Machine learning-based prediction of internal checks in weathered thermally modified timber

摘要

This study investigated possibilities to predict the presence of internal checks in thermally modified Norway spruce timber after 2.5 years of weathering based on the initial properties of the boards. Machine-learning classification enabled sorting the input parameters based on their relative importance for accurate predictions. The parameters of thermally modified timber with the highest relative importance were annual ring width followed by initial moisture content, density and dynamic stiffness. Whereas after kiln drying these were, density, annual ring width, initial moisture content and acoustic velocity. The results showed that predictions are possible, and an accuracy of 67% was achieved by using annual ring width combined with density and initial moisture content, or acoustic velocity that can be determined after either kiln drying or thermal treatment. (C) 2020 Published by Elsevier Ltd.