View
1
Download
0
Category
Preview:
Citation preview
linnaeus university press
Lnu.seISBN: 978-91-88898-22-7 (print), 978-91-88898-23-4 (pdf )
Fredrik Ahlgren
Linnaeus University DissertationsNo 339/2018
Fredrik Ahlgren
Reducing ships’ fuel consumption and emissions by learning from data
Reducing ships’ fuel consum
ption and emissions by learning from
data
Fredrik Ahlgren was born on 5th November 1980, and he started his career in the Royal Swedish Navy as an engineering officer. In the navy, he sailed fast attack crafts, corvettes and submarines, which gave practical experience working with diesel engines, gas turbines and electric battery propulsion. He is also a lecturer, teaching the Marine Engineering programme at Kalmar Maritime Academy. Fredrik has a keen interest in everything concerning new technology and computers and has also been
an amateur runner for many years. Fredrik lives in Kalmar, is married to Madeleine and they have two kids, Gustav and Louise. In his spare time, when he is not with his family or running, he is always busy with hobby projects or reading books. He is also an active secular humanist and is acting as the chairman of the Kalmar Humanist society.
Reducing ships' fuel consumption and emissions by learning from data
Linnaeus University Dissertations
No 339/2018
REDUCING SHIPS' FUEL CONSUMPTION
AND EMISSIONS BY LEARNING FROM DATA
FREDRIK AHLGREN
LINNAEUS UNIVERSITY PRESS
Linnaeus University Dissertations
No 339/2018
REDUCING SHIPS' FUEL CONSUMPTION
AND EMISSIONS BY LEARNING FROM DATA
FREDRIK AHLGREN
LINNAEUS UNIVERSITY PRESS
Abstract Ahlgren, Fredrik (2018). Reducing ships' fuel consumption and emissions by learning from data, Linnaeus University Dissertations No 339/2018, ISBN: 978-91-88898-22-7 (print), 978-91-88898-23-4 (pdf). Written in English. In the context of reducing both greenhouse gases and hazardous emissions, the shipping sector faces a major challenge as it is currently responsible for 11% of the transport sector’s anthropogenic greenhouse gas emissions. Even as emissions reductions are needed, the demand for the transport sector rises exponentially every year. This thesis aims to investigate the potential to use ships’ existing internal energy systems more efficiently. The thesis focusses on making existing ships in real operating conditions more efficient based logged machinery data. This dissertation presents results that can make ship more energy efficient by utilising waste heat recovery and machine learning tools. A significant part of this thesis is based on data from a cruise ship in the Baltic Sea, and an extensive analysis of the ship’s internal energy system was made from over a year’s worth of data. The analysis included an exergy analysis, which also considers the usability of each energy flow. In three studies, the feasibility of using the waste heat from the engines was investigated, and the results indicate that significant measures can be undertaken with organic Rankine cycle devices. The organic Rankine cycle was simulated with data from the ship operations and optimised for off-design conditions, both regarding system design and organic fluid selection. The analysis demonstrates that there are considerable differences between the real operation of a ship and what it was initially designed for. In addition, a large two-stroke marine diesel was integrated into a simulation with an organic Rankine cycle, resulting in an energy efficiency improvement of 5%. This thesis also presents new methods of employing machine learning to predict energy consumption. Machine learning algorithms are readily available and free to use, and by using only a small subset of data points from the engines and existing fuel flow meters, the fuel consumption could be predicted with good accuracy. These results demonstrate a potential to improve operational efficiency without installing additional fuel meters. The thesis presents results concerning how data from ships can be used to further analyse and improve their efficiency, by using both add-on technologies for waste heat recovery and machine learning applications.
Reducing ships' fuel consumption and emissions by learning from data Doctoral Dissertation, Kalmar Maritime Academy, Linnaeus University, Kalmar, 2018 Omslagsbild: Gunilla Hägglund Johnson ISBN: 978-91-88898-22-7 (print), 978-91-88898-23-4 (pdf) Published by: Linnaeus University Press, 351 95 Växjö Printed by: DanagårdLiTHO, 2018
Abstract Ahlgren, Fredrik (2018). Reducing ships' fuel consumption and emissions by learning from data, Linnaeus University Dissertations No 339/2018, ISBN: 978-91-88898-22-7 (print), 978-91-88898-23-4 (pdf). Written in English. In the context of reducing both greenhouse gases and hazardous emissions, the shipping sector faces a major challenge as it is currently responsible for 11% of the transport sector’s anthropogenic greenhouse gas emissions. Even as emissions reductions are needed, the demand for the transport sector rises exponentially every year. This thesis aims to investigate the potential to use ships’ existing internal energy systems more efficiently. The thesis focusses on making existing ships in real operating conditions more efficient based logged machinery data. This dissertation presents results that can make ship more energy efficient by utilising waste heat recovery and machine learning tools. A significant part of this thesis is based on data from a cruise ship in the Baltic Sea, and an extensive analysis of the ship’s internal energy system was made from over a year’s worth of data. The analysis included an exergy analysis, which also considers the usability of each energy flow. In three studies, the feasibility of using the waste heat from the engines was investigated, and the results indicate that significant measures can be undertaken with organic Rankine cycle devices. The organic Rankine cycle was simulated with data from the ship operations and optimised for off-design conditions, both regarding system design and organic fluid selection. The analysis demonstrates that there are considerable differences between the real operation of a ship and what it was initially designed for. In addition, a large two-stroke marine diesel was integrated into a simulation with an organic Rankine cycle, resulting in an energy efficiency improvement of 5%. This thesis also presents new methods of employing machine learning to predict energy consumption. Machine learning algorithms are readily available and free to use, and by using only a small subset of data points from the engines and existing fuel flow meters, the fuel consumption could be predicted with good accuracy. These results demonstrate a potential to improve operational efficiency without installing additional fuel meters. The thesis presents results concerning how data from ships can be used to further analyse and improve their efficiency, by using both add-on technologies for waste heat recovery and machine learning applications.
Reducing ships' fuel consumption and emissions by learning from data Doctoral Dissertation, Kalmar Maritime Academy, Linnaeus University, Kalmar, 2018 Omslagsbild: Gunilla Hägglund Johnson ISBN: 978-91-88898-22-7 (print), 978-91-88898-23-4 (pdf) Published by: Linnaeus University Press, 351 95 Växjö Printed by: DanagårdLiTHO, 2018
"Nobody ever figures out what life is all about,and it doesn’t matter. Explore the world.
Nearly everything is really interestingif you go into it deeply enough."
— Richard Feynman
"Nobody ever figures out what life is all about,and it doesn’t matter. Explore the world.
Nearly everything is really interestingif you go into it deeply enough."
— Richard Feynman
Contents
List of publications . . . . . . . . . . . . . . . . . . . . . . . . . vAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . viiPopular summary in English . . . . . . . . . . . . . . . . . . . . viiiPopulärvetenskaplig sammanfattning på svenska . . . . . . . . . . xPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv
1 Introduction 11.1 Purpose and methodology . . . . . . . . . . . . . . . . . . 11.2 Research boundaries . . . . . . . . . . . . . . . . . . . . . 21.3 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 The need for change 32.1 The Atmosphere and fossil fuels . . . . . . . . . . . . . . . 32.2 Emissions from the Maritime Sector . . . . . . . . . . . . . 62.3 Environmental maritime legislation . . . . . . . . . . . . . 82.4 Energy efficiency and measures . . . . . . . . . . . . . . . . 9
3 The story behind the results 113.1 Generalisations from a case study . . . . . . . . . . . . . . 113.2 Description of the ship: M/S Birka Stockholm . . . . . . . . 123.3 Data pre-processing . . . . . . . . . . . . . . . . . . . . . . 15
4 Tools of the trade 194.1 Simulation of physical systems . . . . . . . . . . . . . . . . 194.2 The simulation software IPSEPro . . . . . . . . . . . . . . . 204.3 Dynamic or steady-state . . . . . . . . . . . . . . . . . . . . 214.4 Scientific programming by Python . . . . . . . . . . . . . . 22
5 What it is all about: Energy 235.1 A well known but abstract concept: Energy . . . . . . . . . 245.2 First Law - Energy conservation . . . . . . . . . . . . . . . 245.3 The Ship as a closed system . . . . . . . . . . . . . . . . . 255.4 Where is the waste heat coming from? . . . . . . . . . . . . 25
Contents
List of publications . . . . . . . . . . . . . . . . . . . . . . . . . vAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . viiPopular summary in English . . . . . . . . . . . . . . . . . . . . viiiPopulärvetenskaplig sammanfattning på svenska . . . . . . . . . . xPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv
1 Introduction 11.1 Purpose and methodology . . . . . . . . . . . . . . . . . . 11.2 Research boundaries . . . . . . . . . . . . . . . . . . . . . 21.3 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 The need for change 32.1 The Atmosphere and fossil fuels . . . . . . . . . . . . . . . 32.2 Emissions from the Maritime Sector . . . . . . . . . . . . . 62.3 Environmental maritime legislation . . . . . . . . . . . . . 82.4 Energy efficiency and measures . . . . . . . . . . . . . . . . 9
3 The story behind the results 113.1 Generalisations from a case study . . . . . . . . . . . . . . 113.2 Description of the ship: M/S Birka Stockholm . . . . . . . . 123.3 Data pre-processing . . . . . . . . . . . . . . . . . . . . . . 15
4 Tools of the trade 194.1 Simulation of physical systems . . . . . . . . . . . . . . . . 194.2 The simulation software IPSEPro . . . . . . . . . . . . . . . 204.3 Dynamic or steady-state . . . . . . . . . . . . . . . . . . . . 214.4 Scientific programming by Python . . . . . . . . . . . . . . 22
5 What it is all about: Energy 235.1 A well known but abstract concept: Energy . . . . . . . . . 245.2 First Law - Energy conservation . . . . . . . . . . . . . . . 245.3 The Ship as a closed system . . . . . . . . . . . . . . . . . 255.4 Where is the waste heat coming from? . . . . . . . . . . . . 25
List of Figures
2.1 Global shares of anthropogenic GHG emissions by sector [2] 42.2 CO2 emissions by sector [2] . . . . . . . . . . . . . . . . . 52.3 World primary energy supply [2] . . . . . . . . . . . . . . . 52.4 Trend in CO2 emissions 1870-2014 [2] . . . . . . . . . . . 62.5 Trends in CO2 emissions for the transport sector 1990-2015 [2] 72.6 CO2 emission share for the transport sector 2015 [2] . . . . 7
3.1 M/S Birka Stockholm. Photo: Kjell Larsson . . . . . . . . 123.2 Distribution of ship speed for one year of operation, at least
one main engine running. . . . . . . . . . . . . . . . . . . 143.3 Propulsion power and ship speed . . . . . . . . . . . . . . . 153.4 Test protocol data ME/AE Wärtsilä. Engine 6LB46B no.
91541 and W6L32 no. 22191 . . . . . . . . . . . . . . . . 163.5 Yearly load distribution Main Engines M/S Birka Stockholm 173.6 Yearly load distribution Auxiliary Engines M/S Birka Stock-
holm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.1 IPSE model used in Paper IV . . . . . . . . . . . . . . . . 21
5.1 Evolution of low speed engines [38] . . . . . . . . . . . . . 275.2 Typical Sankey-diagram of a turbocharged engine [38] . . . 285.3 Sankey diagram of the energy flows in M/S Birka, Paper VI.
Flow values are in GWh/year. . . . . . . . . . . . . . . . . 295.4 Sankey diagram for a MAN 12K98ME/MC engine . . . . . 305.5 Sankey diagram for a MAN 12S90ME engine with and without
WHR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315.6 Temperature entropy diagram of Rankine Cycle (CC-BY
Marcus Thern). . . . . . . . . . . . . . . . . . . . . . . . . 335.7 Boiler process, partial steam or once-through [41]. . . . . . 345.8 Dry fluid, entropy temperature diagram of Isopentane . . . . 36
iii
5.5 Fuel to power - The Diesel engine . . . . . . . . . . . . . . 265.6 Heat to power - The Rankine cycle . . . . . . . . . . . . . . 325.7 Low temperature heat to power - The organic Rankine cycle 355.8 A useful concept of measuring work: Exergy . . . . . . . . 39
6 Learning from data 436.1 Machine learning and artificial intelligence . . . . . . . . . 446.2 Making predictions from data . . . . . . . . . . . . . . . . . 446.3 Basic machine learning concepts . . . . . . . . . . . . . . . 456.4 Choosing the right algorithm . . . . . . . . . . . . . . . . . 466.5 Auto machine learning . . . . . . . . . . . . . . . . . . . . 466.6 Predicting the energy consumption . . . . . . . . . . . . . . 47
7 The impact and context 517.1 Waste heat recovery feasibility . . . . . . . . . . . . . . . . 517.2 The exergy destruction . . . . . . . . . . . . . . . . . . . . 527.3 Measuring energy with machine learning . . . . . . . . . . 537.4 Machine learning applications . . . . . . . . . . . . . . . . 547.5 Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
8 Concluding remarks 578.1 Future research . . . . . . . . . . . . . . . . . . . . . . . . 58
9 Summary of the papers 599.1 Waste Heat Recovery in a Cruise Vessel in the Baltic Sea by
Using and Organic Rankine Cycle: A Case Study . . . . . . 599.2 Auto Machine Learning for predicting Ship Fuel Consumption 599.3 Predicting Dynamic Fuel Oil Consumption using Automated
Machine Learning from Large Time Interval Fuel sums . . . 609.4 Energy integration of Organic Rankine Cycle, Exhaust Gas
recirculation and Scrubber . . . . . . . . . . . . . . . . . . 619.5 Quasi-steady state simulation of an organic Rankine cycle for
waste heat recovery in a passenger vessel . . . . . . . . . . . 619.6 Energy and exergy analysis of a cruise ship . . . . . . . . . 62
List of Figures
2.1 Global shares of anthropogenic GHG emissions by sector [2] 42.2 CO2 emissions by sector [2] . . . . . . . . . . . . . . . . . 52.3 World primary energy supply [2] . . . . . . . . . . . . . . . 52.4 Trend in CO2 emissions 1870-2014 [2] . . . . . . . . . . . 62.5 Trends in CO2 emissions for the transport sector 1990-2015 [2] 72.6 CO2 emission share for the transport sector 2015 [2] . . . . 7
3.1 M/S Birka Stockholm. Photo: Kjell Larsson . . . . . . . . 123.2 Distribution of ship speed for one year of operation, at least
one main engine running. . . . . . . . . . . . . . . . . . . 143.3 Propulsion power and ship speed . . . . . . . . . . . . . . . 153.4 Test protocol data ME/AE Wärtsilä. Engine 6LB46B no.
91541 and W6L32 no. 22191 . . . . . . . . . . . . . . . . 163.5 Yearly load distribution Main Engines M/S Birka Stockholm 173.6 Yearly load distribution Auxiliary Engines M/S Birka Stock-
holm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.1 IPSE model used in Paper IV . . . . . . . . . . . . . . . . 21
5.1 Evolution of low speed engines [38] . . . . . . . . . . . . . 275.2 Typical Sankey-diagram of a turbocharged engine [38] . . . 285.3 Sankey diagram of the energy flows in M/S Birka, Paper VI.
Flow values are in GWh/year. . . . . . . . . . . . . . . . . 295.4 Sankey diagram for a MAN 12K98ME/MC engine . . . . . 305.5 Sankey diagram for a MAN 12S90ME engine with and without
WHR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315.6 Temperature entropy diagram of Rankine Cycle (CC-BY
Marcus Thern). . . . . . . . . . . . . . . . . . . . . . . . . 335.7 Boiler process, partial steam or once-through [41]. . . . . . 345.8 Dry fluid, entropy temperature diagram of Isopentane . . . . 36
iii
5.5 Fuel to power - The Diesel engine . . . . . . . . . . . . . . 265.6 Heat to power - The Rankine cycle . . . . . . . . . . . . . . 325.7 Low temperature heat to power - The organic Rankine cycle 355.8 A useful concept of measuring work: Exergy . . . . . . . . 39
6 Learning from data 436.1 Machine learning and artificial intelligence . . . . . . . . . 446.2 Making predictions from data . . . . . . . . . . . . . . . . . 446.3 Basic machine learning concepts . . . . . . . . . . . . . . . 456.4 Choosing the right algorithm . . . . . . . . . . . . . . . . . 466.5 Auto machine learning . . . . . . . . . . . . . . . . . . . . 466.6 Predicting the energy consumption . . . . . . . . . . . . . . 47
7 The impact and context 517.1 Waste heat recovery feasibility . . . . . . . . . . . . . . . . 517.2 The exergy destruction . . . . . . . . . . . . . . . . . . . . 527.3 Measuring energy with machine learning . . . . . . . . . . 537.4 Machine learning applications . . . . . . . . . . . . . . . . 547.5 Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
8 Concluding remarks 578.1 Future research . . . . . . . . . . . . . . . . . . . . . . . . 58
9 Summary of the papers 599.1 Waste Heat Recovery in a Cruise Vessel in the Baltic Sea by
Using and Organic Rankine Cycle: A Case Study . . . . . . 599.2 Auto Machine Learning for predicting Ship Fuel Consumption 599.3 Predicting Dynamic Fuel Oil Consumption using Automated
Machine Learning from Large Time Interval Fuel sums . . . 609.4 Energy integration of Organic Rankine Cycle, Exhaust Gas
recirculation and Scrubber . . . . . . . . . . . . . . . . . . 619.5 Quasi-steady state simulation of an organic Rankine cycle for
waste heat recovery in a passenger vessel . . . . . . . . . . . 619.6 Energy and exergy analysis of a cruise ship . . . . . . . . . 62
List of publications
This thesis is based on the following publications, referred to by their Romannumerals:
i Waste Heat Recovery in a Cruise Vessel in the Baltic Sea by Usingan Organic Rankine Cycle: A Case StudyF. Ahlgren, M. Mondejar, M. Genrup, M. ThernJournal of Engineering for Gas Turbines and Power 2015;138
ii Auto Machine Learning for predicting Ship Fuel ConsumptionF. Ahlgren, M. ThernProceedings of ECOS 2018 - The 31st International Conference onEfficiency, Cost, Optimization, Simulation and Environmental impactof Energy Systems
iii Predicting Dynamic Fuel Oil Consumption on Ships with Auto-mated Machine LearningF. Ahlgren, M. Mondejar, M. Thern10th International Conference on Applied Energy (ICAE2018), 22-25August 2018, Hong Kong, China
iv Energy integration of Organic Rankine Cycle, Exhaust Gas recir-culation and ScrubberF. Ahlgren, M. Thern, M. Genrup, M. MondejarBook Chapter. Trends and Challenges in Maritime Energy Manage-ment, vol. 6, Springer International Publishing; 2018, p. 157–68.
v Quasi-steady state simulation of an organic Rankine cycle forwaste heat recovery in a passenger vesselM.E. Mondejar, F. Ahlgren, M. Thern, M. GenrupApplied Energy 2017;185
vi Energy and Exergy Analysis of a Cruise ShipF. Baldi, F. Ahlgren, T. van Nguyen, M. Thern, K. AnderssonEnergies 2018;11
All papers are reproduced with the permission of their respective publishers.
