Deep learning-aided preparation and mechanism revaluation of waste wood lignocellulose-based flame-retardant composites.

Wood and its derivatives play a decisive role in traditional Chinese architecture. Waste wood as a major source of garbage in the construction industry represents a valuable source. The efficient recycling of waste wood has become an urgent technical problem in waste recycling research. Herein, we report a facile method to develop a high-performance biomass-based flame-retardant composite from waste wood bonded with isocyanate adhesive. The phytic acid and tannic acid were used as bio-based flame retardants. The effects of flame-retardant type and quantity on the flame retardancy, smoke suppression, and mechanical properties of the composites were investigated. Furthermore, the flame-retardant properties of the composite were also predicted using a deep-learning model. The optimal flame-retardant addition of 9 wt% endows the composites with enhanced flame retardancy, smoke suppression, and superior mechanical properties. A heat release rate prediction model was developed using a long short-term memory network with R ranging from 0.94 to 0.99, indicating that the model can effectively predict the combustion performance of materials. This study supports the high-value utilization of waste wood through deep learning, contributing to the green and low-carbon development of the construction industry.