Engineering is ruled over by many ironclad laws: Kirchoff’s Laws, Boyle’s Law, the laws of thermodynamics, and the list goes on. But when engineering principles encounter real-world situations, these laws cannot predict all aspects of a problem. That’s why, in the spring semester of 2019, Victor Zavala, the Baldovin-DaPra Associate Professor in chemical and biological engineering, introduced a handful of students to a way of thinking that doesn’t come naturally to many engineers.
They explored uncertainty, data analysis, and risk.
Victor Zavala
Zavala says the idea for a stats class tailored to chemical engineers was brewing for a while. In recent years, alumni surveys and anecdotes from practicing engineers revealed that they felt they lacked the tools for data analysis, risk modeling, and machine learning needed to tackle industrial problems. “On a trip to visit alumni in California, a recent graduate working in industry asked me whether the department offered online classes in data analytics and statistical analysis for chemical engineers,” says Kreuz-Bascom Professor and R. Byron Bird Department Chair Regina Murphy. “He mentioned the critical role of data analytics in his current position, and the lack of opportunities for practicing chemical engineers to learn about the field. The conversation reinforced my absolute conviction that courses in statistical analysis and data analytics are sorely needed by both our current students and our recent alumni and that the department should take a leadership role in developing the educational materials for this fast-blossoming field.”
With the support of the department, Zavala developed his three-module curriculum to try and fill that gap. While many engineering students take statistics courses, Zavala says existing classes don’t always approach the concepts in a way that makes sense to them. His course looks at the topic from an engineer’s perspective. “What is important in engineering is that we have this mixture of fundamental knowledge that comes from physics and chemistry and you need to reconcile that with real data,” he says. “So, you need to be careful that statistics are presented in the right context, or engineering students aren’t going to see why it is relevant and how this connects with other topics in the curriculum.”
The topic is not an easy one. “Thinking in statistical terms is a very different way of thinking than engineers are trained for, and it can be very challenging for students,” Zavala says. “Normally, engineers are trained to think, ‘This is the data. Solve the problem.’ They’re not taught to ask questions like, ‘What was the source of the data? Can you trust the sensor collecting the data?’ If you cannot fully trust the data, how confident are you in your decision?”
Zavala hopes that the course will eventually become part of the core curriculum for chemical and biological engineers. Currently, he’s stepped back from teaching the material and is putting together a textbook on the topic which is being supported by the CBE department’s Hougen Scholar program. He hopes to complete the book by the end of summer 2020. That way, he says, more faculty at UW-Madison and in different universities across the world will have the resources to teach this important topic to a wider swath of students.
Ultimately, he envisions the course would be taught in three installments, with the first module presenting concepts including modeling uncertainty during freshman year. “The idea is that, right off the bat, we teach students about how to go from data—things that you observe—into modeling uncertainty and how you can use uncertainty to characterize phenomena that you cannot model from first principles,” he says.
Then, as students’ math skills and knowledge of the field increase, follow-up modules would show them how to reconcile fundamental models with data. A final course would have students work on making decisions while designing a complex system. “For instance, they could design a chemical plant that has a lot of uncertainty surrounding markets or surrounding the equipment performance itself,” Zavala says. “They have to ask, ‘How do I characterize that uncertainty?’ ‘Will my design tolerate such uncertainty?’”
Erik Doershing, who graduated in the spring of 2019 and took the class in his final semester says it’s already had a big impact on his work as a chemical engineer at Honeywell UOP in Barrington, Illinois. “I felt that the course was instrumental in rounding out my education. It was essential for making the leap from theoretical classroom engineering problems to real-world applications in industry,” he says. “In the first six months alone of working in the chemical industry, I’ve found that understanding uncertainty and how data is analyzed is essential to effectively communicate with other engineers—Professor Zavala’s class provided the foundation for this understanding.”
In the end, Zavala believes it’s necessary for engineers to think more about data and risk, and he’s taught the class to working engineers in informal short courses. He would like to put together a more formal online offering as well. Zavala anticipates teaching the course at UW-Madison again next year, after completing his textbook.
“I feel that engineers have to adopt this way of thinking because there are severe issues when you don’t take risk into consideration,” he says. “It’s important for students to understand that decisions can be ambiguous when uncertainty is not properly taken into account. I think, ultimately for me, that’s what is key.”