Fiber reinforced composites (such as carbon and glass) consist of long fibers oriented in a resin matrix. On the microscale (fiber/matrix) the position of fibers is random and subject to variability. Groups of fibers tend closer together to form fiber clusters and disperse to form matrix pockets. Microscale variability in fiber arrangement leads to variability in strength, often caused by stress concentrations forming between close fibers and low stiffness regions in matrix pockets. The problem is current modeling techniques fall short of capturing real life fiber arrangements, leading to less confidence in prediction of part performance under loading. This work aims at developing a robust method to characterize 2-D microscale fiber arrangements based on localized regions of fiber clusters and matrix pockets. With the aid of our NASA collaborators, this work can be integrated into an existing multiscale tool and used to create representative models to actual as-manufactured parts.

The characterization method was developed in MATLAB and works by reading in a microstructure and classifying it based on groups of close fibers (fiber clusters) or areas where there are little to no fibers (matrix-rich clusters). Cluster geometries are used as statistical descriptors to characterize each microstructure. In order to determine how these descriptors are made, a random RVE generator is used which has a range of inputs that control the final fiber positions. By randomly sampling these inputs and working with colleagues in the computer science department at UMass Lowell, machine learning models are being made to establish correlations between inputs and outputs. Finally, a reduced order model is being developed to simulate strength of randomly generated RVE’s. This method can be validated and verified with real scans of composite parts and research performed by collaborators on strength tests.

The thermal, mechanical, and moisture properties of clothing depend on textile architecture. Through exploring these properties across different architectures, we can compile a library of textile models based on their structure. Using this library, the design and manufacture of more durable and comfortable military wear can be accomplished

Our steps are as follows: First, we manufacture a variety of textile samples on the Stoll Knitting machine which vary in stitch length and knit structure. After doing so, the samples undergo multiple tests to characterize their different physical properties. The tests include microscope measurements to quantify the textile architecture, biaxial tests to characterize structural properties, and moisture testing to characterize moisture permeability of the samples. The data produced will be used quantify the isolated efect of textile structure on fabric properties in addition to tuning and validating computational models.

Wind turbine blades have received lots of interest lately for their ability to produce clean and sustainable energy. The share of total electricity generated in the United States increased from less than 1% in the 1990s to over 8% in 2018. This dramatic increase in interest and investment has led to the design, manufacturing, and maintenance of wind energy infrastructure to become a major topic for public and private research.

The two halves of fiber glass wind turbine blades are held together by adhesively bonded joints. These joints take several hours to cure, and this time can be directly translated into a manufacturer’s operating costs. In the pursuit of shorter cure times and lower costs, manufacturers want to know how fast a wind turbine blade can cure. Specifically, will everything be cured If the time limits are decreased? Also, how high can the heating rate and temperatures be increased without the exothermic reaction driving temperatures to levels detrimental for the material properties? In the Windstar project funded by the National Science Foundation (NSF), we seek to answer these questions through numerical analysis and physical validation of the cure cycle temperatures and degrees of cure.

For the numerical study, a 2-D cross section finite element ABAQUS model of the wind turbine blade was created. Additionally, a Fortran subroutine utilizing the Prout-Tompkins autocatalytic cure model was assessed at each cure time step in the cycle. This method allowed for rapid iterations of temperatures and degrees of cure for different cycle times, temperatures, and heating rates. Two different adhesives, manufactured by colleagues at Hexion and Olin, are examined in this work. These adhesives have low coefficients of thermal expansion, and their cure kinetics were characterized using differential scanning calorimetry (DSC). The model will be validated by analyzing the temperatures of a simplified geometry in the lab and an actual turbine blade during manufacturing.

One of the major issues with using Fiber Reinforced Composites (FRCs), such as carbon fiber composites, is accurately predicting when they will break. While useful failure theories do exist, the main way in which the behavior of FRCs is predicted is through Finite Element Modeling (FEM). A large contributor to this is the fact that the individual fibers in these materials are typically not oriented the way they were intended to be prior to manufacturing. This causes fibers to become entangled, leading to stress concentrations and failure prone fiber arrangements.

Most of my research has been on the quantification of fiber entanglement through numerous geometric and spatial descriptors. Using cross-sectional images at resolutions fine enough to make out individual fibers, we can detect the centers of these fibers and recreate them digitally. Using these digital fiber models, we can measure metrics such as the orientation of the fibers as well as their density with respect to one another (i.e., Local Volume Fraction). The aim of my work is to use these metrics to understand fiber entanglement and help inform future composite models.

Our collaborators in this effort include researchers at the Air Force Research Laboratory, NASA Glenn Research Center, and Brigham Young University. Through this project we have obtained both micro-CT and serial section images from the UES ROBO-MET System at NASA Glenn Research Center, the LEROY system at the Air Force Research Laboratory, and the European Synchrotron Radiation Facility.

Creating models of textile-based materials with realistic fiber volume fraction has proven to be difficult in the past. This difficulty is generally due to the unentangled fiber used to make up the tows within the textile model not providing as much resistance to compact as the entangled fibers found in experimentally investigated textiles. For this project an entanglement scheme was created to be able to control the among of entanglement within the textile with the goal of creating textile composite models with a more realistic volume fraction response. The entangled textiles are compressed using the Virtual Textile Morphology Suite provided by the Air Force Research Laboratory and University of Dayton Research Institute. The textiles created using the modeling scheme are compared to data collected from scanned textiles provided by a team at North Carolina State University and used for modeling the textiles performance by a team at Texas A&M.

Designing helmet systems to provide better protection against brain injuries caused by blunt impact is an important but difficult task. For this project in collaboration with the US Army Combat Capabilities Development Command Soldier Center, a computer based (LS DYNA finite element) model has been created that allows for the performance of the helmet using different foam pads and configurations to be tested at different impact velocities and locations on the helmet shell. The creation and processing of these models is accomplished using an automation scheme utilizing MATLAB to control SOLIDWORKS and HyperMesh to create the geometry for the model. The outputs of these models are shared with a data analytics team that using machine learning to predict where in the design space to concentrate further sampling in the search of a more optimum design. The optimum designs found through the sampling and machine learning process are then planned to be manufactured to be experimentally tested by Gentex Corporation.

Adhesively bonded materials are advantageous because their decreased weight and complexity can lead to improvements in efficiency and performance. Additional benefits can be realized for adhesively bonded composite joints because fastener-associated machining of the fibers and matrix is not necessary, so the induced defects such as delamination and cracking are avoided. Despite these substantial benefits, adhesive joints suffer from stress concentrations at the ends of the overlap since the joint loading tends to be eccentric. This causes induced moments that result in high shear and peel stresses at the ends of the joint. The problem is that the current methods of manufacturing tougher joints that reduce these stress concentrations are too difficult and or inconsistent to be used for any real composite joint structure. Without being able to manufacture bond lines that can better redistribute stresses based on modern simulation techniques, engineers are forced to use heavier, more complex, and less reliable assemblies on structures for aerospace.

In this project, we are utilizing a dual cure epoxy resin system that can change stiffness with increased radiation exposure. This property of the resin makes it possible to change the stiffness of the adhesive bond at different points in the overlap. This is helpful for designing structural adhesive joints because it allows for more compliant adhesive to be put on the overlap ends to decrease the peal and or shear stress concentrations. A convenient way to implement this is to use thicker radiation shielding in the locations that need to be more compliant. This will prevent the energy from getting through to the adhesive that is needed to induce additional crosslinking and increased stiffness.

The radiation sensitive epoxy resin formulations were developed in collaboration with Dr. Daniel Schmidt at the Luxemburg institute of technology.