PSI - Issue 78

Angelo Aloisio et al. / Procedia Structural Integrity 78 (2026) 25–32

26

that present visual grading rules were written for fresh wood and do not consider holes and cuts from earlier service, so many sound beams are rejected [12, 3, 25]. Most work on reclaimed timber has looked at sti ff ness and density under static loads [21, 19, 20, 18, 17, 1]. Only a few papers have checked how holes change bending strength, and they show that a hole near the tension face is very damaging, especially in the cyclic loading that earthquakes create [14, 11]. Recent non destructive surveys of old wood still skip this kind of defect [16, 7]. Here the authirs o ff er a simple rule for seismic reuse. They ran many finite-element models of beams with random holes and built a data set that links the sum of hole diameters in two depth zones to the drop in bending capacity. The field check is easy: if the sum in either edge zone is above 12 mm, or the sum in the middle zone is above 28 mm, the beam should not be reused in seismic areas. The Norwegian draft standard for reclaimed softwood [24] swaps the familiar C-grades for a new trio—R24, R18 and R14. Each one mirrors a conventional grade (C30 → R24, C24 → R18, C18 → R14) and assumes that dropping a step cuts the characteristic bending strength by about 20%. An 80% capacity ratio therefore marks the pass / fail line. To check whether a reclaimed beam with small round holes still meets that line the authors built a binary classifier. The route, sketched in Fig. 1, is: 1. Random-hole FE model: Generate many hole patterns and run planar bending analyses in Abaqus. 2. Lab validation: Compare the model to bending tests reported in [22]. 3. Monte Carlo sweep: Convert every pattern into a predicted strength-loss ratio. 4. Classifier training: Store hole geometry and capacity drop, then test simple rules against standard machine-learning algorithms to find a reliable yes / no decision tool for grading reclaimed timber. 2. Problem formulation

Fig. 1. Workflow adopted in this study. Stochastic finite-element analyses create a large synthetic set of hole patterns; experimental bending tests verify the model; Monte Carlo runs generate a capacity-loss database that feeds rule-based and machine-learning classifiers to label beams as Reusable or Non-reusable.

3. Bending strength estimation from the FE model

Running a full fracture–mechanics analysis for every possible hole pattern would be far too heavy for large Monte Carlo studies. Because lumber behaves almost linearly up to a rather brittle failure in bending, the authors use a lighter scheme: a purely elastic finite-element (FE) model provides the stress field, and that stress field is then converted into a loss-of-capacity factor. The elastic route is quick and still gives a solid data set for the classifiers. The check assumes that failure starts when the local stress at a hole reaches the local strength. For each hole h i the design inequalities read   σ 11 , m ≤ f m , σ i j , h i ≤ f i j , ∀{ i , j } ∈ { 1 , 2 } , ∀ h i , (1)