Various AI applications reported in the literature have revealed an increasing use of expert systems and neural networks for blast furnace control. -Both- -model-free and model-based ANNs have been developed to predict the coke rate, with an accuracy of 9 to 13 kg. The trained ANN model works well in the fault diagnosis mode to reveal unstable periods of operation and/or faulty instrument measurements. ANN-based models offer an immense scope for improving process control in blast furnaces, even under noisy conditions. The application of ANNs and the genetic adaptive search for desulfurization of iron in torpedo, desulfurization and dephosphorization of steel in combined blown converters, and prediction of Ae_3 temperature for multicomponent steels has been discussed elsewhere. These examples demonstrate the enormous potential of adopting model-based AI techniques in steel plants. The term "model-based" implies that the process models are made an integral part of the control scheme. Complex process models also can be simplified and adapted to a noisy situation through ANNs.
展开▼
