Engineering Education: Learning the “Black Box”

Students in FIRST Robotics programs use system modeling software, engineering calculation packages and CAD programs to solve engineering challenges. Image courtesy of FIRST Robotics.

The engineering director of Simulation-Based Design at Whirlpool Corporation, John Mannisto, reveals what he looks for in new recruits. A Ph.D. isn’t necessarily going to persuade him to give you a job. “The ones I’m interested in may only have a bachelor’s degree or a master’s, but they’ve had experience designing airflows in manifolds and engines,” he says. “They don’t necessarily understand how the computation works in the software code, but they can get the job done.”

Mannisto’s approach reflects the pragmatism of a hiring manager. He wants new hires who can hit the ground running. But university educators say they need to be teaching more than software skills in engineering curriculums.

“Students are using numerically based software, but they’re using them with very little numeric background,” says David Auslander, professor of Mechanical Engineering at UC Berkeley. To him, that’s a disconnect between the tool and the user. He’d like to teach students “to include numerical approaches from the beginning for such basics as integrals and derivatives” so they understand the underlying principles of simulation software.

Most employers prefer graduates with software mastery. But many educators feel that, without a thorough understanding of the mathematical principles and engineering rules, software mastery rings hollow, perhaps even dangerous. At the heart of this debate are the two different views of the software’s role: Trust the software like a black box that can spit out the correct answer; or peek inside the black box to learn how it produces the answer.

Recalibrating the Curriculum

Auslander specializes in automatic control system analysis and design, computation for physical system simulation and for control system design and implementation, building energy management, and mechanical system control. He has served as associate dean and acting dean of UC Berkeley’s College of Engineering.

“In the past, you needed ways to generalize and analyze problems because dealing with them purely with computation was too expensive,” Auslander says. “In Newton’s era, computation was expensive. In the Manhattan Project [1940s], they had to employ human computers—people who manually performed computing—to deal with the intractable calculation problems they encountered.”

However, the cost of computation drastically decreased once computers became widely available. Harvesting the horsepower of multicore processors, software like Wolfram Mathematica or MathWorks MATLAB can now compute what was previously considered impossible or impractical with hand calculation. Yet “basic engineering curriculums still feature largely a bunch of special cases for which computation was not available at the time. They’re about 50 years out of date,” says Auslander.

To realign engineering education with the type of complex problems students will likely be called upon to solve, Auslander suggests turning the curriculums inside out. The role of software training in engineering education is “a hot button issue,” he says. “The curriculum is not preparing the students to know how to use these types of analytical software.”

He’d like students to have an intuitive understanding of the math working behind the scene, the analytical algorithms in software. “You have to teach things that are numerically based from the beginning,” he says. Otherwise, he asks, “How would they know when they’re getting nonsense and when they’re getting real results? How can they formulate a problem in numeric terms?”

Auslander’s concern is echoed by Dr. Paul Lethbridge, academic program manager at ANSYS. “If you teach someone how to use the product only, you’re teaching them the black box approach,” he says. “If they don’t understand what’s going on inside the black box, then it’s garbage in, garbage out.”

Teaching Skills While Remaining Brand-Agnostic

Most software vendors have outreach initiatives targeting colleges and universities. From the vendors’ point of view, building brand familiarity among aspiring engineers is a worthwhile investment.

Autodesk, for example, offers free education licenses of its design and engineering software. Dassault Systèmes, Siemens PLM Software and PTC have robust educational licensing policies to provide their software to students at heavily discounted rates.

Despite access to commercial code, educational institutions have to narrow the number of software they teach in a course or the volume of software could overwhelm the materials. Some universities have adopted a brand they feel represents the norm.

“We have effectively adopted ANSYS as our CAE platform,” says Dr. Rajesh Bhaskaran, senior lecturer for Cornell University’s Mechanical and Aerospace Engineering department. “We’re using it in about 11 courses, both at graduate and undergraduate levels. But we don’t teach CAE as a separate course. It’s integrated with everything, from solid mechanics and heat transfer to fluid mechanics.”

Cornell University’s SimCafe.org features a collection of simulation tutorials based on ANSYS software. The content is distributed via Creative Commons licensing. It’s an e-learning portal to “integrate industry-standard simulation tools into courses and to provide a resource for supplementary learning outside the classroom,” according to the site.

“I floated through quite a few toolsets in my career, everything from ANSYS to Abaqus [simulation package from Dassault Systèmes],” says Mannisto. “Generally, the concept of constraints and loads—where to hold the model, where to put the load, what a good quality mesh is—that’s all the same. Those are the skills—not the peculiarities of specific software.”

Mannisto looks at FIRST Robotics Programs as fertile ground for talent scouting. Here, throughout a series of robotic contests, Mannisto gets to observe how students with a knack for tinkering work together to solve their electrical, mechanical and software engineering conundrums. “It’s like getting to interview a kid over the course of four [to] five years,” he says.

Software as Learning Platforms

In a commentary for DE (“Make CAE Mainstream”), Mannisto expressed his opinion. He wrote: “It is my belief that there is too much emphasis on the inner workings of a computer program and not enough on the practical application.” He recommended, “Teach CAE not as a course, but instead build it into the entire curriculum. Teach the beam equation, teach Mc/I, then apply it using finite element analysis. Teach continuity, momentum and Navier-Stokes, then show how to define an entrance condition and a pressure boundary in a computational fluid dynamics program.”

Bhaskaran says studying the software can be an education unto itself. “You can use the software to teach the underlying math by relating it to user inputs and outputs. For instance, by meshing the model, the user is marking out the points where she or he would like the tool to compute displacements directly and is also setting the number of algebraic equations that need to be solved.”

ANSYS’ Lethbridge said, “What Dr. Bhaskaran does is to use the black box and then teaches students to probe that black box so they understand what’s going on and how to interpret the results.”

“You don’t need to know how to code, but you need to know how the tool is taking differential equations and converting them into algebraic equations. What strategies does it use? What are the errors introduced in that process? What is it calculating directly and what’s derived through post-processing? When you do something in the software, you should know if you’re affecting the boundary conditions, the governing equations, the numerical solution strategies or post-processing,” Bhaskaran says.

The different views on how to treat the black box are not so far apart. In the end, both the hiring managers and the educators want the same thing: Aspiring engineers who can use the black box correctly. If there’s a gap between the two sides, it’s one that can easily be bridged with ongoing dialog.

Bhaskaran points out the crucial distinction between software expertise and engineering skills. “The software doesn’t solve the physical problem for you. It solves a mathematical model of the physical problem. Students need to know what the mathematical model is—the governing equations and boundary conditions as well as the physical principles and assumptions embedded in the model,” he says.

More Info:

• ANSYS
• Cornell University Sibley School of Mechanical and Aerospace Engineering
• Modelica
• National Instruments
• UC Berkeley Jacobs Institute for Design Innovation
• Whirlpool