Analysis of MOR and MOE prediction model of Quercus mongolica wood by near infrared spectroscopy.

The detection of modulus of rupture (MOR) and modulus of elasticity (MOE) of Quercus mongolica as an important structural material is of scientific significance with engineering application value. In order to undertake nondestructive detection rapidly, a portable near infrared spectrometer with 900-1 700 nm is used. Firstly, the first derivative and S-G convolution smoothing method was implemented for signal pre-processing; and then the model dataset was built by the average value of two sections, i.e., radial plane and tangent plane; finally, the MOR and MOE were estimated by using the isometric mapping partial least squares (Isomap-PLS) algorithm. The experiments used 135 small clear wood samples whose size was 300 mm×20 mm×20 mm, of which 90 were calibration samples and 45 were prediction samples. The experimental results show that the first order derivative is able to eliminate the spectral background and high frequency noise, and the average value of S-G can be used to filter out the high frequency noise. Whats more, the average section model is better than the single section, which owns a higher correlation and lower standard error. Compared with partial least squares (PLS), interval partial least squares (iPLS), moving window partial least squares (MWPLS), changeable size moving window partial least squares (CSMWPLS), back interval partial least squares (BiPLS) and locally linear embedding partial least squares (LLE-PLS), Isomap-PLS is the best, whose prediction correlations of MOR and MOE are 0.89 and 0.88 respectively, standard errors of prediction (SEP) are 11.43 and 2.73 respectively, ratios of performance to standard deviation (RPD) are 2.55 and 2.58 respectively, and RPD is more than 2.5. Hence, this near infrared model can be used to complete the fast and effective nondestructive testing of Q. mongolica wood with no defect.