Using Differential Evolution for Global Optimization with Constraints: Applications to the Design of Mechanical Systems

Lecture / Panel
For NYU Community

Smart Materials and Systems Seminar Series Presents

Professor K. Ming Leung

Department of Computer Science and Engineering
Polytechnic Institute of NYU

Global nonlinear optimization problems are ubiquitous in science, engineering, industry and commerce. They are rather difficult to solve especially when the number of unknown parameters is not small and the objective function possesses numerous local extrema. The problem becomes even more difficult if it involves unknowns consisting of a mixture of continuous, discrete, integer, and binary parameters. In addition, the presence of nonlinear constraints often makes the problem more challenging by making the search terrain highly fragmented and convoluted. This talk deals with the use of a nature-inspire method known as differential evolution to find global nonlinear optimal solutions to optimization problem with constraints. The method is further extended to treat problems with both hard and soft constraints whose priorities can be specified qualitatively and quantitatively. Applications in mechanical engineering such as the design of a pressurized vessel, a gear-train system, and a mechanical spring will be discussed.

[USER:333|profilelink] received his PhD in Physics from the University of Wisconsin at Madison in 1979. He was a postdoctoral fellow at the University of California at Santa Barbara before joining the Polytechnic in 1982. He became a full professor in physics in 1991 and in 1998 became a full professor in computer science. He was a research collaborator at the Brookhaven National Laboratory in Long Island, and has worked as a consultant at Bell Communications Research in Red Bank, Hughes Communication Research in Los Angeles, and at the NASA Jet Propulsion Laboratory in Pasadena. His recent research interests include nature-inspired algorithms such as differential evolution and their use in solving constrained global optimization problems in a variety of topics such as the smart power grid and cancer treatment planning.