The drive to more power-efficient, light-weight and/or performant systems results in more and more nonlinear dynamic system behavior. An example of nonlinear behavior can be retrieved in mechatronic systems, e.g. due to friction and actuation nonlinearities, or in large-scale mechanical systems where nonlinearities often present themselves at the interconnection of the subsystems. Data-driven nonlinear modelling approaches are often highly complex, requiring an expert user, and resulting in highly complex model descriptions. However, for a successful adaptation in a wide range of applications, data-driven nonlinear modelling tools and the resulting nonlinear models need to be interpretable by the practitioners.
The aim of this project is to integrate and extend the recently developed machine- and deep learning tools for the identification of nonlinear dynamical systems with a specific focus on (deep) neural networks. Whereas (deep) neural networks have proven their capabilities in many classification and regression tasks, the use of these estimation methods have been less well studied for the purpose of modelling nonlinear dynamical engineering systems. This PhD project aims to surmount these challenges by establishing an innovative synergy between the machine learning and the nonlinear system identification communities. The goal is to develop computationally efficient model learning approaches capable of supporting control synthesis and system design with special attention towards explainable AI approaches. The flexibility of the machine learning framework in defining learning objectives (aim-relevant estimation) and its ability to facilitate optimal recovery of structural relationships (model structure selection) provide novel perspectives in terms of developing dedicated methods to solve the limiting problems the current nonlinear identification theory.
Tasks:
- Study the literature of machine learning, nonlinear identification and modelling.
- Development of (control-oriented) interpretable data-driven nonlinear modeling approaches.
- Stochastic analysis of consistency and convergence of the results and empirical validation of the techniques on complex physical/chemical and/or electrical/mechatronic systems.
- Exploration of the steps of the identification cycle for the developed methods from experiment design to verification of model completion (validation).
- Dissemination of the results of your research in international and peer-reviewed journals and conferences.
- Writing a successful dissertation based on the developed research and defending it.
- Assume educational tasks like the supervision of Master students and internships.