Sie sind auf Seite 1von 2

F O C U S O N E D U C AT I O N

Teaching Mathematics to Control Engineers


Richard D. Braatz

ow can mathematics courses be taught so as to be more effective for the training of control engineers? Without doubt, the range of mathematical techniques relevant to at least one branch of control engineering is large and includes multivariable calculus, ordinary and partial differential equations, real and complex analysis, linear algebra, multivariable statistics, convex optimization, functional analysis, and differential geometry. To be effective in applications, a control engineer should also be versed in at least one other engineering discipline (such as electrical, mechanical, or aeronautical) as an understanding of a system is required before an effective control system can be designed. A control engineer should also be trained in systems and control problems and techniques, such as model identification, experimental design, fault detection and diagnosis, dynamic optimization, model predictive control, robust control, and nonlinear control. Collectively, the number of courses needed to cover all of these topics would be too high to fit into one curriculum, or even two curricula, which means that most control engineers take courses in only some of these topics. There is not enough time in a control engineering curriculum for students to take all of the mathematics courses that could be potentially relevant, so it would be best if the mathematics courses that are taken by students have content that is most relevant to the training of control engineers. Probably the most obvious example of a mismatch between the topics that are covered and the needs of control engineers is in linear algebra. A well-trained engineer should have some knowledge of linear algebra, and so most engineering degree programs require its students to take an introductory course in linear algebra, early in their studies. The coverage in most of these courses is poorly aligned with the needs of most engineers, and the alignment is even poorer for training control engineers. A typical linear algebra course covers matrix multiplication and addition, eigenvalues, eigenvectors, matrix inverses, and determinants, and a great deal of time is spent teaching students how to determine matrix inverses and solve systems of linear equations for matrices of arbitrary dimensions by hand
Digital Object Identifier 10.1109/MCS.2013.2249471 Date of publication: 16 May 2013

using the method of elimination of variables, Gaussian and Gauss-Jordan elimination, and Cramers rule. The concepts in these methods are important for some students, but rarely will a control engineer solve a large system of linear equations by hand. A different selection of material would increase the usefulness of the linear algebra course for control engineers. For example, instead of spending weeks learning Gaussian elimination, that time would be better spent covering quadratic forms and positive-definite matrices so students can understand the definition of a Lyapunov function. Due to the mismatch between the content of an introductory linear algebra course and the linear algebra most useful for educating control engineers, the subsequent introductory control course usually begins with a review of basic concepts of linear algebra that were not taught in the earlier linear algebra course, such as matrix identities, singular value decompositions, and quadratic forms. A typical student learns much more linear algebra useful to control engineering from reading the appendix of a control textbook [1] or skimming through a reference book on matrix theory [2] than from reading an entire introductory textbook on linear algebra. The textbook on convex optimization by Stephen Boyd and Lieven Vandenberghe [3] includes many examples from control theory, including an optimal control problem in which the objective is to minimize the amount of fuel consumed to move a linear dynamical system from one value of the state vector to another, an optimal control problem to compute smooth and small manipulated variable signals for tracking a desired target output trajectory over a finite time horizon, the design of experimental inputs to maximize the information content of the resulting experimental data, and the estimation of parameters in the presence of uncertainties. This textbook is written in a manner that a student interested only in convex optimization can retain that focus, without being distracted by the control examples as they all flow naturally from the discussion of topics in convex optimization. Also, many of the control examples are in the homework problems, so an instructor who is not interested in control can just avoid assigning those problems as homework. For a student interested in both convex optimization and control theory, the textbook is optimal.
1066-033X/13/$31.002013ieee

66 IEEE CONTROL SYSTEMS MAGAZINE june 2013 

The recently published textbook on functional analysis by Yutaka Yamamoto [4] is tuned to the needs of educating control engineers. The first five chapters provide the background in vector spaces, normal linear spaces and Banach spaces, inner product and Hilbert spaces, normed dual spaces, and the space of linear operators useful for understanding the subsequent four chapters on distributions, Fourier series and transforms, Laplace transforms, and Hardy spaces. These chapters provide the strong foundation for the last chapter, which covers linear systems and control, controllability and observability analysis, state-space realization, and H-control. Most control engineers learn bits of functional analysis along the way as part of their control courses, whereas [4] provides a single cohesive treatment of functional analysis that would serve as background for all of the control courses. It seems likely that some of the mathematically oriented students learning from the textbook will become attracted to careers in control engineering and that a control engineer is more likely to take a course on functional analysis if taught from this textbook.

Perhaps I am overly biased by being a control engineer, but I feel that mathematics courses in such topics as linear algebra, convex optimization, and functional analysis designed for teaching control engineers would provide a better education for students regardless of whether they are in control engineering or in some other mathematically oriented discipline. I believe that mathematics is taught best when in concert with practically relevant examples, and what other discipline has a richer variety of mathematical examples than control engineering?

References
[1] T. Kailath, Linear Systems. Piscataway, NJ: Prentice Hall, 1980. [2] D. S. Bernstein, Matrix Mathematics: Theory, Facts, and Formulas, 2nd ed. Princeton, NJ: Princeton Univ. Press, 2009. [3] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.: Cambridge Univ. Press, 2004. [4] Y. Yamamoto, From Vector Spaces to Function Spaces: Introduction to Functional Analysis with Applications. Philadelphia, PA: SIAM Press, 2012.

Think you know about the latest technology?


You havent even scratched the surface.
See all the layers of technology with Proceedings of the IEEE. From outlining new uses for existing technology to detailing cutting-edge innovations in a variety of disciplines, youll nd the breadth of content and depth of knowledge that only IEEE can provide.

Go beyond the surfacesubscribe today.

www.ieee.org/proceedings

JUNE 2013 IEEE CONTROL SYSTEMS MAGAZINE 67

Das könnte Ihnen auch gefallen