Solid propellant mechanical property prediction system based on artificial intelligence

Abstract

The invention provides a solid propellant mechanical property prediction system based on artificial intelligence, and the system comprises a data input module which is used for achieving the multi-format import and visual preview of an experiment data file; the feature engineering module is used for executing data standardization, abnormal value elimination and key parameter extraction; the intelligent modeling module is used for configuring an adjustable neural network topological structure, including layer number setting, node configuration, dual-stage activation function selection and training parameter optimization; the prediction output module is used for dynamically displaying a mechanical property prediction value and a training error convergence curve; and the verification and evaluation module is used for providing test set performance index calculation and visual analysis functions. The prediction system supports localized deployment of edge equipment, can optimize a process window in the propellant production process in real time, and greatly improves the prediction precision compared with a traditional prediction method.

Description

Technical Field

[0001] The present invention belongs to the technical field of solid propellants and relates to solid propellant performance prediction, in particular to a solid propellant mechanical property prediction system based on artificial intelligence. Background Art

[0002] The prediction of the mechanical properties of solid propellants is based on experiments and numerical simulations using traditional formulation design methods. These methods are costly, require long simulation cycles, and consume a great deal of manpower, material resources, and time. Traditional propellant performance prediction relies on empirical formulas and finite element simulations, which present three major technical bottlenecks:

[0003] First, the nonlinear relationship between component-process-performance is difficult to model.

[0004] Second, the utilization rate of experimental data is insufficient (<30%).

[0005] Third, the process adjustment response cycle is long (usually ≥72 hours). Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide an artificial intelligence-based solid propellant mechanical properties prediction system to solve the technical problem that the prediction accuracy of the model in the existing technology needs to be further improved.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] An artificial intelligence-based solid propellant mechanical properties prediction system includes the following functional modules:

[0009] First, the data input module is used to realize multi-format import and visual preview of experimental data files.

[0010] Second, the feature engineering module is used to perform data standardization, outlier removal, and key parameter extraction.

[0011] Third, the intelligent modeling module is used to configure an adjustable neural network topology, including layer number setting, node configuration, two-stage activation function selection, and training parameter optimization.

[0012] Fourth, the prediction output module is used to dynamically display the predicted values ​​of mechanical properties and the training error convergence curve.

[0013] Fifth, the verification and evaluation module is used to provide test set performance indicator calculation and visual analysis functions.

[0014] Compared with the prior art, the present invention has the following technical effects:

[0015] (I) The prediction system of the present invention implements multivariate component-process-performance correlation analysis by deploying a lightweight neural network model. This system innovatively constructs a three-tier prediction system: first, a database of raw material ratios and process parameters is established, and experimental data on mechanical indicators such as tensile strength and elongation at break are collected.

[0016] (II) The prediction system of this invention uses feature engineering to extract key influencing factors, including multiple core parameters such as oxidant particle size distribution, binder content, and mixing temperature. Ultimately, a graph convolutional neural network model with adaptive learning capabilities was developed to achieve end-to-end prediction of key indicators such as elastic modulus and Poisson's ratio under different formulations.

[0017] (III) The prediction system of the present invention supports local deployment of edge devices and can optimize the process window during propellant production in real time, significantly improving prediction accuracy compared to traditional prediction methods. The prediction system of the present invention is suitable for aerospace propellant formulation development and process optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the architecture of the solid propellant mechanical properties prediction system based on artificial intelligence.

[0019] Figure 2 Schematic diagram of the data input module.

[0020] Figure 3 Schematic diagram of the feature engineering module.

[0021] Figure 4 Schematic diagram of the intelligent modeling module.

[0022] Figure 5 Schematic diagram of the prediction output module.

[0023] Figure 6 Schematic diagram of the verification and evaluation module.

