Solid rocket engine storage performance degradation analysis method based on sample enhancement

Abstract

The invention relates to the technical field of storage performance of rocket engines, in particular to a solid rocket engine storage performance degradation analysis method based on sample enhancement, and the method comprises the steps: obtaining original sample data; constructing a generative adversarial network, and training to obtain a target pair generative adversarial network; obtaining an enhanced new sample data set; constructing a degradation model, and training to obtain a target degradation model; and predicting current storage performance data. According to the invention, the precision of predicting the storage performance data is improved.

Description

Technical Field

[0001] The invention relates to the technical field of storage performance of rocket engines, and in particular to a storage performance degradation analysis method for solid rocket engines based on sample enhancement. Background Art

[0002] The existing solid rocket motor storage performance degradation analysis method usually uses the Arrhenius equation for linear fitting to construct performance degradation models at different temperatures. The Arrhenius equation is mainly used to describe the relationship between chemical reaction rate and temperature. Based on its principle, the model fitting relationship corresponding to the propellant aging test data at 50℃, 60℃, 70℃ and 80℃ is obtained.

[0003] Through the Alineux equation and linear fitting method, we can get the curve of each physical quantity changing with time. By analyzing the curve and substituting it into a specific storage temperature and time, we can predict the size of the physical quantity at the storage temperature and time, thereby realizing the degradation analysis of the performance of the solid rocket engine after storage.

[0004] However, the prior art only considers linear fitting of small samples and multiple batches of storage data through the Alineux equation. For some data, the prior art cannot get a good fit, and some data changes are not linear. Blindly considering the linear model will invisibly produce excessive deviations. Moreover, the degradation model obtained based on the Alineux equation can only predict the engine performance data under a certain storage time at a fixed temperature. The prediction accuracy of the degradation model is far from meeting the requirements. That is, the performance degradation model constructed using the prior art has the problems of overly single model and low accuracy.

[0005] Therefore, it is necessary to provide a solid rocket motor storage performance degradation analysis method based on sample enhancement to solve the above problems. Summary of the invention

[0006] The purpose of the present invention is to solve the problems of the simplification and low precision of the linearized model of the Alineuz equation by applying the deep learning method to the degradation analysis of the storage performance of solid rocket motors. In view of the multi-batch and small sample characteristics of storage data, a method for storage performance degradation analysis of solid rocket motors based on sample enhancement is proposed.

[0007] The present invention discloses a method for analyzing storage performance degradation of a solid rocket motor based on sample enhancement, which adopts the following technical scheme, including: Acquire original sample data, the original sample data including: storage performance data and storage environment data in the engine storage small sample data; Construct a generative adversarial network. The generative adversarial network includes a generator and a discriminator. The generator is used to generate random sample data, and the discriminator is used to determine whether the random sample data is real data. The generative adversarial network is trained to obtain a trained target generative adversarial network. The data judged as "real" by the discriminator of the target generative adversarial network is fused with the original sample data to obtain an enhanced new sample data set. Constructing a degradation model, taking the storage performance data in the new sample data set as the output of the degradation model, taking the storage environment data in the new sample data set as the input of the degradation model, and training the degradation model to obtain a trained target degradation model; The current storage environment data of the solid rocket motor is input into the target degradation model to obtain the current storage performance data.

[0008] Preferably, the storage performance data are mechanical performance parameters of the engine charge after storage, and the mechanical performance parameters include stress, nominal fracture strain and actual fracture strain; the storage environment data include storage time and storage temperature.

[0009] Preferably, the steps of obtaining the original sample data are: Obtaining initial sample data at multiple sets of temperatures; Perform linear interpolation on the storage time in the initial sample data under the same set of temperatures to obtain the storage temperature and storage time after expansion; The stress, nominal fracture strain, actual fracture strain, storage temperature and storage time after expansion are taken as the original sample data.

[0010] Preferably, strided convolution is used to replace the pooling layer of the generator; fractional strided convolution is used to replace the pooling layer of the discriminator; different learning rates are selected for the generator and the discriminator.

[0011] Preferably, the steps of obtaining the trained target pair to generate the anti-network are: Fix the discriminator of the generative adversarial network and train the generator of the generative adversarial network; Fixed the target generator and trained the discriminator of the generative adversarial network; The training can be stopped when the loss of the generator G approaches 0 infinitely in 1-2 consecutive training rounds, and the trained target generative adversarial network is obtained.

