A method and system for establishing a nozzle profile ablation regression prediction model based on deep learning

Abstract

The present invention proposes a method and system for establishing a nozzle profile ablation regression prediction model based on deep learning. This method constructs a nozzle profile ablation regression prediction model based on the two-dimensional profile ablation data of the existing test nozzle and the wall data obtained by numerical simulation of the corresponding test nozzle under ignition conditions, effectively reducing the prediction time cost and calculation cost while ensuring the prediction accuracy; this method also proposes a method for extracting data points for a single two-dimensional ablation profile curve to form a data set. This method takes points according to the horizontal coordinate of the nozzle, and forms a data set by extracting the wall data obtained by numerical simulation of the points on the two-dimensional profile line corresponding to the horizontal coordinate and the wall regression amount of the ignition experiment. Based on limited experimental data and limited computing resources, the present invention reduces the prediction time cost and calculation cost while ensuring accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of nozzle profile prediction, and in particular to a method and system for establishing a nozzle profile ablation regression prediction model based on deep learning. Background Art

[0002] In solid rocket propulsion systems, the nozzle is a core component for thrust generation. The shape of the nozzle surface directly impacts its performance, making it a key challenge for nozzle designers to select materials and design the nozzle surface based on specific requirements. During solid rocket operation, high-temperature, high-pressure gases are ejected directly from the combustion chamber through the nozzle, subjecting the nozzle surface to erosion by these gases. The combustion chamber exhaust is not only characterized by high temperature and pressure but also contains significant amounts of carbon monoxide, carbon dioxide, water vapor, and solid particles. To reduce weight and increase structural strength, nozzle surface materials are often constructed from carbon-carbon composites. This results in chemical reactions in the nozzle surface material under the influence of the high-temperature gases, as well as material abrasion caused by collisions with solid particles. Consequently, the nozzle surface inevitably changes during operation. Predicting changes in the nozzle surface provides a reference for nozzle design and is an inevitable process.

[0003] The existing method for predicting nozzle profiles mainly involves writing UDF to describe the relationship between the nozzle profile recession and flow field parameters, embedding it into numerical simulation software, and combining dynamic meshes and numerical calculation methods to iteratively simulate the nozzle profile recession. Simulation using numerical methods can effectively meet design requirements within a certain accuracy range. However, due to the limitations of computer hardware conditions and the contradiction between numerical calculation accuracy and computational complexity, numerical simulation methods face the problems of long simulation time and high computational cost. Therefore, how to provide a nozzle profile ablation recession prediction method that can reduce prediction cost and time while ensuring prediction accuracy has become a focus of designers. Summary of the Invention

[0004] The purpose of the present invention is to address the problems of long simulation time and high computational cost faced by the prediction of nozzle surface erosion and regression, and to propose a method and system for establishing a nozzle surface erosion and regression prediction model based on deep learning.

[0005] The present invention is achieved through the following technical solutions. The present invention proposes a method for establishing a nozzle profile ablation regression prediction model based on deep learning, the method comprising the following steps:

[0006] Step 1: Obtain nozzle experimental data based on the predicted nozzle experimental conditions and structure, including geometric model, ignition conditions, and profile curve data before and after the test;

[0007] Step 2: Perform steady-state calculation based on the DPM model and ignition conditions to obtain the wall distribution;

[0008] Step 3: extracting data points from a single two-dimensional ablation profile curve to form a data set;

[0009] Step 4: training the nozzle profile ablation regression prediction model;

[0010] Step 5: Perform steady-state numerical calculations on the nozzle to be predicted, extract the numerical calculation results, form a verification set, and substitute it into the trained nozzle profile ablation regression prediction model to obtain prediction data.

[0011] Furthermore, in step 1, based on the experimental conditions and structure of the predicted nozzle, experimental data of four nozzles with similar experimental conditions and structures to the predicted object but different inlet pressures are obtained, including nozzle inlet pressure, inlet temperature, ambient pressure, ambient temperature, relative molecular mass of gas, specific heat of gas, total mass of gas particles, average diameter of gas particles, maximum particle size, minimum particle size, and profile curve data before and after the nozzle experiment; the nozzles are divided into high-pressure inlet and low-pressure inlet groups according to the inlet pressure, and steps 2 to 3 are all applied to the above four nozzles.