[0024] The specific contents of the present invention are further described in detail below with reference to the accompanying drawings and embodiments. DETAILED DESCRIPTION

[0025] It should be noted that, unless otherwise specified, the algorithms, functions, units, mechanisms and methods in the present invention all adopt algorithms, functions, units, mechanisms and methods known in the prior art.

[0026] In accordance with the above technical solution, specific embodiments of the present invention are given below. It should be noted that the present invention is not limited to the following specific embodiments, and all equivalent changes made on the basis of the technical solution of this application fall within the protection scope of the present invention.

[0027] Example:

[0028] This embodiment provides a solid propellant mechanical properties prediction system based on artificial intelligence. The system in this embodiment is designed according to the three-layer model MVC (Model-View-Controller) pattern. The software is divided into application layer, function layer and data layer with interface-oriented programming thinking. Each layer exchanges data through standard interfaces, so that the software achieves a high cohesion and low coupling state, which is convenient for subsequent maintenance and upgrades. Figure 1 shown.

[0029] First, the application layer: This layer describes how users interact with the software, including the creation of training networks and the model training process. This involves interactively creating a training network, entering network parameters, selecting a network algorithm, executing the training model, and comparing training results with experimental data to optimize and adjust network model parameters to ultimately achieve optimal parameters.

[0030] Second, the functional module layer: The functional layer is the specific implementation of the application layer and its interaction with the data layer, including network model algorithm integration, graphical interface implementation and software functional interface support.

[0031] Third, the data support layer: The data layer describes the total set of experimental data, input data, and result data. This module structures the user's existing experimental data and provides a data interface to the functional layer. It persists the data and enables data reading, writing, and storage.

[0032] Specifically, the system includes the following functional modules:

[0033] First, the data input module is used to implement multi-format import and visual preview of experimental data files; in this embodiment, the data input module is as follows Figure 2 As shown in the figure, the data set is imported into the system. The data set contains data related to propellant component content, process parameters and performance. The data input supports EXCEL / CSV format parsing and has data dimension verification function.

[0034] Second, the feature engineering module is used to perform data standardization, outlier removal, and key parameter extraction.

[0035] Specifically in this embodiment, the feature engineering module is as follows Figure 3 As shown, it includes a process parameter normalization unit, a component feature encoding unit, and a data distribution visualization unit.

[0036] In this embodiment, the process parameter normalization unit adopts a parameter classification processing method, adopts a minimum-maximum normalization method for continuous parameters, and implements label encoding for discrete parameters.

[0037] In this embodiment, the component characteristic coding unit adopts a hierarchical coding structure, with the first-level classification being oxidant / binder / metal fuel, the second-level classification being oxidant particle size classification (10 / 50 / 200 μm), and the third-level classification being binder functional group type.

[0038] In this embodiment, the data distribution visualization unit can display the results of the normalized input data and output data.

[0039] Third, the intelligent modeling module is used to configure an adjustable neural network topology, including layer number setting, node configuration, two-stage activation function selection, and training parameter optimization.

[0040] In this embodiment, the intelligent modeling module is as follows: Figure 4 As shown in the figure, it is possible to use an adjustable network depth architecture, implement a 1-5 layer structure, and adopt a mixed activation function combination (ReLU / Sigmoid / Tanh). At the same time, an adaptive training termination mechanism is added to ensure model accuracy.

[0041] In step three, the network model is constructed using hierarchical generation technology to achieve dynamic layer construction (LayerFactory class), and supports the definition of network depth (1-5 layers) and each layer dimension through JSON (Java Script Object Notation, a lightweight data exchange format). It has a built-in dimension automatic inference algorithm, establishes an activation function resource pool (ReLU / Tanh / Sigmoid), and implements the allocation and call of activation functions through an adaptive allocation algorithm.

[0042] In step three, the intelligent training control function dynamically evaluates the termination conditions. The system monitors the F1-score fluctuations of the validation set in real time and adopts a dual-threshold trigger mechanism. The absolute threshold is immediately terminated when acc ≥ 98%, and the relative threshold is triggered when the improvement is < 0.2% for 5 consecutive epochs.

