Trainee Project
Title:
How to choose well the data collection experiment in order to get an accurate model of the system dynamics
for control purposes ?
Dates:
2025/04/01 - 2025/09/30
Supervisor(s):
Description:
Type of research and context: This master project proposes to conduct theoretical research on identification
for model-based control of linear time-invariant systems. The topic focuses on the data informativity property,
which is crucial in order to guarantee an accurate identified model. The key scientific challenge of
this thesis project is to propose finite-time methods to verify data informativity. For adaptive control, it will
consider approaches such as linear quadratic control, quadratic predictive control, etc.
National positioning: The Ampère Laboratory (Lyon), LIAS (Poitiers), and IMS (Bordeaux) have research themes
in system identification. IMS and LIAS are particularly interested in identifying fractional-order systems, which
is not the case for this project. LIAS also conducts research on subspace identification. The Ampère Laboratory
has expertise in identification for control (through Xavier Bombois, CNRS Research Director). Xavier Bombois
has also worked extensively with Kévin Colin on data informativity. For both research axes, the assumption of
an infinite amount of data was considered. In this thesis project, we aim to develop tools for finite-time (non-
asymptotic) excitation design in order to verify the data informativity and synthesize optimal excitation to
ensure minimal regret for adaptive control.
International positioning: KTH Royal Institute of Technology (Stockholm, Sweden), Eindhoven University of
Technology (Eindhoven, Netherlands), Politecnico di Milano (Milan, Italy), and ETH (Zurich, Switzerland) have
strong research components in system identification. In relation to the first challenge covered by the thesis
project, the University of Gröningen has an active expertise in data informativity for data-driven control (e.g.,
through the works of Henk Van Waarde). However, these studies assume data are noise-free, which is not the
case in this master project.
for model-based control of linear time-invariant systems. The topic focuses on the data informativity property,
which is crucial in order to guarantee an accurate identified model. The key scientific challenge of
this thesis project is to propose finite-time methods to verify data informativity. For adaptive control, it will
consider approaches such as linear quadratic control, quadratic predictive control, etc.
National positioning: The Ampère Laboratory (Lyon), LIAS (Poitiers), and IMS (Bordeaux) have research themes
in system identification. IMS and LIAS are particularly interested in identifying fractional-order systems, which
is not the case for this project. LIAS also conducts research on subspace identification. The Ampère Laboratory
has expertise in identification for control (through Xavier Bombois, CNRS Research Director). Xavier Bombois
has also worked extensively with Kévin Colin on data informativity. For both research axes, the assumption of
an infinite amount of data was considered. In this thesis project, we aim to develop tools for finite-time (non-
asymptotic) excitation design in order to verify the data informativity and synthesize optimal excitation to
ensure minimal regret for adaptive control.
International positioning: KTH Royal Institute of Technology (Stockholm, Sweden), Eindhoven University of
Technology (Eindhoven, Netherlands), Politecnico di Milano (Milan, Italy), and ETH (Zurich, Switzerland) have
strong research components in system identification. In relation to the first challenge covered by the thesis
project, the University of Gröningen has an active expertise in data informativity for data-driven control (e.g.,
through the works of Henk Van Waarde). However, these studies assume data are noise-free, which is not the
case in this master project.
Keywords:
data-driven modeling, system identification, data informativity.
Conditions:
Duration: April 2025 - September 2025 (6 months)
Location: CRAN, Polytech site, 2 rue Jean Lamour, 54514 Vandoeuvre cedex, France
Desired Profile: We are looking for an M2 Master student in control engineering, machine learning,
or applied mathematics who wants to do a PhD after this internship at CRAN. The chosen
candidate will then be presented for the doctoral competition at the University of Lorraine in order
to get a funding. A good level in English is required; knowledge of French is not obligatory.
Location: CRAN, Polytech site, 2 rue Jean Lamour, 54514 Vandoeuvre cedex, France
Desired Profile: We are looking for an M2 Master student in control engineering, machine learning,
or applied mathematics who wants to do a PhD after this internship at CRAN. The chosen
candidate will then be presented for the doctoral competition at the University of Lorraine in order
to get a funding. A good level in English is required; knowledge of French is not obligatory.
Department(s):
| Control Identification Diagnosis |
Funds:
Via CRAN/UL funding allocation
Publications:
