Search for a command to run...
Reclaiming load-bearing timber from demolition sites can significantly reduce the environmental footprint of new construction; however, the absence of grading rules tailored to reclaimed members often forces high-value material down the waste hierarchy. This work introduces a visual-based classification framework that quantifies the bending-strength loss caused by small throughholes, typical defects of previous service, and assigns each beam to either a reusable or non-reusable category. A stochastic, plane-stress finite-element model, experimentally validated on four-point bending tests, generated more than 1.5 Ø 104 random hole configurations in beams of three common cross-sections. From the elastic stress fields, a dimensionless reduction factor, kred, was derived to relate hole geometry to residual capacity. The simulated dataset fed two classifier families: (i) machine-learning algorithms (logistic regression, linear SVM, random forest and XGBoost) and (ii) an intentionally simple conditional rule based only on the sum of hole diameters in two visually identifiable depth zones. The conditional rule, with thresholds t 1 = 12 mm for the outer quarter-depth strips and t 2 = 28 mm for the central strip (holes with d < 3 mm ignored), achieved a balanced accuracy of 0.80 and an F2, score of 0.82, on par with the best machine-learning models while remaining transparent and easy to apply. All misclassifications were conservative: no under-strength beam was ever labelled reusable. The resulting procedure provides a fast and camera-friendly grading aid that can unlock the safe and high-value reuse of reclaimed timber, advancing circular economy goals in the construction sector.