[0012] Preferably, the degradation model includes: Input layer, used for inputting input data of the degradation model; Convolutional layer, which includes an initial convolutional layer and multiple small convolutional layers, used to extract feature maps of input data; Pooling layer, used to reduce the dimension of feature map.

[0013] Fully connected layer, used to predict storage performance data; And an output layer for outputting predicted storage performance data.

[0014] Preferably, the model.fit function is used to train the degradation model, the loss function of the degradation model adopts the mean square error loss function, the optimizer adopts Adam, and the training lasts for 2000 epochs.

[0015] A solid rocket motor storage performance degradation analysis system based on sample enhancement, comprising: The original sample data acquisition module is used to acquire the original sample data, the original sample data including: storage performance data and storage environment data in the engine storage small sample data; The adversarial generative network module is used to construct an adversarial generative network. The adversarial generative network includes a generator and a discriminator. The generator is used to generate random sample data, and the discriminator is used to determine whether the random sample data is real data. The adversarial generative network is trained to obtain a trained target adversarial generative network. The data judged as "real" by the discriminator of the target adversarial generative network is fused with the original sample data to obtain an enhanced new sample data set. The degradation model module is used to construct a degradation model, take the storage performance data in the new sample data set as the output of the degradation model, take the storage environment data in the new sample data set as the input of the degradation model, train the degradation model to obtain a trained target degradation model; The storage performance prediction module is used to input the current storage environment data of the solid rocket engine into the target degradation model to obtain the current storage performance data.

[0016] An electronic device, a processor, a memory and a computer program stored in the memory, wherein the steps of the method of the present invention are implemented when the computer program is executed by the processor.

[0017] A program product comprises a computer program, which is used to execute the steps of the method of the present invention when the computer program is executed.

[0018] A storage medium stores a computer program, which is used to execute the steps of the method of the present invention when the computer program is executed.

[0019] The beneficial effects of the present invention are: The present invention achieves sample expansion and enhancement by fusing data judged as "true" in the generated data of the trained target generative adversarial network with the original sample data to obtain an enhanced new sample data set. Then, a degradation model is constructed based on the new sample data set, and the degradation model is trained to obtain a trained target degradation model. After testing the target degradation model, it is found that the target degradation model is effective, and the accuracy of storage performance data predicted by the target degradation model is higher than that of storage performance data obtained by the existing storage performance degradation analysis method of the Alineux equation, that is, the present invention improves the prediction accuracy of storage performance data. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 A flow chart of a method for analyzing storage performance degradation of a solid rocket motor based on sample enhancement according to the present invention; Figure 2 It is a general step block diagram of a method for analyzing storage performance degradation of a solid rocket motor based on sample enhancement of the present invention; Figure 3 A schematic diagram of feature extraction in step S1 in an embodiment of the present invention; Figure 4 A schematic diagram of a generative adversarial network (GAN) in an embodiment of the present invention; Figure 5 A schematic diagram of obtaining a new sample data set in an embodiment of the present invention; Figure 6 Schematic diagram of feature classification in the new sample data set in step 41 in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of the F-CNN network in an embodiment of the present invention; Figure 8 4 is a loss graph (a) of the model ε_d / % training process and an error rate histogram (b) during the test process in step 431 of the embodiment of the present invention; Fig. 9 4 is a loss graph (a) of the model ε_m / % training process and an error rate histogram (b) during the test process in step 431 of the embodiment of the present invention; Fig.10 The figure is a loss graph (a) of the model σ / Mpa training process and an error rate histogram (b) during the testing process in step 431 of the embodiment of the present invention; Fig.11 A comparison chart of the predicted values ​​of each model and the true values ​​in step 431 in an embodiment of the present invention; Fig.12 4 is a loss graph (a) of the model ε_d / % training process and an error rate histogram (b) during the test process in step 432 of the embodiment of the present invention; Fig.13 4 is a loss graph (a) of the model ε_m / % training process and an error rate histogram (b) during the test process in step 432 of the embodiment of the present invention; Fig.14 The figure is a loss graph (a) of the model σ / Mpa training process and an error rate histogram (b) during the testing process in step 432 of the embodiment of the present invention; Fig.15 This is a comparison chart of the predicted values ​​of each model and the true values ​​in step 432 in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] An embodiment of a method for analyzing storage performance degradation of a solid rocket motor based on sample enhancement of the present invention is as follows: Figure 1 As shown, including: S1. Obtain original sample data; Specifically, original sample data is obtained, and the original sample data includes: storage performance data and storage environment data in the engine storage small sample data.

