ARTIFICIAL INTELLIGENCE AND COLORIMETRY AS A COMBINED NON-DESTRUCTIVE METHOD TO PREDICT PROPERTIES OF HEAT-TREATED WOOD

Abstract

Colorimetric evaluation is practical, accurate and fast. Starting from the generally established fact that a heat treatment changes the wood properties, the present paper aimed to predict the properties of heat-treated wood by using colorimetry and artificial neural networks (ANNs). Eucalyptus grandis and Pinus caribaea wood samples were heat-treated to evaluate their color, as well as physical and mechanical properties. The relationship between the wood color and its physical and mechanical properties was evaluated through multilayer perceptron (MLP) neural network. The heat treatment darkened the wood, increased its dimensional stability and reduced its mechanical resistance. Artificial neural networks based on colorimetric and temperature parameters were efficient in modeling the wood properties, with better results to predict its physical parameters. The coefficient of determination (R2) of the models was high and the root mean squared error (RMSE%) low – with homogeneous distribution. The findings suggest that colorimetry is adequate as a non-destructive tool to evaluate heat-treated wood.