Development and application of neural network algorithms for process diagnostics

The following three problems are addressed: (1) multiple-input single-output heteroassociative networks for signal validation for distributed sensor arrays; (2) multiple-input multiple-output autoassociative networks for plant-wide monitoring of a set of process variables for diagnostics; and (3) artificial neural networks for online estimation of chemical composition from spectroscopy data. Both static and dynamic forms of the backpropagation network (BPN) have been developed and applied to the solution of these problems. Chemometric data from Raman FT (Fourier transform) spectroscopy was used to estimate chemical sample composition. Several features of network training and implementations are presented, including adaptive updating of the sigmoidal threshold function during training, an optimal choice of hidden layer nodes using Shannon's information theory approach, and automatic scaling of network inputs and outputs for data encoding and decoding. The details of the development and implementation of the multilayer perceptrons and applications to industrial problems are highlighted.<<ETX>>