flow fields under ground and high-altitude conditions.

[0028] Step 2.3.4) Evaluate the accuracy of ground and upper-air flow field reconstruction, continuously adjust the network branch type, number of layers, number of nodes, learning rate and batch size, and train and test the intelligent flow field reconstruction model;

[0029] Furthermore, the specific steps of step 3) are as follows:

[0030] Step 3.1) Record the current total temperature and total pressure at the fan inlet, as well as the fan's equivalent speed, pressure ratio, and the location of the reconstructed cross section;

[0031] Step 3.2) Record the static temperature, static pressure, total temperature, and total pressure of the measurement point signals in the probe sensor;

[0032] Step 3.3) Based on the normalized input data, the flow field distribution of static temperature, static pressure, total temperature and total pressure of the cross section is reconstructed using the multi-parameter intelligent reconstruction model of the flow field, and the multi-parameter distribution of the current fan component outlet flow field cross section is obtained by inverse normalization.

[0033] The present invention has the following beneficial effects:

[0034] (1) The intelligent reconstruction model architecture of the flow field multi-parameter in this invention can realize the mapping of the relationship between the distribution of multiple physical quantities in the high-dimensional flow field and small sample points, thereby reducing the number of sensors required for multi-parameter reconstruction of the flow field.

[0035] (2) In this invention, the loss function integrates physical mechanism constraints with data-driven approaches, and relies on training with only a limited number of ground working points to break through the limitation of the generalization ability of the reconstructed model for ground and high-altitude working points.

[0036] (3) In this invention, the model input adopts a design that combines probe-type temperature and pressure measurement point signals with fan performance characteristics. This design minimizes interference with the airflow field, improves the model's ability to reconstruct multiple parameters of the flow field, and is robust to measurement point noise and installation errors. Attached Figure Description

[0037] Figure 1 This is a flowchart of a multi-parameter intelligent reconstruction method for the flow field at the outlet of an aero-engine fan under small sample conditions.

[0038] Figure 2 This is a verification diagram of the mesh independence of the 3D numerical simulation model of the fan.

[0039] Figure 3 This is a schematic diagram of the flow field distribution in a single channel at the fan outlet.

[0040] Figure 4 This is a schematic diagram showing the location of static temperature, static pressure, total temperature, and total pressure signals at probe-type measuring points.

[0041] Figure 5It is a multi-parameter intelligent reconfiguration model architecture for the flow field at the outlet of an aero-engine fan.

[0042] Figure 6 It is the loss function curve during the training process of the multi-parameter intelligent reconstruction model of the flow field.

[0043] Figure 7 This is a diagram showing the static temperature reconstruction effect of the multi-parameter intelligent reconstruction model of the flow field.

[0044] Figure 8 This is a static pressure reconstruction effect diagram of the multi-parameter intelligent reconstruction model of the flow field.

[0045] Figure 9 This is a diagram showing the total temperature reconstruction effect of the multi-parameter intelligent reconstruction model of the flow field.

[0046] Figure 10 This is a diagram showing the total pressure reconstruction effect of the multi-parameter intelligent reconstruction model of the flow field.

[0047] Figure 11 This is a diagram showing the total pressure reconstruction effect of the multi-parameter intelligent reconstruction model of the flow field under noise. Detailed Implementation

[0048] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] To achieve multi-parameter reconstruction of the flow field at the outlet of an aero-engine fan, this invention proposes an intelligent multi-parameter reconstruction method for the flow field at the outlet of an aero-engine fan under small sample conditions. This method can be used for high-precision reconstruction of static temperature, static pressure, total temperature, and total pressure at the outlet of an aero-engine fan. A specific embodiment of this invention takes a certain type of single-stage axial fan as an example. The intelligent multi-parameter reconstruction method for the flow field at the outlet of this aero-engine fan component includes the following steps:

[0050] Step 1) Based on a three-dimensional numerical simulation model of a single-stage fan component of an aero-engine, obtain a three-dimensional distribution feature dataset of static temperature, static pressure, total temperature, and total pressure flow fields at multiple cross-sections of the fan outlet under various operating conditions, including the following steps:

[0051] Step 1.1) For the three-dimensional numerical simulation model of a single-stage fan component of a certain type of aero-engine, mesh generation and mesh independence verification were carried out to obtain a high-quality structured mesh. The mesh of the single-stage fan three-dimensional numerical simulation model was divided into five types according to the number of mesh nodes: mesh 1 (158w), mesh 2 (220w), mesh 3 (286w), mesh 4 (354w), and mesh 5 (425w). The fan efficiency under different mesh node numbers is as follows: Figure 2As shown, the accuracy of the calculation results gradually improves with the increase of the number of grid nodes. When the number of grid nodes reaches 4 (354w), the calculation results tend to stabilize, and at this point, adding more nodes has little impact on the results. Therefore, under the premise of ensuring calculation accuracy, considering the computer hardware configuration and calculation time, a grid partitioning scheme with 4 (354w) grid nodes is selected.