v
5.9 Net power output versus vessel speed for the three most optimalfluids (s - Simple ORC, R - Regenerated, av - averaged overoperating time) [23] . . . . . . . . . . . . . . . . . . . . . 37
5.10 Temperature enthalpy diagram from ORC integration with amarine diesel engine [27] . . . . . . . . . . . . . . . . . . 38
5.11 Grassman diagram of M/S Birka, Paper VI. Flow values arein GWh/year. . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.1 Scikit-learn algorithm cheat-sheet [54]. . . . . . . . . . . . 496.2 Dynamic fuel oil consumption, SVR-algorithm 96h sum
average . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.1 IFO380 and MDO prices, adopted from Ship and Bunker [72] 56
iv
List of publications
This thesis is based on the following publications, referred to by their Romannumerals:
i Waste Heat Recovery in a Cruise Vessel in the Baltic Sea by Usingan Organic Rankine Cycle: A Case StudyF. Ahlgren, M. Mondejar, M. Genrup, M. ThernJournal of Engineering for Gas Turbines and Power 2015;138
ii Auto Machine Learning for predicting Ship Fuel ConsumptionF. Ahlgren, M. ThernProceedings of ECOS 2018 - The 31st International Conference onEfficiency, Cost, Optimization, Simulation and Environmental impactof Energy Systems
iii Predicting Dynamic Fuel Oil Consumption on Ships with Auto-mated Machine LearningF. Ahlgren, M. Mondejar, M. Thern10th International Conference on Applied Energy (ICAE2018), 22-25August 2018, Hong Kong, China
iv Energy integration of Organic Rankine Cycle, Exhaust Gas recir-culation and ScrubberF. Ahlgren, M. Thern, M. Genrup, M. MondejarBook Chapter. Trends and Challenges in Maritime Energy Manage-ment, vol. 6, Springer International Publishing; 2018, p. 157–68.
v Quasi-steady state simulation of an organic Rankine cycle forwaste heat recovery in a passenger vesselM.E. Mondejar, F. Ahlgren, M. Thern, M. GenrupApplied Energy 2017;185
vi Energy and Exergy Analysis of a Cruise ShipF. Baldi, F. Ahlgren, T. van Nguyen, M. Thern, K. AnderssonEnergies 2018;11
All papers are reproduced with the permission of their respective publishers.
v
5.9 Net power output versus vessel speed for the three most optimalfluids (s - Simple ORC, R - Regenerated, av - averaged overoperating time) [23] . . . . . . . . . . . . . . . . . . . . . 37
5.10 Temperature enthalpy diagram from ORC integration with amarine diesel engine [27] . . . . . . . . . . . . . . . . . . 38
5.11 Grassman diagram of M/S Birka, Paper VI. Flow values arein GWh/year. . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.1 Scikit-learn algorithm cheat-sheet [54]. . . . . . . . . . . . 496.2 Dynamic fuel oil consumption, SVR-algorithm 96h sum
average . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.1 IFO380 and MDO prices, adopted from Ship and Bunker [72] 56
iv
Acknowledgements
I acknowledge the engine crew of M/S Birka Stockholm for their openness andsupport and Rederiaktiebolaget Eckerö for sharing data. I also acknowledge theSwedish Maritime Administration and Linnaeus University for their financialsupport.
vii
The publications not included in this thesis.
Optimal load allocation of complex ship power plants
F. Baldi, F. Ahlgren, M. Francesco, C. Gabrielii, K. AnderssonEnergy Conversion and Management 2016;124
The application of process integration to the optimisation ofcruise ship energy systems: a case studyF. Baldi, T-V. Nguyen F. AhlgrenProceedings of ECOS 2016 - The 29th International Conference onEfficiency, Cost, Optimization, Simulation and Environmental impactof Energy SystemsA social sustainability perspective on an environmental interven-tion to reduce ship emissionsF. Ahlgren, C. Österman47th Nordic Ergonomics Society annual conference Creating Sustain-able Work Environments, 2015Waste heat recovery in a cruise vessel in the Baltic Sea by usingan organic Rankine cycle: a case studyF. Ahlgren, M.E. Mondejar, M. Genrup, M. ThernASME Turbo Expo 2015: Turbine Technical Conference and Exposi-tion, 2015Study of the on-route operation of a waste heat recovery systemin a passenger vesselM.E. Mondejar, F. Ahlgren, M. Thern, M. GenrupThe 7th International Conference on Applied Energy (ICAE2015),Clean, Efficient and Affordable Energy for a Sustainable Future, 2015
vi
Acknowledgements
I acknowledge the engine crew of M/S Birka Stockholm for their openness andsupport and Rederiaktiebolaget Eckerö for sharing data. I also acknowledge theSwedish Maritime Administration and Linnaeus University for their financialsupport.
vii
The publications not included in this thesis.
Optimal load allocation of complex ship power plants
F. Baldi, F. Ahlgren, M. Francesco, C. Gabrielii, K. AnderssonEnergy Conversion and Management 2016;124
The application of process integration to the optimisation ofcruise ship energy systems: a case studyF. Baldi, T-V. Nguyen F. AhlgrenProceedings of ECOS 2016 - The 29th International Conference onEfficiency, Cost, Optimization, Simulation and Environmental impactof Energy SystemsA social sustainability perspective on an environmental interven-tion to reduce ship emissionsF. Ahlgren, C. Österman47th Nordic Ergonomics Society annual conference Creating Sustain-able Work Environments, 2015Waste heat recovery in a cruise vessel in the Baltic Sea by usingan organic Rankine cycle: a case studyF. Ahlgren, M.E. Mondejar, M. Genrup, M. ThernASME Turbo Expo 2015: Turbine Technical Conference and Exposi-tion, 2015Study of the on-route operation of a waste heat recovery systemin a passenger vesselM.E. Mondejar, F. Ahlgren, M. Thern, M. GenrupThe 7th International Conference on Applied Energy (ICAE2015),Clean, Efficient and Affordable Energy for a Sustainable Future, 2015
vi
fuel consumption, which means lower emissions of carbon dioxide and otherharmful gases such as sulphur and nitrogen oxides. The work presentedin this dissertation has been carried out both on board a vessel and withcomputer-assisted simulations. The vessel’s data was analysed by statisticalmethods for mapping and summing all energy flows also in the amount ofpotential each flow has to be used.
ix
Popular summary in English
Ship data and waste heat can reduce the environmental impact of ship-ping
The shipping industry faces significant challenges in reducing its emissions.Transport is increasing and currently, dirty oil is primarily used, releasingemissions harmful to both nature and people. New requirements mean thatemissions must be drastically reduced, which can be accomplished by makingvessels more efficient.
This dissertation presents results that demonstrate how vessels can be mademore energy efficient. By utilising the heat available in the engine exhaust andmore efficient means of measuring energy, emissions can be reduced by up to5%. Exhaust gases from the engines are hot, and this heat can be used to makeelectricity. The electricity thus generated results in a ship’s diesel generatorssaving fuel. The dissertation presents results which highlight techniques ofgreat potential on board ships.
The research has been based on data coming from ships in real-time operationand has found a significant difference between the way a ship drives in realityand what it was initially built to handle. This fact shows the importance ofusing realistic data. The challenge has been to find the best method for makinga real ship more energy efficient.
A comprehensive analysis of the entire energy system aboard a cruise shiphas been conducted, mapping all energy flows to result in better knowledgeof how to reduce energy losses. The analysis also identified the amount ofenergy that is actually useful for ships.
To drive more efficiently, the crew must know the ship’s current level offuel consumption, which can vary widely depending on how the ship isrunning. New methods have been developed to show the current level offuel consumption without installing additional measurement sensors, whichcan include using computer models for machine learning to calculate fuelconsumption, which provides better support for the crew. The results showthat machine learning is an efficient and cost-effective tool in energy efficiencyas the cost is lower than traditional methods such as installing more fuelmeters.
The results mean that the shipping industry will gain new means of reducing
viii
fuel consumption, which means lower emissions of carbon dioxide and otherharmful gases such as sulphur and nitrogen oxides. The work presentedin this dissertation has been carried out both on board a vessel and withcomputer-assisted simulations. The vessel’s data was analysed by statisticalmethods for mapping and summing all energy flows also in the amount ofpotential each flow has to be used.
ix
Popular summary in English
Ship data and waste heat can reduce the environmental impact of ship-ping
The shipping industry faces significant challenges in reducing its emissions.Transport is increasing and currently, dirty oil is primarily used, releasingemissions harmful to both nature and people. New requirements mean thatemissions must be drastically reduced, which can be accomplished by makingvessels more efficient.
This dissertation presents results that demonstrate how vessels can be mademore energy efficient. By utilising the heat available in the engine exhaust andmore efficient means of measuring energy, emissions can be reduced by up to5%. Exhaust gases from the engines are hot, and this heat can be used to makeelectricity. The electricity thus generated results in a ship’s diesel generatorssaving fuel. The dissertation presents results which highlight techniques ofgreat potential on board ships.
The research has been based on data coming from ships in real-time operationand has found a significant difference between the way a ship drives in realityand what it was initially built to handle. This fact shows the importance ofusing realistic data. The challenge has been to find the best method for makinga real ship more energy efficient.
A comprehensive analysis of the entire energy system aboard a cruise shiphas been conducted, mapping all energy flows to result in better knowledgeof how to reduce energy losses. The analysis also identified the amount ofenergy that is actually useful for ships.
To drive more efficiently, the crew must know the ship’s current level offuel consumption, which can vary widely depending on how the ship isrunning. New methods have been developed to show the current level offuel consumption without installing additional measurement sensors, whichcan include using computer models for machine learning to calculate fuelconsumption, which provides better support for the crew. The results showthat machine learning is an efficient and cost-effective tool in energy efficiencyas the cost is lower than traditional methods such as installing more fuelmeters.
The results mean that the shipping industry will gain new means of reducing
viii
metoder för att kartlägga och summera alla energiflöden också sett i hur storpotential varje flöde har att användas.
xi
Populärvetenskaplig sammanfattning på svenska
Fartygsdata och spillvärme kan minska sjöfartens miljöpåverkan
Sjöfarten står inför en stor utmaning att minska sina utsläpp. Transporternaökar och idag används till största del smutsig tjockolja som ger skadligautsläpp för både natur och människor. Nya krav innebär att utsläppen drastisktmåste minska och en metod är att göra fartygen mer effektiva.
I denna avhandlingen presenteras resultat som kan göra fartygen mer ener-gieffektiva. Genom att ta tillvara på värmen som finns i motorns avgaseroch en effektivare mätning av energin kan utsläppen minska med upp till5 %. Avgaserna från motorerna är varma och denna värme kan användasför att göra elektricitet. Elen som genereras från avgaserna gör att fartygetsdieselgeneratorer sparar bränsle. I avhandlingen presenteras resultat som visarpå en stor potential för dessa tekniker ombord på fartyg.
Forskningen har baserats på data som kommit från fartyg i verklig drift. Dethar visat sig skilja mycket mellan hur ett fartyg verkligen kör och vad det frånbörjan vad byggt att klara. Utmaningen har då varit att kunna hitta den bästametoden för att göra ett verkligt fartyg mer energieffektivt.
I arbetet har det gjorts en omfattande analys av hela energisystemet ombord påett kryssningsfartyg. Kartläggningen av alla energiflöden har inneburit bättrekunskap för att kunna minska på energiförlusterna. Analysen innehöll ocksåhur stor del av energin som faktiskt är användbar.
Det kan skilja mycket i bränsleförbrukning beroende på hur fartyget körsoch för att kunna köra effektivare behöver besättningen veta den aktuellabränsleförbrukningen. I arbetet har det utvecklats nya metoder för att kunnavisa aktuell bränsleförbrukning utan att installera extra mätsensorer. Genomatt använda datormodeller för maskininlärning har bränsleförbrukningen kun-nat beräknas vilket ger ett bättre stöd för besättningen. Resultaten visar attmaskininlärning är ett effektivt och kostnadseffektivt verktyg i energieffektivi-seringen. Kostnaden är lägre än traditionella metoder såsom att installera flerbränslemätare.
Resultaten innebär att sjöfarten får nya metoder att minska bränsleförbrukning-en vilket innebär lägre utsläpp av koldioxid och andra skadliga utsläpp såsomsvavel och kväveoxider. Arbetet har bedrivits både ombord på fartyget samtmed datorstödda simuleringar. Fartygets data analyserades med statistiska
x
metoder för att kartlägga och summera alla energiflöden också sett i hur storpotential varje flöde har att användas.
xi
Populärvetenskaplig sammanfattning på svenska
Fartygsdata och spillvärme kan minska sjöfartens miljöpåverkan
Sjöfarten står inför en stor utmaning att minska sina utsläpp. Transporternaökar och idag används till största del smutsig tjockolja som ger skadligautsläpp för både natur och människor. Nya krav innebär att utsläppen drastisktmåste minska och en metod är att göra fartygen mer effektiva.
I denna avhandlingen presenteras resultat som kan göra fartygen mer ener-gieffektiva. Genom att ta tillvara på värmen som finns i motorns avgaseroch en effektivare mätning av energin kan utsläppen minska med upp till5 %. Avgaserna från motorerna är varma och denna värme kan användasför att göra elektricitet. Elen som genereras från avgaserna gör att fartygetsdieselgeneratorer sparar bränsle. I avhandlingen presenteras resultat som visarpå en stor potential för dessa tekniker ombord på fartyg.
Forskningen har baserats på data som kommit från fartyg i verklig drift. Dethar visat sig skilja mycket mellan hur ett fartyg verkligen kör och vad det frånbörjan vad byggt att klara. Utmaningen har då varit att kunna hitta den bästametoden för att göra ett verkligt fartyg mer energieffektivt.
I arbetet har det gjorts en omfattande analys av hela energisystemet ombord påett kryssningsfartyg. Kartläggningen av alla energiflöden har inneburit bättrekunskap för att kunna minska på energiförlusterna. Analysen innehöll ocksåhur stor del av energin som faktiskt är användbar.
Det kan skilja mycket i bränsleförbrukning beroende på hur fartyget körsoch för att kunna köra effektivare behöver besättningen veta den aktuellabränsleförbrukningen. I arbetet har det utvecklats nya metoder för att kunnavisa aktuell bränsleförbrukning utan att installera extra mätsensorer. Genomatt använda datormodeller för maskininlärning har bränsleförbrukningen kun-nat beräknas vilket ger ett bättre stöd för besättningen. Resultaten visar attmaskininlärning är ett effektivt och kostnadseffektivt verktyg i energieffektivi-seringen. Kostnaden är lägre än traditionella metoder såsom att installera flerbränslemätare.
Resultaten innebär att sjöfarten får nya metoder att minska bränsleförbrukning-en vilket innebär lägre utsläpp av koldioxid och andra skadliga utsläpp såsomsvavel och kväveoxider. Arbetet har bedrivits både ombord på fartyget samtmed datorstödda simuleringar. Fartygets data analyserades med statistiska
x
Nomenclature
BMEP Brake Mean Effective Pressure
ECA Emission Control Area
EEDI Energy Efficiency Design Index
GHG Greenhouse Gases
GT Gross tonne
GWP Global warming potential
HFO Heavy Fuel Oil
IPCC Intergovernmental Panel on Climate Change
LNG Liquefied natural gas
MARPOL The International Convention for the Prevention of Pollutionfrom Ships
MEPC Marine Environmental Protection Committee
ML Machine Learning
ORC Organic Rankine cycle
PM Particulate matter
SECA Sulphur Emission Control Area
SFOC Specific Fuel Oil Consumption
WHR Waste Heat Recovery
xiii
Nomenclature
BMEP Brake Mean Effective Pressure
ECA Emission Control Area
EEDI Energy Efficiency Design Index
GHG Greenhouse Gases
GT Gross tonne
GWP Global warming potential
HFO Heavy Fuel Oil
IPCC Intergovernmental Panel on Climate Change
LNG Liquefied natural gas
MARPOL The International Convention for the Prevention of Pollutionfrom Ships
MEPC Marine Environmental Protection Committee
ML Machine Learning
ORC Organic Rankine cycle
PM Particulate matter
SECA Sulphur Emission Control Area
SFOC Specific Fuel Oil Consumption
WHR Waste Heat Recovery
xiii
Hult and Kjell Larsson. We might be a small and diverse research group,but the discussions we have are often engaging. I would like to mention JanSnöberg, who created the PhD programme at Kalmar Maritime Academy.
A special warm thanks to all my supervisors, Marcus Thern, Maria E. Mondejar,Cecilia Österman and Ann-Charlotte Larsson, who have all supported me invarious ways during these years. Even though none of you have shared thesame campus, it has worked out well, and the meetings in Lund have alwaysbeen productive. Marcus has always pushed me in the right direction, andwith great flexibility and openness to change plans. Maria and Cecilia I havedepended on for honest and though feedback which has always pushed me todo better.
A special thanks also to Magnus Genrup, who has acted as one of the influentialpeople who motivated my PhD studies. Magnus also played a crucial role instarting the PhD-programme in Kalmar, and it was Magnus encouraged mademe continue my PhD studies when I got an exciting job offer following myLicentiate degree. Without him, I would likely never have finished my PhD.
This thesis would never be possible without support from the many people Ihave met in research group meetings at DTU, in Chalmers and Lund and atconferences and workshops; even if your name is not listed here, you are notforgotten.
And last but not least, my wife Madeleine has been my greatest supporter,and none of this would have been possible without you. I would also like tomention our two children, Gustav and Louise, who have grown up with theirfather always doing something with eyes glued to a computer screen. Despiteall that, I have had five fantastic years, and even though it is often challengingand time-consuming, I would do it all over again in a heartbeat.
xv
Preface
My work began in 2013 when I decided to study how to make ships more energyefficient under operational conditions. Many studies do not consider realdata from ships in operation, often focussing instead on specific technologieswhich are based on data from the manufacturer or on listed ship data. A ship’soperational profile can be vastly different from what it was designed and builtfor, and it is therefore essential to investigate how to make ships more energyefficient under operational conditions.
This thesis describes the academic journey of a Marine Engineer who startedhis career as a Navy Engineering Officer and then used that background in theNavy as an added advantage to pursue a PhD, bringing knowledge of actuallybeing on board a ship and near the real operation and engine crew. Thismarine engineering background has been shown as a significant advantage,bringing prior operational knowledge of an engine room and being trusted asa ’kind of like’ when engaging in discussions with the crew. I have managedto tie together onboard operational experiences with academic theories incollaboration with several universities. To create knowledge and explain theworld, evidence is needed which can be found as onboard logged data as wellas crew experiences. The ship M/S Birka Stockholm has been a valuable datasource, and because of numerous collaborations with several universities andmany good ideas, this data source has resulted in many published papers.
The work in this thesis was performed in collaboration with the KalmarMaritime Academy Linnaeus University, researchers at Lund TechnicalUniversity, Chalmers Technical University and the Technical University ofDenmark. I would like to thank Rederiaktiebolaget Eckerö for their supportand especially the engine crew of M/S Birka Stockholm. The project wasfunded by the Swedish Maritime Administration and Linnaeus University.
I must mention the fantastic collaborations with my two of my fellow co-authors, Francesco Baldi and Tuong-Van Nguyen. I met Francesco at thebeginning of my PhD when he was halfway finished and presenting hisLicentiate thesis. That meeting led to a successful collaboration whichcontinues to the present. Through him, I have also gained a good friend andhiking companion.
