PhD Students
BIRPOUTSOUKIS Georgios (Starting date: January 2014)
Experiment design and Identification of Nonlinear Systems in the presence of Prior Knowledge
Thesis advisor: Johan Schoukens
Experiment design and identification of a dynamic system given a nonlinear structure constitutes one of the most challenging topics in the field of system identification. A critical step in the process of modeling a nonlinear system is the choice of the model structure. This problem is circumvented by focusing on nonparametric identication. The price to pay lies in the fact that the size of the model can very easily explode such that very long data records are needed in order to obtain a reliable model. Due to this fact, nonlinear identification is studied under the presence of prior knowledge about the system to be identified, aiming at reducing the effort and the number of data required during the modeling step. Apart from this, informativity of the identification experiment is directly linked to the accuracy of the model. Design of experiments for nonlinear systems has been studied till now for specific cases of nonlinearity when no prior knowledge about the system is available. Therefore, the optimal input for the identification experiment will be studied given a nonparametric model structure together with prior information about the system under study.
BRONDERS Piet (Starting date : October 2014)
Integration of state-of-the-art analysis and modeling techniques in the early design stages of RF PA design
Thesis Advisor : Gerd Vandersteen
Efficient power amplifier designs trade off the power efficiency and the NL distorsion of the amplifier. Current design strategies of complex amplifier structures use relatively simple design strategies, usually based on one or two tone tests. State-of-the-art analysis techniques, such as the use of multisines and the Best Linear Approximation (BLA), are only used in the final stage of the design to verify the nonlinear performance of the complete system. The aim of this PhD research is to integrate the recently developed BLA modeling techniques (developed for both time-invariant and time-varying systems) in the early stages of the design. The nonlinear interaction between two amplifier modules can be seen as either a NL or a time-varying effect. Hence, he general BLA theory for time-varying systems is an ideal candidate to setup a theoretical framework that enables the systematic design of complex amplifiers in a more efficient5 way. This will require the extension of the BLA theory for time-varying systems, as well as the bpractical realization and verification of the proposed design strategy.
BUSSCHOT Cedric (Starting date : October 2015)
Identification methods for low-power sensors
Thesis Advisor : Gerd Vandersteen
At this moment, system identification has already several industrial applications. Examples can be found in measurement devices, the medical sector and the car- and aircraft industry. Up until now, PC’s are being used to carry out these identification algorithms. Implementing these algorithms on a microprocessor opens the door for applications where power consumption and portability play an important role. A good balance between the computing speed, the available memory, the power consumption and the numeric stability is crucial when using a microcontroller. These elements will be taken into account when designing a test implementation, the final goal of this thesis.
CAENEPEEL Matthias (Starting date : October 2012)
A Hierarchical Framework for Design Oriented Modeling
Thesis advisor : Yves Rolain
When electronic components are designed, physical (white box) models are used intensively. As the systems become more complex, the complexity of these models grows as well. Because of this not only the simulation time rises, it also becomes harder to find potential flaws or imperfections in the design. Figures of merit or performance indicators are used to check whether the design satisfies the specifications. Nevertheless they do not always allow to identify the flaws or imperfections in the design when the system doesn’t satisfy the specifications. The growing complexity of the white box models and the restriction of the performance indicators may reduce the efficiency of the design process. To overcome these problems, a framework of behavioral (black box) models will be developed. The aim is to introduce a hierarchical framework of models that allows to split the design into levels and that offers the most appropriate model for each level. Less accurate but also less complex models increase the insight in the design and can be used to localize imperfections in an efficient way. To resolve the imperfections a more detailed or even the physical model can be used. In this way insight in the design is kept without loosing accuracy.
To lessen the gap between the system identification and the designer community, this framework of models will be illustrated on the design of several microwave systems. Eventually the aim is to obtain a general method to create a framework of models for the design of an electronic system.
COOMAN Adam (Starting date : October 2012)
Study and development of analysis and design techniques in a model-based framework
Thesis supervisor : Gerd Vandersteen
Current nonlinear modelling and design techniques rely on complicated models in which insight is easily lost. Therefore, these techniques are not always suitable for the design of such circuits. At ELEC, approximate models for nonlinear systems have been developed and verified. These ‘Best Linear Approximations’ (BLA) describe the nonlinear behaviour of a system in an intuitive way close to the concepts known by designers. Therefore, the BLA could be suitable in design. The goal of my research is to introduce the techniques of the Best Linear Approximation into a broad context of nonlinear analysis and design strategies. This consists of applying existing modelling techniques and, where necessary, extending them so that they can be used optimally in analysis and design problems. The developed techniques will be verified using both simulations and measurements on hardware realisations. To demonstrate the developed design and analysis techniques, (micro-) electronic design problems will be used. First, low-frequent circuits will be considered. Afterwards, the findings will be extended for the use in design of high-frequent electronic circuits.
