PSI - Issue 78

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

31

Table 4. Performance metrics of the conditional model in Eq. 8 for the usable and non-reusable classes assuming the thresholds for t 1 associated with the maximum of the F1 score and t 2 corresponding to the minimum value associated with the F1 score stabilization. The values t 1 and t 2 are in mm. B. Accuracy stands for Balanced Accuracy. Reusable Non-Reusable Model label Precision Recall F1-score Precision Recall F1-score B. accuracy Accuracy AUC-ROC Conditional model 0.97 0.79 0.87 0.34 0.82 0.48 0.80 0.79 0.80

The area under the ROC curve (AUC-ROC) and the balanced accuracy confirm that, overall, the rule makes correct decisions most of the time. Although boosting precision for the non-reusable label would reduce waste, it is not a safety issue; the current errors are conservative because they only reject material that could have been kept.

6. Conclusions

Re-using timber removed from old buildings is only realistic if engineers can grade the material quickly and safely. The authors propose a simple “keep / reject” rule for beams marked by small round holes left from earlier service.

• A stochastic FE model, assuming clear wood and elastic–brittle behaviour, was run on thousands of random hole layouts. Model predictions match the bending tests in [22]. • The key descriptor that separates safe from unsafe cases is the total hole diameter in two depth zones: Zone 1 = outer quarter-depth strips, Zone 2 = the middle half. • The pass / fail rule is Classify =   k red ≥ 80% if � i ∈ Zone 1 d i > t 1 = 12  or � i ∈ Zone 2 d i > t 2 = 28  k red < 80% otherwise (10) where k red comes from Eq. (7) and holes under 3 mm are ignored. • Thresholds t 1 and t 2 were tuned to catch almost every unsafe beam (few false negatives) while keeping accuracy high. Skip holes smaller than 3 mm; measure over the pure-bending zone only; make sure no single hole exceeds 30 mm and the combined diameter in the two edge strips stays under 100 mm. Within those limits the rule o ff ers a fast, conservative check that can slot straight into on-site visual grading.

Acknowledgements

The authors acknowledge the Norwegian Research Council, Innovation Norway and Siva, the actors behind the program Green Platform and the public financing of SirkTRE (project nr. 328731).

References

[1] L. Acun˜a, L.A. Basterra, M.M. Casado, G. Lo´pez, G. Ramo´n-Cueto, E. Relea, C. Mart´ınez, and A. Gonza´lez. Application of resistograph to obtain the density and to di ff erentiate wood species [aplicacio´n del resisto´grafo a la obtencio´n de la densidad y la diferenciacio´n de especies de madera]. Materiales de Construccion , 61(303):451–464, 2011. [2] R. Arbelaez, L. Schimleck, J. Dahlen, and S. Wood. Evaluation of lumber from deconstructed portland residential buildings. Wood and Fiber Science , 51(4):432–440, 2019. [3] F. Arriaga, C. Osuna-Sequera, M. Esteban, G. ´In˜iguez Gonza´lez, and I. Bobadilla. In situ assessment of the timber structure of an 18th century building in madrid, spain. Construction and Building Materials , 304, 2021.