I also want to mention my colleagues Mats Hammander, Fredrik Hjorth, JohnOhlson, Pär Karlsson, Magnus Boström, Gesa Praetorius, Carl Sandberg, Carl
xiv
Hult and Kjell Larsson. We might be a small and diverse research group,but the discussions we have are often engaging. I would like to mention JanSnöberg, who created the PhD programme at Kalmar Maritime Academy.
A special warm thanks to all my supervisors, Marcus Thern, Maria E. Mondejar,Cecilia Österman and Ann-Charlotte Larsson, who have all supported me invarious ways during these years. Even though none of you have shared thesame campus, it has worked out well, and the meetings in Lund have alwaysbeen productive. Marcus has always pushed me in the right direction, andwith great flexibility and openness to change plans. Maria and Cecilia I havedepended on for honest and though feedback which has always pushed me todo better.
A special thanks also to Magnus Genrup, who has acted as one of the influentialpeople who motivated my PhD studies. Magnus also played a crucial role instarting the PhD-programme in Kalmar, and it was Magnus encouraged mademe continue my PhD studies when I got an exciting job offer following myLicentiate degree. Without him, I would likely never have finished my PhD.
This thesis would never be possible without support from the many people Ihave met in research group meetings at DTU, in Chalmers and Lund and atconferences and workshops; even if your name is not listed here, you are notforgotten.
And last but not least, my wife Madeleine has been my greatest supporter,and none of this would have been possible without you. I would also like tomention our two children, Gustav and Louise, who have grown up with theirfather always doing something with eyes glued to a computer screen. Despiteall that, I have had five fantastic years, and even though it is often challengingand time-consuming, I would do it all over again in a heartbeat.
xv
Preface
My work began in 2013 when I decided to study how to make ships more energyefficient under operational conditions. Many studies do not consider realdata from ships in operation, often focussing instead on specific technologieswhich are based on data from the manufacturer or on listed ship data. A ship’soperational profile can be vastly different from what it was designed and builtfor, and it is therefore essential to investigate how to make ships more energyefficient under operational conditions.
This thesis describes the academic journey of a Marine Engineer who startedhis career as a Navy Engineering Officer and then used that background in theNavy as an added advantage to pursue a PhD, bringing knowledge of actuallybeing on board a ship and near the real operation and engine crew. Thismarine engineering background has been shown as a significant advantage,bringing prior operational knowledge of an engine room and being trusted asa ’kind of like’ when engaging in discussions with the crew. I have managedto tie together onboard operational experiences with academic theories incollaboration with several universities. To create knowledge and explain theworld, evidence is needed which can be found as onboard logged data as wellas crew experiences. The ship M/S Birka Stockholm has been a valuable datasource, and because of numerous collaborations with several universities andmany good ideas, this data source has resulted in many published papers.
The work in this thesis was performed in collaboration with the KalmarMaritime Academy Linnaeus University, researchers at Lund TechnicalUniversity, Chalmers Technical University and the Technical University ofDenmark. I would like to thank Rederiaktiebolaget Eckerö for their supportand especially the engine crew of M/S Birka Stockholm. The project wasfunded by the Swedish Maritime Administration and Linnaeus University.
I must mention the fantastic collaborations with my two of my fellow co-authors, Francesco Baldi and Tuong-Van Nguyen. I met Francesco at thebeginning of my PhD when he was halfway finished and presenting hisLicentiate thesis. That meeting led to a successful collaboration whichcontinues to the present. Through him, I have also gained a good friend andhiking companion.
I also want to mention my colleagues Mats Hammander, Fredrik Hjorth, JohnOhlson, Pär Karlsson, Magnus Boström, Gesa Praetorius, Carl Sandberg, Carl
xiv
Chapter 1
Introduction
1.1 Purpose and methodology
The aim of this thesis is to investigate the improvement of operational energyefficiency in ships by examining how the ships’ systems can be optimised. Thisthesis is largely based on logged machinery data from actual ship operations.In the first sections of the thesis, we analyse ship energy balance and calculatethe feasibility of waste heat recovery devices. An organic Rankine cycle wassimulated via software, and the energy flows were measured and calculated(Papers I and V). From this work, we integrate a waste heat recovery devicewith a modern marine two-stroke diesel engine fitted with both exhaust gasrecirculation and a scrubber (Paper IV), utilising as much waste heat aspossible. These studies were conducted with the simulation software IPSEpro.To investigate and thoroughly understand the energy flows, an extensive energyand exergy analysis conducted (Paper VI). From the ship data, we also usedexisting machine learning tools to predict the energy flow using minimalmeasuring points (Paper II and III).
The perspective of this thesis is the view from the inside of the ship, specificallyinvolving how to improve the functionality and efficiency of things insidethe hull. The thesis describes the feasibility of and challenges concerningwaste heat recovery integration on existing ships and demonstrates methods ofutilising machine learning for predicting energy flows. This provides a betterunderstanding of how ships can use less fuel as well as providing a means of
1
Chapter 1
Introduction
1.1 Purpose and methodology
The aim of this thesis is to investigate the improvement of operational energyefficiency in ships by examining how the ships’ systems can be optimised. Thisthesis is largely based on logged machinery data from actual ship operations.In the first sections of the thesis, we analyse ship energy balance and calculatethe feasibility of waste heat recovery devices. An organic Rankine cycle wassimulated via software, and the energy flows were measured and calculated(Papers I and V). From this work, we integrate a waste heat recovery devicewith a modern marine two-stroke diesel engine fitted with both exhaust gasrecirculation and a scrubber (Paper IV), utilising as much waste heat aspossible. These studies were conducted with the simulation software IPSEpro.To investigate and thoroughly understand the energy flows, an extensive energyand exergy analysis conducted (Paper VI). From the ship data, we also usedexisting machine learning tools to predict the energy flow using minimalmeasuring points (Paper II and III).
The perspective of this thesis is the view from the inside of the ship, specificallyinvolving how to improve the functionality and efficiency of things insidethe hull. The thesis describes the feasibility of and challenges concerningwaste heat recovery integration on existing ships and demonstrates methods ofutilising machine learning for predicting energy flows. This provides a betterunderstanding of how ships can use less fuel as well as providing a means of
1
Chapter 2
The need for change
2.1 The Atmosphere and fossil fuels
Anthropogenic greenhouse gas (GHG) emissions are driven by economic andpopulation growth due to the increased use of fossil fuels for energy needs.According to the Intergovernmental Panel on Climate Change (IPCC), theUN body for assessing science related to climate change, GHG emissionshave very probably been the primary cause of global warming since themid-twentieth century [1]. In the pre-industrial era, the amount of carbondioxide (CO2) in the atmosphere was about 280 ppm and has since risen toover 403 ppm by 2016. The CO2 concentration in the atmosphere is still risingwith an additional average growth of 2 ppm per year over the past ten years[2]. The worlds’ CO2 emissions from fossil fuel combustion were over 33 Gtin 2015; however, to provide a fifty-fifty chance of keeping the global target ofa maximum 2 ◦C temperature rise until 2050, the total amount of CO2 emittedin the atmosphere must be below 1100 Gt between 2011 and 2050 [3, 2].
To slow down and reverse the trend of rising CO2 emissions, we must stopemitting greenhouse gases into the atmosphere. In Paris in 2015, a substantialproportion of world leaders came to the Paris agreement, stating that weshould aim to reduce emissions within the 2 ◦C limit, the threshold seen as atipping point at which climate change becomes dangerous.
Since 1970, the process of reducing emissions has been more or less ‘businessas usual’, fossil fuel resources are known to be limited and CO2 emissions
3
doing so, thereby making ships less polluting.
The purpose of this thesis is to answer the question of how to better integrateexisting components in the ship during operations under realistic conditions.
1.2 Research boundaries
This thesis studies the energy system within a ship’s hull. All the workpublished in the thesis takes the ship hull as the boundary and the energysystem within as the focus. The advantage with this approach is that theresults are more focussed on optimising the energy system regardless of theroute the ship is sailing or the weather conditions. That is, regardless of theboundary conditions, the energy system which must be optimised is the same.
1.3 Outline
The thesis is divided into several sections, a description of which follows:
Chapter 2 presents a brief background of shipping and the climate, as well asa definition of the energy used for the transport and marine sector.
In Chapter 3 and 4, the data collection process is described, as well as themethods and software used to produce the results.
In Chapters 5 and 6, the basic concepts of waste heat recovery and machinelearning are address, as both topics comprise the basis of this thesis.
In Chapter 7, the papers are discussed in the context of the shipping industry,recent trends and the research community.
Chapter 8 sketches concluding remarks and summaries the main results ofthis thesis.
Chapter 9 presents a summary of all papers, with their results and contributions.
2
Chapter 2
The need for change
2.1 The Atmosphere and fossil fuels
Anthropogenic greenhouse gas (GHG) emissions are driven by economic andpopulation growth due to the increased use of fossil fuels for energy needs.According to the Intergovernmental Panel on Climate Change (IPCC), theUN body for assessing science related to climate change, GHG emissionshave very probably been the primary cause of global warming since themid-twentieth century [1]. In the pre-industrial era, the amount of carbondioxide (CO2) in the atmosphere was about 280 ppm and has since risen toover 403 ppm by 2016. The CO2 concentration in the atmosphere is still risingwith an additional average growth of 2 ppm per year over the past ten years[2]. The worlds’ CO2 emissions from fossil fuel combustion were over 33 Gtin 2015; however, to provide a fifty-fifty chance of keeping the global target ofa maximum 2 ◦C temperature rise until 2050, the total amount of CO2 emittedin the atmosphere must be below 1100 Gt between 2011 and 2050 [3, 2].
To slow down and reverse the trend of rising CO2 emissions, we must stopemitting greenhouse gases into the atmosphere. In Paris in 2015, a substantialproportion of world leaders came to the Paris agreement, stating that weshould aim to reduce emissions within the 2 ◦C limit, the threshold seen as atipping point at which climate change becomes dangerous.
Since 1970, the process of reducing emissions has been more or less ‘businessas usual’, fossil fuel resources are known to be limited and CO2 emissions
3
doing so, thereby making ships less polluting.
The purpose of this thesis is to answer the question of how to better integrateexisting components in the ship during operations under realistic conditions.
1.2 Research boundaries
This thesis studies the energy system within a ship’s hull. All the workpublished in the thesis takes the ship hull as the boundary and the energysystem within as the focus. The advantage with this approach is that theresults are more focussed on optimising the energy system regardless of theroute the ship is sailing or the weather conditions. That is, regardless of theboundary conditions, the energy system which must be optimised is the same.
1.3 Outline
The thesis is divided into several sections, a description of which follows:
Chapter 2 presents a brief background of shipping and the climate, as well asa definition of the energy used for the transport and marine sector.
In Chapter 3 and 4, the data collection process is described, as well as themethods and software used to produce the results.
In Chapters 5 and 6, the basic concepts of waste heat recovery and machinelearning are address, as both topics comprise the basis of this thesis.
In Chapter 7, the papers are discussed in the context of the shipping industry,recent trends and the research community.
Chapter 8 sketches concluding remarks and summaries the main results ofthis thesis.
Chapter 9 presents a summary of all papers, with their results and contributions.
2
Figure 2.2: CO2 emissions by sector [2]
Figure 2.3: World primary energy supply [2]
Since 1990, an increase in CO2 emissions is observable in the transport sector,with a total increase of 68 % between 1990–2015. As demonstrated in Figure2.5, the most significant emission contributor is the road sector, accountingfor three quarters of emissions. However, in relative trends, the emissionsfor both marine and aviation bunker have been rising more, with aviation by105 % and marine 77 %. This growth implies that even though the absolutenumbers are less, the upward trend is stronger.
5
must be reduced [4]. An estimated one third of all known oil reserves, thatis, half of all gas reserves and 80 % of all coal reserves, must remain unusedbetween 2010 and 2050 to meet the 2 ◦C goal [5]. This fact can indicate afuture with even more uncertainty in the fuel price market. As seen in Figure2.1, the energy sector is by far the largest source of anthropogenic globalGHG. In the figure, the sector others include large-scale biomass burning,post-burn decay, peat decay, indirect N2O emissions from non-agriculturalemissions of NOx and NH3, waste and solvent use. This statistic includesthe total influence of the global warming potential (GWP) of emissions on a100-year global warming potential. Ninety percent of all energy emissionscome from burning fossil fuels. By looking at the shares of emissions foreach sector, as demonstrated in Figure 2.2 the transport sector accounts for 24% of the total emissions.
Even though agriculture and industrial processes contribute to CO2 emissions,the primary source is the energy sector, which accounts for two thirds of allemissions. A shift towards more non-fossil fuel energy production is currentlyoccurring in the world, but the increase in fossil fuel combusted each yearpersists due to the rising global energy demand.
Figure 2.1: Global shares of anthropogenic GHG emissions by sector [2]
4
Figure 2.2: CO2 emissions by sector [2]
Figure 2.3: World primary energy supply [2]
Since 1990, an increase in CO2 emissions is observable in the transport sector,with a total increase of 68 % between 1990–2015. As demonstrated in Figure2.5, the most significant emission contributor is the road sector, accountingfor three quarters of emissions. However, in relative trends, the emissionsfor both marine and aviation bunker have been rising more, with aviation by105 % and marine 77 %. This growth implies that even though the absolutenumbers are less, the upward trend is stronger.
5
must be reduced [4]. An estimated one third of all known oil reserves, thatis, half of all gas reserves and 80 % of all coal reserves, must remain unusedbetween 2010 and 2050 to meet the 2 ◦C goal [5]. This fact can indicate afuture with even more uncertainty in the fuel price market. As seen in Figure2.1, the energy sector is by far the largest source of anthropogenic globalGHG. In the figure, the sector others include large-scale biomass burning,post-burn decay, peat decay, indirect N2O emissions from non-agriculturalemissions of NOx and NH3, waste and solvent use. This statistic includesthe total influence of the global warming potential (GWP) of emissions on a100-year global warming potential. Ninety percent of all energy emissionscome from burning fossil fuels. By looking at the shares of emissions foreach sector, as demonstrated in Figure 2.2 the transport sector accounts for 24% of the total emissions.
Even though agriculture and industrial processes contribute to CO2 emissions,the primary source is the energy sector, which accounts for two thirds of allemissions. A shift towards more non-fossil fuel energy production is currentlyoccurring in the world, but the increase in fossil fuel combusted each yearpersists due to the rising global energy demand.
Figure 2.1: Global shares of anthropogenic GHG emissions by sector [2]
4
Figure 2.5: Trends in CO2 emissions for the transport sector 1990-2015 [2]
Road. 75%
Marine (intl. +
domestic). 11%
Aviation (intl. + domestic). 11%
Other. 3%
CO2 shares from transport sector, 2015
Figure 2.6: CO2 emission share for the transport sector 2015 [2]
In terms of the world’s total anthropogenic CO2 emissions, the shippingsector contributed a total of approximately 2.7 % [10]. Relating shipping
7
Figure 2.4: Trend in CO2 emissions 1870-2014 [2]
2.2 Emissions from the Maritime Sector
There are 50,732 ships with a gross tonnage above 1,000 gross tonnes (GT, anon-linear measure of the ship internal volume) in the world, and the shippingsector is responsible for about 80 % of international trade (trade betweencountries, as opposed to domestic trade) in terms of cargo weight [6]. Itis estimated that 96 % of all merchant ships above 100 GT are driven bydiesel engines, often large, two-stroke diesel engines, which have a typicalmechanical efficiency of about 50 %, which is considered energy efficientby today’s standards [7]. Moreover, because ships also carry vast amountsof cargo, the fuel use per tonne of cargo can be less intensive compared toother transport methods such as aviation and road transport. Nevertheless,the shipping sector faces several challenges and is responsible for severalproblems concerning the environment, not only CO2 emissions but also theneed to reduce particulate matter (PM), sulphur oxides (SOx) and nitrogenoxides (NOx) [8, 9].
Carbon dioxide is the most significant GHG emitted by ships. Accordingto the third IMO GHG Study 2014, the total level of CO2 emissions fromshipping was 949 Mt in 2012. International shipping contributes to 796 Mt,which means that domestic shipping accounts for 153 Mt of CO2 emissions.
6
Figure 2.5: Trends in CO2 emissions for the transport sector 1990-2015 [2]
Road. 75%
Marine (intl. +
domestic). 11%
Aviation (intl. + domestic). 11%
Other. 3%
CO2 shares from transport sector, 2015
Figure 2.6: CO2 emission share for the transport sector 2015 [2]
In terms of the world’s total anthropogenic CO2 emissions, the shippingsector contributed a total of approximately 2.7 % [10]. Relating shipping
7
Figure 2.4: Trend in CO2 emissions 1870-2014 [2]
2.2 Emissions from the Maritime Sector
There are 50,732 ships with a gross tonnage above 1,000 gross tonnes (GT, anon-linear measure of the ship internal volume) in the world, and the shippingsector is responsible for about 80 % of international trade (trade betweencountries, as opposed to domestic trade) in terms of cargo weight [6]. Itis estimated that 96 % of all merchant ships above 100 GT are driven bydiesel engines, often large, two-stroke diesel engines, which have a typicalmechanical efficiency of about 50 %, which is considered energy efficientby today’s standards [7]. Moreover, because ships also carry vast amountsof cargo, the fuel use per tonne of cargo can be less intensive compared toother transport methods such as aviation and road transport. Nevertheless,the shipping sector faces several challenges and is responsible for severalproblems concerning the environment, not only CO2 emissions but also theneed to reduce particulate matter (PM), sulphur oxides (SOx) and nitrogenoxides (NOx) [8, 9].
Carbon dioxide is the most significant GHG emitted by ships. Accordingto the third IMO GHG Study 2014, the total level of CO2 emissions fromshipping was 949 Mt in 2012. International shipping contributes to 796 Mt,which means that domestic shipping accounts for 153 Mt of CO2 emissions.
6
content of today’s heavy fuel oil (HFO) is approximately 2.7 %, but the globallimit will be lowered to 0.5 % on 1 January 2020, which impact the choice offuel or the treatment of exhaust gases [17, 18].
2.4 Energy efficiency and measures
It is wholly a confusion of ideas tosuppose that the economical use offuel is equivalent to a diminishedconsumption. The very contrary isthe truth.
William Stanley Jevons
Knowledge concerning how the energy system works, where the losses areand how it can be optimised is essential to maximise the efficiency of existingships. Current energy systems must be optimised, and knowledge from thatexperience must be transferred to new ship designs. Reducing emissions canbe done in several ways, and many measures must be undertaken to realisethis goal. Making ships more energy efficient means that they consumeless fuel – and thereby produce fewer emissions – for the same amount ofwork. Operational measures, for example, weather routing or slow steaming,are purely operational and can save vast amounts of fuel with few or noinvestment costs. Energy efficiency is an important way to mitigate and reducecarbon emissions from the shipping sector, and the drivers are compliant withregulations, economic incentives and requirements from customers [19]. Theability to reduce emissions by energy efficiency measures is a vital part of thesolution.
Energy must be optimised to reduce total energy usage in the transport sector.The main drivers of this optimisation process are political for greenhousegases and other emissions, as well as economic incentives. A ship whichconsumes less fuel for the same transport work, regardless of whether thisfuel comes from a renewable source, does have an economic advantage andless environmental impact. Ships are built with a vast number of components,all of which have their optimum efficiency.
9
emissions to total transport sector emissions (accounting for 24 % of the total),as shown in Figure 2.6, indicates that the maritime sector contributes 11 % ofadmissions [2].
