Modeling and optimization of frontal polymerization-based manufacturing of polymers and composites

Frontal polymerization (FP) is a self-propagating reaction sustained by the heat released during the conversion of the monomer to polymer. A local thermal stimulus initiates FP such that the heat released during polymerization diffuses to the monomer further downstream causing it to polymerize. This process continues until all the monomer is converted into polymer. Recently, FP has been studied as an energy-efficient, out-of-autoclave/oven alternative for manufacturing fiber-reinforced-polymer composites (FRPCs). In this work, we study the FP-based manufacturing process numerically using a combination of finite element analysis (FEA) and analytical methods. In particular, we study the effect of the cure kinetics, initial conditions, boundary conditions, and fiber volume fraction on the speed of the reaction front. FEA is also used to simulate the manufacturing of carbon FRPCs using a through-thickness thermal trigger. The thermal-trigger profile is then expressed in terms of heat-flux design variables and optimized using gradient descent where the adjoint sensitivities of the design variables are analytically computed. The optimal trigger is then validated experimentally. For optimal manufacturing, material selection plays a pivotal role. This issue is explored by using a 1D convolutional neural network (CNN) to link the heat flow vs. temperature curves obtained using differential scanning calorimetry (DSC) tests of various resins/formulations directly to their corresponding reaction front velocity. Due to a lack of adequate experimental data, we generate the data numerically by perturbing the cure kinetics parameters. Two approaches for training the CNN are considered. The first method involves classifying the dataset into certain bins based on front velocity values whereas the other directly predicts the front velocity. Both approaches work well since the trained CNN achieves > 90% accuracy for the training and test datasets. The latter is chosen as the method of choice since it directly predicts the front velocity from a given DSC curve. The trained CNN is then tested on available experimental data giving mixed results. The influence of the cure kinetics, initial conditions, and boundary conditions on the unstable propagation of FP is also studied numerically. A range of governing parameters is considered to perform 1D FEA simulations and the resulting instability patterns are recorded. The obtained results are classified into 4 classes: quenched front, stable front, unstable mode with 1 temperature peak, and unstable mode with 2 temperature peaks. The inverse problem of predicting the optimal cure kinetics and process parameters to achieve a prescribed instability pattern is solved using a machine learning (ML) surrogate trained using the outputs of the FEA-based parametric study. The patterns recorded are encoded to contain key information like the maximum temperature and wavelength associated with unstable FP. Two neural networks (NNs)/multi-layer perceptrons (MLPs), one for each instability mode, are trained to predict the encoded pattern specifications based on the input of cure kinetics and process parameters. The fully trained surrogates are then embedded in a traditional gradient descent optimization loop to predict optimal values for cure kinetics and process parameters for a target pattern. Traditional manufacturing methods for FRPCs involve bulk polymerization (BP) to avoid nonuniform curing which leads to warping of the finished component. The FP-based manufacturing process is significantly different as the monomer is cured locally in an incremental fashion. As a result, FP-based manufacturing of a morphing composite structure is studied in the context of unbalanced carbon/pDCPD cantilever beams. The effect of the location of the applied thermal trigger and the direction of front travel on the final shape of the composite cantilever beam is explored experimentally. A mechanical model is then developed in the small strain setting and coupled to the existing thermo-chemical model to numerically capture the deformation of the composite cantilever. The numerical model correctly captures the mode of deformation.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Aerospace Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Vyas, Sagar Ketan

Contributors dc:contributor

• Geubelle, Philippe H
• Sottos, Nancy R
• Baur, Jeffery W
• Chew, Huck Beng

• Frontal Polymerization
• Fiber-reinforced Composites
• Numerical Modeling
• Process Optimization
• Machine Learning

Statement dc:rights

• Copyright 2024 Sagar Ketan Vyas

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/127376

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F127376