Flexural behavior of wood in the transverse direction investigated using novel computer vision and machine learning approach

Abstract

Abstract A deep-learning-based semantic segmentation approach (U-Net) was used to partition the anatomical features in the cross-section of hinoki (Chamaecyparis obtusa) wood during a micro three-point bending test. Using the Crocker–Grier linking algorithm, thousands of cells were successfully extracted, and several parameters (area, eccentricity, fitted ellipse aspect ratio, bounding box aspect ratio) were used to evaluate the intensity of the cells’ deformation. Thus, the 2D map of the deformation intensity distribution was constructed. By analyzing flat-sawn, quarter-sawn, and rift-sawn specimens, it was confirmed that the annual ring orientation affects the flexural behavior of wood in the transverse direction. The quarter-sawn specimens exhibited the largest modulus of elasticity (MOE) and modulus of rupture (MOR). The ray tissue aligned against the load may have contributed to the restriction of cell deformation. The rift-sawn specimens exhibited the smallest MOE and MOR, possibly owing to the loading of the specimen in the in-plane off-axial direction, which induced the shear deformation of the cell wall. For all three specimen types, the fracture had high occurrence probability in the tension part of the specimen, which exhibited large cell deformation. Therefore, the proposed method can be adapted to the prediction of wood specimen fractures. With different test wood species, this approach can be of great help in elucidating the relationship between the anatomical features and the mechanical behavior of wood to improve the effective utilization of wood resources.