X-ray Scattering Analysis of Spruce Aided by Machine Learning

Norway spruce wood exhibits structural variations in its different tissue types. The main part of the tree consists of earlywood, which forms at the beginning of the growing season and has larger, thinner-walled cells, and latewood, which develops later and has smaller, thicker-walled cells. These differences affect the wood’s properties and applications. Additionally, the moisture content of wood can vary, further influencing its structure and uses. Our project delves into these complexities using advanced structural characterization techniques like X-ray scattering and machine learning to gain new insights into these variations.
X-ray scattering techniques, such as Small-Angle X-ray Scattering (SAXS) and Wide-Angle X-ray Scattering (WAXS), are powerful tools that probe the internal structure of materials. When X-rays interact with the crystalline cellulose microfibrils in the wood, they scatter in ways that reveal detailed information about the wood’s nanostructure. By analysing the scattering patterns, we can infer structural parameters that describe the arrangement, size and crystal structure of the microfibrils.
In our study, we collected extensive scanning X-ray scattering data from three Norway spruce wood samples: two fully wet and one room-dry. The large dataset, consisting of 101×101 scans of each of the three samples in WAXS and SAXS, allowed us to apply machine learning algorithms, specifically Principal Component Analysis (PCA) and clustering, to identify and categorize earlywood and latewood within these samples. This analysis provided mean structural parameters values for each type of tissue: wet earlywood, dry earlywood, wet latewood, and dry latewood.
To complement the scattering results, we conducted a parallel experiment using Near-Infrared (NIR) hyperspectral imaging on the dry sample, which provided detailed spatially-resolved spectral data. Applying the same PCA and clustering methods on the NIR spectra, we identified earlywood and latewood regions consistent with the X-ray scattering results. This cross-validation provides support for the reliability of our findings. NIR complements the physical structural results with spectral information, that provides information about the chemical differences in the wood tissue types.
Our research provides new understanding by integrating sophisticated imaging techniques with machine learning to enhance our understanding of wood structure. This study highlights the potential of these technologies in wood science and reflects a highly multidisciplinary project that brings out new results with innovative research.