Machine learning-based prediction of surface checks and bending properties in weathered thermally modified timber
Joran van Blokland, Vahid Nasir, Julie Cool, Stavros Avramidis, Στέργιος Αδαμόπουλος
Abstract
Machine learning (ML)-based models, decision tree and ANFIS, were used to predict the degree of surface checking and bending properties of 30-month weathered thermally modified timber. The results showed that the investigated initial board properties did not allow accurate predictions of surface checks. ML regression and clustering analysis confirmed important variables for accurate predictions of bending properties were dynamic stiffness, acoustic velocity, density and lowest local bending modulus. ML models performed better than conventional regression models used for timber grading, and a prediction accuracy of 80–90% for bending stiffness and 50–70% for bending strength could be achieved.