Search

CN-122024959-A - Rapid forecasting method for thermal-force-electric response of quartz/epoxy resin composite material under VQ-VAE-based laser irradiation

CN122024959ACN 122024959 ACN122024959 ACN 122024959ACN-122024959-A

Abstract

The invention relates to the technical field of composite material damage assessment and performance prediction, and discloses a method and a system for quickly forecasting thermal-force-electric response of a quartz/epoxy resin composite material under laser irradiation based on a vector quantization variation self-encoder (VQ-VAE). The VQ-VAE adopts a vector quantization mechanism to extract material ablation phase change and damage characteristics, and adjusts the visual field resolution by improving the size of a convolution layer in the encoder, so that the extraction of small-scale damage characteristics is realized. The method realizes second-level prediction from parameter input to full-field output, remarkably improves calculation efficiency and prediction precision, and is suitable for rapid performance evaluation and design optimization of the composite material under laser irradiation.

Inventors

  • TAN ZHUHUA
  • Mu Haitian

Assignees

  • 河北工业大学

Dates

Publication Date
20260512
Application Date
20260127

Claims (10)

  1. 1. A rapid forecasting method of a quartz/epoxy resin composite material thermo-mechanical-electrical response under VQ-VAE-based laser irradiation is characterized by comprising the following steps: Step 1, constructing a multi-physical field data set covering a target working condition based on high-precision numerical simulation; Step 2, constructing and training a cascade forecast model consisting of a VQ-VAE reconstruction network and a DNN forecast network; and 3, rapidly forecasting and verifying the new laser working condition by using the trained cascade forecasting model.
  2. 2. The method of claim 1, wherein the step 1 of simulating the construction of the multi-physical field dataset comprises: sampling points in a laser loading working condition parameter space by using a Latin hypercube sampling method; Performing thermal-force-electric coupling finite element simulation on each sample point, and extracting data of temperature field, stress field and dielectric constant parameters; and carrying out normalization and size unification on the extracted data to construct a multi-physical data set, wherein the multi-physical data set comprises a training set and a testing set.
  3. 3. The method according to claim 1, wherein the VQ-VAE reconstruction network in step 2 comprises: an encoder adopting a multi-layer small-size convolution kernel stacking structure; A vector quantization layer comprising a learnable coding table for mapping the continuous features into discrete codes; And a decoder for reconstructing the physical field data from the discrete codes.
  4. 4. A method according to claim 3, wherein the convolution kernel size in the encoder in step 2 is 3 x 3 and the network depth is 4 layers.
  5. 5. The method of claim 4, wherein the DNN prediction network in step 2 is a fully connected neural network having an output dimension that matches a continuous eigenvector dimension of the encoder output in the VQ-VAE reconstruction network.
  6. 6. The method of claim 5, wherein the DNN prediction network is trained using a joint loss function comprising a feature space mean square error loss and a reconstructed mean square error loss.
  7. 7. The method according to claim 1, wherein the fast forecasting process in step 3 is: Inputting the laser loading working condition parameters into a DNN prediction network, and outputting continuous feature vectors; Converting the vector quantization layer into discrete codes; and inputting the frozen VQ-VAE to a decoder in the reconstruction network, and outputting a multi-physical field prediction result.
  8. 8. The method of claim 1, wherein the composite material is a quartz fiber reinforced epoxy composite material.
  9. 9. A rapid forecasting system for implementing the steps of the method according to any one of claims 1 to 8, characterized in that it comprises the following modules: the data generation module is used for executing high-precision coupling simulation and data extraction; The model training module is used for constructing and training a VQ-VAE reconstruction network and a DNN prediction network; And the forecast executing module is used for receiving the laser parameters and outputting a physical field forecast result.
  10. 10. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program, when executed, is capable of carrying out the steps of the method according to any one of claims 1 to 8.

Description

Rapid forecasting method for thermal-force-electric response of quartz/epoxy resin composite material under VQ-VAE-based laser irradiation Technical Field The invention relates to the technical field of composite material damage assessment and performance prediction, in particular to a rapid forecasting method of a thermal-force-electric response of a quartz/epoxy resin composite material under VQ-VAE-based laser irradiation. Background The quartz fiber reinforced epoxy resin composite material has excellent specific strength, heat resistance and electromagnetic wave transmission performance, and is widely applied to high-end fields such as aerospace, military defense, high-speed aircrafts, radar systems and the like. Particularly, in key functional components such as radomes, electromagnetic windows and the like, the materials are required to bear complex pneumatic load and thermal environment, and stable electromagnetic wave permeability is required to be maintained so as to ensure realization of core functions such as radar detection, communication transmission and the like. In practical use environments, particularly in military countermeasure or high energy weapon test scenarios, such composite components may be subject to irradiation attacks by high energy laser weapons. The laser beam has the characteristics of concentrated energy and rapid action, and can lead the surface and the inside of the material to generate severe temperature rise, heat stress concentration, matrix pyrolysis, fiber damage, even ablation and other phenomena in a very short time. These impairments not only weaken the structural integrity of the material, but also significantly alter its dielectric properties, leading to reduced or even complete failure of the wave-transparent properties, thereby severely affecting the functioning of the overall radar or communication system. Therefore, the accurate and rapid prediction of the thermal-force-electric multi-physical field coupling response of the composite material under the laser irradiation becomes an indispensable technical link in material design, damage assessment, protection strategy formulation and system reliability guarantee. Traditionally, this forecast has relied on numerical simulations, experimental tests or simplified models, but these methods have significant drawbacks in terms of efficiency, accuracy or applicability. At present, in the aspect of forecasting the performance of the composite material under laser irradiation, the following three types of technical paths and the corresponding technical problems mainly exist: 1. The method can obtain more accurate physical field distribution by establishing a fine finite element model, and solving heat conduction, thermoelastic mechanics and electromagnetic field equations through coupling. However, to capture the high gradient phenomena of ablation fronts, stress concentration areas, etc., extremely fine grids and extremely small time steps must be employed, resulting in a single complete simulation that takes up to hours or even days. The computing efficiency is seriously low, so that the computing efficiency cannot meet the actual engineering requirements of engineering design optimization, real-time damage evaluation or rapid parameter scanning and the like; 2. The traditional machine learning agent model method utilizes a neural network and other models to establish the mapping from laser parameters to key responses, and the method can improve the calculation speed, but has the following general problems: only limited scalar or low-dimensional vector results (such as highest temperature and maximum stress) can be predicted, complete high-resolution physical field distribution can not be reconstructed, and a great amount of local damage details are lost; The traditional self-encoder (VAE) adopts a continuous potential variable space, so that physical processes with mutability and discrete characteristics such as material phase change, ablation pit formation and the like are difficult to effectively characterize, and a prediction result is blurred in a key area; The standard Convolutional Neural Network (CNN) structure is not optimized for micron-to-millimeter-level characteristics of the laser damage of the composite material, and small-scale details such as the edges of ablation pits, microcracks and the like are difficult to effectively extract and retain, so that the accuracy of electromagnetic performance prediction is affected; 3. The test method relies on the physical sample and special test equipment, and the response is measured by laser irradiation experiment and by means of a temperature recorder, a vector network analyzer and other equipment. Limitations include: The experimental period is long, the cost is high, and the experimental period is obviously influenced by factors such as sample preparation consistency, test environment interference and the like; the parameterization research and design iterat