Artificial Neural Networks as a Decision-Support System for Predicting the Quality Attributes of Thermally Modified Wood

Thermal modification represents a key, environmentally safe way to enhance dimensional stability and biological resistance of wood. Yet, the complex, non-linear synergy of the hydrothermal process variables often renders traditional quality prediction models inadequate. This paper aimed to tackle this challenge by developing a robust decision-support model driven by an ANN, specifically a Multi-Layer Perceptron (MLP), to model the thermo-chemical properties of the wood cell wall matrix. A high-throughput meta-dataset comprising 200 validated experimental data points from the peer-reviewed literature (2000–2025) was used to investigate behavioural variability across softwood and hardwood taxa. The proposed MLP model combines 5 key input parameters (Species Type, Density, Temperature, Duration, and Medium) to predict 3 key performance metrics: ML, EMC, and MOR loss. Statistical validations highlight the model's strong generalizability, yielding R ² > 0.935 across all outputs, with Mass Loss achieving the highest precision (R ²  = 0.962). More refined sensitivity analysis (ANN weight partitioning) quantified the hierarchical influence of process variables and pinpointed temperature (45.2%) and treatment time (25.1%) as the predominant kinetic predictors of wood degradation. By integrating classical wood science theories with recent computational intelligence, this study has developed a powerful virtual prototyping engine that avoids destructive sampling and aligns with new Smart Manufacturing and global bioeconomy paradigms.