The maritime sector must reduce carbon emissions to meet the climate goals.The IMO Marine Environment Protection Committee (MEPC) announcedin April 2018 that member states agreed to cut the shipping sector’s CO2emissions by 50 % by 2050 [11]. Notably, the shipping sector was not includedin the Paris agreement [12].
According to Horvath et al. seven focus areas will allow the achievementof a decarbonised shipping sector, mission refinement, resistance reduction,propulsor selection, propulsor-hull-prime mover optimisation, prime moverselection, propulsion augments, and using new fuels [13].
2.3 Environmental maritime legislation
The International Maritime Organization (IMO) is the UN organisationresponsible for regulating ship safety, pollution and security. InternationalConvention for the Prevention of Pollution from Ships, 1973, as modified bythe Protocol of 1978 relating thereto and by the Protocol of 1997 (MARPOL),regulates shipping emissions and pollution, and the conventions become lawwhen they are ratified by the member states [14].
The focus so far has addressed the need not only to reduce CO2 emissionsbut also sulphur emissions, which marine transportation also significantlycontribute to. Before stricter sulphur regulations were enforced, shipping wasresponsible in 2009 for a total of 124 000 t of SOx emissions in the Baltic Sea[15]. About 15 % of all global NOx and about 5 % of all SOx emissions werein 2005 attributable to the shipping sector [16].
At present, four emission control areas (ECA) have been defined by the IMO:the Baltic Sea, the North Sea, the North American and the United StatesCaribbean Sea areas. In the sulphur emission control areas (SECA), themaximum amount of sulphur in the fuel cannot exceed 0.1 % (notably, this isstill 100 times more than the EU directive for diesel fuel in trucks), whichbecame enforceable in January 2015. Sulphur emissions currently have aglobal limit of 3.5 %, which means that no fuel with more than 3.5 % sulphurcan be used. In theory, this is not a problem today as the average sulphur
8
content of today’s heavy fuel oil (HFO) is approximately 2.7 %, but the globallimit will be lowered to 0.5 % on 1 January 2020, which impact the choice offuel or the treatment of exhaust gases [17, 18].
2.4 Energy efficiency and measures
It is wholly a confusion of ideas tosuppose that the economical use offuel is equivalent to a diminishedconsumption. The very contrary isthe truth.
William Stanley Jevons
Knowledge concerning how the energy system works, where the losses areand how it can be optimised is essential to maximise the efficiency of existingships. Current energy systems must be optimised, and knowledge from thatexperience must be transferred to new ship designs. Reducing emissions canbe done in several ways, and many measures must be undertaken to realisethis goal. Making ships more energy efficient means that they consumeless fuel – and thereby produce fewer emissions – for the same amount ofwork. Operational measures, for example, weather routing or slow steaming,are purely operational and can save vast amounts of fuel with few or noinvestment costs. Energy efficiency is an important way to mitigate and reducecarbon emissions from the shipping sector, and the drivers are compliant withregulations, economic incentives and requirements from customers [19]. Theability to reduce emissions by energy efficiency measures is a vital part of thesolution.
Energy must be optimised to reduce total energy usage in the transport sector.The main drivers of this optimisation process are political for greenhousegases and other emissions, as well as economic incentives. A ship whichconsumes less fuel for the same transport work, regardless of whether thisfuel comes from a renewable source, does have an economic advantage andless environmental impact. Ships are built with a vast number of components,all of which have their optimum efficiency.
9
emissions to total transport sector emissions (accounting for 24 % of the total),as shown in Figure 2.6, indicates that the maritime sector contributes 11 % ofadmissions [2].
The maritime sector must reduce carbon emissions to meet the climate goals.The IMO Marine Environment Protection Committee (MEPC) announcedin April 2018 that member states agreed to cut the shipping sector’s CO2emissions by 50 % by 2050 [11]. Notably, the shipping sector was not includedin the Paris agreement [12].
According to Horvath et al. seven focus areas will allow the achievementof a decarbonised shipping sector, mission refinement, resistance reduction,propulsor selection, propulsor-hull-prime mover optimisation, prime moverselection, propulsion augments, and using new fuels [13].
2.3 Environmental maritime legislation
The International Maritime Organization (IMO) is the UN organisationresponsible for regulating ship safety, pollution and security. InternationalConvention for the Prevention of Pollution from Ships, 1973, as modified bythe Protocol of 1978 relating thereto and by the Protocol of 1997 (MARPOL),regulates shipping emissions and pollution, and the conventions become lawwhen they are ratified by the member states [14].
The focus so far has addressed the need not only to reduce CO2 emissionsbut also sulphur emissions, which marine transportation also significantlycontribute to. Before stricter sulphur regulations were enforced, shipping wasresponsible in 2009 for a total of 124 000 t of SOx emissions in the Baltic Sea[15]. About 15 % of all global NOx and about 5 % of all SOx emissions werein 2005 attributable to the shipping sector [16].
At present, four emission control areas (ECA) have been defined by the IMO:the Baltic Sea, the North Sea, the North American and the United StatesCaribbean Sea areas. In the sulphur emission control areas (SECA), themaximum amount of sulphur in the fuel cannot exceed 0.1 % (notably, this isstill 100 times more than the EU directive for diesel fuel in trucks), whichbecame enforceable in January 2015. Sulphur emissions currently have aglobal limit of 3.5 %, which means that no fuel with more than 3.5 % sulphurcan be used. In theory, this is not a problem today as the average sulphur
8
Chapter 3
The story behind the results
You should take the approach thatyou’re wrong. Your goal is to be lesswrong.
Elon Musk
The thesis makes use of several methods in the different publications whichcomprise it. In this chapter, a brief overview of the methods in the publicationsis presented. This chapter is written as an overall addition, covering thatwhich is less apparent in the studies alone so as to give the reader a holisticview of the methods used and why they were chosen. The details of eachmethod are more thoroughly described in the publications themselves. Section3.2 concerns the ship and provides an overview of the ship’s specifics anddescribes its operational conditions, while Chapter 4 concerns the modellingand simulation.
3.1 Generalisations from a case study
The data in this thesis is largely based on a dataset from the ship M/SBirka Stockholm. The data has been valuable for feeding simulations withoperational data; however, the results from these studies are consideredgeneralisable, as they pertain not only to a single ship but rather the integration
11
This thesis investigates the energy system within the ship, regardless of theconditions in which it operates. There are many ways of reducing the fuelconsumption of an existing ship, either with add-on technologies for or newerdesigns. The list is adopted from Bouman et al. [20].
• Power and Propulsion systems
– Hybrid propulsion– Waste heat recovery
• Hull design
– Vessel size– Hull shape– Lightweight materials– Air lubrication– Hull coating
• Alternative fuels
– Biofuels– Liquefied natural gas (LNG)
• Alternative energy sources
– Wind power– Fuel cells– Cold ironing– Solar power
• Operation
– Speed optimisation– Capacity utilisation– Voyage optimisation– Trim/draft and energy management optimisation
Energy system optimisation involves the search for the minimum energyconsumption for necessary work, that is, using as little fuel as possible for aspecific trip. When a ship is designed, many factors must considered: it mustbe safe, it must take a certain amount of cargo, it should be usable for manyyears, it must be easy and cheap to maintain and it must be versatile if itsoperation area or purpose changes [21]. As many factors influence the designchoices, and given that the market, legislation and fuel prices are continuouslychanging, the conditions in which a ship operates are likely not those forwhich it was designed.
10
Chapter 3
The story behind the results
You should take the approach thatyou’re wrong. Your goal is to be lesswrong.
Elon Musk
The thesis makes use of several methods in the different publications whichcomprise it. In this chapter, a brief overview of the methods in the publicationsis presented. This chapter is written as an overall addition, covering thatwhich is less apparent in the studies alone so as to give the reader a holisticview of the methods used and why they were chosen. The details of eachmethod are more thoroughly described in the publications themselves. Section3.2 concerns the ship and provides an overview of the ship’s specifics anddescribes its operational conditions, while Chapter 4 concerns the modellingand simulation.
3.1 Generalisations from a case study
The data in this thesis is largely based on a dataset from the ship M/SBirka Stockholm. The data has been valuable for feeding simulations withoperational data; however, the results from these studies are consideredgeneralisable, as they pertain not only to a single ship but rather the integration
11
This thesis investigates the energy system within the ship, regardless of theconditions in which it operates. There are many ways of reducing the fuelconsumption of an existing ship, either with add-on technologies for or newerdesigns. The list is adopted from Bouman et al. [20].
• Power and Propulsion systems
– Hybrid propulsion– Waste heat recovery
• Hull design
– Vessel size– Hull shape– Lightweight materials– Air lubrication– Hull coating
• Alternative fuels
– Biofuels– Liquefied natural gas (LNG)
• Alternative energy sources
– Wind power– Fuel cells– Cold ironing– Solar power
• Operation
– Speed optimisation– Capacity utilisation– Voyage optimisation– Trim/draft and energy management optimisation
Energy system optimisation involves the search for the minimum energyconsumption for necessary work, that is, using as little fuel as possible for aspecific trip. When a ship is designed, many factors must considered: it mustbe safe, it must take a certain amount of cargo, it should be usable for manyyears, it must be easy and cheap to maintain and it must be versatile if itsoperation area or purpose changes [21]. As many factors influence the designchoices, and given that the market, legislation and fuel prices are continuouslychanging, the conditions in which a ship operates are likely not those forwhich it was designed.
10
The ship has an intended top speed of 21 knots, and the propulsion systemconsists of four 4-stroke medium speed Wärtsilä 6L46 main engines (ME) witha nominal power of 5850 kW, each at 500 rpm. The engines are turbochargedand intercooled with direct fuel injection and, according to manufacturerdata, a rated minimum specific fuel consumption (SFOC) of 170 g/kWh at75 % load, which corresponds to a thermal efficiency of 49.6 % [22]. Thesefour MEs are connected to two gearboxes, which in turn drive two outgoingshafts with controllable pitch propellers (CPP). This means that at least oneengine on each propulsion side must be running to produce full manoeuvringcapability. For the onboard electricity demand there are four 4-stroke mediumspeed Wärtsilä 6L32 auxiliary engines (AE), each rated for a nominal powerof 2760 kW.
The ship seldom runs at the design point of 21 knots, which must be carefullyconsidered when making plans and decisions about optimising the energysystem. In Figure 3.2 the speed distribution for a full year of operation isshown, filtered for at least one main propulsion engine running with rpmabove 10. The most common speeds are around 9-15 knots, which was thebase for the chosen speed interval at 12-14 knots for the design point in thefirst organic Rankine cycle (ORC) waste heat recovery study ‘Waste HeatRecovery in a Cruise Vessel in the Baltic Sea by Using an Organic RankineCycle: A Case Study’ [23].
Figure 3.3 indicates that most of the time, the ship operates on two engines,which accounts for most of speeds between 0 and 17 knots. Speeds up to 5knots with zero load and no engines on are also observable, occurring becausethe ship often drifts with engines off at night.
The yearly load distribution presented in Figure 3.5 for the main engines, andin Figure 3.6 for the auxiliary engines, demonstrates that the main enginespredominantly run 40 % load, and the auxiliaryx engines at 30 %. This, incombination with the speed intervals shown in Figure 3.2, indicates thatthe propulsion and auxiliary systems primarily run in off-design conditions.A diesel engine generally operates at a maximum efficiency of between 80and 90 %. According to the test protocol data for these specific engines, a30 % to 40 % load corresponds to a specific fuel consumption of less than186 g/(kW h) for the main engines and less than 198 g/(kW h) for the auxiliaryengines, as demonstrated in Figure 3.4. Even though a diesel engine hasa much flatter efficiency curve (it can run on lower loads and continue tomaintain a decent efficiency) compared to a gas turbine, running the engines
13
of waste heat recovery (WHR) technologies in a complex system. What is notat first apparent is that the operational profile of a ship is far from the dataspecs of for which it was originally designed.
The validity and accuracy of each single data point are not known, but theresults are considered valid due to the number of readings in the system, whichprovides a holistic view. The fuel measurements, which could be consideredthe most important as they provide the total energy input, are those whichhave been validated with bunker data, mass flow meters and dynamic readingsof engine data. The exact physical locations of the thousands of individualsensors is not known, but accounting for this was deemed too time-consuminggiven that no or little extra accuracy was likely to result.
3.2 Description of the ship: M/S Birka Stockholm
The cruise ship M/S Birka Stockholm, seen in Figure 3.1 is a passenger vesselwhich operates on daily leisure cruises with passengers between Stockholmand Mariehamn. She was built in Aaker Finnyards, Raumo in 2004 andhas a capacity of 1,800 passengers, is 177 m long and has a beam length of28.6 m. The vessel can therefore be considered a relatively small cruise shipby international comparison.
Figure 3.1: M/S Birka Stockholm. Photo: Kjell Larsson
12
The ship has an intended top speed of 21 knots, and the propulsion systemconsists of four 4-stroke medium speed Wärtsilä 6L46 main engines (ME) witha nominal power of 5850 kW, each at 500 rpm. The engines are turbochargedand intercooled with direct fuel injection and, according to manufacturerdata, a rated minimum specific fuel consumption (SFOC) of 170 g/kWh at75 % load, which corresponds to a thermal efficiency of 49.6 % [22]. Thesefour MEs are connected to two gearboxes, which in turn drive two outgoingshafts with controllable pitch propellers (CPP). This means that at least oneengine on each propulsion side must be running to produce full manoeuvringcapability. For the onboard electricity demand there are four 4-stroke mediumspeed Wärtsilä 6L32 auxiliary engines (AE), each rated for a nominal powerof 2760 kW.
The ship seldom runs at the design point of 21 knots, which must be carefullyconsidered when making plans and decisions about optimising the energysystem. In Figure 3.2 the speed distribution for a full year of operation isshown, filtered for at least one main propulsion engine running with rpmabove 10. The most common speeds are around 9-15 knots, which was thebase for the chosen speed interval at 12-14 knots for the design point in thefirst organic Rankine cycle (ORC) waste heat recovery study ‘Waste HeatRecovery in a Cruise Vessel in the Baltic Sea by Using an Organic RankineCycle: A Case Study’ [23].
Figure 3.3 indicates that most of the time, the ship operates on two engines,which accounts for most of speeds between 0 and 17 knots. Speeds up to 5knots with zero load and no engines on are also observable, occurring becausethe ship often drifts with engines off at night.
The yearly load distribution presented in Figure 3.5 for the main engines, andin Figure 3.6 for the auxiliary engines, demonstrates that the main enginespredominantly run 40 % load, and the auxiliaryx engines at 30 %. This, incombination with the speed intervals shown in Figure 3.2, indicates thatthe propulsion and auxiliary systems primarily run in off-design conditions.A diesel engine generally operates at a maximum efficiency of between 80and 90 %. According to the test protocol data for these specific engines, a30 % to 40 % load corresponds to a specific fuel consumption of less than186 g/(kW h) for the main engines and less than 198 g/(kW h) for the auxiliaryengines, as demonstrated in Figure 3.4. Even though a diesel engine hasa much flatter efficiency curve (it can run on lower loads and continue tomaintain a decent efficiency) compared to a gas turbine, running the engines
13
of waste heat recovery (WHR) technologies in a complex system. What is notat first apparent is that the operational profile of a ship is far from the dataspecs of for which it was originally designed.
The validity and accuracy of each single data point are not known, but theresults are considered valid due to the number of readings in the system, whichprovides a holistic view. The fuel measurements, which could be consideredthe most important as they provide the total energy input, are those whichhave been validated with bunker data, mass flow meters and dynamic readingsof engine data. The exact physical locations of the thousands of individualsensors is not known, but accounting for this was deemed too time-consuminggiven that no or little extra accuracy was likely to result.
3.2 Description of the ship: M/S Birka Stockholm
The cruise ship M/S Birka Stockholm, seen in Figure 3.1 is a passenger vesselwhich operates on daily leisure cruises with passengers between Stockholmand Mariehamn. She was built in Aaker Finnyards, Raumo in 2004 andhas a capacity of 1,800 passengers, is 177 m long and has a beam length of28.6 m. The vessel can therefore be considered a relatively small cruise shipby international comparison.
Figure 3.1: M/S Birka Stockholm. Photo: Kjell Larsson
12
Figure 3.3: Propulsion power and ship speed
3.3 Data pre-processing
The data from the ship described in Section 3.2 was exported from themachinery logging system during a total of four trips between 2014 and 2015.Since the logging system onboard was a proprietary database from Valmarine,the data could only be exported via a Microsoft Excel-plugin.
The data was originally stored as one-minute snapshots in the database, withthe export tool automatically exporting a 15-minute average of the snapshots.Moreover, as the export tool was manually operated with check-boxes, someof the exported data-points overlapped in time and were duplicated. ThePython Pandas library was used to merge all the data and processing alldata headers into a consistent format. Pandas is an extensive library forstatistical computing as well as a framework for storing large amounts ofdata in a dataframe (a two-dimensional data structure) [25]. All Excel fileswere initially imported into a large Pandas DataFrame in a time series dataformat, at which point the data was resampled to a consistent time series witha 15-minute average.
15
Figure 3.2: Distribution of ship speed for one year of operation, at least one main engine running.
on the best efficiency point can nevertheless increase fuel efficiency by about5 %.
Before 2015, a low sulphur (0.5 %) residual marine grade fuel oil (RMG IF 380)was used, which is considered a heavy fuel oil (HFO). The residual fuel isused for all engines apart from one auxiliary engine and boiler which usemarine gas oil (MGO), in order to limit emission when the ship is in port.After the implementation of stricter SECA legislation in January 2015, a newfuel was introduced on board which complied with the new sulphur emissionlegislation. This low sulphur residual (RMB 30) fuel had a sulphur content of0.1 %.
Schematics for both machinery and electricity were used to determine theship’s energy system, as well as discussions with the machinery crew toestimate unknown data points. For example, neither the fuel oil fired boilerrunning time nor the heat demand for ventilation were logged, so both had tobe estimated in Paper VI, ‘Energy and Exergy Analysis of a Cruise Ship’ [24].
14
Figure 3.3: Propulsion power and ship speed
3.3 Data pre-processing
The data from the ship described in Section 3.2 was exported from themachinery logging system during a total of four trips between 2014 and 2015.Since the logging system onboard was a proprietary database from Valmarine,the data could only be exported via a Microsoft Excel-plugin.
The data was originally stored as one-minute snapshots in the database, withthe export tool automatically exporting a 15-minute average of the snapshots.Moreover, as the export tool was manually operated with check-boxes, someof the exported data-points overlapped in time and were duplicated. ThePython Pandas library was used to merge all the data and processing alldata headers into a consistent format. Pandas is an extensive library forstatistical computing as well as a framework for storing large amounts ofdata in a dataframe (a two-dimensional data structure) [25]. All Excel fileswere initially imported into a large Pandas DataFrame in a time series dataformat, at which point the data was resampled to a consistent time series witha 15-minute average.
15
Figure 3.2: Distribution of ship speed for one year of operation, at least one main engine running.
on the best efficiency point can nevertheless increase fuel efficiency by about5 %.
Before 2015, a low sulphur (0.5 %) residual marine grade fuel oil (RMG IF 380)was used, which is considered a heavy fuel oil (HFO). The residual fuel isused for all engines apart from one auxiliary engine and boiler which usemarine gas oil (MGO), in order to limit emission when the ship is in port.After the implementation of stricter SECA legislation in January 2015, a newfuel was introduced on board which complied with the new sulphur emissionlegislation. This low sulphur residual (RMB 30) fuel had a sulphur content of0.1 %.
