Data-Rich Multivariable Control of Heavy-Duty Engines
(2012) In PhD Thesis TFRT-1092- Abstract
- The combustion engine is today the dominant technology for transportation of goods and people world-wide. Concerns for global warming, toxic exhaust emissions, as well as cost and availability of fuel have in recent years created incentives for technological evolution of combustion engines. More sophisticated engine instrumentation with additional degrees of freedom has been added to the engine design to reduce emissions and fuel consumption. But, as engines become more complex, the task of calibration and control becomes more challenging.
This thesis investigates approaches to utilize rich sensor information for multivariable engine control. Different combustion modes, and different combinations of sensors and... (More) - The combustion engine is today the dominant technology for transportation of goods and people world-wide. Concerns for global warming, toxic exhaust emissions, as well as cost and availability of fuel have in recent years created incentives for technological evolution of combustion engines. More sophisticated engine instrumentation with additional degrees of freedom has been added to the engine design to reduce emissions and fuel consumption. But, as engines become more complex, the task of calibration and control becomes more challenging.
This thesis investigates approaches to utilize rich sensor information for multivariable engine control. Different combustion modes, and different combinations of sensors and actuators have been studied and evaluated experimentally on a full-scale six-cylinder heavy-duty engine. The work is divided into four areas: virtual emissions sensing, dynamic emissions models, optimal engine control, and control of sensitive combustion modes. The theme of the thesis is to show how feedback control based on rich sensor information can be exploited to improve the engine operation and reduce the off-line calibration effort.
The virtual sensing work presents a data-mining method for predicting exhaust emissions from cylinder pressure data. Principal component analysis was used to reduce the dimensionality of the high-resolution data, and a neural network model was trained to predict emissions on a cycle-to-cycle, cylinder-individual basis.
The work on dynamic models investigates how system identification can be used to find multivariable dynamic models from a set of engine actuators to a set of variables related to high-level engine specifications, namely emissions, work output, combustion phasing, and peak pressure derivative. It was shown how fairly simple Wiener models can capture the main dynamics of the engine at a grid of operating points.
One of the identified multivariable models was used for optimal control of the engine. In contrast to most previous work in the field, integration of fuel- and gas-path control into a single framework was pursued. A model predictive controller was designed based on a cost function expressed in terms of high-level engine control objectives, and feedback was based on measured emissions as well as cylinder pressure data.
The final part of the thesis presents work on two sensitive combustion modes, HCCI and dual-fuel operation. Here, feedback control is necessary to achieve robust operation. For both types of combustion, it was shown how a combination of two actuators can be used to successfully control the combustion process. (Less) - Abstract (Swedish)
- Popular Abstract in Swedish
See attachment in related material
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/2439853
- author
- Henningsson, Maria LU
- supervisor
-
- Rolf Johansson LU
- Per Tunestål LU
- Bo Bernhardsson LU
- opponent
-
- Professor Glover, Keith, University of Cambridge, United Kingdom
- organization
- publishing date
- 2012
- type
- Thesis
- publication status
- published
- subject
- keywords
- Multivariable Control, Model Predictive Control, LQG Control, Virtual Sensing, System Identification, Diesel engines, HCCI engines, Dual-Fuel Combustion, Cylinder Pressure Sensors
- in
- PhD Thesis TFRT-1092
- pages
- 190 pages
- publisher
- Department of Automatic Control, Lund Institute of Technology, Lund University
- defense location
- Room M:B, M-building, Ole Römers väg 1, Lund University Faculty of Engineering
- defense date
- 2012-05-11 10:15:00
- ISSN
- 0280-5316
- 0280-5316
- project
- Diesel HCCI in a Multi-Cylinder Engine
- language
- English
- LU publication?
- yes
- id
- 8fd75c20-c3eb-4e98-a562-53a197bda47c (old id 2439853)
- date added to LUP
- 2016-04-01 12:57:19
- date last changed
- 2019-05-23 16:04:44
@phdthesis{8fd75c20-c3eb-4e98-a562-53a197bda47c,
abstract = {{The combustion engine is today the dominant technology for transportation of goods and people world-wide. Concerns for global warming, toxic exhaust emissions, as well as cost and availability of fuel have in recent years created incentives for technological evolution of combustion engines. More sophisticated engine instrumentation with additional degrees of freedom has been added to the engine design to reduce emissions and fuel consumption. But, as engines become more complex, the task of calibration and control becomes more challenging. <br/><br>
<br/><br>
This thesis investigates approaches to utilize rich sensor information for multivariable engine control. Different combustion modes, and different combinations of sensors and actuators have been studied and evaluated experimentally on a full-scale six-cylinder heavy-duty engine. The work is divided into four areas: virtual emissions sensing, dynamic emissions models, optimal engine control, and control of sensitive combustion modes. The theme of the thesis is to show how feedback control based on rich sensor information can be exploited to improve the engine operation and reduce the off-line calibration effort.<br/><br>
<br/><br>
The virtual sensing work presents a data-mining method for predicting exhaust emissions from cylinder pressure data. Principal component analysis was used to reduce the dimensionality of the high-resolution data, and a neural network model was trained to predict emissions on a cycle-to-cycle, cylinder-individual basis.<br/><br>
<br/><br>
The work on dynamic models investigates how system identification can be used to find multivariable dynamic models from a set of engine actuators to a set of variables related to high-level engine specifications, namely emissions, work output, combustion phasing, and peak pressure derivative. It was shown how fairly simple Wiener models can capture the main dynamics of the engine at a grid of operating points. <br/><br>
<br/><br>
One of the identified multivariable models was used for optimal control of the engine. In contrast to most previous work in the field, integration of fuel- and gas-path control into a single framework was pursued. A model predictive controller was designed based on a cost function expressed in terms of high-level engine control objectives, and feedback was based on measured emissions as well as cylinder pressure data.<br/><br>
<br/><br>
The final part of the thesis presents work on two sensitive combustion modes, HCCI and dual-fuel operation. Here, feedback control is necessary to achieve robust operation. For both types of combustion, it was shown how a combination of two actuators can be used to successfully control the combustion process.}},
author = {{Henningsson, Maria}},
issn = {{0280-5316}},
keywords = {{Multivariable Control; Model Predictive Control; LQG Control; Virtual Sensing; System Identification; Diesel engines; HCCI engines; Dual-Fuel Combustion; Cylinder Pressure Sensors}},
language = {{eng}},
publisher = {{Department of Automatic Control, Lund Institute of Technology, Lund University}},
school = {{Lund University}},
series = {{PhD Thesis TFRT-1092}},
title = {{Data-Rich Multivariable Control of Heavy-Duty Engines}},
url = {{https://lup.lub.lu.se/search/files/3068636/2439874.pdf}},
year = {{2012}},
}