CN-121980678-A - Propellant stretching compression mapping relation establishment method based on machine learning

CN121980678ACN 121980678 ACN121980678 ACN 121980678ACN-121980678-A

Abstract

The invention relates to a method for establishing a propellant stretching compression mapping relation based on machine learning, which comprises the steps of extracting propellant stretching characteristic parameters from a mechanical property curve of a stretching test, extracting propellant compression characteristic parameters from the mechanical property curve of a surface compression property test, respectively carrying out average treatment on the propellant stretching characteristic parameters and the propellant compression characteristic parameters under the same working condition as a sample, establishing a sample library to divide a training set and a testing set, respectively adopting three machine learning models of Gaussian process regression, support vector regression and random forest regression, training the models by using the training set to establish a mapping relation between the surface compression characteristic parameters and the body stretching characteristic parameters, outputting the propellant stretching characteristic parameters, carrying out comparison and mutual inspection among the models by using the testing set, and predicting the propellant uniaxial stretching characteristic parameters by using the selected models. The method can solve the problem that the tensile mechanical property of the propellant cannot be directly obtained in the long-term storage process of the solid engine.

Inventors

LI SHASHA
Yuan Jiuzuo
XU XUN
HUANG JIANGLIU
CHEN DIAN
YIN CHAO
ZHANG HUIKUN

Assignees

上海航天化工应用研究所

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (8)

1. The method for establishing the propellant stretching compression mapping relation based on machine learning is characterized by comprising the following steps of: according to the mechanical property characteristics of the propellant, a test matrix is designed, and a tensile property test of a propellant sample is carried out to obtain a mechanical property curve of the tensile test; Carrying out in-situ detection-based surface compression performance test on the same propellant sample to obtain a mechanical performance curve of the surface compression performance test, wherein the compression rate of the propellant sample in the compression process corresponds to the stretching rate; Extracting propellant stretching characteristic parameters from a mechanical property curve of a stretching test, wherein the propellant stretching characteristic parameters comprise elastic modulus, tensile strength and elongation of a propellant sample in a uniaxial stretching test, extracting propellant compression characteristic parameters from the mechanical property curve of a surface compression property test, wherein the propellant compression characteristic parameters comprise propellant surface compression ratio, maximum indentation load and maximum indentation displacement, respectively carrying out average treatment on the propellant stretching characteristic parameters and the propellant compression characteristic parameters under the same working condition to serve as a sample, establishing a sample library, and dividing the sample library into a training set and a test set; Three machine learning models of Gaussian process regression, support vector regression and random forest regression are adopted respectively, the three models are trained by a machine learning method through a training set, a mapping relation between surface compression and body stretching characteristic parameters is established, and propellant stretching characteristic parameters are output; comparing and verifying the prediction precision of the three machine learning models by using a test set, and selecting a machine learning model with the largest average relative error and the smallest overall average relative error; Inputting the propellant surface compression characteristic parameters based on in-situ detection into a selected machine learning model, and predicting according to the mapping relation established by the model to obtain the propellant uniaxial tension characteristic parameters.
2. The method for establishing a propellant tensile compression mapping relation based on machine learning according to claim 1, wherein when carrying out a propellant sample tensile property test and a surface compression property test based on in-situ detection, the requirements include: the test temperature is-40 to +70 ℃, the stretching rate is 20mm/min and 100mm/min, and the probe pressing-in rate of the in-situ detection equipment corresponding to the mechanical properties of the surface is 0.125m/s and 0.33m/s; The propellant test sample state is the initial state, the propellant test sample is respectively subjected to different ageing temperatures including +40, +60, +70 ℃, different ageing times including 0, 30 days, 60 days, 90 days, 120 days, 240 days, 300 days, 360 days, 480 days, 540 days and 600 days, and different strain conditions including 0, 10% and 15%, and the test amount of each test sample is not less than 3 groups.
3. The method for establishing the propellant stretching compression mapping relation based on machine learning according to claim 1, wherein the ratio of the training set to the testing set is 8:2.
4. The method for establishing the propellant stretching compression mapping relation based on machine learning according to claim 1, wherein a kernel function of Gaussian process regression adopts a mixed kernel form, and a rational secondary kernel function, a radial basis kernel function, a Matern kernel function and a basic kernel function are selected as constituent components of the mixed kernel function, so that the generalization capability of a model is enhanced.
5. The method for establishing the propellant stretching compression mapping relation based on machine learning according to claim 1, wherein the prediction precision of three machine learning models is compared and verified by using a test set, and the method is specifically as follows: Randomly dividing the training set and the test set for K times, and comparing the performance of all prediction errors of each test set divided for K times; After each machine learning model is trained by the ith training set, the jth sample in the ith test set is predicted to obtain a predicted value Assuming that the sample is true According to Obtaining the relative error of the jth sample Suppose that the ith test set shares According to the test sample Obtaining the average relative error of the ith test set Calculating the relative errors of all samples of each test set according to Obtaining the maximum average relative error According to Obtaining the overall average relative error Finally, respective three machine learning models are obtained And (3) with And mutually checking and evaluating the accuracy of the prediction effect.
6. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor performs the steps of the method according to any one of claims 1 to 5.
7. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any one of claims 1-5 when the computer program is executed.
8. A computer program product comprising a computer program which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 5.