Schematics for both machinery and electricity were used to determine theship’s energy system, as well as discussions with the machinery crew toestimate unknown data points. For example, neither the fuel oil fired boilerrunning time nor the heat demand for ventilation were logged, so both had tobe estimated in Paper VI, ‘Energy and Exergy Analysis of a Cruise Ship’ [24].
14
Figure 3.5: Yearly load distribution Main Engines M/S Birka Stockholm
17
165
175
185
195
205
215
225
20% 40% 60% 80% 100%
SFO
C g/
kWh
ISO
304
6/1
Engine load
Main engine
Auxiliary engine
Poly. (Main engine)
Poly. (Auxiliary engine)
Figure 3.4: Test protocol data ME/AE Wärtsilä. Engine 6LB46B no. 91541 and W6L32 no. 22191
A Python script was developed to pre-process and filter the data accordingto rules manually set in a separate Excel-sheet. Duplicate data pointsand overlapping time-series which were mixed in different Excel-files wereautomatically merged in a consistent format. All data headers were filtered outfor characters with the American Standard Code for Information Interchange(ASCII), code above 255. A Hierarchal Data Format (HDF) was used,and all data was compressed with Blosc (a high-performance compressoroptimised for binary data) for efficient and fast access. This data format hasthe advantage of being rapid, as the loading times compared to an Excel or aCSV-file are significantly faster using Python-code. We used an HDF databaseto consolidate all data in Papers II, III and VI, which means that the data wasaccessible directly for array and matrix calculations.
16
Figure 3.5: Yearly load distribution Main Engines M/S Birka Stockholm
17
165
175
185
195
205
215
225
20% 40% 60% 80% 100%
SFO
C g/
kWh
ISO
304
6/1
Engine load
Main engine
Auxiliary engine
Poly. (Main engine)
Poly. (Auxiliary engine)
Figure 3.4: Test protocol data ME/AE Wärtsilä. Engine 6LB46B no. 91541 and W6L32 no. 22191
A Python script was developed to pre-process and filter the data accordingto rules manually set in a separate Excel-sheet. Duplicate data pointsand overlapping time-series which were mixed in different Excel-files wereautomatically merged in a consistent format. All data headers were filtered outfor characters with the American Standard Code for Information Interchange(ASCII), code above 255. A Hierarchal Data Format (HDF) was used,and all data was compressed with Blosc (a high-performance compressoroptimised for binary data) for efficient and fast access. This data format hasthe advantage of being rapid, as the loading times compared to an Excel or aCSV-file are significantly faster using Python-code. We used an HDF databaseto consolidate all data in Papers II, III and VI, which means that the data wasaccessible directly for array and matrix calculations.
16
Chapter 4
Tools of the trade
Beautiful is better than ugly.Explicit is better than implicit.Simple is better than complex.Complex is better than complicated.Now is better than never.Although never is often better thanright now.If the implementation is hard toexplain, it’s a bad idea.If the implementation is easy toexplain, it may be a good idea.
Excerpt from the Zen of Python byTim Peters
4.1 Simulation of physical systems
To simulate means to imitate, replicate, mimic or reproduce something. Akey aspect in modern engineering research is using computers to simulatephysical systems. By building a replica of a machine (a model) in computersoftware, we can then experiment on that model and observe what occurs.Thus, if we manage to build a model which behaves the same way as the realphysical system, we can optimise the modelled system in the computer withoutbuilding the machine and without the risk of destroying it in the process.
19
Figure 3.6: Yearly load distribution Auxiliary Engines M/S Birka Stockholm
18
Chapter 4
Tools of the trade
Beautiful is better than ugly.Explicit is better than implicit.Simple is better than complex.Complex is better than complicated.Now is better than never.Although never is often better thanright now.If the implementation is hard toexplain, it’s a bad idea.If the implementation is easy toexplain, it may be a good idea.
Excerpt from the Zen of Python byTim Peters
4.1 Simulation of physical systems
To simulate means to imitate, replicate, mimic or reproduce something. Akey aspect in modern engineering research is using computers to simulatephysical systems. By building a replica of a machine (a model) in computersoftware, we can then experiment on that model and observe what occurs.Thus, if we manage to build a model which behaves the same way as the realphysical system, we can optimise the modelled system in the computer withoutbuilding the machine and without the risk of destroying it in the process.
19
Figure 3.6: Yearly load distribution Auxiliary Engines M/S Birka Stockholm
18
Exhaust EGR
Exhaust after turbo
Air CAC
FW Jacket CW
mass[kg/s] h[kJ/kg]p[bar] t[°C]
FGC
Figure 4.1: IPSE model used in Paper IV
4.3 Dynamic or steady-state
All calculations and simulations in the included studies are steady-statesimulations, which means that they are solved in an equilibrium state whenthe system is at rest and in equilibrium within its boundaries. This typeof simulation does not account for dynamic behaviours of the system, fasttransitions between power loads or changing temperatures. Given that thesystem has inertia, both in mass and in thermal energy, it is essential to factorin time if a system is to function in a dynamic environment. A dynamic modelis more complicated to build and more resource intensive to solve, and as it isdependent on time, all equations must be solved for that time moment.
21
Even simpler systems consisting of only a few heat sources and componentsare computer intensive to optimise. As every component has a subset ofdesign choices and constraints and might operate in different temperatures orthermal loads, the number of available system setups increases exponentiallywith each component. For example, simulating a simple Rankine cycle withonly a boiler and a condenser can still be complex despite the relatively fewcomponents. As variables, the boiler component has heat input, the size ofthe boiler, friction losses, the heat transfer coefficient and many more. Theseall have an impact on the total efficiency when combined with the rest of thesystem, and the entire system quickly becomes complex.
4.2 The simulation software IPSEPro
IPSEpro (SimTech, 2013) is a simulation software use to calculate processesand heat balances as well as plant efficiencies and performance. It can furtherbe used to analyse thermal power systems, both steady-state and off-design,as demonstrated in Figure 4.1. Its main characteristic is that it is an openequation modelling environment, which means that the engineering equationsare not part of the source code and therefore can be extended by the users.The solver in IPSEpro is a Newton-Raphson solver, which attempts to find thesolution by going down the slope of a function until it reaches the minimum.User-defined models can be developed in the model development kit, andintegrated into the simulation setup. The software does not contain eitherlibraries for all available fluids in different ORC setups or binary mixtures, soRefprop was used as the database for fluid properties. The connection wasaccomplished via a DLL developed by Maria E. Mondejar originally derivedfrom Mälardalens University [26]. In this way, the IPSEpro model couldbe optimised by using the optimisation function in Matlab, and with fluidproperties which were retrieved from the Refprop database. Both IPSEproand Refprop are Windows software.
IPSEpro was used as the simulation software for the ORC in Paper I, ‘WasteHeat Recovery in a Cruise Vessel in the Baltic Sea by Using an OrganicRankine Cycle: A Case Study’, and Paper IV, ‘Energy integration of OrganicRankine Cycle, Exhaust Gas recirculation and Scrubber’ [23, 27].
20
Exhaust EGR
Exhaust after turbo
Air CAC
FW Jacket CW
mass[kg/s] h[kJ/kg]p[bar] t[°C]
FGC
Figure 4.1: IPSE model used in Paper IV
4.3 Dynamic or steady-state
All calculations and simulations in the included studies are steady-statesimulations, which means that they are solved in an equilibrium state whenthe system is at rest and in equilibrium within its boundaries. This typeof simulation does not account for dynamic behaviours of the system, fasttransitions between power loads or changing temperatures. Given that thesystem has inertia, both in mass and in thermal energy, it is essential to factorin time if a system is to function in a dynamic environment. A dynamic modelis more complicated to build and more resource intensive to solve, and as it isdependent on time, all equations must be solved for that time moment.
21
Even simpler systems consisting of only a few heat sources and componentsare computer intensive to optimise. As every component has a subset ofdesign choices and constraints and might operate in different temperatures orthermal loads, the number of available system setups increases exponentiallywith each component. For example, simulating a simple Rankine cycle withonly a boiler and a condenser can still be complex despite the relatively fewcomponents. As variables, the boiler component has heat input, the size ofthe boiler, friction losses, the heat transfer coefficient and many more. Theseall have an impact on the total efficiency when combined with the rest of thesystem, and the entire system quickly becomes complex.
4.2 The simulation software IPSEPro
IPSEpro (SimTech, 2013) is a simulation software use to calculate processesand heat balances as well as plant efficiencies and performance. It can furtherbe used to analyse thermal power systems, both steady-state and off-design,as demonstrated in Figure 4.1. Its main characteristic is that it is an openequation modelling environment, which means that the engineering equationsare not part of the source code and therefore can be extended by the users.The solver in IPSEpro is a Newton-Raphson solver, which attempts to find thesolution by going down the slope of a function until it reaches the minimum.User-defined models can be developed in the model development kit, andintegrated into the simulation setup. The software does not contain eitherlibraries for all available fluids in different ORC setups or binary mixtures, soRefprop was used as the database for fluid properties. The connection wasaccomplished via a DLL developed by Maria E. Mondejar originally derivedfrom Mälardalens University [26]. In this way, the IPSEpro model couldbe optimised by using the optimisation function in Matlab, and with fluidproperties which were retrieved from the Refprop database. Both IPSEproand Refprop are Windows software.
IPSEpro was used as the simulation software for the ORC in Paper I, ‘WasteHeat Recovery in a Cruise Vessel in the Baltic Sea by Using an OrganicRankine Cycle: A Case Study’, and Paper IV, ‘Energy integration of OrganicRankine Cycle, Exhaust Gas recirculation and Scrubber’ [23, 27].
20
Chapter 5
What it is all about: Energy
Thermodynamics is a funny subject.The first time you go through it, youdon’t understand it at all. Thesecond time you go through it, youthink you understand it, except forone or two small points. The thirdtime you go through it, you knowyou don’t understand it, but by thattime you are so used to it, it doesn’tbother you anymore.
Arnold Sommerfeld, when askedwhy he had never written a book on
the subject (1950)
Thermodynamics is not only the science behind power cycles but also concernsalmost everything in this world. This section covers the most basic fundamentsof this thesis. There are four thermodynamic laws, which are numbered fromzero to three (similarly to the programming language Python, wherein allvectors start from zero).
23
4.4 Scientific programming by Python
The software base for much of the work conducted this thesis is Open Source,meaning that the software source code is free for anyone to participate in.Using Open Source software tools assures transparency at the most basiclevel and also supports the idea that everyone should be able to reproduce theresults without needing to buy expensive software.
Python is a programming language initially released in 1991 and built sothat the code should be easily readable by people. Due to the open standardand large community, and as it runs on all different platforms and is freelyavailable, the language develops rapidly. Python can be run on everythingfrom supercomputers to tiny embedded devices, and currently, Python isa popular language for scientific computing, and with the library Numpy(numerical Python), it is a direct competitor to Matlab, bringing with it asimilar feature set. Numerous other libraries add functionality to the coreprogramming language, and the majority of these are free and open source,from which both academics and industry draw benefit.
In the original work behind Paper VI, the entire model was first written inMatlab but later rewritten to Python and published on Github [28, 29].
In the earlier articles in this thesis, the tools for analysing all data was MicrosoftExcel and the Pivot-table tool, but such became burdensome when the Excel-files surpassed 100 MB in size [23, 30, 31]. Saving a 100-MB Excel-filecan take up to a minute, and with the risk of crashing the system. Returningand reproducing a method is also a complex task after spending much timeworking with a certain file (it is also worth noting that in the author’s opinion,Microsoft products does not function very well on a Macintosh computer),which led to the use of Python for the data analysis.
A Linux-server was used as the base for all calculations in Papers II and IIIusing machine learning (ML) to predict fuel consumption. The Linux serverwas set up with a slim Ubuntu installation with the Anaconda scientific Pythondistribution for managing Python libraries [32]. The frontend for the code wasJupyter Notebook, which is an interactive frontend for scientific programming[33]. The main advantages of this are that the notebooks can contain a mix ofcode, results, graphs and text, which makes it ideal for writing the methodand simultaneously producing results.
22
Chapter 5
What it is all about: Energy
Thermodynamics is a funny subject.The first time you go through it, youdon’t understand it at all. Thesecond time you go through it, youthink you understand it, except forone or two small points. The thirdtime you go through it, you knowyou don’t understand it, but by thattime you are so used to it, it doesn’tbother you anymore.
Arnold Sommerfeld, when askedwhy he had never written a book on
the subject (1950)
Thermodynamics is not only the science behind power cycles but also concernsalmost everything in this world. This section covers the most basic fundamentsof this thesis. There are four thermodynamic laws, which are numbered fromzero to three (similarly to the programming language Python, wherein allvectors start from zero).
23
4.4 Scientific programming by Python
The software base for much of the work conducted this thesis is Open Source,meaning that the software source code is free for anyone to participate in.Using Open Source software tools assures transparency at the most basiclevel and also supports the idea that everyone should be able to reproduce theresults without needing to buy expensive software.
Python is a programming language initially released in 1991 and built sothat the code should be easily readable by people. Due to the open standardand large community, and as it runs on all different platforms and is freelyavailable, the language develops rapidly. Python can be run on everythingfrom supercomputers to tiny embedded devices, and currently, Python isa popular language for scientific computing, and with the library Numpy(numerical Python), it is a direct competitor to Matlab, bringing with it asimilar feature set. Numerous other libraries add functionality to the coreprogramming language, and the majority of these are free and open source,from which both academics and industry draw benefit.
In the original work behind Paper VI, the entire model was first written inMatlab but later rewritten to Python and published on Github [28, 29].
In the earlier articles in this thesis, the tools for analysing all data was MicrosoftExcel and the Pivot-table tool, but such became burdensome when the Excel-files surpassed 100 MB in size [23, 30, 31]. Saving a 100-MB Excel-filecan take up to a minute, and with the risk of crashing the system. Returningand reproducing a method is also a complex task after spending much timeworking with a certain file (it is also worth noting that in the author’s opinion,Microsoft products does not function very well on a Macintosh computer),which led to the use of Python for the data analysis.
A Linux-server was used as the base for all calculations in Papers II and IIIusing machine learning (ML) to predict fuel consumption. The Linux serverwas set up with a slim Ubuntu installation with the Anaconda scientific Pythondistribution for managing Python libraries [32]. The frontend for the code wasJupyter Notebook, which is an interactive frontend for scientific programming[33]. The main advantages of this are that the notebooks can contain a mix ofcode, results, graphs and text, which makes it ideal for writing the methodand simultaneously producing results.
22
5.3 The Ship as a closed system
In all studies in this thesis, the ship is considered to be the control volume, withthe hull as its boundaries. A control volume in thermodynamics is defined asa region in space wherein mass may flow. The heat flows and mass leavingthe system are a negative mass or energy balance [35].
5.4 Where is the waste heat coming from?
All energy conversions are less than 100 % efficient, and the energy that isnot used for anything purposeful is called waste heat, which can be utilised.For instance, the heat released from a diesel engine can be utilised to heat thefuel, cabins or cargo, or can be converted back to mechanical energy with anORC, at which point it called waste heat recovery (WHR).
In Papers I and V, we considered the recovery of energy from the waste heatin the exhaust gases, which is at a higher temperature (197 ◦C to 417 ◦C) thanthe engine jacket cooling water (95 ◦C). The conversion efficiency using anORC would be significantly lower utilising the lower temperature.
As stated by the first law of thermodynamics, in a control volume Eq. 5.1, thetotal energy input (
∑in �Hin) equals the total energy output (
∑out �Hout). Broken
down into fractions, as seen in Eq. 5.2, the energy input into the enginesis in fuel ( �Hfuel) and air ( �Hair), which equals usable work for either heating( �Qheating), or electricity and propulsion power ( �Wel).
∑in
�Hin =∑out
�Hout (5.1)
�Hfuel + �Hair =∑waste
�Hwaste + �Wel + �Qheating (5.2)
In Paper VI, the full energy system of M/S Birka Stockholm was analysedduring a year of operation and is presented in Figure 5.3, page 29. In thefigure, the green represents the electric and mechanical energy, which isused for propulsion and for all electrical appliances on board. The yellowrepresents the exergy (described more in Section 5.8) in the exhaust gas. The
25
5.1 A well known but abstract concept: Energy
The scientific basis for all calculations and results is the laws of thermody-namics [34]. The simple question, ‘How can I make my ship consume lessfuel’, has a more complex answer from the energy perspective. The conceptof energy is abstract, which has salience in calculations but cannot be touched,tasted, felt or seen.
5.2 First Law - Energy conservation
A Swedish television programme for children called Tippen (the Swedishword for the waste dump) ran during the summer of 1994. Every episodestarted with the catchy slogan ‘Ingenting försvinner, allt finns kvar’, whichmeans ‘Nothing is disappearing, everything is still there’. This is a usefulcatchphrase not only for objects but also for energy. Energy is conserved in allprocesses, and it cannot by definition be destroyed. All energy which enters asystem, regardless of the form of fuel, electricity or heat, must all add up to atotal which is constant.
Energy can exist in various forms:
• Mechanical energy• Electrical energy• Magnetic energy• Radiant energy• Chemical energy (diesel fuel, batteries)• Nuclear energy• Heat energy
As energy can exist in various forms, it is an essential to know how energytransformation occurs within a system.
24
5.3 The Ship as a closed system
In all studies in this thesis, the ship is considered to be the control volume, withthe hull as its boundaries. A control volume in thermodynamics is defined asa region in space wherein mass may flow. The heat flows and mass leavingthe system are a negative mass or energy balance [35].
5.4 Where is the waste heat coming from?
All energy conversions are less than 100 % efficient, and the energy that isnot used for anything purposeful is called waste heat, which can be utilised.For instance, the heat released from a diesel engine can be utilised to heat thefuel, cabins or cargo, or can be converted back to mechanical energy with anORC, at which point it called waste heat recovery (WHR).
In Papers I and V, we considered the recovery of energy from the waste heatin the exhaust gases, which is at a higher temperature (197 ◦C to 417 ◦C) thanthe engine jacket cooling water (95 ◦C). The conversion efficiency using anORC would be significantly lower utilising the lower temperature.
As stated by the first law of thermodynamics, in a control volume Eq. 5.1, thetotal energy input (
∑in �Hin) equals the total energy output (
∑out �Hout). Broken
down into fractions, as seen in Eq. 5.2, the energy input into the enginesis in fuel ( �Hfuel) and air ( �Hair), which equals usable work for either heating( �Qheating), or electricity and propulsion power ( �Wel).
∑in
�Hin =∑out
�Hout (5.1)
�Hfuel + �Hair =∑waste
�Hwaste + �Wel + �Qheating (5.2)
In Paper VI, the full energy system of M/S Birka Stockholm was analysedduring a year of operation and is presented in Figure 5.3, page 29. In thefigure, the green represents the electric and mechanical energy, which isused for propulsion and for all electrical appliances on board. The yellowrepresents the exergy (described more in Section 5.8) in the exhaust gas. The
25
5.1 A well known but abstract concept: Energy
The scientific basis for all calculations and results is the laws of thermody-namics [34]. The simple question, ‘How can I make my ship consume lessfuel’, has a more complex answer from the energy perspective. The conceptof energy is abstract, which has salience in calculations but cannot be touched,tasted, felt or seen.
