Interview with Bojana Rosic, University of Twente, Enschede/NL

Please briefly explain what your work is about.

The creation of virtual representations of physical systems inherently relies on stochastic physics-based modeling (often of the multi-physics type) and Bayesian analysis. This approach enables components of the system, and the system as a whole, to simulate a range of possible behaviors and outcomes rather than relying solely on deterministic predictions. In the context of digital system simulation, the aim of using Bayesian analysis is to integrate new data as it becomes available, ensuring that the model accurately reflects the current state of the physical system. In my research, I work on the Bayesian integration platform by addressing challenges such as real-time data fusion through the development of efficient Bayesian procedures, real-time prediction of system behavior using machine/deep learning and other types of surrogate models, and the creation of procedures for model order reduction and domain decomposition of high-dimensional systems. In other words, my focus is on effectively handling uncertainties and ensuring scalability for complex, high-dimensional systems, achieving real-time simulation for stochastic optimization purposes. In the application domain I work in areas such as additive manufacturing, predictive maintenance, system and manufacturing process optimization, and decision-making in adaptive and autonomous systems.

Where do you expect the largest impact of stochastic optimization?

Stochastic optimization is set to have a significant impact in engineering by enabling more robust and efficient decision-making under uncertainty. Its greatest value lies in optimizing complex systems where variability in process parameters, machine/system performance, environmental conditions, and operator behavior can all influence process outcomes. Applications such as process design, its control and reliability benefit from stochastic methods that account for these uncertainties, leading to safer and more efficient products. By incorporating probabilistic models, engineers can design processes and systems that are not only optimal on average but also resilient to variability, reducing risks and improving long-term performance.

Where are the bottlenecks?

Bottlenecks in process and system optimization frequently occur where variability and uncertainty have a significant impact on production flow and quality, particularly in high-dimensional systems with numerous interconnected parameters and processes. In such complex, high-dimensional environments, small uncertainties can propagate and amplify throughout multiple stages, making accurate modeling and control especially difficult. This complexity is further compounded by the challenges of digital twinning, where creating a precise and real-time virtual representation of the system is hindered by sparse, noisy, or incomplete data. Human factors, such as differences in operator skill or adherence to procedures, add yet another layer of uncertainty. Addressing these high-dimensional bottlenecks through advanced stochastic optimization methods is crucial for enhancing robustness, minimizing waste, and improving overall manufacturing productivity.

Organiser and Contact

  #EcceEcab25