Multivariate Image Regression for Quality Control of Natural Fiber Composites

Abstract

This study reports on using near-infrared (NIR) hyperspectral imaging for nondestructive spectral-spatial characterization of natural fiber composites. To illustrate the approach, maleic anhydride grafted polyethylene (MAPE)/hemp fiber composites were produced with different filler contents between 0 and 60%. Two different chemometrics methods based on (1) the traditional multivariate PLS calibration and (2) multivariate image analysis and regression (MIA/MIR) were tested to predict tensile properties using NIR images. The results show good agreement between the measured properties and their predictions by both of these methods. The MIR ability to map chemical composition was compared to that of multivariate curve resolution (MCR) and the results were found similar. The proposed MIR approach was found very promising for quality control of polymer composites because of its combined ability to quantify filler content and dispersion within the material, to distinguish compositional variations from physical defects, and to predict the end-user properties of the product all with a single MIR model.