For industry, advanced control methods boost bottom line
In a chemical plant, a single raw product such as crude oil feedstock might be refined into many products, including gasoline, jet fuel and asphalt. How one product becomes many requires extremely precise, tightly controlled integrated processes, and Paul A. Elfers Professor of Chemical and Biological Engineering James Rawlings and his students develop the theory and algorithms to hone this control.
For many years, standard industrial controllers, coupled together, controlled multistage processes. Over time, these processes became more complex, and different parts of a process affected others. “It was hard to design local controllers to work together well to control a large, integrated system,” says Rawlings.
Rawlings and his students study the underlying theory of a control method called model-predictive control, which considers all variables in a large integrated system. The model uses plant data to forecast what the system will do, and then repeatedly optimizes the system over time. Rawlings’ group conducts research to verify and improve how accurately this happens. “Can you prove that the system will go to the set point you’ve selected?” he says. “Can you prove that it will be what’s called ‘robust’ to disturbances and model errors?”
His group also has developed algorithms that take thousands of variables into account and repeatedly and efficiently solve optimization problems in real time.
In many industries, these advanced control methods lead to more efficient processes, energy savings, improved products and an increased bottom line, says Rawlings. “This is having an impact of savings on the order of $1 billion per year, just in the chemical process industries in North America, compared with the previous technology used for this,” he says.
His collaborators include Computer Sciences Professor Steven Wright and industrial partners in the Texas-Wisconsin-California Control Consortium. Consortia members fund a variety of Rawlings’ applied research, while National Science Foundation enables the fundamental research Rawlings can extend to larger classes of optimization and control problems.