[0024] In this embodiment, feature extraction is first performed, such as Figure 3 As shown, the extracted features are the storage performance data and storage environment data in the engine storage small sample data, wherein the storage performance data are the mechanical performance parameters of the engine charge after storage, the mechanical performance parameters include stress σ / Mpa, nominal fracture strain ε_m / % and actual fracture strain ε_d / %; the storage environment data include storage time and storage temperature.

[0025] In order to facilitate the subsequent sample enhancement of performance data, the steps for obtaining the original sample data in this embodiment are: obtaining initial sample data under multiple groups of temperatures; performing linear interpolation on the storage time in the initial sample data under the same group of temperatures to obtain the storage temperature and storage time after expansion; and taking the stress, nominal fracture strain, actual fracture strain, and storage temperature and storage time after expansion as the original sample data.

[0026] S2. Construct a generative adversarial network and train it to obtain a target pair generative adversarial network; Specifically, the principle diagram of generating adversarial networks is as follows Figure 4 As shown, the entire generative adversarial network includes a generator and a discriminator. The generator is used to generate random sample data, and the discriminator is used to determine whether the random sample data is real data. The training of the generative adversarial network obtains a trained target pair generative adversarial network.

[0027] It should be noted that the generator generates new data through machine learning. The basis of the generator's generated data is often a set of noise or random numbers, the purpose of which is to "fool" the discriminator as much as possible. The generated sample is recorded as G(z). The generator has no label and is an unsupervised network; the discriminator is used to determine whether the data is the original sample data or the data generated by the generator, the purpose of which is to find out the "fake data" created by the generator as much as possible. Its input parameter is x, x represents the data, and the output D(x) represents the probability that x is real data. If it is 1, it means that it is 100% real data, and if the output is 0, it means that it cannot be real data. Therefore, the discriminator has a label and belongs to a supervised network.

[0028] Step 21: Build a generative adversarial network: The core of the entire GAN (Generative Adversarial Network) lies in the selection and construction of the generator and the discriminator. For small sample data, if the number of convolution layers used to construct the generator and the discriminator is too small, serious training distortion will occur. If it is too large, serious overfitting will occur. Therefore, it is necessary to determine through continuous debugging that the most suitable number of convolution layers for the generator G is 128 layers, and the most suitable number of convolution layers for the discriminator D is 64 layers; at the same time, for the discriminator, strided convolution is used to replace the pooling layer; for the generator, fractional-strided convolutions are used to replace the pooling layer. For the core parameters such as learning rate and loss function, first select the dynamic learning rate, that is, select different learning rates for the generator and the discriminator, and select the cross entropy function for the loss function. Add the loss function for generating data discrimination to the two loss functions of the original generator and discriminator, and also select the cross entropy function.

[0029] Step 22: Train the Generative Adversarial Network: Step 221, fix the discriminator D and train the generator G: Using a discriminator D with good performance, the generator G continuously generates "fake data", and then the discriminator D judges the "fake data" generated by the generator G. At the beginning, the data generated by the generator G is very weak, so it is easy to be judged by the discriminator D. But as the training continues, the quality of the data generated by the generator G improves, and eventually it deceives the discriminator D. At this time, the discriminator D is basically in a "blind guessing" state, and the probability of judging whether it is fake data is 50%.

[0030] Step 222, fix the generator G and train the discriminator D: After passing the first stage, it is meaningless to continue training the generator G. At this time, the generator G is fixed, and then the discriminator D is trained. Through continuous training, the discriminator D improves its identification ability, and finally the discriminator D can accurately identify false data.

[0031] Step 223, obtaining a target generative adversarial network; Repeat step 221 and step 222 through continuous cycles until the loss of generator G approaches 0 infinitely in 1-2 consecutive training rounds, then stop training and finally obtain the target generative adversarial network. The generator G of the target generative adversarial network can be used to generate data.

