A new approach based on a combination of capacitance and near-infrared spectroscopy for estimating the moisture content of timber

Abstract

The moisture content (MC) of wood influences its material properties. Determination of MC is essential in both the research and manufacturing fields. This study examined a nondestructive method for estimating MC rapidly and effectively. A capacitance sensor and a near-infrared (NIR) spectrometer were used to measure the MC of Japanese cedar and Japanese cypress timber. High-frequency capacitance (20 MHz) and NIR spectral absorption (908–1676 nm) data were collected for cross section and tangential section, as well as for the whole-sample average, in two MC ranges: from the green to the fiber saturation point (FSP) and from FSP to air-dried state. The results indicated that when standard error of prediction (SEP) is compared, the performance in [FSP to air-dried state] was better; when coefficient of determination in cross-validation ( $$R_{\text{val}}^{2}$$ ) and residual predictive deviation in cross-validation (RPDval) were compared, the performance in [Green to FSP] was better. Statistical analysis was performed using multiple linear regression and partial least squares. Combining capacitance and NIR absorbance at two wavelengths (Capacitance + NIR-MLR calibration) from the green to FSP was the best calibration yielding the most promising results: $$R_{\text{val}}^{2}$$ = 0.96, SEP = 5.20% and RPDval = 4.97 on the cross section of samples. The results were higher than those of other calibrations in R2 and SEP and RPD values. The NIR-PLS calibration performed better than others with quite good R2, lower SEP and higher RPD in the MC range from FSP to air-dried state. The first calibration using only capacitance of wood was good in the first range of MC, but it is not good in the second range (R2 under 0.5). Depending on the MC range, the performance of each calibration was different. In both MC ranges, the results on the cross section were higher than on the tangential section due to the anisotropic characteristics of wood material. From Capacitance + NIR-MLR calibration, the predicted models were developed using multiple linear regression and logarithmic regression. Results suggest the possibility of developing a new portable device combining a capacitance sensor and NIR spectroscopy to accurately predict the MC of wood.