Using virtual data effects to stabilize pilot run neural network modeling

Executing pilot runs before mass production is a common strategy in manufacturing systems. Using the limited data obtained from pilot runs to shorten the lead time to predict future production is this worthy of study. Since a manufacturing system is usually comprehensive, Artificial Neural Networks are widely utilized to extract management knowledge from acquired data for its non-linear properties; however, getting as large a number of training data as needed is the fundamental assumption. This is often not achievable for pilot runs because there are few data obtained during trial stages and theoretically this means that the knowledge obtained is fragile. The purpose of this research is to utilize virtual sample generation techniques and the corresponding data effects to stabilize the prediction model. This research derives from using extreme value theory to estimate the domain range of a small data set, which is used for virtual sample production to fill the information gaps of sparse data. Further, for the virtual samples, a fuzzy-based data effect calculation system is developed to determine the comprehensive importance of each datum. The results of this research indicate that the prediction error rate can be significantly decreased by applying the proposed method to a very small data set.