A smart predictor for material property testing
Wilson Wang, Derek Kanneg · Smart Materials and Structures · 2008
A reliable predictor is very useful for real-world industrial applications to forecast the future behavior of dynamic systems. A smart predictor, based on a novel recurrent neural fuzzy (RNF) scheme, is developed in this paper for multi-step-ahead prediction of material properties. A systematic investigation based on two benchmark data sets is conducted in terms of performance and efficiency. Analysis results reveal that, of the data-driven forecasting schemes, predictors based on step input patterns outperform those based on sequential input patterns; the RNF predictor outperforms those based on recurrent neural networks and ANFIS schemes in multi-step-ahead prediction of nonlinear time series. An adaptive Levenberg–Marquardt training technique is adopted to improve the robustness and convergence of the RNF predictor. Furthermore, the proposed smart predictor is implemented for material property testing. Investigation results show that the developed RNF predictor is a reliable forecasting tool for material property testing; it can capture and track the system's dynamic characteristics quickly and accurately. It is also a robust predictor to accommodate different system conditions.