If you want to talk about real-time process control, try the folks at Neural Applications Corp. Or perhaps the process engineers at the St. Paul plant of North Star Steel Corp., which partnered with Neural in prototyping advanced "intelligent systems" technology to optimize total energy input in electric-arc-furnace steelmaking. The Intelligent Total Energy Controller, based on neural-network modeling techniques, simultaneously analyzes and reacts to data from four separate energy-related subsystems -- which control the electric arc, gas burners, carbon, and oxygen input -- to achieve the most advantageous mix of energy inputs. A look at the numbers suggests the magnitude of the task: Process data collected by more than 200 sensors are analyzed by the computer system at a rate of up to 3,600 samples per second. And process-control feedback to the subsystems is updated three times every second -- making constant adjustments to the steelmaking process, notes Bill Staib, vice president for technology at Neural Applications. Unlike conventional engineering approaches to complex systems -- which control each subsystem separately -- the Neural Applications technology takes advantage of the adaptive and data-mining capabilities of a neural network to optimally coordinate the once-separate process loops. The value of neural-net technology, Staib explains, is that "the process-control model is based on data, rather than assumptions. It gives you a more exact model to begin with, and, as the process changes over time, you are able to track those changes." The Intelligent Total Energy Controller can play a role in reducing air emissions and also affords managers increased flexibility in responding to variations in the cost and availability of different energy sources. But the primary benefit is a reduction in energy-related costs. Neural estimates that aggregate savings could total $50 million a year for the U.S. steel industry if the technology were to be deployed across the electric-arc steelmaking sector. That's a conservative estimate, based on a 5% increase in overall energy efficiency. An earlier product, the Intelligent Arc Furnace controller, developed in 1991 and now in use at more than 30 sites, achieved energy savings of 8% by optimizing electrical input alone. The latest configuration extends the scope of the system to include chemical components of energy, such as oxygen input, as well. Still under development is a product that addresses the "pure chemistry" of the furnace, including the mix of scrap, Staib points out. The long-term impact could extend well beyond the steel industry. Intelligent systems technology -- which is a hybrid combining elements of neural networks, fuzzy logic, genetic algorithms, and expert systems -- also has potential in other industries, including chemicals and food processing. "Most process-control applications have some room for improvement," Staib points out. "The value of neural networks and intelligent systems technology comes into play in those processes where, say, a 5% improvement in efficiency is worth a lot of money, because it takes a lot of engineering power to solve it." Neural Applications will continue prototype development of its system through 1997. The technology is based on development work led by Edward Wilson, an intelligent-systems scientist who is a visiting scholar and lecturer at Stanford University. Dane Meredith, melt-shop superintendent at North Star Steel, played a key role in the prototyping stage. A commercial product is expected to be available sometime next year.