Predição de propriedades mecânicas da madeira termicamente modificada por redes neurais artificiais

Abstract

Treatments aimed at improving wood characteristics have been widely employed to enhance its value in the industry. Among these treatments, thermal modification stands out. However, after treatment, the evaluation of mechanical properties becomes crucial, as there is often a reduction in strength. Concurrently, artificial neural networks (ANNs) have played a significant role in wood quality assessment and processing, offering fundamental potential for predicting its characteristics. This study aimed to investigate the effect of thermal treatment at different temperatures on the mechanical properties and colorimetric characteristics of Pinus wood and develop prediction models using Artificial Neural Networks (ANNs) to estimate the Modulus of Elasticity (MOE) and Modulus of Rupture (MOR) based on thermal treatment parameters and colorimetric characteristics. The study involved static bending mechanical tests, colorimetric analyses, and thermal treatments on Pinus elliottii wood specimens. The results demonstrated that thermal treatment significantly affected the mechanical properties of wood, with an increase in MOE up to 160°C and a gradual reduction at higher temperatures, while MOR showed a decreasing trend from 180°C onwards. Thermal treatment of Pinus elliottii wood had significant impacts on colorimetric characteristics. Brightness (L*) exhibited a positive relationship at 160°C and then decreased, while red hue (a*) and yellow hue (b*) showed specific variations at different temperatures. Color saturation (C*) and hue angle (h*) were also affected by thermal treatment, with results varying according to temperature conditions and exposure time. The ANN prediction models proved effective in estimating MOE and MOR based on thermal treatment parameters and wood colorimetric characteristics. The results indicated that the ANN models achieved satisfactory levels of accuracy, with MAPE of 11.90% for MOE and 14.41% for MOR. Additionally, R² values were 0.82 for MOE and 0.81 for MOR, demonstrating the models' ability to explain variations in mechanical properties. In terms of practical implications, the developed prediction models can be applied in the thermally treated wood industry to optimize production processes, predict product lifespan, and enhance product quality. However, it is important to consider the models' sensitivity to extreme data variations and identify potential improvements, such as the inclusion of additional variables related to wood manufacturing and thermal treatment.