Prediction of water absorption and swelling of thermally modified fir wood by artificial neural network models

Classification is a useful tool for analyzing the quality control of thermally modified timber. This study aimed to predict water absorption and swelling of thermally modified solid fir wood (Abies sp.) by single and multiple input artificial neural networks (ANN) models based on the lightness difference (∆L*), the total color difference (∆E*), contact angle, and mass loss. Water absorption, swelling in longitudinal (α_L), radial (α_R), and tangential (α_T) directions, as well as volumetric swelling (α_V), were measured after 24 h of water soaking. The lowest mean absolute percentage error (MAPE) for the prediction of water absorption by the single input ANN model was 3.910%, based on mass loss. Moreover, for the prediction of αR, αT, and αV by the single input ANN model, it was 3.104%, 3.386%, and 2.755% based on ∆E*, mass loss, and contact angle, respectively. While MAPE values for the prediction of water absorption, α_R, α_T, and α_V by multiple input ANN models were 2.424%, 3.152%, 3.115%, and 2.067%, respectively. These levels of errors are satisfactory for predicting water absorption and swelling of thermally modified fir wood by the ANN models (MAPE < 10%). The difference between MAPE values of the single and multiple input ANN models based on different predictors was smaller than 3%. In terms of time and cost measurement, simple measurement, single input ANN model with color change, especially ∆L* and ∆E* as a predictor, could be the best item. From this viewpoint, the next preferred predictor could be contact angle.