DE COCK Alexander (Starting date: 1st October 2012)
Optimal input design for nonlinear systems
Thesis Advisor : Johan Schoukens
The quality of an identified model strongly depends upon the design of the experiment. For linear dynamic systems and for static nonlinear systems the optimal input design problem is well understood. However for nonlinear dynamic systems the picture is completely different. Not only the power spectrum but also the amplitude distribution (to compute the multivariate higher order moments) of the excitation signal are involved and should be simultaneously optimized. Additionally, due to the intrinsic complexity of nonlinear systems, it is very hard to find a model that is valid everywhere. As a result most nonlinear models are developed to describe the system in a restricted domain that covers the later use of the model. Outside that domain larger model errors are tolerated. For this reason it is also important to take into account on the intended use of the model during the input design. Finally, like for every input design, the robustness with respect to the variation on the model parameters needs to be taken in consideration.
DECUYPER Jan (Starting date: October 2013)
Nonlinear identification of vortex-induced oscillators
Thesis Advisor : Johan Schoukens, Tim De Troyer
Vortex-induced vibrations (VIV) are a frequently found form of fluid-structure interaction. Due to flow instabilities, vortices are shed in the wake of the structure. As a result, the structure is excited by an alternating lift force. Despite a growing amount of research, a full physical comprehension of the complex dynamical system has yet to be achieved. For now, one has to rely on cumbersome time-consuming computational fluid dynamic (CFD) simulations to be able to predict the kinematics of the system with high fidelity. What lacks is an accurate, efficient model that comes at a low computational cost. The inherent nonlinear nature of the relationship between oscillation and lift force, makes modelling a challenging task. Being highly nonlinear, the model must for instance be able to reproduce the autonomous oscillation of the lift force, even in the static non-excited case. Apart from this, also the typical phase-locking behaviour must be imbedded in the model. In this work a polynomial nonlinear state space (PNLSS) model structure is used to estimate the device under test. Construction of the model generally requires three major steps. First data is obtained by means of CFD simulations where the DUT is excited with both rich multi sine excitations as mono sines. From this data the BLA transfer function is constructed. Ultimately the linear state space representation is extended with nonlinear (polynomial) functions and a nonlinear optimization is performed, using the initial BLA as starting values.
EGON GEERARDYN David (Starting date : October 2011)
Development of User-friendly System Identification Techniques
Thesis Advisor : Johan Schoukens
The aim of this thesis is to develop system identification techniques that do not request an extended theoretical knowledge of system edentification methods from the user. Different sub-problems need to be solved+: the design of well-suited experiments, proper (pre-)processing of the measured signals, the selection of model complexity and architecture, and the validation of these models. This research aims to make system identification an easier and more accessible tool for experts in other domains who may have little experience with system identification.
ERTVELDT Julien (Starting date : October 2011)
Measurement and modelling of nonlinear flutter
Thesis Advisors : Patrick Guillaume and Rik Pintelon
Identification of aircraft aeroelastic behaviour is critical for certification of the aircraft. Nonlinear behaviour of the structure and aerodynamics, as well as time-varying conditions are difficulties that have to be overcome during the identification procedure. This thesis deals with the design and validation of an active aeroelastic test bench (AATB) to perform wind tunnel tests of both linear and non-linear aeroelastic systems. The test bench is also used for the validation of system identification methods for time-varying systems and their application to aeroelastic problems.
FAKHRIZADEH ESFAHANI Alireza (Starting date: September 2013)
Identification of Nonlinear Systems
Thesis Advisor : Johan Schoukens
Identification of highly structured non-linear models is highly required to fill the gap between designers and modellers. In this PhD, I will focus mainly on block-oriented structure identification. There are many challenges in the field, such as identifiability of blocks, the selection of the excitation signal characteristic for identification purposes, and block structure simplification.
