Predicting Moisture Content in Timber Drying
Abstract
The forest industry is a cornerstone of Sweden's economy, yet it accounts for roughly 14% of the country's total electricity consumption; a substantial share driven largely by kiln drying of timber. Current drying schedules in sawmills are conservative by design, resulting in excessive energy use and prolonged drying cycles. Reducing energy consumption while maintaining uniform target moisture content requires more accurate, data-driven predictions of how moisture evolves inside wood during drying.
This project brings together materials science expertise (Uppsala University, WISE) and machine learning expertise (Linköping University, WASP) to develop predictive models for moisture content and moisture gradients in timber. The project started with the WISE side designing experiments and collecting data. In November 2025, we obtained beamtime at the ForMAX beamline at MAX IV, Lund, and used industrial X-ray computed tomography (XCT) to scan timber samples at multiple stages of the drying process, creating a high-quality labelled dataset. Large Swedish sawmills already deploy XCT scanners for non-destructive strength grading; our work targets the largely untapped potential of this data for drying optimisation.
Building on this dataset, the ML side will now train and evaluate deep learning models, convolutional neural networks (CNNs) and Vision Transformer (ViT)-based architectures, to predict material density and moisture content directly from XCT scans. The goal is a model accurate enough to inform adaptive, energy-efficient drying schedules, reducing unnecessary drying time and energy waste.