[0012] Furthermore, in step 2, based on the profile curve data and experimental data provided in step 1, Fluent is used to perform steady-state numerical calculations on the experimental nozzle based on the DPM model to obtain the nozzle wall temperature distribution, nozzle wall pressure distribution, nozzle wall shear force distribution, and Oka erosion rate distribution.

[0013] Furthermore, the specific process of step 2 is as follows:

[0014] Step 21: Process the nozzle geometry model;

[0015] Step 22: Discretize the computational domain;

[0016] Step 2 and 3: Set the calculation conditions and perform steady-state numerical calculations;

[0017] Step 24: Adaptive mesh refinement;

[0018] Step 25: Superimpose particle flow for steady-state calculation.

[0019] Furthermore, in step three, the surface ablation regression amount is extracted based on the structure of the experimental nozzle and the numerical calculation data obtained in step two to form a data set. The specific process is as follows:

[0020] Step 31: Mark data points on the profile curve data graph before and after the nozzle test and measure the radial erosion regression;

[0021] Step 32: interpolate to obtain a large number of sampling points and record the horizontal coordinates of the sampling points;

[0022] Step 33: Calculate the horizontal coordinate according to the extracted value of the sampling point horizontal coordinate;

[0023] Steps 3 and 4: Compose a training-test dataset. Use nozzle wall temperature, wall pressure, wall shear force, nozzle material, Oka erosion rate, nozzle area ratio, nozzle throat diameter, and nozzle location as training parameters, and nozzle erosion regression as the prediction object. Combine the experimental data of the four nozzles to form a high- and low-pressure mixed training-test dataset.

[0024] Furthermore, in step 4, the nozzle profile ablation regression prediction model is trained based on the PyCharm platform equipped with PyTorch. The specific steps for generating the nozzle profile ablation regression prediction model are as follows:

[0025] Step 41: Use the fully connected layer as the training network, select the initial parameters of the training network according to the size of the input data set and the number of input parameters, and obtain the fully connected layer architecture;

[0026] Step 42: Set the sample learning rate and the maximum number of training rounds, randomly select 80% of the data points as the training set to train the neural network, and train until the residual is less than 1×10 -4 ;

[0027] Step 43: Use the remaining 20% ​​of data points as the test set and calculate the average error, maximum error, and minimum error between the test set and the experimental data;

[0028] Step 44: Repeat step 42 and step 43 alternately until the maximum error is less than 4%, the average error is less than 0.5%, and the iteration converges, thereby obtaining the ablation regression prediction model.

[0029] Furthermore, in step five, the nozzle to be predicted is subjected to steady-state numerical calculation according to the process shown in step two, and the numerical calculation results of the position to be predicted are extracted according to step three to form a verification set, which is substituted into the nozzle profile ablation regression prediction model obtained in step four to obtain predicted data.

[0030] The present invention also proposes a system for establishing a nozzle profile ablation regression prediction model based on deep learning, the system comprising:

[0031] Acquisition module: To predict the experimental conditions and structure of the nozzle, obtain the nozzle experimental data, including the geometric model, ignition conditions, and profile curve data before and after the test;

[0032] Calculation module: Perform steady-state calculations based on the DPM model and ignition conditions to obtain wall distribution;

[0033] Dataset formation module: extracts data points from a single two-dimensional ablation profile curve to form a data set;

[0034] Training module: training nozzle profile ablation regression prediction model;

[0035] Prediction module: Perform steady-state numerical calculations on the nozzle to be predicted, extract the numerical calculation results, form a verification set, and substitute it into the trained nozzle profile ablation regression prediction model to obtain prediction data.

[0036] The present invention also proposes an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of a method for establishing a nozzle profile ablation and regression prediction model based on deep learning.

[0037] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of a method for establishing a nozzle profile ablation and regression prediction model based on deep learning.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention proposes a method for establishing a nozzle profile ablation regression prediction model based on deep learning. This method constructs a nozzle profile ablation regression prediction model based on the two-dimensional profile ablation data of the existing test nozzle and the wall data obtained by numerical simulation of the corresponding test nozzle under ignition conditions, effectively reducing the prediction time cost and calculation cost while ensuring the prediction accuracy; this method also proposes a method for extracting data points for a single two-dimensional ablation profile curve to form a data set. This method takes points according to the horizontal coordinate of the nozzle, and forms a data set by extracting the wall data obtained by numerical simulation of the points on the two-dimensional profile line corresponding to the horizontal coordinate and the wall regression amount of the ignition experiment. Based on limited experimental data and limited computing resources, the present invention reduces the prediction time cost and calculation cost while ensuring accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0041] Figure 1 This is a flow chart of a method for establishing a nozzle profile ablation and regression prediction model based on deep learning according to the present invention.

