Core Concepts
Developing a physics-informed deep learning model for short fiber/polymer nanocomposites to predict mechanical behavior accurately.
Abstract
The article proposes a physics-informed deep learning model for short fiber/polymer nanocomposites to predict viscoelastic-viscoplastic behavior under various conditions. It combines LSTM and feed-forward neural networks to enforce thermodynamic principles and predict internal variables. The model is trained using synthetic data and experimental tests, showing accurate predictions for different material compositions and environmental conditions.
Directory:
Introduction
Focus on lightweight and durable materials in materials science.
Importance of composites with nanoparticles and nanotubes.
Physics-Informed Deep Learning Model
Combines LSTM and feed-forward neural networks.
Trained to enforce thermodynamic principles and predict internal variables.
Constitutive Modeling
Incorporates viscoelastic-viscoplastic behavior for nanocomposites.
Calibration using experimental data and synthetic data generation.
Numerical Results
Calibration of classical model for fiber-reinforced nanocomposites.
Comparison of model predictions with experimental data.
Stats
The deep-learning model is trained using synthetic data from a classical constitutive model.
The model predicts the mechanical behavior of epoxy-based nanocomposites accurately.
Quotes
"The PIDL model can accurately predict the mechanical behavior of epoxy-based nanocomposites for different volume fractions of fibers and nanoparticles under various hygrothermal conditions."