Predicting Fiberboard Physical Properties using Multilayer Perceptron Neural Network

Medium Density Fiberboard (MDF) is an engineered wood used in furniture industry as an alternative to solid wood. Bes ides using forest wood and rubber wood as the main source of fiber, oil palm biomass was proven as an excellent substitute. Regardless of any fiber used, identifying its strength level is the main issue. Therefore, prior to releasing processed fiberboards for manufacturing use, boards need to undergo test procedures for mechanical and physical properties as set by the standard. These tests are timely, especially to research institutions which involve various characteristics of boards. The most extensive procedures of BS EN standard are 24-hour thickness swelling, 24-hour water absorption and 48-hour moisture content. The aim of this research is to reduce testing time by excluding these lengthy tests. A model of each is produced to predict the properties of omitted tests using other properties of MDF, including fiberboard density and percentage of empty fruit bunch fiber. A prediction model was produced by the multilayer perceptron Neural Network containing seven input neurons for seven predictors. Only one hidden layer used with four neurons. Output layer contains three output neurons, one for each target. WA24hours obtained smallest SSE for both training and testing with 0.113 and 0.108 respectively. Prediction model has contributed to the increase in MDF testing efficiency based on British Standard European Norm (BS EN).