| Popis: |
Engineered wood products of today's construction industry require predictable mechanical properties of the used structural timber. At sawmills, automated strength grading is used to assess the stiffness and strength of sawn and dried timber boards. This process can be based on various technologies, e.g. surface scanning, dynamic excitation, flat-wise bending, which are used to derive so-called indicating properties, i.e. simplified numerical values. Heuristically derived statistical models can then be used to predict the stiffness and strength based on indicating properties. However, statistical strength grading can only exploit a small fraction of the potential strength of a single board, since it assesses the properties of a board in relation to its population. A growing number of sawmills in Sweden use computed tomography (CT) scanners to assess the incoming logs to optimise their positioning prior to sawing. CT scans provide high-quality data of the cross-sectional density distribution along the length of a log, which could also be used to derive continuum mechanical models of the yet unsawn boards and with that assess their mechanical properties. If the stiffness and strength of a virtual board can be predicted before it is sawn, then it could be pre-classified into a strength class or its specific use as a specific construction part could be predetermined already at the log stage, which would lead to a more efficient material usage. Additionally, the predictive power of the existing statistical strength grading processes could be improved for the final boards. The goals of this study were to i) derive 3D quasi-continuum and finite element (FE) models of CT scanned timber boards using different material laws for local stiffness based on measured density and ii) compare their capabilities for predicting stiffness and strength of the boards. The experimental material consisted of dried softwood boards (12% moisture content) of nominal cross-sectional dimensions 50x100mm with different lengths, scanned with a medical high-resolution CT scanner. The boards underwent an eigenfrequency measurement by dynamic excitation and were tested until failure in a four-point bending test, where both the local and global displacement were recorded. A previously developed algorithm was used to derive 3D quasi-continuum reconstructions from the CT scans and subsequently finite element (FE) models. The algorithm reconstructed the board geometry, pith, knots and local fibre directions (material coordinate system) on a volume grid of material points spaced 0.68mm apart. The stiffness tensor in each material point was made locally dependent on the measured density by different mathematical laws, e.g. constant, linear or power laws. Furthermore, material laws which scaled the stiffness tensor based on the ratio between the simulated and measured eigenfrequency were tested for comparison. The bending stiffness profile was calculated for each board along its length and different indicating properties for predicting stiffness and strength were derived and compared with respect to the experimental results. With the FE model, strain distributions in the cross-sections were studied and local stress states around the experimentally observed points of initial failure were investigated to determine whether similar dominant failure stress states existed among boards. The results showed high coefficients of determination between predicted stiffness and strength for material laws based on power laws and low values for linear laws. Nevertheless, the four-point bending tests only provided point-wise data (mid points) that could be used to validate the numerical model. It is therefore recommended to use field-based evaluations in the future, e.g. the surface strain obtained with DIC under four-point testing. ReadIStrength |