[0042] Figure 2 It is the flow chart of steady-state numerical calculation.

[0043] Figure 3 It is a cross-sectional diagram of the computational domain.

[0044] Figure 4 The flowchart of the method for extracting data points from a single two-dimensional ablation profile curve to form a data set.

[0045] Figure 5 It is a schematic diagram of marking and measurement methods.

[0046] Figure 6 This is a comparison diagram of the displacement-axial distance distribution before and after interpolation, where (a) is before interpolation and (b) is after interpolation.

[0047] Figure 7 It is a schematic diagram of the method of forming a sampling circle by slicing.

[0048] Figure 8 This is a flow chart of the ablation-regression model training method.

[0049] Figure 9 This is a schematic diagram of the neural network structure used for training.

[0050] Figure 10 It is a schematic diagram of the residual curve.

[0051] Figure 11 It is a schematic diagram of error distribution.

[0052] Figure 12 It is a schematic diagram of the test set fitting situation.

[0053] Figure 13 It is a schematic diagram of the validation set fitting situation. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0055] Combine Figures 1-13 The present invention proposes a method for establishing a nozzle profile ablation regression prediction model based on deep learning, the method comprising the following steps:

[0056] Step 1: Obtain nozzle experimental data based on the predicted nozzle experimental conditions and structure, including geometric model, ignition conditions, and profile curve data before and after the test;

[0057] In step 1, based on the experimental conditions and structure of the predicted nozzle, experimental data of four nozzles with similar experimental conditions and structures to the predicted object but different inlet pressures are obtained, including nozzle inlet pressure, inlet temperature, ambient pressure, ambient temperature, relative molecular mass of gas, specific heat of gas, total mass of gas particles, average diameter of gas particles, maximum particle size, minimum particle size, and profile curve data before and after the nozzle experiment; the nozzles are divided into high-pressure inlet and low-pressure inlet groups according to the level of inlet pressure. Steps 2 to 3 are all applied to the above four nozzles.

[0058] Step 2: Perform steady-state calculation based on the DPM model and ignition conditions to obtain the wall distribution;

[0059] In step 2, based on the profile curve data and experimental data provided in step 1, a steady-state numerical calculation of the experimental nozzle is performed using Fluent based on the DPM model to obtain the nozzle wall temperature distribution, nozzle wall pressure distribution, nozzle wall shear force distribution, and Oka erosion rate distribution; the specific process of step 2 is as follows:

[0060] Step 21: Process the nozzle geometry model; extract the internal closed geometry to extract the 3D model of its internal flow field. Set the computational domain length to 10 times the nozzle length and the computational domain radius to 10 times the nozzle outlet radius. Add an extension section at the nozzle inlet with a length of 3 times the nozzle inlet radius. Considering the computational convergence, group the large-scale and small-scale geometric regions of the flow field, and establish a BOI encrypted domain at the center of the flow field to obtain the flow field structure.

[0061] Step 22: Discrete the computational domain. By simply estimating the flow velocity within the nozzle, determine the first layer of mesh height for the wall boundary layer, discretize the computational domain using an unstructured meshing method, and refine the mesh in the BOI refinement domain. The mesh is then optimized using mesh orthogonality and aspect ratio as metrics, ultimately yielding the nozzle flow mesh. The mesh volume and mesh quality reports for the four nozzles are shown in Table 1.

[0062] Table 1 Mesh quality report

[0063]

[0064] Steps 2 and 3: Set the calculation conditions and perform steady-state numerical calculations. Considering that the fluid inside the nozzle is a high-temperature, high-pressure gas, the material properties of the fluid are ideal gas according to the kinetic molecular theory, and the viscosity of the gas fluid is calculated using the Sutherland model. Considering that the fluid inside the nozzle meets the compressible flow conditions, the pressure inlet is used as the inlet boundary condition, and the nozzle inlet pressure and inlet temperature are used as the boundary condition parameters; the pressure outlet is used as the outlet boundary condition, and the ambient pressure and ambient temperature are used as the boundary condition parameters. For the wall, a no-slip boundary condition is used. Considering that the later stage focuses on the distribution of physical quantities on the nozzle wall, the calculation uses the k-ωSST turbulence model. Numerical calculations are performed to obtain the calculation results without considering particle erosion at the given inlet pressure and temperature.

