@article{ee4b0609b3a942d889514ca59836f16f,
title = "In-situ monitoring of reinforcement compaction response via MXene-coated glass fabric sensors",
abstract = "In this study, MXene-coated glass fabric sensors were used to obtain reinforcement compaction and stress relaxation data within a closed mold. A layer of MXene-coated sensor was embedded within a multilayer glass fiber preform to monitor compaction forces under both dry and wet conditions i.e., when the stack was fully impregnated with resin or a test fluid. The effect of the test fluid type on compressibility and sensor piezo-resistivity was also determined. The sensors showed excellent sensitivity in both dry and impregnated states and were able to successfully monitor different loading conditions such as peak stresses carried by the reinforcement, long-term stress relaxation and cyclic loads. Polynomial data fitting and machine learning models were used to calibrate the sensors to predict the compaction response. An electro-mechanical based model, on the lines of traditional viscoelastic stress relaxation model, was used to represent piezo-resistivity for long term relaxation. The proposed technique has great potential of in-situ monitoring of mold clamping forces and part thickness by measuring piezo-resistive changes taking place throughout a molding cycle using MXene based embedded smart sensors.",
keywords = "A. Graphene and other 2D-materials, A. Multifunctional composites, B. Electro-mechanical behavior, C. Stress relaxation, E. Resin transfer molding (RTM)",
author = "Ali, {M. A.} and Irfan, {M. S.} and T. Khan and F. Ubaid and K. Liao and Rehan Umer",
note = "Funding Information: The conventional or parametric regression models have a fixed functional form and limited number of parameters. On the other hand, machine learning models are nonparametric (i.e. not limited by a functional form) with theoretically a finite number of parameters [70,71]. Hence, machine learning models are more capable of recognizing hidden patterns and capturing the variability within the data. Support Vector Regression (SVR) [70,72] and Gaussian Process Regression (GPR) [71,73] are examples of widely used machine learning algorithms for regression tasks. Both these models are considered nonparametric techniques because they rely on kernel functions. SVR is one of the most robust prediction methods, being based on statistical learning frameworks that tries to fit the best line/hyperplane to the available data within a threshold value. GPR [71,73] models are nonparametric kernel-based probabilistic models that can infer a probability distribution over all possible values, rather than the exact values for every parameter in the functions. GPR has several benefits, working well on small datasets and having the ability to provide uncertainty measurements on the predictions. The SVR and GPR machine learning models were implemented and trained using MATLAB{\textregistered} Statistics and Machine Learning Toolbox [72,73]. The SVR model was implemented with a Gaussian kernel, and the sequential minimal optimization (SMO) algorithm was used for its training. Similarly, the GPR model was implemented with a squared-exponential kernel and the quasi-Newton algorithm was used for its training.This publication is based on work supported by the Khalifa University of Science and Technology internal grant CIRA-2020-007. Funding Information: This publication is based on work supported by the Khalifa University of Science and Technology internal grant CIRA-2020-007 . Publisher Copyright: {\textcopyright} 2022 Elsevier Ltd",
year = "2022",
month = aug,
day = "18",
doi = "10.1016/j.compscitech.2022.109623",
language = "British English",
volume = "227",
journal = "Composites Science and Technology",
issn = "0266-3538",
publisher = "Elsevier B.V.",
}