[0052] Step 1.2) Select operating points of the fan component under different inlet total temperature, inlet total pressure, flow rate and speed, and set the three-dimensional numerical simulation SA one-equation flow model, inlet and outlet boundary conditions and calculation parameters of the fan component;

[0053] Step 1.3) Obtain the fan pressure ratio and the static temperature, static pressure, total temperature, and total pressure distribution at its outlet flow field cross-section;

[0054] Step 1.4), repeat steps 1.2) through 1.3) until the multi-parameter three-dimensional distribution characteristic data of the flow field at 44 operating points are obtained. The flow field simulation operating points are shown in Table 1.

[0055] Table 1 Simulation Operating Conditions for Temperature, Pressure, and Flow Field

[0056]

[0057] Step 1.5) Normalizes the multi-parameter three-dimensional distribution characteristic data of the flow field, where pressure ratio, reduced rotational speed, and cross-sectional position are normalized using a maximum-minimum method, and static temperature... static pressure Total temperature Total pressure The normalization method is as follows:

[0058]

[0059] In the formula, These are the parameters for static temperature, static pressure, total temperature, and total pressure. and For all operating conditions The maximum and minimum values, These are the normalized static temperature, static pressure, total temperature, and total pressure parameters. The similarity coefficient. Under the condition of similar flows, The calculation formula is:

[0060]

[0061] In the formula, and These are the total temperature and total pressure at the fan inlet, respectively. Next, regarding... Figure 3The single-channel flow field distribution shown is reduced in dimension to obtain The size of the single-channel, single-section flow field data is represented. Here, 4 represents the number of physical parameters, 50 represents the width of the cross-sectional data distribution, and 120 represents the length of the cross-sectional data distribution. Due to the irregularity of the fan's single-cycle airflow channel, the flow field data outside the single-cycle airflow channel is set to 0 for simplified calculation.

[0062] Step 2) Construct a multi-parameter intelligent reconstruction model architecture for the flow field at the aero-engine fan outlet, design a loss function, train and optimize the multi-parameter intelligent reconstruction model for the flow field, including the following steps:

[0063] Step 2.1), the intelligent reconstruction model architecture for the multi-parameter flow field at the aero-engine fan outlet consists of two input branches: a measurement point upsampling branch and a cross-sectional feature branch. The latter branch is merged to output the multi-parameter distribution of the single-channel cross-sectional flow field, including the following steps:

[0064] Step 2.1.1), the input for the sampling branch at the measurement point is... Figure 4 The static temperature, static pressure, total temperature, and total pressure signals of the probe-type measuring points shown have a branch structure including modules such as interpolation, transposed convolution, convolution, pooling, dropout, batch normalization, and fully connected layers.

[0065] Step 2.1.2), the input for the cross-sectional feature branch is the fan's reduced speed, pressure ratio, and cross-sectional position, and the branch structure includes a fully connected module;

[0066] Step 2.1.3): After merging the sampling branch and the cross-sectional feature branch at the measuring point, the single-channel cross-sectional static temperature, static pressure, total temperature, and total pressure model is output through the merged branch to reconstruct the flow field. The final determined architecture of the multi-parameter intelligent reconstruction model of the aero-engine fan outlet flow field is as follows: Figure 5 As shown.

[0067] Step 2.2), Loss function of the multi-parameter intelligent reconstruction model of the air-engine fan outlet flow field Data constraints Mechanism constraints Two parts, namely

[0068]

[0069] In the formula, make and They are on the same order of magnitude. Data constraints enable the model to reconstruct the flow field parameters. With CFD simulation flow field parameters To minimize the numerical error between them, the root mean square error can be selected. To characterize, that is

[0070]

[0071] In the formula, This refers to the number of reconstructed flow field data points for a single cross-section. Mechanistic constraints include the reconstructed static temperature at the same point in the model's reconstructed flow field. No more than the total reconstructed temperature Reconstructing static pressure No more than the total reconfiguration pressure The Mach number obtained from the reconstructed static temperature and reconstructed total temperature. Mach number obtained from reconstructed static pressure and reconstructed total pressure The numerical error between them should be minimized, which can be expressed as:

[0072]

[0073]

[0074]

[0075]

[0076]

[0077]

[0078] In the formula, , , These are the mechanistic losses related to the temperature, pressure, and Mach number terms, respectively.