5.2 First Law - Energy conservation
A Swedish television programme for children called Tippen (the Swedishword for the waste dump) ran during the summer of 1994. Every episodestarted with the catchy slogan ‘Ingenting försvinner, allt finns kvar’, whichmeans ‘Nothing is disappearing, everything is still there’. This is a usefulcatchphrase not only for objects but also for energy. Energy is conserved in allprocesses, and it cannot by definition be destroyed. All energy which enters asystem, regardless of the form of fuel, electricity or heat, must all add up to atotal which is constant.
Energy can exist in various forms:
• Mechanical energy• Electrical energy• Magnetic energy• Radiant energy• Chemical energy (diesel fuel, batteries)• Nuclear energy• Heat energy
As energy can exist in various forms, it is an essential to know how energytransformation occurs within a system.
24
xii Introduction: A Century of Diesel Progress
trunk piston engines supplied by Sulzer. Each 310 mm bore/460 mm stroke engine delivered 280 kW at 250 rev/min.
The year 1910 also saw the single-screw 1179 dwt Anglo-Saxon tanker Vulcanus enter service powered by a 370 kW Werkspoor six-cylinder four-stroke crosshead engine with a 400 mm bore/600 mm stroke. The Dutch-built vessel was reportedly the first oceangoing motor ship to receive classification from Lloyd’s Register.
In 1911 the Swan Hunter-built 2600 dwt Great Lakes vessel Toiler crossed the Atlantic with propulsion by two 132 kW Swedish Polar engines. Krupp’s first marine diesel engines, six-cylinder 450 mm bore/800 mm stroke units developing 920 kW at 140 rev/min apiece, were installed the same year in the twin-screw 8000 dwt tankers Hagen and Loki built for the German subsidiary of the Standard Oil Co. of New Jersey.
FIGURE I.3 Main lines of development for direct-drive low-speed enginesFigure 5.1: Evolution of low speed engines [38]
a turbocharged engine involves the exhaust gases. Depending on the enginesize, revolutions per minute and engine design, the fractions may be different,but the most efficient diesel engines are about 50 % effective, which meanshalf of the chemical energy from the fuel is not utilised in mechanical work.If this energy is not being used for anything useful, then it is considered wasteheat and is therefore a loss. For instance, a two-stroke MAN diesel enginehas an efficiency of 50 %, and even with this relatively high efficiency, alot of wasted energy could still be utilised for mechanical work. Figure 5.4
27
cooling flows are represented by orange, and red represents the energy losses,which are released into the atmosphere either by the heat from the exhaustgases or sea-water cooling. The Sankey diagram of the energy flows canprovide a helpful indication as to where the losses are but do not give a fullpicture, as it might not always be possible to utilise heat sources with too lowa temperature.
5.5 Fuel to power - The Diesel engine
When the chemical bonds in fuel react with air under both high pressureand temperature, heat is released and new chemical bonds form. A largetwo-stroke slow-speed diesel engine is the most efficient machine for marinepropulsion and power about 96 % of the worlds ships above 100 GT [36].These engines can run directly on less expensive heavy fuel oil (HFO), whichis a low-grade residual fuel (meaning it is not distilled and dirty). An engine isconsidered a low-speed engine if the rpm is below 300. The relatively slowerrotation speeds and large engine sizes (the largest diesel engine is the SulzerRT-flex 96 has a dry weight of over 2300 t). Medium-speed diesel enginesare generally not as efficient as the largest two-stroke engines, but the newestmedium-speed four-stroke engines are closing the gap. The Wärsilä 31 engine,achieved a world record in 2015 as the most efficient four-stroke engine, ratedspecific fuel oil consumption (SFOC) at 165 g/kWh [37]. As shown in Eq.5.3, this corresponds to a thermal efficiency of 51 %.
ηthermal =3600
42.7 · SFOC(5.3)
The diesel engine has evolved significantly in efficiency since the beginningof the twentieth century, from thermal efficiencies of 25 % to 30 % to thecurrent 50 %. As seen in Figure 5.1, the turbo-charged two-stroke engineboosted the brake mean effective pressure (BMEP) significantly, which led tohigher efficiencies. Currently, exhaust gases are not only utilised to compresscylinder inlet air but can also power a generator turbine, which further increasesefficiency.
Though the diesel engine is more efficient than an otto-engine or a gas turbine,a substantial amount of heat is nevertheless released into the surroundings.As demonstrated in Figure 5.2, the primary part of the waste heat energy for
26
xii Introduction: A Century of Diesel Progress
trunk piston engines supplied by Sulzer. Each 310 mm bore/460 mm stroke engine delivered 280 kW at 250 rev/min.
The year 1910 also saw the single-screw 1179 dwt Anglo-Saxon tanker Vulcanus enter service powered by a 370 kW Werkspoor six-cylinder four-stroke crosshead engine with a 400 mm bore/600 mm stroke. The Dutch-built vessel was reportedly the first oceangoing motor ship to receive classification from Lloyd’s Register.
In 1911 the Swan Hunter-built 2600 dwt Great Lakes vessel Toiler crossed the Atlantic with propulsion by two 132 kW Swedish Polar engines. Krupp’s first marine diesel engines, six-cylinder 450 mm bore/800 mm stroke units developing 920 kW at 140 rev/min apiece, were installed the same year in the twin-screw 8000 dwt tankers Hagen and Loki built for the German subsidiary of the Standard Oil Co. of New Jersey.
FIGURE I.3 Main lines of development for direct-drive low-speed enginesFigure 5.1: Evolution of low speed engines [38]
a turbocharged engine involves the exhaust gases. Depending on the enginesize, revolutions per minute and engine design, the fractions may be different,but the most efficient diesel engines are about 50 % effective, which meanshalf of the chemical energy from the fuel is not utilised in mechanical work.If this energy is not being used for anything useful, then it is considered wasteheat and is therefore a loss. For instance, a two-stroke MAN diesel enginehas an efficiency of 50 %, and even with this relatively high efficiency, alot of wasted energy could still be utilised for mechanical work. Figure 5.4
27
cooling flows are represented by orange, and red represents the energy losses,which are released into the atmosphere either by the heat from the exhaustgases or sea-water cooling. The Sankey diagram of the energy flows canprovide a helpful indication as to where the losses are but do not give a fullpicture, as it might not always be possible to utilise heat sources with too lowa temperature.
5.5 Fuel to power - The Diesel engine
When the chemical bonds in fuel react with air under both high pressureand temperature, heat is released and new chemical bonds form. A largetwo-stroke slow-speed diesel engine is the most efficient machine for marinepropulsion and power about 96 % of the worlds ships above 100 GT [36].These engines can run directly on less expensive heavy fuel oil (HFO), whichis a low-grade residual fuel (meaning it is not distilled and dirty). An engine isconsidered a low-speed engine if the rpm is below 300. The relatively slowerrotation speeds and large engine sizes (the largest diesel engine is the SulzerRT-flex 96 has a dry weight of over 2300 t). Medium-speed diesel enginesare generally not as efficient as the largest two-stroke engines, but the newestmedium-speed four-stroke engines are closing the gap. The Wärsilä 31 engine,achieved a world record in 2015 as the most efficient four-stroke engine, ratedspecific fuel oil consumption (SFOC) at 165 g/kWh [37]. As shown in Eq.5.3, this corresponds to a thermal efficiency of 51 %.
ηthermal =3600
42.7 · SFOC(5.3)
The diesel engine has evolved significantly in efficiency since the beginningof the twentieth century, from thermal efficiencies of 25 % to 30 % to thecurrent 50 %. As seen in Figure 5.1, the turbo-charged two-stroke engineboosted the brake mean effective pressure (BMEP) significantly, which led tohigher efficiencies. Currently, exhaust gases are not only utilised to compresscylinder inlet air but can also power a generator turbine, which further increasesefficiency.
Though the diesel engine is more efficient than an otto-engine or a gas turbine,a substantial amount of heat is nevertheless released into the surroundings.As demonstrated in Figure 5.2, the primary part of the waste heat energy for
26
Propeller shaft
Shaft losses
Losses
Losses
Losses
AG losses
Losses
Losses
ThrustersHVAC
HRSG
HRSG
Cylinder
Turbine
Cylinder
LOC
JWC
HTC
LTC
Compressor
Bypass valve
CAC-HT
CAC-LT
Compressor
CAC-HT
CAC-LT
LOC
JWC
Turbine
Boiler
Auxiliary boiler
Environment
Environment
0.8
25
0.4
Cooling water5.5
Exhausts2.8
Exhausts2.2
1.3
64
3.2
0.21.0
HRHT
Exhausts3.5
2.3
0.5
0.8
Fuel67
Fuel43
Fuel5
Switchboard Others
Propeller
Preheater
Reheater
Hot water
Machinery space heaters
HFO tank heating
Tank heatingGalley
Other tanks
1.0
Figure 5.3: Sankey diagram of the energy flows in M/S Birka, Paper VI. Flow values are in GWh/year.
29
Efficiency
5
Naturallyaspirated engines
100% Fuel40%To work
100% Fuel37%To exhaust
Turbocharged engines
5% Radiation10%
Chargeair
15%5% To coolant
5% Lub oil
10% Coolant50%
Exhaustto turbine
35–40%Exhaust from turbine
5%Chargecooler
Heat fromwalls
Recirc
To lub oil8% Radiation
35% To work
Pumping
FIGURE 1.4 Typical Sankey diagrams
Figure 5.2: Typical Sankey-diagram of a turbocharged engine [38]
28
Propeller shaft
Shaft losses
Losses
Losses
Losses
AG losses
Losses
Losses
ThrustersHVAC
HRSG
HRSG
Cylinder
Turbine
Cylinder
LOC
JWC
HTC
LTC
Compressor
Bypass valve
CAC-HT
CAC-LT
Compressor
CAC-HT
CAC-LT
LOC
JWC
Turbine
Boiler
Auxiliary boiler
Environment
Environment
0.8
25
0.4
Cooling water5.5
Exhausts2.8
Exhausts2.2
1.3
64
3.2
0.21.0
HRHT
Exhausts3.5
2.3
0.5
0.8
Fuel67
Fuel43
Fuel5
Switchboard Others
Propeller
Preheater
Reheater
Hot water
Machinery space heaters
HFO tank heating
Tank heatingGalley
Other tanks
1.0
Figure 5.3: Sankey diagram of the energy flows in M/S Birka, Paper VI. Flow values are in GWh/year.
29
Efficiency
5
Naturallyaspirated engines
100% Fuel40%To work
100% Fuel37%To exhaust
Turbocharged engines
5% Radiation10%
Chargeair
15%5% To coolant
5% Lub oil
10% Coolant50%
Exhaustto turbine
35–40%Exhaust from turbine
5%Chargecooler
Heat fromwalls
Recirc
To lub oil8% Radiation
35% To work
Pumping
FIGURE 1.4 Typical Sankey diagrams
Figure 5.2: Typical Sankey-diagram of a turbocharged engine [38]
28
5W
aste
Hea
t R
ecov
ery
Sys
tem
(WH
RS
) for
Red
uctio
n of
Fue
l Con
sum
ptio
n, E
mis
sion
and
EE
DI
5
Was
te H
eat R
ecov
ery
Syst
em (W
HRS)
fo
r Red
uctio
n of
Fue
l Con
sum
ptio
n, E
mis
sion
and
EED
I
Sum
mar
y
The
incr
easi
ng i
nter
est
in e
mis
sion
re-
duct
ion,
shi
p op
erat
ing
cost
s re
duct
ion
and
the
new
ly a
dapt
ed IM
O E
ED
I rul
es
calls
for
mea
sure
s th
at e
nsur
e op
timal
utili
satio
n of
the
fue
l use
d fo
r m
ain
en-
gine
s on
boa
rd s
hips
.
Mai
n en
gine
exh
aust
gas
ene
rgy
is b
y
far t
he m
ost a
ttra
ctiv
e am
ong
the
was
te
heat
sou
rces
of
a sh
ip b
ecau
se o
f th
e
heat
flo
w a
nd t
empe
ratu
re.
It is
pos
si-
ble
to g
ener
ate
an e
lect
rical
out
put
of
up t
o 11
% o
f th
e m
ain
engi
ne p
ower
by u
tilis
ing
this
exh
aust
gas
ene
rgy
in
a w
aste
hea
t re
cove
ry s
yste
m c
ompr
is-
ing
both
st
eam
an
d po
wer
tu
rbin
es,
and
com
bine
d w
ith u
tilis
ing
scav
enge
air
ener
gy fo
r ex
haus
t bo
iler
feed
-wat
er
heat
ing.
This
pa
per
desc
ribes
th
e te
chno
logy
behi
nd
was
te
heat
re
cove
ry
and
the
pote
ntia
l fo
r sh
ip-o
wne
rs t
o lo
wer
fue
l
cost
s, c
ut e
mis
sion
s, a
nd t
he e
ffect
on
the
EE
DI o
f the
shi
p.
Intr
oduc
tion
Follo
win
g th
e tr
end
of a
req
uire
d hi
gher
over
all s
hip
effic
ienc
y si
nce
the
first
oil
cris
is in
197
3, th
e ef
ficie
ncy
of m
ain
en-
gine
s ha
s in
crea
sed,
and
tod
ay t
he fu
el
ener
gy e
ffici
ency
is
abou
t 50
%.
This
high
effi
cien
cy h
as, a
mon
g ot
her t
hing
s,
led
to lo
w S
FOC
val
ues,
but
als
o a
cor-
resp
ondi
ngly
low
er e
xhau
st g
as t
em-
pera
ture
afte
r th
e tu
rboc
harg
ers.
Eve
n th
ough
a m
ain
engi
ne f
uel e
nerg
y
effic
ienc
y of
50%
is
rela
tivel
y hi
gh,
the
prim
ary
obje
ctiv
e fo
r th
e sh
ip-o
wne
r is
still
to
low
er s
hip
oper
atio
nal c
osts
fur
-
ther
, as
the
tot
al f
uel
cons
umpt
ion
of
the
ship
is s
till t
he m
ain
targ
et. T
his
may
lead
to
a fu
rthe
r re
duct
ion
of C
O2
emis
-
sion
s –
a ta
sk,
whi
ch i
s ge
ttin
g ev
en
mor
e im
port
ant
with
the
new
IMO
EE
DI
rule
s in
pla
ce fr
om 2
013.
The
prim
ary
sour
ce o
f w
aste
hea
t of
a
mai
n en
gine
is th
e ex
haus
t gas
hea
t dis
-
sipa
tion,
whi
ch a
ccou
nts
for
abou
t ha
lf
of t
he t
otal
was
te h
eat,
i.e.
abo
ut 2
5%
of t
he t
otal
fue
l ene
rgy.
In t
he s
tand
ard
high
-effi
cien
cy e
ngin
e ve
rsio
n, t
he e
x-
haus
t ga
s te
mpe
ratu
re is
rel
ativ
ely
low
afte
r th
e tu
rboc
harg
er,
and
just
hig
h
enou
gh
for
prod
ucin
g th
e ne
cess
ary
stea
m f
or t
he h
eatin
g pu
rpos
es o
f th
e
ship
by
mea
ns o
f a
stan
dard
exh
aust
gas
fired
boi
ler
of t
he s
mok
e tu
be d
e-
sign
.
How
ever
, th
e M
AN
B
&W
tw
o-st
roke
ME
mai
n en
gine
tun
ed f
or W
HR
S w
ill
incr
ease
the
pos
sibi
litie
s of
pro
duci
ng
elec
tric
ity f
rom
the
exh
aust
gas
. Th
e
resu
lt w
ill b
e an
impr
ovem
ent i
n to
tal e
f-
ficie
ncy
but
a sl
ight
red
uctio
n of
the
ef-
ficie
ncy
of th
e m
ain
engi
ne w
ill b
e se
en.
Fig.
1 sh
ows
a co
mpa
rison
of
engi
ne
heat
bal
ance
s, w
ith a
nd w
ithou
t WH
RS
.
The
figur
e sh
ows
that
for
the
eng
ine
in
com
bina
tion
with
WH
RS
the
tot
al e
ffi-
cien
cy w
ill in
crea
se t
o ab
out
55%
.
The
IMO
EE
DI
form
ula
allo
ws
for
con-
side
ring
addi
ng W
HR
S i
nto
the
ship
,
anal
yse
EE
DI e
ffect
s an
d E
ED
I set
tings
.
As
an e
ven
low
er C
O2
emis
sion
lev
el
can
be a
chie
ved
by i
nsta
lling
a w
aste
heat
rec
over
y sy
stem
the
EE
DI,
whi
ch
is a
mea
sure
for
CO
2 em
issi
ons,
will
also
be
low
ered
.
Fuel
100
%(1
68.7
g/k
Wh)
Heat
radi
atio
n
0.6%
Air c
oole
r
14.2
%
Exha
ust g
as a
nd c
onde
nser
22.9
% (2
2.3%
)
Jack
et w
ater
cool
er 5
.2%
Lubr
icat
ing
oil
cool
er 2
.9%
Elec
tric
prod
uctio
n of
WHR
S 5.
1% (5
.7%
)
Gain
= 1
0.4%
(11.
6%)
Tota
l pow
er o
utpu
t 54.
3% (5
5.0%
)
Shaf
t pow
erOu
tput
49.
1%
12S9
0ME-
C9.2
eng
ine
for W
HRS
SMCR
: 69,
720
kW a
t 84
rpm
ISO
ambi
ent r
efer
ence
con
ditio
ns
WHR
S: s
ingl
e pr
essu
re (D
ual p
ress
ure)
Shaf
t pow
erOu
tput
49.
3%
Lubr
icat
ing
oil
cool
er 2
.9%
Jack
et w
ater
cool
er 5
.2%
Exha
ust g
as
25.5
%
Air c
oole
r
16.5
%
Heat
radi
atio
n
0.6%
Fuel
100
%(1
67 g
/kW
h)
12S9
0ME-
C9.2
sta
ndar
d en
gine
SMCR
: 69,
720
kW a
t 84
rpm
ISO
ambi
ent r
efer
ence
con
ditio
ns
Fig.
1: H
eat
bal
ance
for
larg
e-b
ore
MA
N B
&W
eng
ine
typ
es w
ithou
t an
d w
ith W
HR
S
Figure 5.5: Sankey diagram for a MAN 12S90ME engine with and without WHR
31
presents the heat balance of a MAN 12K98ME/MC Mk 6 engine, a largelow rpm two-stroke engine used for propulsion in ships. The engine outputs49.3 % of the released chemical energy to mechanical shaft power, but therest (50.7 %) is released into the atmosphere as exhaust gases, or is cooledby sea water. The exhaust gases are often in the range of 200 ◦C to 250 ◦Cfor a two-stroke engine, heat which could be used to either drive a Rankinecycle or an exhaust turbine. In this way, system efficiency could be increasedby approximately 5 %, as demonstrated in Figure 5.5, which demonstrates acomparison between the same MAN 12S90ME engine fitted with and withouta waste heat recovery device.
