Researcher in the Spotlight: Daming Lou

I’m working on Next Generation Model Reduction Techniques for Complex Systems.

Hello, my name is Daming Lou and I’m working on Next Generation Model Reduction Techniques for Complex Systems. This takes place in the Control System research group of the Department of Electrical Engineering. Taking an interpolation approach to system dynamics (time domain and frequency domain), I’m able to capture and preserve the most relevant dynamics in the low-rank model. This is paired with two other projects within the Impuls II program, the work of Ruben Merks on optimal control and Bijan Goshayeshi on rarefied gas flows in small-scale devices. By sharing a common setup, we can discuss the nature of our problems together.

Going beyond conventional approaches

Figure 1. Projection-based model reduction for parameter-dependent systems.

What are these problems? The biggest challenge in this project is the dimensionality of the model. Due to the accuracy requirements of the system, the mathematical models tend to be large-scale and contain as many as 10e3. A huge amount of computational power is needed, meaning that conventional methods and approaches are not suitable for such models. 

We’ve proposed an industrial-approved method to get around this, and an associated toolbox is under development. This proposed method – projection-based model reduction for parameter-dependent systems – is depicted in Fig 1.  Together with algorithms and the toolbox, it will enable users to handle models with ranks of up to half a million.


Honing in on precision

Through my work, the computational complexity of systems dynamics can hopefully be reduced. As a direct result of this, it will be possible to design better and more accurate controllers. I believe that this project will be very interesting to people who are working in the high-precision domain, who deal with large-scale systems/models and who manipulate matrices – these are the people whose work is set to become more powerful and precise than ever before.