[0065] Step 24: Adaptive mesh encryption: Based on the flow field calculated in steps 2 and 3, the computational domain mesh is manually adaptively encrypted using the shock wave capture method. Without changing the calculation settings, a steady-state numerical calculation of the flow field is performed to obtain more accurate calculation results.

[0066] Step 25: Superimpose particle flow rates for steady-state calculations. Considering that the nozzle gas contains a large amount of solid particles, the DPM model is used for calculations based on the flow field calculated in step 24. Since a one-time input of a given particle flow rate will cause calculation divergence, the present invention uses a method of superimposing particle flow rates for numerical calculations. Specifically, 0.05 times the target particle flow rate is taken as the particle flow rate applied for the first time. After the gas flow rate reaches a stable value, the particle flow rate is continuously superimposed with a step size of 0.05 times the target particle flow rate until the target value is reached.

[0067] Step 3: extracting data points from a single two-dimensional ablation profile curve to form a data set;

[0068] In step three, the surface ablation regression amount is extracted based on the structure of the experimental nozzle and the numerical calculation data obtained in step two to form a data set. The specific process is as follows:

[0069] Step 31: Mark the data points on the profile curve data graph before and after the nozzle test, and measure the radial ablation regression; the marking and measurement steps are as follows: Figure 5 shown.

[0070] Step 32: Interpolate to obtain a large number of sampling points and record the horizontal coordinates of the sampling points; sort out the data points in the previous step, and use the amount of retreat as the vertical coordinate, the nozzle axis as the horizontal coordinate axis, the nozzle inlet as the starting point of the coordinate axis, and the airflow ejection direction as the positive horizontal coordinate to obtain a visual relationship between the amount of retreat and the horizontal coordinate. Due to the small number of valid sample point data, linear interpolation is used to expand the number of samples to ensure that the number of samples for each material is greater than 100. The sample distribution before and after interpolation is as follows: Figure 6Recording the horizontal coordinates of all sample points makes it easier to extract sample data from the numerical calculation results.

[0071] Step 3: Extract the numerical calculation coordinates according to the sampling point coordinates; import the numerical calculation results into CFD-Post, and generate a series of cross sections intersecting with the nozzle wall according to the recorded sampling point coordinates to obtain the following: Figure 7 A series of sampling circles are shown. The temperature, pressure, Oka erosion rate, and wall shear force in the X, Y, and Z directions are extracted from the sampling circles. The temperature, pressure, and erosion rate are linearly averaged, and the axial, radial, and circumferential shear forces are obtained through coordinate transformation.

[0072] Steps 3 and 4: Compose a training-test dataset. Use nozzle wall temperature, wall pressure, wall shear force, nozzle material, Oka erosion rate, nozzle area ratio, nozzle throat diameter, and nozzle location as training parameters, and nozzle erosion regression as the prediction object. Combine the experimental data of the four nozzles to form a high- and low-pressure mixed training-test dataset.

[0073] Step 4: training the nozzle profile ablation regression prediction model;

[0074] In step 4, the nozzle profile ablation regression prediction model is trained based on the PyCharm platform equipped with PyTorch. The specific steps for generating the nozzle profile ablation regression prediction model are as follows:

[0075] Step 41: Use the fully connected layer as the training network, select the initial parameters of the training network according to the size of the input data set and the number of input parameters, and obtain the fully connected layer architecture; Figure 9 shown.

[0076] Step 42: Set the sample learning rate and the maximum number of training rounds, randomly select 80% of the data points as the training set to train the neural network, and train until the residual is less than 1×10 -4 ;

[0077] Step 43: Use the remaining 20% ​​of data points as the test set and calculate the average error, maximum error, and minimum error between the test set and the experimental data;

[0078] Step 44: Repeat step 42 and step 43 alternately until the maximum error is less than 4%, the average error is less than 0.5%, and the iteration converges, thereby obtaining the ablation regression prediction model.

