Machine learning-based prediction of internal moisture variation in kiln-dried timber

Abstract

ABSTRACT Monitoring the moisture content uniformity in kiln-dried wood and preventing large gradients is vital as nonuniformity renders dried timbers susceptible to warpage and degrade. This research uses a gradient-boosting machine learning model to model kiln drying by providing a predictive approach to estimate moisture levels and gradients. A population of 378 western hemlock square timbers was assigned into nine drying batches, each undergoing a different drying schedule. Inputs were four timber attributes, i.e, initial and final moisture, initial weight, and basic density, and three drying parameters, i.e. drying schedule, end-schedule conditioning, and dried timber post-storage. The results revealed that drying schedules and post-storage significantly impacted moisture gradients, while the effect of conditioning was insignificant. All the input parameters were crucial in developing the predictive machine-learning model, where wood attributes had relatively higher importance than drying parameters. Also, outputs highly depend on final moisture after drying. The best training and testing performances were achieved when predicting the shell moisture, followed by the core moisture and moisture gradient. Further research is required to enhance the predictive performance of the moisture gradient predictive model. Future studies could also develop classification models for the moisture gradient beneficial to sawmills.