Prediction of Fiber Quality Using Refining Parameters in Medium-density Fiberboard Production via the Support Vector Machine Algorithm

Abstract

Fiber quality greatly influences the performance of medium-density fiberboard (MDF). To evaluate the fiber quality more accurately during refining, a novel quantitative parameter-property relationship model was developed based on the support vector machine (SVM) algorithm. Based on the mill production conditions, a total data set of 1173 experimental fiber quality data points under a wide range of five refining parameters was employed to train and verify the model. By comparing the effectiveness between the model using nonlinear SVM and the model based on multiple linear regression (MLR), the values of mean absolute error (MAE), mean relative error (MRE), root mean square error (RMSE), and Theil’s inequality coefficient (TIC) were reduced 92.19%, 92.36%, 87.29%, and 87.21%, respectively. The results showed that the performance of the predictive model developed using SVM was superior to the MLR model. Furthermore, the variations of the percentage of qualified fibers with each production parameter were predicted using the established model. The prediction model that resulted can be applied to predict the fiber quality during the refining process in an MDF production mill.