Session: 03-05-01 Design, Material Processing, and Applications of Polymer Composites

Paper Number: 98829

98829 - Design of Experiment and Machine Learning-Based Framework to Investigate Mechanical Strength of Soft and Lightweight Hybrid Composites 

3D printing has recently emerged as an effective method to manipulate microstructure to enhance mechanical properties of lightweight composite materials in applications of aerospace, automotive, and other fields. A lightweight hybrid composite has heterogeneous microstructure benefiting flexible structural design, multifunctionality, reducing the weight and energy consumption of the materials. A critical aspect of developing advanced composites is to design the optimal microstructure that impacts the mechanical properties and maximizes their performance. To understand the structure-property relationships, mechanical properties parametric data is often correlated with the structural parameters of heterogeneous composites. The mechanical properties of advanced composites highly rely on the filler spatial distribution, filler geometry, density, fraction of filler, etc. With the advancement of 3D printing, the structural characteristics of composites can be influenced by the manufacturing process parameters. Optimization of the process parameters such as the orientation of the fillers can enhance the mechanical properties of composites like stiffness and failure properties caused by load transfer and stress distribution. Therefore, mechanical properties of materials can be significantly enhanced by focusing on dominant parameters. Due to the numerous degrees of freedom to select optimal process parameters which correlates with enhanced mechanical properties, the use of the ML model can simplify the process by predicting the tensile properties, such as ultimate strength.

In this work, we analyze the mechanical properties of 3D printed hybrid lightweight composites through development of a micromechanical model along with a prediction-based machine learning (ML) model. This study explores that the composites can be strengthened in terms of qualitative performance of their high design space, anisotropy properties, load carrying capacity, low density. The composites are 3D printed using soft and hyper elastic polydimethylsiloxane (PDMS) as matrix and short glass microfibers and porous glass microparticles as fillers. To understand the relation between mechanical properties and process parameters, we conduct a statistical analysis on the 3D printing process parameters based on the experimental tensile testing data by using classic Taguchi method and ANOVA analysis. The volume fraction of fibers and the orientation of the fibers are significant for strength and stiffness, respectively. A theoretical mechanics model for the hybrid composites is developed based upon the combination of the spatial orientation of fiber, volume fraction of fibers and particles, and critical interfacial mechanical properties between fillers and matrix. The micromechanical model harnesses a combination of the Halpin-Tsai, the rule of mixture, and the Tsai-Hill criterion models using Weibull probability density function in order to incorporate the fraction and spatial orientation of the fillers and to define the failure criteria for the hybrid composite model. Furthermore, we perform a ML model in order to predict the tensile strength based on a Deep Neural Network (DNN) algorithm. The micromechanical model with combination of the predictive ML model is found to capture the mechanical strength of hybrid composites through the variation of the 3D printing parameters.

Presenting Author: Sanjida Ferdousi University of North Texas

Presenting Author Biography: I am a doctoral student in the Mechanical Engineering Department at the University of North Texas, supervised by Dr. Yijie Steven Jiang. My research focuses on 3D printing and characterization of composite materials with the aid of artificial intelligence. In particular, I study how to optimize the composite micro-structure of elastic materials based on the material performance and characterizing the interfacial relations by harnessing advance machine learning techniques.

Authors:

Sanjida Ferdousi University of North Texas
Yijie Jiang University of North Texas