[0032] S3, obtain the enhanced new sample data set; Specifically, the data judged as "true" by the discriminator of the target generative adversarial network is fused with the original sample data to obtain an enhanced new sample data set.

[0033] In this embodiment, the process of obtaining a new sample data set is as follows: Figure 5 As shown, that is, Figure 4 On this basis, the data judged as "true" by the discriminator of the target anti-network is fused with the original data of the sample to obtain the enhanced sample data set.

[0034] S4, constructing a degradation model and training to obtain a target degradation model; Specifically, a degradation model is constructed, the storage performance data in the new sample data set is used as the output of the degradation model, the storage environment data in the new sample data set is used as the input of the degradation model, and the degradation model is trained to obtain a trained target degradation model.

[0035] Step 41: Construct a degradation model: like Figure 6As shown, the feature data in the new sample data set are divided into two categories, input features and output features. In this embodiment, the input features are storage environment factors with a size of m, and the output features are storage performance data with a size of n. The construction of the degradation model can be understood as how to use these n input features to predict the data of the m output features under specific circumstances (determined storage temperature and storage time).

[0036] Specifically, the construction structure is as follows: the degradation network model adopts Figure 7 The structure of the F-CNN network (two-dimensional convolutional network) shown in FIG. 4 is shown in FIG. 4 ; the F-CNN network includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer, wherein the input layer is used to input the input data of the degradation model; that is, the shape of the degradation model input is (n, 1, 1), indicating that the input data has n features (storage temperature, storage time, and storage humidity), and the remaining dimensions are used for convolution operations; the convolutional layer includes an initial convolutional layer (64 convolution kernels, size 4x4) and 23 small convolutional layers of size 1x1, which are used to extract the feature map of the input data; the pooling layer is used to reduce the dimension of the feature map, using two-dimensional maximum pooling (size 2×2), using Flatten to flatten the pooled multidimensional data, and then using a fully connected layer (64 neurons, activation function is ReLu) and the final output layer (m neurons) to predict m physical quantities to predict storage performance data.

[0037] Step 42: Degradation model training: Specifically, the storage performance data in the new sample data set is used as the output of the degradation model, and the storage environment data in the new sample data set is used as the input of the degradation model to train the degradation model. The loss function of the degradation model adopts the mean square error (MSE), and the optimizer is Adam. The training lasts for 2000 epochs, and the loss values ​​of the training set and the validation set are saved for subsequent drawing.

[0038] Step 43: Test the trained degradation model: n The input characteristics of the storage environment data correspond to the output m The output characteristics of the stored performance data of the engine, m , n The size can be adjusted according to the actual situation. Here, two input characteristics (storage time and storage temperature) and three output characteristics (stress, nominal fracture strain and actual fracture strain) are taken as examples to carry out model test analysis.

[0039] Step 431: Test the accuracy of the degradation model on real data: Remove the original sample data in the new sample data set, and only use the data judged as "true" by the discriminator of the target pair generation network in S3 to train the degradation model, and record the model trained by F-CNN as degradation model A. When testing, use the input features (storage time and storage temperature) in the real sample as the test, output the corresponding storage performance prediction data (stress, nominal fracture strain and actual fracture strain), and compare the error between the predicted data and the output features of the real sample. The test results of this embodiment are as follows: Figures 8 to 11 As shown, the test numbers are 1-42; where σ / Mpa, ε_m / % and ε_d / % represent the mechanical properties of the engine grain after storage: stress, nominal fracture strain and actual fracture strain, where Figure 8 (a) is the loss graph of the model ε_d / % training process, Figure 8 (b) is the error rate histogram during the test; Fig. 9 (a) is the loss diagram of the model ε_m / % training process, Fig. 9 (b) Error rate histogram during testing; Fig.10 (a) is the loss diagram of the model σ / Mpa training process, Fig.10 (b) Error rate histogram during testing; Fig.11 This is a comparison chart of the predicted value and the true value of each model. The accuracy is determined by the error rate. The smaller the error rate, the more accurate the model prediction. Figure 8-11 In the (a) part of the figure, it is found that the degradation models of various physical quantities can converge quickly during the training process, and the convergence loss value is very small, which means that the training effect of the degradation models of various physical quantities is very good; observing the test error rate histogram of each model, that is, Figure 8-11 In the (b) part of the figure, it is found that the model has better generalization ability on the new test set and higher overall prediction accuracy. Fig.11 The comparison results of can also illustrate this conclusion. In summary, the degradation models of various physical quantities not only show good fitting ability and training effect in the training process, but also show excellent generalization ability in the test process on real data. So far, the test of the GAN network has been completed, indicating that the enhanced samples are usable.

