Making models deal with the complexity of modern engineering systems

In engineering systems, models are used for the model-based design to predict the system's future performance, without the need for prototyping, controller deployment, or time-consuming experimentation.

Subsequently, designs can be improved, leading to enhanced performance with fast design cycles. Such engineering systems have been key enablers of uncountable many technological breakthroughs.

Complex and non-linear

There is a drawback though. As engineering systems are becoming increasingly complex and modeling accuracy requirements increasingly stringent, non-linear models derived from physical laws of nature are often too complex for model-based design.

For example, using such a complex model for prediction can be computationally demanding and sometimes even infeasible due to limited computational and data storage capabilities as well as time restrictions. To reduce the computational expense, the complex model is often replaced by a reduced model that is significantly less complex, yet sufficiently accurate.

Driven by data

An alternative to physics-based modeling is data-driven modeling. Such an approach uses recorded input and output data from the system to directly construct compact models that closely describe this data.

Therefore, data-driven modeling is typically fast and enables system modeling if first-principle modeling is infeasible. For example, data-driven modeling provides basic understanding of our brain function, whereas first-principle modeling of our brain function is considered a challenging task.

New modeling tools

In his research, Fahim Shakib has developed a set of methods to reduce the complexity of the models while preserving their key properties, with minimal accuracy loss.

He also presents fast data-driven modeling methods that use experimental data obtained from the system and deliver guaranteed model stability.

In both cases, certified model stability is a key enabler in safely using these models in unseen and generalized scenarios for model-based system-and-control design. The developed methodologies have been validated experimentally and numerically.

More information

Fahim Shakib, Data-driven modeling and complexity reduction for nonlinear systems with stability guarantees, Supervisors: Nathan van der Wouw, Alexey Pavlov and Alexander Progromskiy.