Intelligent prediction of fire retardancy and smoke suppression in waste wood composites with a hybrid deep learning model

Abstract

The green and high-value utilization of waste wood is crucial for sustainable development in the wood industry. In this study, waste wood served as the primary raw material, modified with bio-derived tannic acid (TA) and phytic acid (PA) as flame retardants, to fabricate high-performance biomass-based composite. The synergistic mechanisms between waste wood, modifiers, and adhesives were elucidated using techniques including FTIR, XPS, and SEM-EDS. Deep learning models—RNN, LSTM, and CNN-LSTM-A—were innovatively applied to predict flame retardancy (heat release rate (HRR), total heat release (THR)) and smoke suppression (total smoke release (TSR), total smoke production (TSP)) of the composites. The results indicate that the addition of 9 wt% TA/PA significantly enhances the flame retardancy and smoke suppression properties of the composite material. Specifically, the PHRR and THR were reduced by 49.03 % and 11.52 %, respectively, while the TSR and TSP both decreased substantially by 87.20 %. The modulus of rupture (MOR), modulus of elasticity (MOE), and internal bond strength (IB) increased by 34.01 %, 35.52 %, and 84.62 %, enabling simultaneous enhancement of flame retardancy and mechanical properties. The CNN-LSTM-A model demonstrated superior prediction accuracy, with R² values between 0.9769 and 0.9949 (Testing sets) and 0.9776–0.9981 (Training sets). This work provides both theoretical and practical support for developing high-performance green composites from waste wood via TA/PA modification and introduces a deep learning-based intelligent prediction approach for performance-oriented design and development of advanced wood composites.