[0079] Step 2.3) Initialize the network structure parameters of the multi-parameter intelligent reconstruction model for the flow field, design the temperature and pressure multi-parameter flow field reconstruction accuracy evaluation index, and based on the dataset obtained in Step 1), train, test, and optimize the multi-parameter intelligent reconstruction model for the flow field, including the following steps:

[0080] Step 2.3.1), based on experience and trial and error, select the branch structure, number of layers, and number of nodes of the multi-parameter intelligent reconstruction model network for the flow field, and set the dropout regularization ratio to 0.2;

[0081] Step 2.3.2) divides the dataset shown in Table 1 into a ground-based training set and a ground-to-high-altitude testing set. The ground-based testing set includes conditions 3, 7, 14, 16, 25, 32, and 35, while the high-altitude testing set includes conditions 21-22 and 40-44. The remainder is the ground-based training set. The Adam optimizer is used, with an initial learning rate of 0.005, decreasing to 0.1 every 50 rounds, and training is performed using the mini-batch gradient descent algorithm. The loss function during training is as follows: Figure 6 As shown;

[0082] Step 2.3.3) involves reconstructing the flow field under ground and high-altitude conditions, including static temperature, static pressure, total temperature, and total pressure, and determining the surface relative errors of different cross-sections. Root mean square error Peak signal-to-noise ratio Structural similarity index Evaluate accordingly;

[0083] Step 2.3.4) Evaluate the accuracy of ground and upper-air flow field reconstruction, continuously adjust the network branch type, number of layers, number of nodes, learning rate and batch size, and train and test the multi-parameter intelligent flow field reconstruction model until the required reconstruction accuracy is obtained;

[0084] Step 3) Reconstructs the outlet flow field temperature and pressure using the temperature and pressure information from sparse measuring points, including the following steps:

[0085] Step 3.1) Record the current fan inlet total temperature and total pressure operating conditions, as well as the fan equivalent speed, pressure ratio and reconstruction section position. For example, the inlet total temperature is 288K, the inlet total pressure is 101325Pa, the equivalent speed is 0.9, the pressure ratio is 1.436, and the reconstruction section position is 0.5.

[0086] Step 3.2) Record the static temperature, static pressure, total temperature, and total pressure of the measurement point signals in the probe-type composite sensor;

[0087] Step 3.3) Based on the normalized input data, the static temperature, static pressure, total temperature, and total pressure distributions of the flow field cross section are reconstructed using a multi-parameter intelligent reconstruction model of the flow field, such as... Figures 7-10 As shown. It can be seen that in the reconstructed flow field of the multi-parameter intelligent reconstruction model of the flow field, The maximum value is 1.31% of the total pressure in the reconstructed flow field, and the minimum value is 0.24% of the total temperature in the reconstructed flow field; the average relative reconstruction error The maximum value is 1.55% of the total pressure in the reconstructed flow field, and the minimum value is 0.62% of the total temperature in the reconstructed flow field; the maximum relative reconstruction error The maximum value is 8.81% of the static temperature in the reconstructed flow field, and the minimum value is 2.44% of the total temperature in the reconstructed flow field. These values ​​indicate that the multi-parameter intelligent reconstruction model of the flow field has excellent capabilities in reconstructing the static temperature, static pressure, total temperature, and total pressure of the flow field across the entire cross section.

[0088] In addition, peak signal-to-noise ratio Similarity index Peak signal-to-noise ratio (PSNR) is a commonly used metric for measuring the similarity between two images. A value between 30dB and 40dB indicates good reconstructed image quality; distortion is perceptible but acceptable. A value greater than 40dB indicates excellent reconstructed image quality. In comparison, the structural similarity index... It better reflects the human eye's perception of image quality. When A value greater than 0.9 indicates that the two images are similar, and the changes are perceptible but acceptable. The closer the value is to 1, the more identical the two images are. This is important for reconstructing the static temperature, static pressure, total temperature, and total pressure parameters of the flow field. The minimum value is 35.85 dB. The minimum value of 0.9797 further demonstrates that the intelligent multi-parameter flow field reconstruction model can reconstruct the static temperature, static pressure, total temperature, and total pressure of the flow field with high accuracy. Finally, after inverse normalization of the reconstructed flow field, the multi-parameter distribution of the current fan component outlet flow field cross-section can be obtained.

[0089] To verify the robustness of the method described in this invention, the positions of the measurement points in the model input were changed, randomly shifting them longitudinally by ±3 coordinate points and laterally by ±6 coordinate points, thereby simulating measurement point noise and installation errors. In reconstructing the multi-parameter distribution of the flow field, using... Figure 11 Taking the total pressure distribution of the reconstructed flow field as an example, it can be seen that... , , They were 1.62%, 2.05%, and 9.30%, respectively. , The values ​​were 33.98 dB and 0.9754 dB, respectively. Compared with the total pressure distribution in the noiseless reconstructed flow field, although... , , It has increased somewhat. , While the impact has decreased somewhat, the reconstruction effect remains quite good. Similarly, the reconstruction effect of static temperature, static pressure, and total temperature parameters of the flow field can be analyzed, and the results show that the multi-parameter intelligent reconstruction model of the flow field is robust to measurement point noise and installation errors.

