Researchers have shown that a vibration sensor paired with machine learning can measure timber moisture with near‑lab accuracy, offering a fast, non‑destructive alternative that could transform quality control in sawmills and panel lines. If factory trials confirm the results, sawmills could gain a low‑cost, contactless system that delivers near‑lab moisture readings in real time, cutting waste and improving product consistency.
The paper, Non‑destructive wood moisture prediction using vibration signals and XGBoost, published by researchers at China’s Nanjing Forestry University, recorded tiny vibrations from Populus tomentosa boards, extracted time‑ and frequency‑domain features from accelerometer signals, and trained an XGBoost regression model to identify which vibration features predict moisture content.
Moisture alters density, stiffness and internal damping, leaving distinct fingerprints in a board’s vibration. The team filtered raw data, pulled out spectral and temporal metrics, and compared several regression algorithms. XGBoost outperformed gradient boosting, random forest, and support vector regression, and generalised well on held-out samples.
Industry implications are immediate. Oven‑dry tests remain the gold standard but are slow and destructive, while many inline alternatives are expensive or sensitive to production noise. A vibration‑based system could provide contactless, board‑by‑board moisture estimates at production speed, enabling tighter kiln control, fewer rejects and more consistent engineered wood products, with lower capital and maintenance costs than some optical or radio‑frequency systems.
The study was carried out in controlled laboratory conditions using uniform poplar samples, so wider adoption will require further validation. Models must be tested and recalibrated across species, board sizes and surface conditions, and sensor hardware must be hardened for mill environments. Recalibration or transfer‑learning strategies will likely be necessary to retain accuracy across diverse timbers and geometries. If factory trials validate the lab findings, vibration sensing combined with explainable machine learning could become a practical, low‑cost tool for real‑time moisture control that boosts yield, cuts rejects and raises product consistency.
For more information: Jiawen Shi, Jiawei Zhang, Aoyun Li, Yongqi Liu, Bin Na,
Vibration-based non-destructive prediction of wood moisture content using machine learning models, Microchemical Journal, Volume 218, 2025, 115679, ISSN 0026-265X, https://doi.org/10.1016/j.microc.2025.115679.
