Non-destructive testing of mechanical properties of solid wood panel based on partial least squares structural equation modeling transfer method

Abstract

Calibration transfer between near infrared (NIR) spectrometers is a subtle issue in the chemometrics and process industry. Similar instruments may generate strongly different spectral responses, and regression models developed on a first NIR system can rarely be used with spectra collected by a second apparatus. In this work, two novel methods based on Structural Equation Modeling (SEM), called Enhanced Feature Extraction Approaches for factor analysis (EFEA-FA) and Enhanced Feature Extraction Approaches for spectral space transformation (EFEA-SST), were proposed to perform calibration transfer between NIR spectrometers. They were applied to a NIR nondestructive testing model for solid wood panels mechanical properties. Four different standardization algorithms were evaluated for transferring solid wood panels quality databases between a portable NIRS (InGaAs)-array spectrometer (NIRquest512) and a HSI Camera (SPECIM FX17). The results showed that EFEA-SST yielded the best model evaluation metrics (R2 and Root Mean Square Error of Prediction (RMSEP)) values for tensile strength (RMSEP=11.309, R2=0.865) parameters, while EFEA-FA gave the best fit for flexural strength (RMSEP=10.653, R2=0. 912). These results suggest the potential of two novel quality parameters prediction methods based on spectral databases transferred between diverse NIRS spectrometers.