[0090] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent reconstruction of multi-parameter flow field at the outlet of an aero-engine fan under small sample conditions, characterized in that, Includes the following steps: Step 1) Based on the three-dimensional numerical simulation model of the aero-engine fan component, the three-dimensional distribution feature dataset of static temperature, static pressure, total temperature and total pressure of the fan outlet flow field under multiple operating conditions is collected and the dataset is normalized. Step 2) Construct a multi-parameter intelligent reconstruction model of the airflow field at the exhaust of an aero-engine fan, and train and optimize the model using the dataset obtained in Step 1); wherein, the model adopts a dual-input architecture consisting of a measurement point upsampling branch and a cross-sectional feature branch; the training of the model adopts a loss function that integrates data constraints and physical mechanism constraints; Step 3) Using the actual collected sparse measurement point temperature and pressure information, combined with the current operating characteristics of the fan, input it into the multi-parameter intelligent reconstruction model of the flow field trained and optimized in Step 2), and output the reconstructed complete flow field distribution of static temperature, static pressure, total temperature and total pressure at the fan outlet section.

2. The method according to claim 1, characterized in that, Step 1) specifically includes: Step 1.1) Perform mesh generation and independence verification on the three-dimensional numerical simulation model of the fan component to obtain a computational mesh that meets the accuracy requirements; Step 1.2) Set up operating conditions covering different inlet total temperature, inlet total pressure, flow rate, and rotational speed, and configure the corresponding flow model and boundary conditions; Step 1.3) Perform simulation calculations to obtain the fan performance parameters and the three-dimensional distribution of static temperature, static pressure, total temperature, and total pressure at the outlet flow field section under various operating conditions; Step 1.4), repeat steps 1.2) to 1.3), and accumulate multiple working condition samples that constitute the three-dimensional distribution feature dataset; Step 1.5) Normalize the three-dimensional distribution feature dataset.

3. The method according to claim 2, characterized in that, Step 1) further includes: Step 1.6) reduces the normalized single-channel three-dimensional flow field data to a multi-section two-dimensional matrix form.

4. The method according to claim 1, characterized in that, In step 2): The input to the upsampling branch of the measuring point is the static temperature, static pressure, total temperature, and total pressure signals of the probe measuring point. This branch performs upsampling and feature extraction on sparse measuring points through a neural network module that includes interpolation, transpose convolution, and convolution. The input to the cross-sectional feature branch is a global feature characterizing the fan's operating state, including the fan's reduced speed, pressure ratio, and reconstructed cross-sectional position. This branch performs feature encoding through a fully connected layer. The features of the two branches, after being merged and processed by a subsequent neural network, ultimately output the intelligent reconstructed flow field of static temperature, static pressure, total temperature, and total pressure of a single-channel cross-section.

5. The method according to claim 4, characterized in that, In step 2), the physical mechanism constraints include at least: The static temperature mechanism constraint is used to ensure that the static temperature value at any point in the reconstructed flow field does not exceed its total temperature value. Static pressure mechanism constraints are used to ensure that the static pressure value at any point in the reconstructed flow field does not exceed its total pressure value; Mach number consistency constraints are used to ensure that the Mach number obtained from the reconstructed static temperature and total temperature calculations is numerically consistent with the Mach number obtained from the reconstructed static pressure and total pressure calculations.

6. The method according to claim 4 or 5, characterized in that, The training and optimization of the model in step 2) specifically includes: Step 2.1) divide the dataset obtained in Step 1) into a ground condition training set and a test set containing high-altitude conditions; Step 2.2) Using the training set, train the model in batches based on the Adam optimizer and adopt a learning rate decay strategy; Step 2.3) Using the test set, the reconstruction accuracy of the model under ground and high-altitude conditions is comprehensively evaluated by multiple indicators, including surface reconstruction relative error, root mean square error, peak signal-to-noise ratio, and structural similarity index. Step 2.4) Based on the evaluation results, repeatedly adjust the model hyperparameters and structure until the model reaches the preset reconstruction accuracy requirements on both the training and test sets.

7. The method according to claim 1, characterized in that, Step 3) specifically includes: Step 3.1) Obtain the current fan's inlet total temperature, inlet total pressure, equivalent speed, pressure ratio, and the location information of the section to be reconstructed; Step 3.2) Simultaneously acquire the static temperature, static pressure, total temperature, and total pressure signals from the sparse measurement points measured by the probe-type sensor installed at the fan outlet; Step 3.3) The input information obtained in Steps 3.1) and 3.2) is normalized according to the same rules as in Step 1) and then input into the intelligent reconstruction model trained in Step 2) to obtain the normalized reconstructed flow field. After inverse normalization, the multi-parameter distribution of the fan outlet flow field section in a physical sense is finally obtained.

Images