[0040] Step 432: Test the accuracy of the degradation model on the test set: During the training process, 30% of the data in the new sample data set is randomly selected, half of the data in the 30% data is used as the validation set, and the other half of the data in the 30% data is used as the test set. The remaining 70% of the data in the new sample data set is used to train the degradation model. The generated model is recorded as degradation model B. During the test, the input features (storage time and storage temperature) in the test set are used to output the corresponding performance prediction data (stress, nominal fracture strain and actual fracture strain). The error between the performance prediction data predicted by the test set and the actual output data of the validation set samples is compared, and the corresponding error rate is calculated. The accuracy of the model prediction is judged by the error rate. The smaller the error rate, the more accurate the model prediction. The test results are as follows: Figure 12 to Figure 15 As shown, the test numbers are 1-101; where σ / Mpa, ε_m / % and ε_d / % represent the mechanical properties of the engine grain after storage: stress, nominal fracture strain and actual fracture strain; where Fig.12 (a) is the loss graph of the model ε_d / % training process, Fig.12 (b) is the error rate histogram during the test; Fig.13 (a) is the loss diagram of the model ε_m / % training process, Fig.13 (b) Error rate histogram during testing; Fig.14 (a) is the loss diagram of the model σ / Mpa training process, Fig.14 (b) Error rate histogram during testing; Fig.15 This is a comparison chart of the predicted value and the true value of each model. The accuracy is determined by the error rate. The smaller the error rate, the more accurate the model prediction. Figure 12-15 In the (a) part of the figure, it is found that the degradation models of various physical quantities can converge quickly during the training process, and the convergence loss value is very small, which means that the training effect of the degradation models of various physical quantities is very good; observing the test error rate histogram of each model, that is, Figure 12-15 In the (b) part of the figure, it is found that the model has better generalization ability on the new test set and higher overall prediction accuracy. Fig.15 The comparison results can also illustrate this conclusion. In summary, the degradation models of various physical quantities not only show good fitting ability and training effect in the training process, but also show excellent generalization ability in the test process on the set test set. So far, the test of F-CNN has been completed, indicating that the construction of the degradation model is effective.

[0041] In summary, the test results show that the prediction error rate of the entire degradation model is much lower than that of the original Alineux equation. Step 431 completes the test of the GAN network, indicating that the enhanced samples are available; Step 432 completes the test of the F-CNN, indicating that the construction of the degradation model is effective, and in terms of model complexity and accuracy, it is much higher than the traditional method. It proves that the proposed method for analyzing the storage performance degradation of solid rocket motors based on sample enhancement is feasible.

[0042] S5. predicting current storage performance data; The current storage environment data of the solid rocket motor is input into the target degradation model to obtain the current storage performance data.

[0043] A solid rocket motor storage performance degradation analysis system based on sample enhancement, comprising: an original sample data acquisition module, an adversarial generation network module, The original sample data acquisition module is used to acquire the original sample data, the original sample data including: storage performance data and storage environment data in the engine storage small sample data; The adversarial generative network module is used to construct an adversarial generative network. The adversarial generative network includes a generator and a discriminator. The generator is used to generate random sample data, and the discriminator is used to determine whether the random sample data is real data. The adversarial generative network is trained to obtain a trained target adversarial generative network. The data judged as "real" by the discriminator of the target adversarial generative network is fused with the original sample data to obtain an enhanced new sample data set. A two-dimensional convolutional network module is used to construct a degradation model, take the storage performance data in the new sample data set as the output of the degradation model, take the storage environment data in the new sample data set as the input of the degradation model, train the degradation model to obtain a trained target degradation model; The storage performance prediction module is used to input the current storage environment data of the solid rocket engine into the target degradation model to obtain the current storage performance data.

[0044] An electronic device, a processor, a memory and a computer program stored in the memory, wherein the steps of the method of the present invention are implemented when the computer program is executed by the processor.

[0045] A program product comprises a computer program, which is used to execute the steps of the method of the present invention when the computer program is executed.