[0079] Step 5: Perform steady-state numerical calculations on the nozzle to be predicted, extract the numerical calculation results, form a verification set, and substitute it into the trained nozzle profile ablation regression prediction model to obtain prediction data.

[0080] In step five, the nozzle to be predicted is subjected to steady-state numerical calculation according to the process shown in step two. According to step three, the numerical calculation results of the position to be predicted are extracted to form a verification set, which is substituted into the nozzle profile ablation regression prediction model obtained in step four to obtain prediction data. By combining the high-pressure nozzle samples and the low-pressure nozzle samples for training, a hybrid nozzle ablation prediction model is obtained. Figure 10-11 As shown, the iterative residual of the model is 10 -5 The following is considered to be iterative convergence. The error of the test set on the model can be obtained, the maximum error is 3.52% and the minimum error is 6.81×10 -5 , the average error is 0.32%. It can be considered that the model can fit the ablation trend of the sample used by the model. In order to verify its generalization performance, the test set and validation set are substituted into the model to obtain the following Figure 12-13 The results shown, compared with the actual experimental data, show that the overall predictions and trends of the model are correct.

[0081] The present invention also proposes a system for establishing a nozzle profile ablation regression prediction model based on deep learning, the system comprising:

[0082] Acquisition module: To predict the experimental conditions and structure of the nozzle, obtain the nozzle experimental data, including the geometric model, ignition conditions, and profile curve data before and after the test;

[0083] Calculation module: Perform steady-state calculations based on the DPM model and ignition conditions to obtain wall distribution;

[0084] Dataset formation module: extracts data points from a single two-dimensional ablation profile curve to form a data set;

[0085] Training module: training nozzle profile ablation regression prediction model;

[0086] Prediction module: Perform steady-state numerical calculations on the nozzle to be predicted, extract the numerical calculation results, form a verification set, and substitute it into the trained nozzle profile ablation regression prediction model to obtain prediction data.

[0087] Relevant abbreviations and English definitions in this invention:

[0088]

[0089]

[0090] The present invention also proposes an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of a method for establishing a nozzle profile ablation and regression prediction model based on deep learning.

[0091] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of a method for establishing a nozzle profile ablation and regression prediction model based on deep learning.

[0092] The memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DRRAM). It should be noted that the memory of the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0093] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated therein. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a high-density digital video disc (DVD)), or a semiconductor medium (eg, a solid state disc (SSD)).

[0094] During implementation, each step of the above method can be completed by an integrated logic circuit of the hardware in the processor or by instructions in the form of software. The steps of the method disclosed in conjunction with the embodiments of the present application can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in conjunction with its hardware. To avoid repetition, it will not be described in detail here.

[0095] It should be noted that the processor in the embodiments of the present application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiment can be completed by an integrated logic circuit of the hardware in the processor or by instructions in the form of software. The above processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. The various methods, steps, and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiments of the present application can be directly embodied as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0096] The above is a detailed introduction to the method and system for establishing a nozzle profile erosion and regression prediction model based on deep learning proposed in the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for establishing a nozzle profile ablation regression prediction model based on deep learning, characterized in that: The method comprises the following steps: Step 1: Obtain nozzle experimental data based on the predicted nozzle experimental conditions and structure, including geometric model, ignition conditions, and profile curve data before and after the test; Step 2: Perform steady-state calculation based on the DPM model and ignition conditions to obtain the wall distribution; Step 3: extracting data points from a single two-dimensional ablation profile curve to form a data set; Step 4: training the nozzle profile ablation regression prediction model; Step 5: Perform steady-state numerical calculations on the nozzle to be predicted, extract the numerical calculation results, form a validation set, and substitute it into the trained nozzle profile ablation regression prediction model to obtain prediction data; In step three, the surface ablation regression amount is extracted based on the structure of the experimental nozzle and the numerical calculation data obtained in step two to form a data set. The specific process is as follows: Step 31: Mark data points on the profile curve data graph before and after the nozzle test and measure the radial erosion regression; Step 32: interpolate to obtain a large number of sampling points and record the horizontal coordinates of the sampling points; Step 33: Calculate the horizontal coordinate according to the extracted value of the sampling point horizontal coordinate; Steps 3 and 4: Compose a training-test dataset. Use nozzle wall temperature, wall pressure, wall shear force, nozzle material, Oka erosion rate, nozzle area ratio, nozzle throat diameter, and nozzle location as training parameters, and nozzle erosion regression as the prediction object. Combine the experimental data of the four nozzles to form a high- and low-pressure mixed training-test dataset.

