X-ray Scattering Analysis of Wood Aided by Machine Learning

Small and wide-angle x-ray and neutron scattering on wood and other (ligno)cellulosic materials can be applied to improve our understanding of their structure [1]. The development of methods for characterization is important for the understanding of the anatomical composition and nanoscale structure of biobased materials— this is fundamental to their utilization for sustainable applications.

Wood scanning with wide-angle and small-angle X-ray scattering (WAXS, SAXS) was done at the ID02 beamline of the ESRF synchrotron. The beam size was of about 30x30 µm² to observe wet and dry Norway spruce samples. The measurements were done at an energy of 12.23 keV, an exposure time of 0.2 s each, and a sample to detector distance of 1.4 m for the SAXS and 14 cm for the WAXS. Detector images were processed and fitted to obtain structural parameters such as the distance between microfibrils [2].

The results of the fitting from dry and wet samples were analyzed with principal component analysis (PCA) algorithms to depict the statistical information related to water content and wood structure. A clustering algorithm was able to distinct types of wood tissue (earlywood/latewood) and samples (wet/dry) based on the PCA output. Experiments of a different nature, such as near infrared (NIR) spectroscopic imaging, can also benefit from PCA and validate the methodology’s findings [3]. Results highlight the sample and tissue type classifications on different experiments featuring different characteristics of the same wood sample but holding comparable classes based on the PCs. The outlook moves towards new automated procedures that will lead to more sophisticated supervised learning algorithms.