Regression and ANN models for predicting MOR and MOE of heat-treated fir wood

Abstract

Modulus of rupture (MOR) and modulus of elasticity (MOE) of heat-treated fir wood (abies sp) were predicted by simple and multiple regression and artificial neural network (ANN) models based on ΔE (total color difference), ΔL (lightness difference), contact angle, and mass loss. The results showed that ANN provided better performance compared to regression models. Predicted values of MOR and MOE by ANN and regression models concur well with measured values. ΔL was the best predictor for predicting MOR in both simple ANN and regression models. The cubic function was the best simple regression model for the prediction of both MOR and MOE. The lowest mean absolute percentage error (MAPE) by simple ANN model for the prediction of MOR and MOE were 2.21% and 0.61%, respectively and their corresponding MAPE values for the simple linear regression model were 3.99% and 0.91%. MAPE for prediction of MOR by multiple regression and ANN models were 2.75% and 0.7% respectively, and their corresponding MAPE for MOE were 1.80% and 0.55%, which were comparable with those of simple ANN and regression models. Since the difference between simple and multiple models is small, simple models based on predictors which could be easily and in a short time measured in-line and be inexpensive to predict MOR and MOE with low MAPE, are recommended. Therefore, from this point of view, the color change could be the best option, which can also be measured inline. Contact angle and mass loss are the next options. Mass loss is measured at the beginning and at the end of the heat treatment, and the contact angle can be measured in 30–60 s after heat treatment.