ELEC workshop on System Identification, 29 May – 23 June 2017



The VUB-ELEC department is proud to announce the 2017 edition of the workshop on System Identification, as a continuation of 9 successful editions of the ELEC Doctoral School on the Identification of Nonlinear Dynamic Systems (with a slightly altered format).

System identification is the engineering discipline which aims at constructing mathematical models of physical systems, based on measured data. The purpose of these estimated models can be very broad: to gain physical insights into the system, to build a robust controller, to detect anomalies, to compensate for unwanted behaviour, to predict future outcomes, etc... Since 1989, the VUB-ELEC department has gathered a lot of experience in system identification.

The ELEC workshop on System Identification brings together researchers in System Identification and researchers/industrials who are in need of modelling tools based on measured data. As a particpant, you will have the opportunity to learn from, and collaborate with experts in (frequency domain) system identification to solve your own modelling problem.

The previous years, the doctoral school has been glad to host participants from 23 different countries across the world, in application areas including robotics and automation, electrochemical impedance spectroscopy, bio mechanics, telecommunication, ...

Example application areas

System identification has applications in a lot of engineering disciplines. Some examples below.

Electrochemical impedance spectroscopy

This includes the detection and characterisation of corrosion of metals, and the characterisation of the state-of-charge (SOC) and state-of-health (SOH) of batteries.

Power electronics

Heavy duty electrical machines involve the use of highly nonlinear electronic components in complex control loops. Reliable models of these components are crucial for the predictability and thus the safety of these devices.

(HF/RF) electronic circuits

The requirements on (HF/RF) electronic circuits like amplifiers and mixers is becoming increasingly stringent, such that advanced modelling and identification tools are required for design and distortion compensation purposes.

Mechanical applications

In robotics, a position dependency of the dynamic behaviour can be observed, which can be captured by using nonlinear or time-varying identification tools. In the study of structural vibrations (modal analysis), the modelling tools must take into account the possible variations of the load (think of a bridge with a variable density of cars driving on top, or a wind turbine subject to different wind speeds).

Biomedical applications

Making black box models of (parts of) living creatures is very challenging, and involves the appearance of many nonlinear and parameter-varying effects. Think of the influence of moisture on the mechanical and electrical properties of biological tissues, or the dependence of the dynamics of joints on the muscular activity.

Courses lectured (with exercises)

During the first two weeks of the workshop, the following topics will be covered during the lectures and exercises.

Frequency response function measurements (Non-parametric tools)

Teachers: John LataireYves RolainRik Pintelon

In this course we learn how the frequency response function (FRF) of a linear dynamic system is measured using periodic and random excitations. The attendees will learn how to

  • design broadband periodic excitations (multisines)
  • make FRF-measurements using periodic excitations and random excitations: how to select the power spectrum? how to deal with transients? How to measure the noise characteristics? How to tackle a situation in feedback? How to work in an errors-in-variables framework?
  • how to design excitation signals for detecting and quantifying nonlinear distortions and time variations?
  • how to distinguish the nonlinear distortions and the time variations from the noise contributions from a single experiment?

Dynamic System Identification (Parametric tools)

Teachers: Rik PintelonIvan Markovsky, Philippe DreesenJohn LataireMaarten SchoukensKoen Tiels

In this course we learn how to build mathematical models starting from noisy data or measurements. Then, we apply this methodology to the parametric identification of linear, nonlinear, and time-varying dynamic systems. The following topics are covered:

  • Why do we need identification methods? A simple example.
  • Properties of estimators: asymptotic unbiased and consistent estimators; efficiency; Cramér-Rao lower bound.
  • A systematic approach to the identification problem: least squares, weighted least squares, Maximum likelihood, Bayes' estimators.
  • Model selection and validation criteria.
  • Formulate estimators for dynamic systems, with a specific attention to the assumptions on the setup (Zero-Order-Hold versus Band-limited), also tackling the problem of missing data.
  • Behavioural theory appplied to system identification, with solution methods in the low rank approximation setting (more info).
  • Nonlinear systems, identified as block structures, and via the PNLSS (Polynomial Nonlinear State Space) approach.
  • The parametric identification of time- and parameter varying systems.

Invited Lecture: Control-oriented system identification

Lecturer: Simone Formentin (Dipartimento di Elettronica, Informazione e Bioingegneria, Politecnico di Milano)

Abstract - For many years, the role of system identification within control projects has been that of computing from data a dynamical model of the plant under control, which is as accurate as possible. A suitable model-based controller is then designed standing on the assumption that the identified model coincides with the real system. In case of a large uncertainty, robust control can be employed to take into consideration not only the nominal model, but also a description of the uncertainty set. Unfortunately, robust controllers may lead to very conservative performance.
This course addresses the interplay between identification and control and shows that, to overcome the major limitations of model-based control, the model and the controller must be designed together. In this way, the identified model does not necessarily fit the data at best, but the related controller maximizes the closed-loop performance. As an effective alternative, the direct mapping of data onto controller parameters is also discussed.
The course ends with a close examination of the most recent advances in the field and a discussion of the main open problems.

Collaboration opportunity

Prospective participants are invited to propose a small project, to be worked out during the last 2 weeks of the workshop, in collaboration with ELEC researchers. More info in the call for participants.


See the call for participants

Back to top