A Neural Network to Predict Particleboard Manufacturing Process Parameters

Abstract

Abstract A methodology to predict the occurrence of out of control conditions, based on current operating conditions, in a particleboard manufacturing facility was developed using neural network theory. Multivariable linear regression and time series analysis techniques were applied to analyze the data set for informational purposes. Back-propagation neural networks were successfully trained to represent bonding treatment process parameters. The inputs to the network included data representing the current process condition, including moisture contents, bulk densities, and dryer temperatures, along with historical values of these same parameters. Neither the regression nor the time series analysis resulted in the development of a valid statistical model. The best trained neural networks were able to successfully predict the development of out of control conditions in the manufacturing process for 70% of the test cases, having been trained using only a relatively small training data set. Thus the potential for the utilization of neural networks in manufacturing process analysis was demonstrated. For. Sci. 37(5):1463-1478.