GOOS Jan (Starting date: October 2011)
Measurement & Modeling of linear parameter varying (LPV) systems
Thesis advisor : Rik Pintelon
Although the linear time-invariant (LTI) system identification framework has already proven its merits for many years, in quite some applications the linearity and time-invariance hypotheses are only approximately true or not valid at all. In the linear parameter varying (LPV) framework, the dynamic relation between the input and output signals is still assumed to be linear, but it is continuously adapted based on the actual value of the scheduling parameters. Besides the excitation one should also choose a periodic or non-periodic scheduling. We will develop algorithms for the identification of such time-varying dynamical systems. In a first step, a non-parametric estimate of the noise and system model is estimated. In a second step, we try to extract a state space model with parameter-varying A,B,C,D matrices.
HOLLANDER Gabriel (Starting date: May 2014)
Modeling non-linear systems with structered models
Thesis advisor : Johan Schoukens
The goal of this research is to model non-linear systems with structered models. This is possible using non-linear state-space models, or with block-oriented non-linear models. It seems that both methods result in the same problem: a static multivariate vector function should be replaced or approximated by a set of decoupled monovariate functions, preceded and followed by linear transformation matrices. Finding the solution of this mathematical problem lies at the heart of the research.
LAZOV Vladimir (Starting date : June 2014)
Development of unified modelling, analysis and design strategy (Model-Aided Engineering) for the compensation of non-idealities using within telecom applications
Thesis advisor : Gerd Vandersteen
Modelling, analysing and developing a design strategy to identify the nonlinearities within telecom applications. Analysis various techniques will provide the necessary insight in the operation and the imperfections of (autonomous and non-autonomous) nonlinear electronic systems. This will be based on the Best-Linear-Approximation (BLA), a methodology that extends linear system theory to get insight in the nonlinear behaviour of the system. The analysis of complete transceivers will be considered. More particularly, the research is focused on cyclostationary signals that often arise due to time varying nature of the blocks used in the communication system such as: mixers, frequency converters and etc.
The aim of my work is to develop a complete description of a cyclostationary noise signals and efficient technique for analyzing it in large RF circuits within telecom applications.
MAES Hannes (Starting date : October 2012)
Development of FOT (forced oscillating technique) methods for non-invasive measurements of the respiratory impedance and the use of (non-linear) modelling for classification of lung pathologies
Thesis advisor : Gerd Vandersteen
At this moment there are two main challenges in the development of FOT methods for non-invasive lung impedance measurements. On one hand there is the challenge to improve the measurement devices so that the lung impedance (air pressure over airflow) can be measured in an accurate way, even when very low-frequent excitation signals are used. To obtain this, the nonlinear distortions of the measurement device and the influence of the breathing of the patient have to be taken into account. On the other hand there is the challenge to use the nonlinear character of the lung impedance to obtain a better classification of lung pathologies.
OLIVA URIBE David (Starting date : January 2011)
Estimation of the Mechanical Characteristics of Biological Tissues applying System Identification Techniques
Thesis Advisors : Johan Schoukens, Jörg Wallaschek (University of Hannover)
The aim of this research is to apply system identification techniques to estimate the mechanical parameters of biological tissues (in particular brain tissue) using a piezoelectric tactile sensor. The purpose of the estimation of the mechanical parameters is to provide the tactile sensor a reliable measurement procedure for the differentiation of two biological materials with slightly differences. A potential application can be found in neurosurgery for brain tumour resection.
PEUMANS Dries (Starting date: October 2015)
BLA-based Design and Analysis of VCO-based Sigma-Delta Modulators
Promoter: Gerd Vandersteen
The purpose of this research project is to unite two different nonlinear analysis techniques. On the one hand the theory of the Describing Functions, which is mostly used in control applications, but also has been proven valuable in contemporary applications like sigma-delta modulators. On the other hand, the state-of-the-art modelling techniques of the Best Linear Approximations, which have been developed and extensively studied during the past two decades in the department ELEC. Contrary to the Describing Functions, this Best Linear Approximation does allow to model dynamic nonlinear behaviour. The goal is to develop a unified nonlinear analysis technique and to illustrate its applicability on VCO-based sigma-delta modulators.
RELAN Rishi (Starting date: November 2013)
Data Driven Structured Modelling of Nonlinear Dynamic Systems (ERC Advanced Grant)
Promoter: Johan Schoukens
To close the gap between the designers and the modellers, we propose a fundamentally new approach to deliver highly structured nonlinear models meeting the designers needs.From a theoretical point of view, the major contribution is the development of a new nonlinear structured system identification framework. From practical point of view, the new nonlinear modelling paradigm will become an enabling technology to further push the performance and efficiency of system and control design.