2. The method according to claim 1, characterized in that In step 1, based on the experimental conditions and structure of the predicted nozzle, experimental data of four nozzles with similar experimental conditions and structures to the predicted object but different inlet pressures are obtained, including nozzle inlet pressure, inlet temperature, ambient pressure, ambient temperature, relative molecular mass of gas, specific heat of gas, total mass of gas particles, average diameter of gas particles, maximum particle size, minimum particle size, and profile curve data before and after the nozzle experiment; the nozzles are divided into high-pressure inlet and low-pressure inlet groups according to the level of inlet pressure. Steps 2 to 3 are all applied to the above four nozzles.

3. The method according to claim 2, characterized in that In step 2, based on the profile curve data and experimental data provided in step 1, Fluent is used to perform steady-state numerical calculations on the experimental nozzle based on the DPM model to obtain the nozzle wall temperature distribution, nozzle wall pressure distribution, nozzle wall shear force distribution, and Oka erosion rate distribution.

4. The method according to claim 3, characterized in that The specific process of step 2 is as follows: Step 21: Process the nozzle geometry model; Step 22: Discretize the computational domain; Step 2 and 3: Set the calculation conditions and perform steady-state numerical calculations; Step 24: Adaptive mesh refinement; Step 25: Superimpose particle flow for steady-state calculation.

5. The method according to claim 4, characterized in that In step 4, the nozzle profile ablation regression prediction model is trained based on the PyCharm platform equipped with PyTorch. The specific steps for generating the nozzle profile ablation regression prediction model are as follows: Step 41: Use the fully connected layer as the training network, select the initial parameters of the training network according to the size of the input data set and the number of input parameters, and obtain the fully connected layer architecture; Step 42: Set the sample learning rate and the maximum number of training rounds, randomly select 80% of the data points as the training set to train the neural network, and train until the residual is less than 1×10 -4 ; Step 43: Use the remaining 20% ​​of data points as the test set and calculate the average error, maximum error, and minimum error between the test set and the experimental data; Step 44: Repeat steps 42 and 43 alternately until the maximum error is less than 4%, the average error is less than 0.5%, and the iteration converges, thereby obtaining the ablation regression prediction model.

6. The method according to claim 5, characterized in that In step five, the nozzle to be predicted is subjected to steady-state numerical calculation according to the process shown in step two. According to step three, the numerical calculation results of the position to be predicted are extracted to form a verification set, which is substituted into the nozzle profile ablation regression prediction model obtained in step four to obtain predicted data.

7. A system for establishing a nozzle profile ablation regression prediction model based on deep learning, characterized in that: The system comprises: Acquisition module: To predict the experimental conditions and structure of the nozzle, obtain the nozzle experimental data, including the geometric model, ignition conditions, and profile curve data before and after the test; Calculation module: Perform steady-state calculations based on the DPM model and ignition conditions to obtain wall distribution; Dataset formation module: extracts data points from a single two-dimensional ablation profile curve to form a data set; Training module: training nozzle profile ablation regression prediction model; Prediction module: Perform steady-state numerical calculations on the nozzle to be predicted, extract the numerical calculation results, form a verification set, and substitute it into the trained nozzle profile ablation regression prediction model to obtain prediction data; In the data set formation module, the surface ablation regression amount is extracted based on the structure of the experimental nozzle and the numerical calculation data obtained in the calculation module to form a data set. The specific process is as follows: Mark the data points on the profile curve data graph before and after the nozzle test and measure its radial ablation regression; Interpolate to obtain a large number of sampling points and record the horizontal coordinates of the sampling points; Calculate the abscissa according to the extracted values ​​of the sampling point abscissa; A training-test dataset was constructed. The nozzle wall temperature, wall pressure, wall shear force, nozzle material, Oka erosion rate, nozzle area ratio, nozzle throat diameter, and nozzle position were used as training parameters, and the nozzle ablation regression was used as the prediction object. The experimental data of four nozzles were combined to form a high- and low-pressure mixed training-test dataset.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium for storing computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images