Prediction of the color change of surface thermally treated wood by artificial neural network

Abstract

Surface thermal treatment (STT) can achieve efficient and successful thermal modification on wood surfaces, resulting in a beautiful, natural, uniform darker color and velvety texture. This study aimed to evaluate the effect of STT on White Ash, Yellow Poplar, and Red Oak specimens using a heated press at varying temperatures and times. To enhance the material utilization, reduce the number of experiments, and optimize the process, we employed artificial neural network (ANN) to model the relationship between the provided color, treatment time, and temperature required to attain the desired color. As the ANN model can simulate the process result very fast with a high degree of accuracy ($R^2$ above 0.96), it allowed us to rule off approximately 95% of the possible combinations, conducting a minimal subset of experiments and thereby saving an enormous amount of time (one experiment takes five hours to be prepared appropriately and more than 20 samples need be tested to get the ideal color). Previous research either investigated how to use ANN or demonstrated other new methodologies for applying thermal treatments. In this study, we propose a novel method to do efficient thermal treatment and train an ANN model which helps eliminate the misdeem experiments. Our ANN model can successfully predict the color change of thermally treated wood. The mean absolute percentage errors (MAPE) from our models were from 10.61 to 10.97% for training and 10.00–10.41% for testing. All obtained determination coefficients ($R^2$) were above 0.96. We have demonstrated our method on White Ash, Yellow Poplar, and Red Oak specimens, compared the findings to previous baselines, and exhibited an improvement of over 30% for $R^2$ in several instances.