Lund University / LTH / Energy Sciences / TPE / Magnus Genrup /2017-07-26 43
Some diesel engine technology…Two-stroke low-speed
Courtesy of MAN B&W
Figure 5.4: Sankey diagram for a MAN 12K98ME/MC engine
A four-stroke engine generally has a higher exhaust gas temperature due to lessscavenging air and is also generally not as efficient as the largest two-strokeengines due to the higher rpm. The M/S Birka, as presented in Section 3.2,has only four-stroke engines both for propulsion and electricity production,which indicates a slightly higher amount of waste heat. The drawback offitting a WHR-device on a smaller engine is not only the added complexity ofspace constraints but also the fact that a more dynamic load condition makesdesigning and operating an ORC-device more difficult. This was discussed in
30
5W
aste
Hea
t R
ecov
ery
Sys
tem
(WH
RS
) for
Red
uctio
n of
Fue
l Con
sum
ptio
n, E
mis
sion
and
EE
DI
5
Was
te H
eat R
ecov
ery
Syst
em (W
HRS)
fo
r Red
uctio
n of
Fue
l Con
sum
ptio
n, E
mis
sion
and
EED
I
Sum
mar
y
The
incr
easi
ng i
nter
est
in e
mis
sion
re-
duct
ion,
shi
p op
erat
ing
cost
s re
duct
ion
and
the
new
ly a
dapt
ed IM
O E
ED
I rul
es
calls
for
mea
sure
s th
at e
nsur
e op
timal
utili
satio
n of
the
fue
l use
d fo
r m
ain
en-
gine
s on
boa
rd s
hips
.
Mai
n en
gine
exh
aust
gas
ene
rgy
is b
y
far t
he m
ost a
ttra
ctiv
e am
ong
the
was
te
heat
sou
rces
of
a sh
ip b
ecau
se o
f th
e
heat
flo
w a
nd t
empe
ratu
re.
It is
pos
si-
ble
to g
ener
ate
an e
lect
rical
out
put
of
up t
o 11
% o
f th
e m
ain
engi
ne p
ower
by u
tilis
ing
this
exh
aust
gas
ene
rgy
in
a w
aste
hea
t re
cove
ry s
yste
m c
ompr
is-
ing
both
st
eam
an
d po
wer
tu
rbin
es,
and
com
bine
d w
ith u
tilis
ing
scav
enge
air
ener
gy fo
r ex
haus
t bo
iler
feed
-wat
er
heat
ing.
This
pa
per
desc
ribes
th
e te
chno
logy
behi
nd
was
te
heat
re
cove
ry
and
the
pote
ntia
l fo
r sh
ip-o
wne
rs t
o lo
wer
fue
l
cost
s, c
ut e
mis
sion
s, a
nd t
he e
ffect
on
the
EE
DI o
f the
shi
p.
Intr
oduc
tion
Follo
win
g th
e tr
end
of a
req
uire
d hi
gher
over
all s
hip
effic
ienc
y si
nce
the
first
oil
cris
is in
197
3, th
e ef
ficie
ncy
of m
ain
en-
gine
s ha
s in
crea
sed,
and
tod
ay t
he fu
el
ener
gy e
ffici
ency
is
abou
t 50
%.
This
high
effi
cien
cy h
as, a
mon
g ot
her t
hing
s,
led
to lo
w S
FOC
val
ues,
but
als
o a
cor-
resp
ondi
ngly
low
er e
xhau
st g
as t
em-
pera
ture
afte
r th
e tu
rboc
harg
ers.
Eve
n th
ough
a m
ain
engi
ne f
uel e
nerg
y
effic
ienc
y of
50%
is
rela
tivel
y hi
gh,
the
prim
ary
obje
ctiv
e fo
r th
e sh
ip-o
wne
r is
still
to
low
er s
hip
oper
atio
nal c
osts
fur
-
ther
, as
the
tot
al f
uel
cons
umpt
ion
of
the
ship
is s
till t
he m
ain
targ
et. T
his
may
lead
to
a fu
rthe
r re
duct
ion
of C
O2
emis
-
sion
s –
a ta
sk,
whi
ch i
s ge
ttin
g ev
en
mor
e im
port
ant
with
the
new
IMO
EE
DI
rule
s in
pla
ce fr
om 2
013.
The
prim
ary
sour
ce o
f w
aste
hea
t of
a
mai
n en
gine
is th
e ex
haus
t gas
hea
t dis
-
sipa
tion,
whi
ch a
ccou
nts
for
abou
t ha
lf
of t
he t
otal
was
te h
eat,
i.e.
abo
ut 2
5%
of t
he t
otal
fue
l ene
rgy.
In t
he s
tand
ard
high
-effi
cien
cy e
ngin
e ve
rsio
n, t
he e
x-
haus
t ga
s te
mpe
ratu
re is
rel
ativ
ely
low
afte
r th
e tu
rboc
harg
er,
and
just
hig
h
enou
gh
for
prod
ucin
g th
e ne
cess
ary
stea
m f
or t
he h
eatin
g pu
rpos
es o
f th
e
ship
by
mea
ns o
f a
stan
dard
exh
aust
gas
fired
boi
ler
of t
he s
mok
e tu
be d
e-
sign
.
How
ever
, th
e M
AN
B
&W
tw
o-st
roke
ME
mai
n en
gine
tun
ed f
or W
HR
S w
ill
incr
ease
the
pos
sibi
litie
s of
pro
duci
ng
elec
tric
ity f
rom
the
exh
aust
gas
. Th
e
resu
lt w
ill b
e an
impr
ovem
ent i
n to
tal e
f-
ficie
ncy
but
a sl
ight
red
uctio
n of
the
ef-
ficie
ncy
of th
e m
ain
engi
ne w
ill b
e se
en.
Fig.
1 sh
ows
a co
mpa
rison
of
engi
ne
heat
bal
ance
s, w
ith a
nd w
ithou
t WH
RS
.
The
figur
e sh
ows
that
for
the
eng
ine
in
com
bina
tion
with
WH
RS
the
tot
al e
ffi-
cien
cy w
ill in
crea
se t
o ab
out
55%
.
The
IMO
EE
DI
form
ula
allo
ws
for
con-
side
ring
addi
ng W
HR
S i
nto
the
ship
,
anal
yse
EE
DI e
ffect
s an
d E
ED
I set
tings
.
As
an e
ven
low
er C
O2
emis
sion
lev
el
can
be a
chie
ved
by i
nsta
lling
a w
aste
heat
rec
over
y sy
stem
the
EE
DI,
whi
ch
is a
mea
sure
for
CO
2 em
issi
ons,
will
also
be
low
ered
.
Fuel
100
%(1
68.7
g/k
Wh)
Heat
radi
atio
n
0.6%
Air c
oole
r
14.2
%
Exha
ust g
as a
nd c
onde
nser
22.9
% (2
2.3%
)
Jack
et w
ater
cool
er 5
.2%
Lubr
icat
ing
oil
cool
er 2
.9%
Elec
tric
prod
uctio
n of
WHR
S 5.
1% (5
.7%
)
Gain
= 1
0.4%
(11.
6%)
Tota
l pow
er o
utpu
t 54.
3% (5
5.0%
)
Shaf
t pow
erOu
tput
49.
1%
12S9
0ME-
C9.2
eng
ine
for W
HRS
SMCR
: 69,
720
kW a
t 84
rpm
ISO
ambi
ent r
efer
ence
con
ditio
ns
WHR
S: s
ingl
e pr
essu
re (D
ual p
ress
ure)
Shaf
t pow
erOu
tput
49.
3%
Lubr
icat
ing
oil
cool
er 2
.9%
Jack
et w
ater
cool
er 5
.2%
Exha
ust g
as
25.5
%
Air c
oole
r
16.5
%
Heat
radi
atio
n
0.6%
Fuel
100
%(1
67 g
/kW
h)
12S9
0ME-
C9.2
sta
ndar
d en
gine
SMCR
: 69,
720
kW a
t 84
rpm
ISO
ambi
ent r
efer
ence
con
ditio
ns
Fig.
1: H
eat
bal
ance
for
larg
e-b
ore
MA
N B
&W
eng
ine
typ
es w
ithou
t an
d w
ith W
HR
S
Figure 5.5: Sankey diagram for a MAN 12S90ME engine with and without WHR
31
presents the heat balance of a MAN 12K98ME/MC Mk 6 engine, a largelow rpm two-stroke engine used for propulsion in ships. The engine outputs49.3 % of the released chemical energy to mechanical shaft power, but therest (50.7 %) is released into the atmosphere as exhaust gases, or is cooledby sea water. The exhaust gases are often in the range of 200 ◦C to 250 ◦Cfor a two-stroke engine, heat which could be used to either drive a Rankinecycle or an exhaust turbine. In this way, system efficiency could be increasedby approximately 5 %, as demonstrated in Figure 5.5, which demonstrates acomparison between the same MAN 12S90ME engine fitted with and withouta waste heat recovery device.
Lund University / LTH / Energy Sciences / TPE / Magnus Genrup /2017-07-26 43
Some diesel engine technology…Two-stroke low-speed
Courtesy of MAN B&W
Figure 5.4: Sankey diagram for a MAN 12K98ME/MC engine
A four-stroke engine generally has a higher exhaust gas temperature due to lessscavenging air and is also generally not as efficient as the largest two-strokeengines due to the higher rpm. The M/S Birka, as presented in Section 3.2,has only four-stroke engines both for propulsion and electricity production,which indicates a slightly higher amount of waste heat. The drawback offitting a WHR-device on a smaller engine is not only the added complexity ofspace constraints but also the fact that a more dynamic load condition makesdesigning and operating an ORC-device more difficult. This was discussed in
30
Figure 5.6: Temperature entropy diagram of Rankine Cycle (CC-BY Marcus Thern).
This means that even though much of the heat is not used for the process (andis therefore wasted), depending on the temperature of the source, it mightnot be feasible for use in a thermodynamic cycle. For example if the heatsource is not of high enough temperature it might not be feasible to use. If thecold reservoir is sea water with a temperature of 10 ◦C and the waste heat isengine jacket cooling water with a temperature of 90 ◦C, then the theoreticalCarnot-efficiency is 22 %, as stated in Eq. 5.5. The theoretical efficiency isthe upper limit, and not including the efficiency of components and frictionlosses, which in this case could be substantial to the overall efficiency.
η = 1 − 10 + 27390 + 273
= 0.22 (5.5)
By adding heat to the boiler and constantly adding new water as water boilsoff as steam, the temperature remain constant in the boiler. There are differenttypes of steam boilers, and the simplest and earliest form is a kettle-type
33
Paper I, ‘Waste Heat Recovery in a Cruise Vessel in the Baltic Sea by Usingan Organic Rankine Cycle: A Case Study’ [39].
5.6 Heat to power - The Rankine cycle
The Rankine cycle is the thermodynamic cycle wherein water is evaporatedin a boiler by a heat source, and then the steam powers a turbine to producemechanical power. As seen in Figure 5.6, between points 1 and 2, condensedwater is pumped into the boiler, at which point heat is added (Qin) and thewater is evaporated into steam between points 3 and 4. The steam expandsinto a turbine where mechanical work is extracted (Wturbine). Between points4 and 5 (Wout), the steam is condensed in a cooler. In this simple Rankinecycle, the steam is not superheated, which means it condenses in the turbine.A superheated cycle is where additional heat is added after the evaporation.A superheated cycle has a greater efficiency due to a higher cycle averagetemperature, as well as the fact that condensation does not occur in the expander(turbine). The cycle is named after William John Macquorn Rankine, oneof the founding fathers of the scientific theory of thermodynamics in themid-nineteenth century. This cycle currently produces about three quarters ofall the world’s electricity, as both coal and nuclear power plants are based onthe principle of boiling water to drive a generator turbine [40].
The obvious advantage of using water as the working medium is that it is non-toxic to both humans and the environment and is easily available everywhere.Water has also the advantage of having a relatively high heat capacity of4.2 kJ kg−1 K−1 and a high enthalpy of vaporisation (also known as the latentheat) of 2257 kJ kg−1 at a standard atmosphere. A higher heat capacity meansless mass flow is needed to transport the same amount of power. To boilwater at a normal atmospheric pressure requires a temperature of 100 ◦C, butthis creates only saturated steam, and as the steam expands in the turbine,it condenses into a liquid, which can damage the turbine. The maximumefficiency of a thermodynamic power cycle transforming into mechanical workheat between two reservoirs is stated by the Carnot theorem in Eq. 5.4 whichthat efficiency can be maximised by using higher temperature differences.
η = 1 − TCTH
(5.4)
32
Figure 5.6: Temperature entropy diagram of Rankine Cycle (CC-BY Marcus Thern).
This means that even though much of the heat is not used for the process (andis therefore wasted), depending on the temperature of the source, it mightnot be feasible for use in a thermodynamic cycle. For example if the heatsource is not of high enough temperature it might not be feasible to use. If thecold reservoir is sea water with a temperature of 10 ◦C and the waste heat isengine jacket cooling water with a temperature of 90 ◦C, then the theoreticalCarnot-efficiency is 22 %, as stated in Eq. 5.5. The theoretical efficiency isthe upper limit, and not including the efficiency of components and frictionlosses, which in this case could be substantial to the overall efficiency.
η = 1 − 10 + 27390 + 273
= 0.22 (5.5)
By adding heat to the boiler and constantly adding new water as water boilsoff as steam, the temperature remain constant in the boiler. There are differenttypes of steam boilers, and the simplest and earliest form is a kettle-type
33
Paper I, ‘Waste Heat Recovery in a Cruise Vessel in the Baltic Sea by Usingan Organic Rankine Cycle: A Case Study’ [39].
5.6 Heat to power - The Rankine cycle
The Rankine cycle is the thermodynamic cycle wherein water is evaporatedin a boiler by a heat source, and then the steam powers a turbine to producemechanical power. As seen in Figure 5.6, between points 1 and 2, condensedwater is pumped into the boiler, at which point heat is added (Qin) and thewater is evaporated into steam between points 3 and 4. The steam expandsinto a turbine where mechanical work is extracted (Wturbine). Between points4 and 5 (Wout), the steam is condensed in a cooler. In this simple Rankinecycle, the steam is not superheated, which means it condenses in the turbine.A superheated cycle is where additional heat is added after the evaporation.A superheated cycle has a greater efficiency due to a higher cycle averagetemperature, as well as the fact that condensation does not occur in the expander(turbine). The cycle is named after William John Macquorn Rankine, oneof the founding fathers of the scientific theory of thermodynamics in themid-nineteenth century. This cycle currently produces about three quarters ofall the world’s electricity, as both coal and nuclear power plants are based onthe principle of boiling water to drive a generator turbine [40].
The obvious advantage of using water as the working medium is that it is non-toxic to both humans and the environment and is easily available everywhere.Water has also the advantage of having a relatively high heat capacity of4.2 kJ kg−1 K−1 and a high enthalpy of vaporisation (also known as the latentheat) of 2257 kJ kg−1 at a standard atmosphere. A higher heat capacity meansless mass flow is needed to transport the same amount of power. To boilwater at a normal atmospheric pressure requires a temperature of 100 ◦C, butthis creates only saturated steam, and as the steam expands in the turbine,it condenses into a liquid, which can damage the turbine. The maximumefficiency of a thermodynamic power cycle transforming into mechanical workheat between two reservoirs is stated by the Carnot theorem in Eq. 5.4 whichthat efficiency can be maximised by using higher temperature differences.
η = 1 − TCTH
(5.4)
32
5.7 Low temperature heat to power - The organicRankine cycle
The organic Rankine cycle (ORC) is the same type of cycle as a Rankine cyclebased on water, the main difference being the working fluid, which is not waterbut rather an organic medium. An organic medium is one of a vast variety ofcarbon-based fluids, an advantage of which include a lower evaporating point,which can lead to higher efficiency than water within specific temperaturespans. There are ORC applications from a few kW up to several MW. Thetotal installed ORC power in the world as of 2016 is estimated to about 2.7GW distributed to 1,754 ORC units [42].
An ORC may be chosen instead of a steam Rankine cycle for temperaturesunder around 400 ◦C, above which temperature the efficiency of steam can behigher and the complex molecule chains in the organic fluid can be brokendown. Creating a steam cycle for smaller power applications is also morechallenging due to smaller mass flows, which makes it harder to design anefficient small turbine for steam as it must be lubricated [43].
The ORC fluid could either be heated directly by the heat source or via anintermediate loop of another medium, such as oil or fresh water. This wasdiscussed in Paper I, where an intermediate loop of freshwater was used dueto long piping arrangements and risk of corrosion of seawater [39]. Thedrawbacks of using an intermediate loop is both the added complexity ofthe system, as well as lower thermal efficiency due to the energy losses andirreversibility in the heat exchangers. It could, however, lead to less risk ofdegradation of the organic fluid (due to high temperatures) and dampening ofmass flow and temperature during dynamic conditions [44]. It must also beaccounted for which fuel is used for the heat exchanger at the exhaust gasses,as the sulphur in the fuel condenses at lower temperatures between 125 ◦C and140 ◦C [45]. As HFO contains an average of 2.7 % sulphur, and even thoughMDO or low sulphur HFO contains 0.1 % it is still a factor to consider [46].Therefore the limiting temperature of the exhaust gasses was set to 150 ◦C inPaper I [39].
The organic fluids are separated into two different categories depending on thefluid properties, namely, wet and dry fluids. As seen in Figure 5.8, a dry fluidhas a positive condensation slope on the right side of the temperature-entropydiagram, meaning that it does not need to be superheated to avoid condensation
35
boiler, wherein water is held inside a drum and heat is applied from below.The other type is a tubular boiler, which circulates water in tubes to whichheat is applied. As seen in Figure 5.7, this type of boiler can either be (a) apartial form of steam generation or (b) a complete form of steam generation.
1-3Steam 42 / Steam Generation – An Overview
The Babcock & Wilcox Company
(see Fig. 7a), or fixed steam-water separation point, and those that do not (see Fig. 7b), identified as once-through steam generators (OTSG).
The most common and simplest to control is the steam drum system. In this system, the drum serves as the point of separation of steam from water throughout the boiler’s load range. Subcooled water (less than boiling temperature) enters the tube to which heat is applied. As the water flows through the tube, it is heated to the boiling point, bubbles are formed, and wet steam is generated. In most boilers, a steam-water mixture leaves the tube and enters the steam drum, where steam is separated from water. The remaining water is then mixed with the replacement water and returned to the heated tube.
Without a steam drum, i.e., for an OTSG system, subcooled water also enters the tube to which heat is ap-plied, but the flowing water turns into steam somewhere along the flow path (length of tube), dependent upon water flow rate and heat input rates. Shown in Fig. 7b, the flow rate and heat input are closely controlled and coordinated so that all of the water is evaporated and only steam leaves the tube. There is no need for the steam drum (fixed steam-water separation point).
CirculationFor both types of boiling systems described above,
water must continuously pass through, or circulate through, the tubes for the system to generate steam con-tinuously. For an OTSG, water makes one pass through the boiler’s tubes before becoming steam to be sent to the turbine-generator. However, for those boilers with a fixed steam-water separation point or steam drum, a molecule of water can make many passes through a circulation loop before it leaves as steam to the turbine-generator. Options for this latter system are shown in Fig. 8.
Two different approaches to circulation are com-monly used: natural or thermal circulation, and forced or pumped circulation. Natural circulation is illustrated in Fig. 8a. In the downcomer, unheated tube segment A-B, no steam is present. Heat addition generates a steam-water mixture in segment B-C. Because the steam and steam-water mixture in segment B-C are less dense than the water segment A-B, gravity will cause the water to flow Fig. 6 Simple kettle boiler.
Fig. 7 Boiling process in tubular geometries.
