What is a neural control system?

Headnote

Neural control systems operate by 'learning' from experience which can be rapidly acquired, rather than relying on complex mathematical modelling. It becomes relatively easy to prioritise the requirements of different steelplants. Where one EAF operator might wish to minimise power use for a given output, another might wish to maximise output for a given power expenditure.

Artificial intelligence has been used to provide significantly better arc furnace control than traditional methods. The Intelligent Arc Furnace (IAF) controller, developed by Neural Applications Corp is said to reduce electric power consumption by up to 8%, reduce electrode consumption by up to 25%, and increase productivity by up to 12%.

The IAF control system is based on a type of artificial intelligence called a neural network, in which the control system makes decisions in a way remotely similar to the way a human brain does. As it gains experience, it teaches itself how to control the furnace more accurately and efficiently.

EMULATING THE BRAIN

The IAF controller uses computers that are programmed to act in a way similar to that of a human neural network. The system includes a general program for controlling an electric arc furnace, a network of software neurons that learns how to evaluate conditions in the furnace and make decisions, and a learning algorithm that helps the network determine the importance of various inputs.

The system has three basic neural network programs (Figi):

* a Furnace Emulator that predicts furnace operations;

* a Regulator Emulator that is initially trained to duplicate the response of a traditional controller;

* a Neural Furnace Controller which combines the outputs of the two emulators to optimise furnace operation.

Like a human, the IAF controller must be taught how the arc furnace operates. The basic teaching process requires about 10min for the controller to observe the furnace being operated. It monitors conditions in the furnace, observes the control actions taken by the existing controller, and evaluates the results of those control actions. As it observes the operation, the controller learns how the furnace reacts to control adjustments and it adjusts the 'weights' of individual neurons accordingly.

COMPLEX INTERRELATIONSHIPS

To control an arc furnace, the following information is collected by the data acquisition system for all three phases (Fig...