Home : Volume 22 : Spring 1996 :
From model predictive control to thin films

While on the faculty of the University of Texas, chemical engineering professor James B. Rawlings concentrated on developing manufacturing methods for the gulf coast petrochemical industries. Now that he has returned to the UW-Madison, where he earned his PhD in 1985, he plans to apply the same techniques to Wisconsin's rapidly modernizing paper and dairy industries.

Rawlings specializes in "model predictive control," a computerized control method that adjusts manufacturing processes in real-time. The computer control system makes its initial decisions based on data and models from previous manufacturing runs. Various types of sensors monitor the product and send data back to the computer, which then fine-tunes the process's settings. This sort of iterative process is, Rawlings says, the advanced control theory most often used in the chemical process industries. In addition to adding to the basic theory of model predictive control, he is focusing on its application to specific industrial problems, such as efficiently manufacturing polymer films.

To make a thin film, a polymer is extruded out of a narrow die and then passed between rollers. As the machine stretches and heats the film to make it even thinner, beta ray sensors scan the film for variations in thickness. "You're trying to produce the film as uniformly as possible," Rawlings says. "If you have much variation, there are going to be unacceptable areas, such as holes and bumps, because the material is so thin. If you can get the variation to be very small, then you can take the average thickness and squeeze it way down. And then you'll get films that can do things that thick films can't do."

A critical point in the process is where the polymer passes through the die. There, a row of actuators--automated levers that push down on the die's lip--controls the film's thickness. By precisely altering the actuators' position, the system can remove high and low spots.

To know where to increase or decrease pressure just enough to make a uniform film, the control system has to accurately gauge the consistency of the film that has already passed under the actuators. Rawlings and graduate student John C. Campbell are working with the 3M Company to find just the right combination of monitoring, starting model and on-going adjustment strategies. The Madison researchers develop the software on campus and send it to a small pilot plant in Minneapolis, to which Campbell travels periodically to collect data. "Then we can say yes or no, that was a big improvement in the variability and gauge and thickness of these films," Rawlings explains.

By using a laser to measure the size of crystals in a chemical slurry and then fine-tuning its temperature, James B. Rawlings' "model predictive control" systems improve manufacturing techniques for chemicals that are produced as fine powders. (37K JPG)

He uses the same sorts of model predictive control techniques to improve the manufacture of chemicals that are produced as fine powders. The last steps in the manufacture of these pure specialty chemicals are usually crystallization from solution, and filtering and drying. Like film, the process of making these particles depends on uniformity. As crystals nucleate and grow in a solution, their size distribution varies. Some of the crystals may become much larger than others. The steps of separating the solid crystals from the solution and drying them, however, are most efficient when the crystals are all about the same size and are distributed evenly, Rawlings says. "If you don't control the crystallization step to make the crystals large and uniform, you can't separate the liquid from the solids in the slurry." The researchers have constructed an experimental crystallization system that uses a laser to measure the crystal size distribution in real time. The laser's light passes through the crystal slurry and scatters as it hits the solid crystals. By measuring the intensity of the diffracted light, the system calculates the crystal size distribution. Based on this measurement, the crystallizer temperature is adjusted to manufacture large crystals of uniform size.

The solutions that Rawlings finds for these chemical process industries also apply to two of Wisconsin's largest enterprises, paper and dairy. Paper is made much like polymer film: a slurry of wood pulp and water is drained and then squeezed between rollers. "They have the same issue. They would like to get the paper to be uniform quality across the sheet," Rawlings says. "In the paper industry I would like to establish the same kind of collaborative efforts where they'll let us come in to their plant, take data on their manufacturing lines, develop our own model, and then design an advanced controller, and install it." But, he cautions, unlike 3M's small, pilot facility, where they accept mistakes as the price of developing a good system, it's hard to find small paper plants. Most paper-making machines cost tens to hundreds of millions of dollars and "making a bad batch would upset the managers."

He has, however, already formed a collaborative project to study milk-fat crystallization with agricultural engineering and food science Associate Professor Richard Hartel. Milk producers need to measure and control the formation of crystals when they are separating fat from whole milk to make products such as skim milk.

When working on problems such as these, Rawlings and his students often use a specialized software package called Octave, which he and graduate student John W. Eaton developed. Octave is a high-level programming language that greatly simplifies tasks, such as creating and manipulating matrices, that are common in chemical engineering. "That frees the students, especially inexperienced students, from the tyranny of the syntax of the programming language, which is so unforgiving," Rawlings says. Although the researchers developed Octave to help students concentrate on chemical engineering issues rather than programming minutiae, the software has found its way into the laboratories of a wide range of physical sciences here and abroad.

One reason Octave has spread so rapidly is that rather than license it commercially, Rawlings and Eaton have made the program available to anyone without charge. Unlike developers of shareware programs, who distribute freely but expect their products to remain in their original form, however, Rawlings and Eaton expect users to customize Octave. The idea is that the software will eventually become self-sustaining as other researchers independently add features.

Now that it is passed freely from lab-to-lab and downloaded from Internet sites worldwide, the software has begun to evolve independently. But the real test of their plan will come when they stop maintaining Octave, Rawlings says. "I hope there is a large enough community of people who do these kinds of calculations and develop an expertise with Octave so that the program has a life of its own."

Content by perspective@engr.wisc.edu

Date last modified: Wednesday, 17-Apr-1996 12:00:00 CDT

Thank you for visiting!