The Role of Drying Schedule and Conditioning in Moisture Uniformity in Wood: A Machine Learning Approach

Area of Science:

• Wood Science and Technology
• Materials Science
• Forest Products Utilization

Background:

• Large moisture content (MC) variation in wood after drying significantly reduces product value, particularly for species with high initial MC spread.
• Effective kiln-drying processes are essential for producing high-quality timber and minimizing post-drying MC deviations.

Purpose of the Study:

• To optimize kiln-drying processes for timber.
• To develop a predictive machine learning model for estimating and classifying target timber moisture content.
• To evaluate the importance of wood attributes and drying parameters in predicting final wood MC.

Main Methods:

• Utilized a gradient-boosting machine learning model to predict and classify timber moisture content.
• Input variables included initial moisture, initial weight, basic density, and drying parameters (schedule, conditioning, post-storage).
• Compared the performance of regression-based prediction with classification into acceptable, over-dried, and under-dried groups.

Main Results:

• Initial weight showed the highest correlation and relative importance for final moisture prediction.
• Drying parameters (schedule, conditioning, post-storage) were significant for model robustness, as their removal decreased training accuracy.
• The regression model had limitations in accurately predicting final MC, but the classification model successfully categorized dried wood.

Conclusions:

• Gradient-boosting machine learning effectively classifies moisture content in kiln-dried western hemlock timber.
• The classification approach is suitable for pre- and post-sorting timber based on MC.
• Accurate MC classification enhances timber quality control and value preservation.