Artificial neural networks applied to online fault diagnosis in chemical plants

Different kinds of artificial neural networks are compared regarding with their application to the fault diagnosis in steady state chemical processes. Their performance is studied taking into account the influence of some design parameters. Faults in sensors are considered separately by using auto-associative neural networks and the proposed algorithm. The developments have been applied to two case studies. The first one corresponds to a chemical plant with recycle. The second one is applied to a fluidized bed coal gasifier, at a pilot plant scale. In this latter case, the performance of the selected and optimized neural network approach is compared with a statistical technique - the principal component analysis. The methodology of implementation and optimization of the artificial neural network approach for fault diagnosis shows promising results. This approach can be used to complement a knowledge-based approach for robust fault detection and diagnosis in chemical plants.