Title: Efficient data-driven modelling of constrained nonlinear processes
Language: English
Authors: Deregnaucourt, Maxime-Vianney 
Qualification level: Doctoral
Advisor: Jakubek, Stefan 
Issue Date: 2015
Number of Pages: 82
Qualification level: Doctoral
Abstract: 
This dissertation summarizes my research activities as a PhD student at the Vienna University of Technology since 2011. My research activities originated with the collaboration between the Institute of Mechanics and Mechatronics (Division of Control and Process Automation) and AVL List GmbH and are part of the research activities of the Christian Doppler Laboratory for Model Based Calibration Methodologies. My research activities concern the efficient modelling of constrained nonlinear processes using data-based models with applications for the calibration of nonlinear systems. The calibration consists in optimizing systems by finding proper controller parameters and is confronted to the increasing complexity of the modern systems. Modern systems are constrained by legal norms or physical limits to operating ranges where the system processes must be operated optimally. For optimizing the system behaviour, modelling complex constrained nonlinear processes with high quality is therefore an important issue. For that purpose, an efficient modelling methodology relying on nonlinear system identification and design of experiments is proposed. For the nonlinear system identification, nonlinear data-driven modelling is used which consists in producing a data-based model relying on nonlinear model architecture. The main objective of nonlinear data-driven modelling is the production of a model structure and model parameters corresponding to the process using training data. The training data are generated from the system by carrying out experiments. In this context, the inputs of a system process are varied and the process output is measured. A prerequisite to data-driven modelling is that the training data describe the process behaviour characteristics. For generating training data describing the process behaviour, model-based design of experiments is used. Model-based design of experiments consists in using the model of a process for exciting the very same process so that the nonlinear process behaviour is revealed and assessed efficiently. The main challenge for model-based design of experiments is the integration of information contained in the model so that high-quality models are parameterised with the training data and process input and output constraints are complied with. In Chapter 1, the problematic, the objectives, and the mains concepts for modelling constrained nonlinear processes with data-based models are described, as well as the contributions to the selected journal publications. In Chapter 2, the selected journal publications and my own contribution are presented.
Keywords: System Identification; Design of Experiments; Nonlinear System
URI: https://doi.org/10.34726/hss.2015.28033
http://hdl.handle.net/20.500.12708/17030
DOI: 10.34726/hss.2015.28033
Library ID: AC12189363
Organisation: E325 - Institut für Mechanik und Mechatronik 
Publication Type: Thesis
Hochschulschrift
Appears in Collections:Thesis

Files in this item:

Show full item record

Page view(s)

12
checked on May 5, 2021

Download(s)

8
checked on May 5, 2021

Google ScholarTM

Check


Items in reposiTUm are protected by copyright, with all rights reserved, unless otherwise indicated.