Near-infrared Spectroscopy Detection Method for Compressive Strength of Fraxinus mandschurica

Abstract

This study used near-infrared (NIR) spectroscopy as a non-destructive test to predict the compressive strength (i.e., modulus of rupture (MOR) and the modulus of elasticity (MOE)) of Fraxinus mandshurica parallel to the wood grain. Tests were conducted with 120 small and clear wood samples to obtain the diffuse NIR reflectance spectra of the radial and tangent surfaces of the wood samples. Standard normal variable transformation (SNV) combined with Savitzky-Golay (SG) convolution smoothing algorithm was used to filter the raw NIR spectra. Uninformative variables elimination (UVE) and a genetic algorithm (GA) were utilized to identify specific wavelengths in the spectra that directly correlated to compression strength. Finally, a partial least squares (PLS) regression model was developed with the identified wavelengths to determine the MOR and MOE of the samples. The results showed the correlation coefficients of the prediction models for MOR and MOE were 0.88 and 0.89, respectively. The root mean square errors of prediction for MOR and MOE models were 7.37 and 0.49, respectively. Based on these results, it is feasible to accurately estimate the compressive strength of Fraxinus mandshurica (parallel to the grain) using NIR spectroscopy.