||dr.ir. X.J.A. Bombois
|Contact Hours / Week x/x/x/x:
||Exam by appointment
||Experimental modelling of dynamic systems; methodology.
Discrete-time signal- and system-analysis. Identification of transferfunctions.
Representations of linear models; black-box models.
Identification of prediction-error-methods; least squares-method.
Approximation modelling; algorithms. Experiment design and
data-analysis. Identification in time- and frequency-domain;
closed-loop identification; model validation; Matlab toolbox;
||General learning objectives
System identification deduces and subsequently validates mathematical models of real-life dynamical systems (industrial processes, mechanical servo-systems, …) based on experimental data collected from those systems. This course can be considered as a follow up of the course Sc4010 Filtering and Identification where different solutions to identify a model are presented (note nevertheless that Sc4010 is in no way a prerequisite for this course). The course Sc4110 selects two widely-used linear identification methodologies: Empirical Transfer Function Estimate (ETFE) and Prediction Error Identification (PEI) and provides the students with engineering and theoretical skills to perform the identification in a suitable way. In particular, after this course, the students are able to set up an experiment, identify a nominal model, assess the accuracy/precision of this model, and make appropriate design choices to arrive at a validated model.
Detailed learning objectives:
1) Based on time-domain input-output data collected on the true system in open loop, the student is able to deduce a frequency-domain model of a system using the ETFE identification method
2) The student is able to specify the bias and variance properties of models identified by the ETFE identification method.
3) For the ETFE identification method, the student is able to interpret the bias and variance properties of identified models, and knows how these properties can be influenced by input signal design and by applying windowing techniques.
4) The student is able to specify different linear model structures, and to characterize their computational and statistical properties in prediction error identification.
5) The student masters the statistical properties (bias, variance, consistency) of prediction error estimators both for the situation of exact plant and noise model sets, and for the situation of exact plant model sets only.
6) The student is able to specify how experiment design and signal to noise ratio affect estimated models. This includes mastering the concept of sufficiently exciting input signals, and the design of appropriate input signals.
7) The student is able to apply and interpret correlation-based model structure validation tests, and to draw conclusions on the (in)validity of model structures, distinguishing between plant models and noise models.
8) For both ETFE and PE identification methods, the student is able to appropriately acquire digital data from a real-life system (choice of sampling frequency, data processing).
Required level for the assignment
1) the student is able to explain in details the presented theory, to demonstrate important properties and to make links and comparisons between the different parts of the course
2) the student is able to use the presented tools in practice on a laboratory setup and to interpret his/her result with a critical attitude
||Lectures and project 0/0/6/0
Assignment form: final project on a laboratory setup followed by an oral examination
|Literature and Study Materials:
||lecture notes and slides
||Basics in linear algebra and signal theory
||Assignment form: final project on a laboratory setup followed by an oral or a written examination (the choice between oral and written exam will depend on the number of students)
||Course load: 14 theory courses, 3 exercise sessions and 2 computer sessions