This website requires JavaScript.

Parameterization-based Neural Network: Predicting Non-linear Stress-Strain Response of Composites

Haotian FengPavana Prabhakar
Dec 2022
摘要
Composite materials like syntactic foams have complex internalmicrostructures that manifest high-stress concentrations due to materialdiscontinuities occurring from hollow regions and thin walls of hollowparticles or microballoons embedded in a continuous medium. Predicting themechanical response as non-linear stress-strain curves of such heterogeneousmaterials from their microstructure is a challenging problem. This is truesince various parameters, including the distribution and geometric propertiesof microballoons, dictate their response to mechanical loading. To that end,this paper presents a novel Neural Network (NN) framework calledParameterization-based Neural Network (PBNN), where we relate the compositemicrostructure to the non-linear response through this trained NN model. PBNNrepresents the stress-strain curve as a parameterized function to reduce theprediction size and predicts the function parameters for different syntacticfoam microstructures. We show that our approach can predict more accuratenon-linear stress-strain responses and corresponding parameterized functionsusing smaller datasets than existing approaches. This is enabled by extractinghigh-level features from the geometry data and tuning the predicted responsethrough an auxiliary term prediction. Although built in the context of thecompressive response prediction of syntactic foam composites, our NN frameworkapplies to predict generic non-linear responses for heterogeneous materialswith internal microstructures. Hence, our novel PBNN is anticipated to inspiremore parameterization-related studies in different Machine Learning methods.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答