Design Optimization

Engineering design is about functional characteristics: Can we build lighter mechanical devices without compromising structural properties? How can we design antennas, such as these in our smartphones, to send and recieve information as efficiently as possible? Are the better ways than traditional rules of thumb to design intricate details in loudspeaker components?  Can we improve the design of other acoustic devices such as mufflers and damping structures in noisy environments? Is it possible to reduce the aerodynamic drag on, for instance, an airplane wing by altering its surface structure to move downstream the point at which the airflow becomes turbulent?

All these are applications that have been addressed in the research developed by the Design Optimization Group. The foundational principle is to combine computer simulations with gradient-based optimization algorithms in a kind of learning methodology. The algorithms successively improves an initial design layout using the data given by computer simulations. A key ingredient is the use of adjoint variables, a techique also crucial in Machine Learning, where it is known as back propagation. This technique allows the efficient simultaneous handling of thousands, millions, and, recently, even billions of design variables. Therefore, these algorithms can handle design complexity beyond what is possible by manual means or trial and error.

The research carried out by the group concerns methods development as well as innovative applications. In the methods development, we aim for a vertical perspective, addressing a whole chain of issues: mathematical modeling of the considered systems, discretization and implementation on modern computing hardware, execution, analysis, post processing, prototype design, and experimental verification. Although we have covered the whole chain in some projects, most research focuses on the initial parts of the chain. Since we aim for novel methods, the software that is needed is usually developed within the research group itself. For the use of the methods in innovative applications, we cooperate with external partners that contribute system knowledge in the application under  consideration.