Macro-performance prediction method for short-fiber reinforced composites based on deeplearning

Abstract

The invention discloses a macro-performance prediction method forshort-fiber reinforced composites based on deep learning. The steps include generating a representative volume unit by using a random adsorption method, calculating the macro-performanceof the material based on the homogenization method of the numerical simulation, and establishing a training sample set corresponding to the macro-performance of the fiber distribution image, and constructing a training convolutionalneural network and the like on the basis of the training sample set. The method combines the advantages of deep learning in the field of image recognition and uses convolutional neural networks to extract features. Through fitting the sample distribution, the accurate and fast response relationship between the fiberdistribution images and the macro-performance is realized.The method solves the problem that the traditional machine learning method is used as a proxy modelin which the extracted the features of thefiber distribution information are incomplete and the training precision is low. Furthermore, considering that the number of network layers is deepened and the training samples are less likely to beoverfitted, the sample is expanded by using the rotation and symmetric transformation of the fiber distribution image, so that the training precision is effectively improved, and the model maintainsgood robustness withina certain range outside the sample space.

Description

technical field [0001] The invention belongs to the field of structural design of composite materials, and relates to a mechanical analysis method of short fiber composite materials and a deep learning theory, in particular to a method for predicting the macroscopic performance of short fiber reinforced composite materials based on deep learning. Background technique technical background: [0002] Short fiber reinforced composites are widely used in aerospace and other defense industries due to their good mechanical and physical properties. Different engineering fields have different requirements on the mechanical properties of composite materials, and an accurate macroscopic performance prediction model is the basis for material design and structural design. It is well known that uncertain factors widely exist in the actual material structure. Affected by processing technology (heat treatment, pressure forming) and external environment changes (temperature, air pressure,...

AI Technical Summary

Problems solved by technology

[0005] The technical problem to be solved in the present invention is: to overcome the deficiency of the traditional proxy model, for the input with complex characteristics such as fiber distribution images, use convolutional neural network to extract features, and fit the sample distribution; for the possible problem of insufficient sample size, A method of expanding the sample set is proposed

Images