Description

Propellant stretching compression mapping relation establishment method based on machine learning Technical Field The invention belongs to the technical field of solid rocket propellants, and relates to a propellant stretching compression mapping relation establishment method based on machine learning. Background The mechanical property of the propellant is a main basis for the design of a loading structure, in order to evaluate the mechanical property and the integrity of the loading structure of the propellant in engineering, a standard dumbbell-shaped sample is usually prepared from a propellant square billet to carry out a uniaxial tensile test, and the tensile mechanical property parameters of the propellant are obtained to evaluate the performance of an engine. However, in the engine, the propellant is in a stress state for a long time, and the result predicted by the propellant square billet test has a certain difference from the actual engine. Constrained by the testing technique and the use characteristics of the engine, the propellant tensile properties cannot be directly tested in a full-size engine, and if the engine is subjected to the anatomical sampling test, the resource waste is caused. With the development of sensing technology, the mechanical properties of the propellant can be evaluated in real time in an engine in an in-situ detection mode. However, in-situ monitoring of the mechanical properties of the propellant charge requires the use of compression-based testing methods. For the particle reinforced composite solid propellant, the tensile and compression mechanical properties of the composite solid propellant show a complex mapping relation. Therefore, a prediction model from compression to stretching is established according to the mapping relation between compression and stretching of the propellant, compression data detected in situ are input, and various mechanical index data of stretching can be predicted accurately, so that engineering cost is reduced, and engineering efficiency is improved. Disclosure of Invention The invention solves the technical problems of overcoming the defects of the prior art, providing a propellant stretching compression mapping relation establishment method based on machine learning, and solving the technical problems that the stretching mechanical property of the propellant cannot be directly tested and obtained in the long-term storage process of the existing solid engine. The invention provides a propellant stretching compression mapping relation establishment method based on machine learning, which comprises the following steps: according to the mechanical property characteristics of the propellant, a test matrix is designed, and a tensile property test of a propellant sample is carried out to obtain a mechanical property curve of the tensile test; Carrying out in-situ detection-based surface compression performance test on the same propellant sample to obtain a mechanical performance curve of the surface compression performance test, wherein the compression rate of the propellant sample in the compression process corresponds to the stretching rate; Extracting propellant stretching characteristic parameters from a mechanical property curve of a stretching test, wherein the propellant stretching characteristic parameters comprise elastic modulus, tensile strength and elongation of a propellant sample in a uniaxial stretching test, extracting propellant compression characteristic parameters from the mechanical property curve of a surface compression property test, wherein the propellant compression characteristic parameters comprise propellant surface compression ratio, maximum indentation load and maximum indentation displacement, respectively carrying out average treatment on the propellant stretching characteristic parameters and the propellant compression characteristic parameters under the same working condition to serve as a sample, establishing a sample library, and dividing the sample library into a training set and a test set; Three machine learning models of Gaussian process regression, support vector regression and random forest regression are adopted respectively, the three models are trained by a machine learning method through a training set, a mapping relation between surface compression and body stretching characteristic parameters is established, and propellant stretching characteristic parameters are output; comparing and verifying the prediction precision of the three machine learning models by using a test set, and selecting a machine learning model with the largest average relative error and the smallest overall average relative error; Inputting the propellant surface compression characteristic parameters based on in-situ detection into a selected machine learning model, and predicting according to the mapping relation established by the model to obtain the propellant uniaxial tension characteristic parameters. Further, when car