[0043] Fourth, the prediction output module is used to dynamically display the predicted values ​​of mechanical properties and the training error convergence curve.

[0044] In this embodiment, the prediction output module is as follows: Figure 5 As shown, the simultaneous prediction of multiple mechanical parameters is achieved, including tensile strength and elastic modulus, and the error change curve is monitored in real time during the training process.

[0045] Fifth, the verification and evaluation module is used to provide test set performance indicator calculation and visual analysis functions.

[0046] In this embodiment, the verification and evaluation module is as follows: Figure 6 As shown, the integrated cross-validation function, the determination coefficient R 2, mean absolute error (MAE), and response delay (RMSE) multi-index calculation, as well as prediction-measurement comparison and analysis functions.

[0047] In this embodiment, the cross-validation adopts the improved Stratified K-Fold method (K=5), and the data distribution consistency is maintained according to the clustering results of the process parameters. The proportion of each fold validation set is strictly controlled at 20±0.5%.

[0048] Comparative Example 1:

[0049] This comparative example adopts a traditional method, using a BP (Back Propagation) neural network with a fixed 3-layer structure and a Sigmoid activation function.

[0050] Verification and comparison:

[0051] Using the same input parameters, predictions were made using the embodiment of the present invention and comparative example 1, and the comparison results of tensile strength predictions are as follows:

[0052] Table 1 Comparison results

[0053] method <![CDATA[Coefficient of determination R 2 > Mean Absolute Error (MAE) Response delay RMSE Embodiments of the present invention 0.925 0.21MPa 80ms Comparative Example 1 0.71 0.49MPa 1.2s

[0054] As shown in Table 1, when comparative example 1 is used for prediction, the decision system is degraded, the model interpretation ability is reduced, the mean absolute error is increased, and the response time is delayed from 80 ms in the embodiment to 1.2 s, indicating that the method of the present invention has improved both accuracy and response performance.

Claims

1. An artificial intelligence-based solid propellant mechanical properties prediction system, characterized in that: The system includes the following functional modules: First, the data input module is used to realize the multi-format import and visual preview of experimental data files; Second, the feature engineering module is used to perform data standardization, outlier removal, and key parameter extraction; Third, the intelligent modeling module is used to configure the adjustable neural network topology, including layer number setting, node configuration, two-stage activation function selection and training parameter optimization; Fourth, the prediction output module is used to dynamically display the predicted values ​​of mechanical properties and the training error convergence curve; Fifth, the verification and evaluation module is used to provide test set performance indicator calculation and visual analysis functions.

2. The artificial intelligence-based solid propellant mechanical properties prediction system according to claim 1, characterized in that: The data input module supports EXCEL / CSV format analysis and has data dimension verification function.

3. The artificial intelligence-based solid propellant mechanical properties prediction system according to claim 1, characterized in that: The feature engineering module includes a process parameter normalization unit, a component feature encoding unit, and a data distribution visualization unit.

4. The artificial intelligence-based solid propellant mechanical properties prediction system according to claim 1, characterized in that: The intelligent modeling module can use an adjustable network depth architecture, can implement a 1-5 layer structure, and adopt a hybrid activation function combination, while adding an adaptive training termination mechanism.

5. The artificial intelligence-based solid propellant mechanical properties prediction system according to claim 1, characterized in that: The prediction output module realizes the synchronous prediction of multiple mechanical parameters, including tensile strength and elastic modulus, and the error change curve is monitored in real time during the training process.

6. The artificial intelligence-based solid propellant mechanical property prediction system according to claim 1, characterized in that: The validation evaluation module integrates cross-validation function, and the determination coefficient R 2 , mean absolute error (MAE), response delay (RMSE) multi-index calculation, and prediction-measurement comparison analysis functions.

Images