TIELS Koen (Starting date : 2015)
User-friendly estimation of nonlinear state-space models
Johan Schoukens
Nonlinear state-space models are flexible model structures that can capture many nonlinear and dynamic system behaviors. This project aims at developing a user-friendly and modular Matlab toolbox to estimate nonlinear state-space models. I will also contribute to retrieving structure in nonlinear state-space models.
VAES Mark (Starting date : February 2013)
Disseminating System identification for non-specialists
Thesis advisors : Johan Schoukens, Yves Rolain
Today, system identification is a mature research field that created many well-developed and highly performing methods. One of the main and recurring problem that system identification faces is to reach a broader audience. This is due to the steepness of the identification learning curve. As a result, some practitioners that need highly performing models stick to outdated or even inadequate methods. The goal of this research is to lower the learning curve for identification methods as much as possible. To get there, a low cost self-study kit will be developed. Though, the main goal in this work is not to teach the theory of system identification in a magisterial way. We intend to train potential users to obtain skills that lead to a good identification practice. For practitioners in the field, a fast return and a practical orientation are mandatory. We intend to use the combination of simple hands-on exercises linked to a number of physical applications stemming from multiple application domains. This is done to show the potential and the weaknesses of the methods at hand. After this training, the practitioner should be able to select an efficient and up-to-date identification method that suits its needs maximally.
VAN NECHEL Evi (Starting date: October 2014)
Model driven design of high performance microwave filters
Thesis advisor: Yves Rolain
Microwave filters are steadily evolving towards the use of more complex transmission line structures as building blocks. This is needed to cope with the highly demanding specifications of modern telecommunication equipment. The design procedures that are used to realize these filters however have not evolved at the same pace. They still rely on the use of empirical design equations that are borrowed from standard line structures. The lack of accuracy that results from this crude approximation is then filled using massive numerical optimization based on EM simulations. This process is both time-consuming and blocks the intuitive insight in the operation of the device that is so useful to any designer. To circumvent these disadvantages, we propose to introduce models in the design process. As the “one model does it all” was already proven to be unsuccessful in previous attempts, we propose to use a different paradigm, where more a specifically tailored model is used for specific design tasks. More specifically, we intend to use a circuit based model to adequately and accurately describe the elementary building blocks of the filter structure. This scalable model will be used to introduce the boundary conditions and the process limitations of the technology in a very early stage of the design. Initially, this allows to close the design cycle early and adapt the prototype. In a second stage, we hope to introduce this additional knowledge in the approximation and the design of the initial prototype. If successful, this will fundamentally modify the way filters are designed.
VASQUEZ RODRIGUEZ Sandra Paola (Starting date: October 2015)
Fault diagnosis in on-shore wind farms based on linear parameter-varying (LPV) models
Thesis advisors: Michel Kinnaert (ULB) and Rik Pintelon (VUB)
This project introduces a fault detection and isolation (FDI) system for wind farms based on SCADA (Supervisory Control and Data Acquisition) data. Instead of the traditional approach of monitoring each turbine individually, this project proposes a model-based FDI system that exploits correlations between measurements associated to neighboring turbines. This feature could enhance the capability for early detection of faults and the reduction of false detection and missed detection occurrences.
The main challenge is to account for the difference in wind conditions that each turbine is subject to, which can significantly modify the relative turbine behaviors. To this end, linear parameter-varying (LPV) models will be used to represent the relative dynamics between the turbines as a function of wind speed and wind direction, notably. These LPV models will be identified in the frequency domain since the modeling can be done in a user defined frequency band. Also, the continuous-time framework will be chosen since models related with physics can facilitate the design of the FDI system.
The monitoring of wind farms will be done with focus on turbine efficiency and overheating, while ensuring a systematic design and tuning of the FDI system. The validation on a wind farm simulator and with SCADA data from an on-shore wind farm made of ten 2.5 MW turbines will be performed.
VERBEKE Dieter (Starting date: October 2014)
Identification Techniques for large scale MIMO-systems
Thesis advisor: Johan Schoukens
The goal of the project is to develop a user friendly identification methodology to identify systems with a large number of inputs and outputs. This results in a large number of transfer functions that need to be validated. Nowadays, this task is done manually, but it becomes infeasible if the number of inputs and outputs is large. A three-step procedure is proposed to deal with this problem: (i) a high quality nonparametric FRF is estimated for each transfer function (without user interaction), together with a nonparametric noise model; (ii) a parametric model is identified using one of the classical MIMO approaches; (iii) these models are validated using the information of the first step. Those transfer functions that fail the validation test should be further analyzed and improved. The identification methodology will be theoretically analyzed and verified on real life examples.