[0046] A storage medium stores a computer program, which is used to execute the steps of the method of the present invention when the computer program is executed.

[0047] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for analyzing storage performance degradation of solid rocket motors based on sample enhancement, characterized in that: include: Acquire original sample data, the original sample data including: storage performance data and storage environment data in the engine storage small sample data; Construct a generative adversarial network. The generative adversarial network includes a generator and a discriminator. The generator is used to generate random sample data, and the discriminator is used to determine whether the random sample data is real data. The generative adversarial network is trained to obtain a trained target generative adversarial network. The data judged as "real" by the discriminator of the target generative adversarial network is fused with the original sample data to obtain an enhanced new sample data set. Constructing a degradation model, taking the storage performance data in the new sample data set as the output of the degradation model, taking the storage environment data in the new sample data set as the input of the degradation model, and training the degradation model to obtain a trained target degradation model; The current storage environment data of the solid rocket motor is input into the target degradation model to obtain the current storage performance data.

2. The method for analyzing storage performance degradation of solid rocket motors based on sample enhancement according to claim 1, characterized in that: The storage performance data are the mechanical performance parameters of the engine grain after storage, which include stress, nominal fracture strain and actual fracture strain; the storage environment data include storage time and storage temperature.

3. The method for analyzing storage performance degradation of solid rocket motors based on sample enhancement according to claim 1, characterized in that: The steps to obtain the original sample data are: Obtaining initial sample data at multiple sets of temperatures; Perform linear interpolation on the storage time in the initial sample data under the same set of temperatures to obtain the storage temperature and storage time after expansion; The stress, nominal fracture strain, actual fracture strain, storage temperature and storage time after expansion are taken as the original sample data.

4. The method for analyzing storage performance degradation of solid rocket motors based on sample enhancement according to claim 1, characterized in that: Use strided convolution to replace the pooling layer of the generator; use fractional strided convolution to replace the pooling layer of the discriminator; use different learning rates for the generator and discriminator.

5. The method for analyzing storage performance degradation of solid rocket motors based on sample enhancement according to claim 1, characterized in that: The steps to obtain the trained target pair to generate the anti-network are: Fix the discriminator of the generative adversarial network and train the generator of the generative adversarial network; Fixed the target generator and trained the discriminator to generate anti-network; The training is completed until the loss of the generator G approaches 0 infinitely, and the trained target generative adversarial network is obtained.

6. The method for analyzing storage performance degradation of solid rocket motors based on sample enhancement according to claim 1, characterized in that: Degradation models include: Input layer, used for inputting input data of the degradation model; Convolutional layer, which includes an initial convolutional layer and multiple small convolutional layers, used to extract feature maps of input data; Pooling layer, used to reduce the dimension of feature map. Fully connected layer, used to predict storage performance data; And an output layer for outputting predicted storage performance data.

7. The method for analyzing storage performance degradation of solid rocket motors based on sample enhancement according to claim 1, characterized in that: The degradation model is trained using the model.fit function. The loss function of the degradation model uses the mean square error, the optimizer uses Adam, and the training lasts for 2000 epochs.

8. A solid rocket motor storage performance degradation analysis system based on sample enhancement, characterized in that: The original sample data acquisition module is used to acquire the original sample data, the original sample data including: storage performance data and storage environment data in the engine storage small sample data; Generative adversarial network module, used to construct a generative adversarial network, which includes a generator and a discriminator. The generator is used to generate random sample data, and the discriminator is used to determine whether the random sample data is real data. The generative adversarial network is trained to obtain a trained target generative adversarial network; the data judged as "true" by the discriminator of the target generative adversarial network is fused with the original sample data to obtain an enhanced new sample data set; The degradation model module is used to construct a degradation model, take the storage performance data in the new sample data set as the output of the degradation model, take the storage environment data in the new sample data set as the input of the degradation model, train the degradation model to obtain a trained target degradation model; The storage performance prediction module is used to input the current storage environment data of the solid rocket engine into the target degradation model to obtain the current storage performance data.

9. An electronic device, characterized in that: A processor, a memory, and a computer program stored in the memory, wherein when the computer program is executed by the processor, the steps of the method described in any one of claims 1 to 7 are implemented.

10. A storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed, it is used to execute the steps of the method described in any one of claims 1 to 7.

Images