The Enhancement of Empirical Model Capability and Optimal/Robust Design of Intractable Processes

Abstract

Most industrial systems are developed and improved more from experimental data than from theoretical analysis. In this work, the robust design and optimal design of products and processes are developed using an information-index-based artificial neural network response surface. The structure and training policy of such an artificial neural network model are determined by the cross-validation information index (CVI for short) developed in this work. In the case of noisy and limited experimental data, this information index is particularly useful for finding the number of nodes in the hidden layer and the weighting of the smoothness factor. The polymer composite pultrusion process is studied. Simulation and experimental results show that this novel approach is highly effective and promising.