An approach to fault diagnosis of chemical processes via neural networks

This article presents an approach to fault diagnosis of chemical processes at steadystate operation by using artificial neural networks. The conventional back-propagation network is enhanced by adding a number of functional units to the input layer. This technique considerably extends a network's capability for representing complex nonlinear relations and makes it possible to simultaneously diagnose multiple faults and their corresponding levels in a chemical process. A simulation study of a heptane-to-toluene process at steady-state operation shows successful results for the proposed approach.