ZYARI Maral (Starting date: October 2013)
Nonlinear RF Measurement and modeling: going beyond the NVNA (Nonlinear Vector Network Analyzer)
Thesis advisor : Yves Rolain
The current state of the art in nonlinear network measurements at RF and microwave frequencies still is focusing either on CW or two-tone measurements with or without load-pull capability. The main reason for behind this is that these measurements allow an easy representation of the information in a format that is readily understood by the users and leads to the figures of merit that are used all over in the literature to specify the systems on the market. Now is the time to think about the next generation of instruments that leverages the capability of the first generation devices and extends it to cope with the challenges created by modern, power save and high capacity telecommunication systems that were sketched before. This requires one to extend the NVNA that is currently available to generate multi-tone modulated signals at all ports, to present variable and controllable impedance profiles over the modulation bandwidth at all ports.
Post-Docs
CSURCSIA Peter (Starting date : November 2015)
Development of a nonlinearity detection method for multiple-input multiple-output systems
Thesis supervisor : Johan Schoukens
The aim of the research task proposal is to develop a framework which provides a user-friendly interpretation of the measured multiple-input multiple-output (MIMO) by extracting the user relevant information. The key idea is to use the gathered knowledge of ELEC and develop a framework which is suitable:
i. to detect the presence of nonlinear distortions
ii. to semi-automatically decide if the linear framework is still safe to be used
iii. to indicate to the user how much can be gained using a nonlinear framework.
DREESEN Philippe (Starting date : October 2013)
Revealing structure in multivariate polynomial models
Host: Johan Schoukens
Nonlinear system identification often makes use of multivariate polynomials to describe the nonlinearities. For a given multivariate mapping, a simpler description, i.e., having decoupled univariate branches, may underlie such a description or provide a good approximation. This research is concerned with developing and understanding the different approaches, ranging from tensor algebra to algebraic geometry, to unravel a given polynomial mapping into a parsimonious structure.
ISHTEVA Mariya (Starting date : January 2013)
Structured low-rank approximation
Host: Ivan Markovsky
Low-rank approximations are widely used in data mining, machine learning, and signal processing, as a tool for dimensionality reduction, feature extraction, and classification. In system identification, signal processing, and computer algebra, in addition to having low rank, the data are often structured, e.g., having Hankel, Toeplitz, or Sylvester structure. The goal of this project is to develop methodology, theory and software for the (best) structure-preserving matrix and tensor low-(multilinear) rank approximation.
LATAIRE John (Starting date : September 2011)
Measuring and modelling weakly nonlinear, slowly time-varying dynamic systems
Host: Rik Pintelon
The framework of LTI (Linear Time Invariant) systems has been shown to provide good approximating models for a large amount of real-life dynamic systems. However, the assumption of time invariance is not always satisfied for some applications, such as systems with a varying set point (e.g. a moving robot arm) or systems with varying parameters (e.g. pitting corrosion or metal deposition). The aim of this research is to extend the well known tools for identifying linear time invariant systems to linear, slowly time-varying systems. It is investigated how insight can be gained into the frequency domain behaviour of time-varying systems. Both non-parametric and parametric identification procedures are considered.
MARCONATO Anna (Starting date : September 2009)
Application of learning-from-examples algorithms to nonlinear system identification
Host: Johan Schoukens
This research aims at studying the application of machine learning techniques to the identification of nonlinear systems. As a first example, we focus on nonlinear state space models, for which the dynamics of the underlying system are captured by means of a subspace identification approach, and the nonlinear behavior can be modeled by employing learning-from-examples algorithms such as Support Vector Machines for Regression.
SCHOUKENS Maarten
Data-driven nonlinear modeling in the presence of model errors
Host: Gerd Vandersteen
This project aims to develop a framework for the data-driven modeling of nonlinear systems in the presence of dominant model errors. Such a framework requires a paradigm shift in the data-driven nonlinear modeling community from noise-model oriented approaches to model-error dominated approaches.
STOEV Julian (Starting date: October 2014)
More information soon.
TIELS Koen
User-friendly estimation of nonlinear state-space models
Host: Johan Schoukens
Nonlinear state-space models are flexible model structures that can capture many nonlinear and dynamic system behaviors. This project aims at developing a user-friendly and modular Matlab toolbox to estimate nonlinear state-space models. I will also contribute to retrieving structure in nonlinear state-space models.