Application of Near-Infrared Spectroscopy for Evaluation of Drying Stress on Lumber Surface: A Comparison of Artificial Neural Networks and Partial Least Squares Regression

This study aimed to examine the feasibility of evaluating the stress level at the surface of lumber during drying using near-infrared (NIR) spectroscopy combined with artificial neural networks (ANNs). Sugi (Cryptomeria japonica D. Don) lumber with an initial moisture content ranging from 41.1 to 85.8% was dried using a commercial drying schedule. An ANN model for predicting surface-released strain (SRS) was developed based on NIR spectra collected from the lumber during drying. The predictive ability of the ANN model was compared with a partial least squares (PLS) regression model.

The ANN model showed good correlation between laboratory-measured SRS and predicted SRS with an R ² of 0.79, a root mean square error of prediction (RMSEP) of 0.0009, and a ratio of performance to deviation (RPD) of 1.81. The PLS regression model gave a lower R ² of 0.69, a higher RMSEP of 0.0010, and a lower RPD of 1.38 than the ANN model, suggesting that the predictive performance of the ANN model was superior to the PLS regression model. The SRS evolution during drying as predicted by the models showed a similar trend to the laboratory-measured one. The predicted elapsed times to reach maximum tensile SRS and stress reversal roughly coincided with the laboratory-measured times. These results suggest that NIR spectroscopy combined with multivariate analysis has the potential to predict the drying stress level on the lumber surface and the critical periods during drying, such as the points of maximum tensile stress and stress reversal.