Fig. 5 Coal-fired circulating fluidized-bed combustion steam generator.
RefractoryLine
AirHeater
Steam CoilAir Heater
FlueGas
Multi-CycloneDust
CollectorEconomizer
Superheater
Feedwaterto Drum
SecondaryAir Duct
PrimaryAir Duct
Air Duct toFluid Bed Cooler
AshRecycleSystem
GravimetricFeeder
FuelChute
Fluid BedCooler
Steam Drum
InternalEvaporativeCircuit
In-FurnaceU-Beams
Wing WallFuel Bunker
Downcomer
ExternalU-Beams
Figure 5.7: Boiler process, partial steam or once-through [41].
34
5.7 Low temperature heat to power - The organicRankine cycle
The organic Rankine cycle (ORC) is the same type of cycle as a Rankine cyclebased on water, the main difference being the working fluid, which is not waterbut rather an organic medium. An organic medium is one of a vast variety ofcarbon-based fluids, an advantage of which include a lower evaporating point,which can lead to higher efficiency than water within specific temperaturespans. There are ORC applications from a few kW up to several MW. Thetotal installed ORC power in the world as of 2016 is estimated to about 2.7GW distributed to 1,754 ORC units [42].
An ORC may be chosen instead of a steam Rankine cycle for temperaturesunder around 400 ◦C, above which temperature the efficiency of steam can behigher and the complex molecule chains in the organic fluid can be brokendown. Creating a steam cycle for smaller power applications is also morechallenging due to smaller mass flows, which makes it harder to design anefficient small turbine for steam as it must be lubricated [43].
The ORC fluid could either be heated directly by the heat source or via anintermediate loop of another medium, such as oil or fresh water. This wasdiscussed in Paper I, where an intermediate loop of freshwater was used dueto long piping arrangements and risk of corrosion of seawater [39]. Thedrawbacks of using an intermediate loop is both the added complexity ofthe system, as well as lower thermal efficiency due to the energy losses andirreversibility in the heat exchangers. It could, however, lead to less risk ofdegradation of the organic fluid (due to high temperatures) and dampening ofmass flow and temperature during dynamic conditions [44]. It must also beaccounted for which fuel is used for the heat exchanger at the exhaust gasses,as the sulphur in the fuel condenses at lower temperatures between 125 ◦C and140 ◦C [45]. As HFO contains an average of 2.7 % sulphur, and even thoughMDO or low sulphur HFO contains 0.1 % it is still a factor to consider [46].Therefore the limiting temperature of the exhaust gasses was set to 150 ◦C inPaper I [39].
The organic fluids are separated into two different categories depending on thefluid properties, namely, wet and dry fluids. As seen in Figure 5.8, a dry fluidhas a positive condensation slope on the right side of the temperature-entropydiagram, meaning that it does not need to be superheated to avoid condensation
35
boiler, wherein water is held inside a drum and heat is applied from below.The other type is a tubular boiler, which circulates water in tubes to whichheat is applied. As seen in Figure 5.7, this type of boiler can either be (a) apartial form of steam generation or (b) a complete form of steam generation.
1-3Steam 42 / Steam Generation – An Overview
The Babcock & Wilcox Company
(see Fig. 7a), or fixed steam-water separation point, and those that do not (see Fig. 7b), identified as once-through steam generators (OTSG).
The most common and simplest to control is the steam drum system. In this system, the drum serves as the point of separation of steam from water throughout the boiler’s load range. Subcooled water (less than boiling temperature) enters the tube to which heat is applied. As the water flows through the tube, it is heated to the boiling point, bubbles are formed, and wet steam is generated. In most boilers, a steam-water mixture leaves the tube and enters the steam drum, where steam is separated from water. The remaining water is then mixed with the replacement water and returned to the heated tube.
Without a steam drum, i.e., for an OTSG system, subcooled water also enters the tube to which heat is ap-plied, but the flowing water turns into steam somewhere along the flow path (length of tube), dependent upon water flow rate and heat input rates. Shown in Fig. 7b, the flow rate and heat input are closely controlled and coordinated so that all of the water is evaporated and only steam leaves the tube. There is no need for the steam drum (fixed steam-water separation point).
CirculationFor both types of boiling systems described above,
water must continuously pass through, or circulate through, the tubes for the system to generate steam con-tinuously. For an OTSG, water makes one pass through the boiler’s tubes before becoming steam to be sent to the turbine-generator. However, for those boilers with a fixed steam-water separation point or steam drum, a molecule of water can make many passes through a circulation loop before it leaves as steam to the turbine-generator. Options for this latter system are shown in Fig. 8.
Two different approaches to circulation are com-monly used: natural or thermal circulation, and forced or pumped circulation. Natural circulation is illustrated in Fig. 8a. In the downcomer, unheated tube segment A-B, no steam is present. Heat addition generates a steam-water mixture in segment B-C. Because the steam and steam-water mixture in segment B-C are less dense than the water segment A-B, gravity will cause the water to flow Fig. 6 Simple kettle boiler.
Fig. 7 Boiling process in tubular geometries.
Fig. 5 Coal-fired circulating fluidized-bed combustion steam generator.
RefractoryLine
AirHeater
Steam CoilAir Heater
FlueGas
Multi-CycloneDust
CollectorEconomizer
Superheater
Feedwaterto Drum
SecondaryAir Duct
PrimaryAir Duct
Air Duct toFluid Bed Cooler
AshRecycleSystem
GravimetricFeeder
FuelChute
Fluid BedCooler
Steam Drum
InternalEvaporativeCircuit
In-FurnaceU-Beams
Wing WallFuel Bunker
Downcomer
ExternalU-Beams
Figure 5.7: Boiler process, partial steam or once-through [41].
34
As described by E. Macchi, the general requirements for choosing an organicfluid are the following [47]:
• Commercially available• Nonflammable• Nontoxic• Compatible with other materials• Environmental factors, global warming potential (GWP) and ozone
depletion potential (ODP)
0-8 8-10 10-12 12-14 14-16 16-21100
200
300
400
500
600
700
800
900
1000
Vessel speed
Wne
t / kW
S-tolueneav-S-tolueneS-c1cc6av-S-c1cc6S-ebenzeneav-S-ebenzeneR-benzeneav-R-benzeneR-dmcav-R-dmcR-cyclohexaneav-R-cyclohexane
Figure 5.9: Net power output versus vessel speed for the three most optimal fluids (s - Simple ORC, R - Regenerated,av - averaged over operating time) [23]
In Paper I, the sink used for the ORC is the sea water, and the most convenientinstallation location was determined to be at the top deck near the exhaustpipes. An intermediate loop between the sea water and the condenser was usedto keep the length of sea water piping length to a minimum given the effectsof corrosion and a requirement to keep the pipe clean. The condensationtemperature was kept at a constant 30 ◦C during all simulations.
The efficiency of the ORC was considered as the fraction of the workoutput (Wnet) from the turbine and heat input from preheater, evaporatorand superheater (Qpre,Qevap,Qsuper) as indicated by Eq. 5.6. Wnet is the
37
in the expander (turbine). In most ORC applications, the expansion process isdry, and therefore no blade erosion due to condensation occurs. Many organicfluids also serve as a lubricant in addition to a working medium. Water alsohas a high freezing point compared to organic fluids, which can be a limitingfactor [43].
-200
-150
-100
-50
0
50
100
150
200
250
-2,5 -1,5 -0,5 0,5 1,5 2,5
Tem
pera
ture
°C
Entropy kJ/(kg K)
Figure 5.8: Dry fluid, entropy temperature diagram of Isopentane
Choosing the optimal fluid depends on the application, as different workingfluids perform better or worse due to the thermodynamic properties incombination with the heat source and sink. In Paper I, an optimisation ofthe best performing fluid was done using the database from Refprop andexperimenting with 112 different organic fluids in the simulation [23]. Inthat particular study, benzene came out as the most efficient working fluidfor the specific speed interval of 12 to 14 knots. As shown in Figure 5.9, thepractical difference between the fluids from the ORC integration in Paper Iwas not highly significant for the regenerated ORC cases; in a speed interval ofbetween 14 and 16 knots, the absolute difference between the best performerand the worst (of the three best fluids that is) was less than 7 kW, or a 1.3 %difference. As benzene is not only flammable but also highly toxic, it stillmight not be the best fluid, despite its higher efficiency for that particular case.
36
As described by E. Macchi, the general requirements for choosing an organicfluid are the following [47]:
• Commercially available• Nonflammable• Nontoxic• Compatible with other materials• Environmental factors, global warming potential (GWP) and ozone
depletion potential (ODP)
0-8 8-10 10-12 12-14 14-16 16-21100
200
300
400
500
600
700
800
900
1000
Vessel speed
Wne
t / kW
S-tolueneav-S-tolueneS-c1cc6av-S-c1cc6S-ebenzeneav-S-ebenzeneR-benzeneav-R-benzeneR-dmcav-R-dmcR-cyclohexaneav-R-cyclohexane
Figure 5.9: Net power output versus vessel speed for the three most optimal fluids (s - Simple ORC, R - Regenerated,av - averaged over operating time) [23]
In Paper I, the sink used for the ORC is the sea water, and the most convenientinstallation location was determined to be at the top deck near the exhaustpipes. An intermediate loop between the sea water and the condenser was usedto keep the length of sea water piping length to a minimum given the effectsof corrosion and a requirement to keep the pipe clean. The condensationtemperature was kept at a constant 30 ◦C during all simulations.
The efficiency of the ORC was considered as the fraction of the workoutput (Wnet) from the turbine and heat input from preheater, evaporatorand superheater (Qpre,Qevap,Qsuper) as indicated by Eq. 5.6. Wnet is the
37
in the expander (turbine). In most ORC applications, the expansion process isdry, and therefore no blade erosion due to condensation occurs. Many organicfluids also serve as a lubricant in addition to a working medium. Water alsohas a high freezing point compared to organic fluids, which can be a limitingfactor [43].
-200
-150
-100
-50
0
50
100
150
200
250
-2,5 -1,5 -0,5 0,5 1,5 2,5
Tem
pera
ture
°C
Entropy kJ/(kg K)
Figure 5.8: Dry fluid, entropy temperature diagram of Isopentane
Choosing the optimal fluid depends on the application, as different workingfluids perform better or worse due to the thermodynamic properties incombination with the heat source and sink. In Paper I, an optimisation ofthe best performing fluid was done using the database from Refprop andexperimenting with 112 different organic fluids in the simulation [23]. Inthat particular study, benzene came out as the most efficient working fluidfor the specific speed interval of 12 to 14 knots. As shown in Figure 5.9, thepractical difference between the fluids from the ORC integration in Paper Iwas not highly significant for the regenerated ORC cases; in a speed interval ofbetween 14 and 16 knots, the absolute difference between the best performerand the worst (of the three best fluids that is) was less than 7 kW, or a 1.3 %difference. As benzene is not only flammable but also highly toxic, it stillmight not be the best fluid, despite its higher efficiency for that particular case.
36
sources into the same cycle, as much of the energy in the lower temperaturesources can be utilised in heating the working fluid in the first step. As shownin the TQ-diagram (temperature enthalpy) in Figure 5.10, the ORC workingfluid is first heated in jacket cooling water, which has a temperature of 80 ◦C.In the first heat exchanger, 500 kJ is transferred before entering the next heatexchanger, which is heated by the charge air cooler. In the third step, theworking fluid is evaporated (hence the constant temperature) with the engineexhaust gases. In the fourth step, the working fluid is superheated with theexhaust gas recirculation. The ORC working fluid must pass through fourheat exchangers, which adds to the complexity of the system, but comparedto either mixing all heat sources or to only using the ones with the highesttemperature, this method utilises more of the available energy.
5.8 A useful concept of measuring work: Exergy
The second law of thermodynamics introduces the property entropy (S).Entropy is a not a conserved property as the First Law states (conservationof energy), this means it is not conserved trough processes. A process canoccur in only one direction, complying with the principle Sgen ≥ 0. Entropyis a property which is always increasing in the universe as all processes areirreversible. Heat is always flowing from hot to cold, never the opposite. Thisfact alone means there always have to be a cold reservoir in all power cyclesconverting heat energy to mechanical energy. And this also implies that bymixing a cold and hot reservoir the entropy is increased. The definitive, andsomewhat depressing, conclusion of this law leads to the heat death of theuniverse.
It can be seen as nature taking a tax on all conversions from heat to mechanical,as some of the energy supplied must be released to the surroundings as heat.Energy as a concept is not always useful, as it does not account for its usability.We also concluded from the first law of thermodynamics that it could not bedestroyed, only transformed. From that statement, we could argue that thereis no need for energy efficiency – energy cannot be destroyed so what is thepoint? However, exergy is not indestructible as energy. The concept of exergytries to unify both the first and second laws of thermodynamics. Exergy, likeenergy, is measured in Joules and is transformed to anergy in an irreversibleprocess. By definition, the sum of exergy and anergy is constant.
39
shaft output work from the turbine, which powers a generator that produceselectricity. The generator efficiency is not accounted for in the Eq. 5.6.
ηORC =Wnet
Qpre +Qevap +Qsuper(5.6)
0
100
200
300
400
0 500 1000 1500 2000
Tem
pera
ture
°C
Energy kJ
Working fluid
Source
Figure 5.10: Temperature enthalpy diagram from ORC integration with a marine diesel engine [27]
In Paper IV, we integrated an ORC with a slow speed two-stroke marine dieselengine [27]. That study demonstrated the potential of integrating several heatsources into one ORC. As the lower temperatures of the cooling water cannotbe utilised in an effective way due to the fact that the Carnot efficiency istoo low, it can still be used as the first step in a heating process of severalsteps. The main advantage of this setup is that it is possible to utilise a moresubstantial part of the waste heat, but it comes with added complexity andissues of performance stability. An added amount of heat exchangers resultsin many new combinations of temperatures depending on the engine load.That fact alone means it is more difficult to optimise this system for off-designoperation, for example, for low loads.
A heat source with a low temperature is less worthwhile in terms of usability toproduce mechanical work, even though the energy content is equal to a highertemperature. This fact makes it even more important to integrate different heat
38
sources into the same cycle, as much of the energy in the lower temperaturesources can be utilised in heating the working fluid in the first step. As shownin the TQ-diagram (temperature enthalpy) in Figure 5.10, the ORC workingfluid is first heated in jacket cooling water, which has a temperature of 80 ◦C.In the first heat exchanger, 500 kJ is transferred before entering the next heatexchanger, which is heated by the charge air cooler. In the third step, theworking fluid is evaporated (hence the constant temperature) with the engineexhaust gases. In the fourth step, the working fluid is superheated with theexhaust gas recirculation. The ORC working fluid must pass through fourheat exchangers, which adds to the complexity of the system, but comparedto either mixing all heat sources or to only using the ones with the highesttemperature, this method utilises more of the available energy.
5.8 A useful concept of measuring work: Exergy
The second law of thermodynamics introduces the property entropy (S).Entropy is a not a conserved property as the First Law states (conservationof energy), this means it is not conserved trough processes. A process canoccur in only one direction, complying with the principle Sgen ≥ 0. Entropyis a property which is always increasing in the universe as all processes areirreversible. Heat is always flowing from hot to cold, never the opposite. Thisfact alone means there always have to be a cold reservoir in all power cyclesconverting heat energy to mechanical energy. And this also implies that bymixing a cold and hot reservoir the entropy is increased. The definitive, andsomewhat depressing, conclusion of this law leads to the heat death of theuniverse.
It can be seen as nature taking a tax on all conversions from heat to mechanical,as some of the energy supplied must be released to the surroundings as heat.Energy as a concept is not always useful, as it does not account for its usability.We also concluded from the first law of thermodynamics that it could not bedestroyed, only transformed. From that statement, we could argue that thereis no need for energy efficiency – energy cannot be destroyed so what is thepoint? However, exergy is not indestructible as energy. The concept of exergytries to unify both the first and second laws of thermodynamics. Exergy, likeenergy, is measured in Joules and is transformed to anergy in an irreversibleprocess. By definition, the sum of exergy and anergy is constant.
39
shaft output work from the turbine, which powers a generator that produceselectricity. The generator efficiency is not accounted for in the Eq. 5.6.
ηORC =Wnet
Qpre +Qevap +Qsuper(5.6)
0
100
200
300
400
0 500 1000 1500 2000
Tem
pera
ture
°C
Energy kJ
Working fluid
Source
Figure 5.10: Temperature enthalpy diagram from ORC integration with a marine diesel engine [27]
In Paper IV, we integrated an ORC with a slow speed two-stroke marine dieselengine [27]. That study demonstrated the potential of integrating several heatsources into one ORC. As the lower temperatures of the cooling water cannotbe utilised in an effective way due to the fact that the Carnot efficiency istoo low, it can still be used as the first step in a heating process of severalsteps. The main advantage of this setup is that it is possible to utilise a moresubstantial part of the waste heat, but it comes with added complexity andissues of performance stability. An added amount of heat exchangers resultsin many new combinations of temperatures depending on the engine load.That fact alone means it is more difficult to optimise this system for off-designoperation, for example, for low loads.
A heat source with a low temperature is less worthwhile in terms of usability toproduce mechanical work, even though the energy content is equal to a highertemperature. This fact makes it even more important to integrate different heat
38
In Paper VI, the exergy flows were calculated for one full year of operation, asshown in Figure 5.11. As discussed earlier in this section 5.4, and relatingto the Sankey diagram Figure 5.3 p. 29, an energy analysis does not containinformation on the usability of the energy. When looking at the energy systemand accounting for the amount of work which can theoretically be used,however, the picture changes. As seen in the Grassman diagram, Figure 5.11,the green is exergy for electricity and propulsion, and the yellow representsthe exergy in the exhaust gases. Notably, that this diagram significantly differsfrom the Sankey diagram as the exergy entering the diagram on the left isnot the same as that which leaves on the right due to the fact that exergyis destroyed in the process. What can be derived from this is that thereare significant exergy losses from the cooling water as well as the exhaustgases, which is an indicator that the system can be further optimised. Whencomparing the energy and exergy diagrams (Figures 5.3 and 5.11), the energylosses for cooling water are 19 GW h, while the exergy losses are 5.5 GW h.The exergy analysis provides a more concise picture of what can be used,and also where the irreversible losses occur. The reference state for theexergy analysis was set to the measured sea water temperature at an standardatmospheric pressure.
41
In an isolated system, such as a room filled with air or a fuel container, noenergy or mass can be transferred outside the boundaries of the room, andaccording to the first law of thermodynamics (energy conservation), the totalenergy is always constant. In Case A, nothing occurs, with the fuel containingchemically stored energy. In Case B, the fuel is ignited, combusting withthe surrounding air and the temperature rising as the chemical energy istransformed into heat. In both Cases A and B, the amounts of energy areidentical, but comparing the actual economic or useful value of the two casesilluminates differences which do not reflect in the measured energy content.The fuel in Case A could be used to power an engine, producing mechanicalwork which in turn can be transformed into electricity. The hot combustedgasses in Case B can, of course, be used for heating, but if we want to power acomputer or drive a car, they are much less useful. Case A thus has a moresignificant potential for use compared to case B.
Exergy is the property that quantifies the potential for mechanical work, whichis in most cases the purpose of a machine. Exergy can be defined as themaximum theoretical useful work (shaft work or electrical work) as the systemis brought into complete thermodynamic equilibrium with the thermodynamicenvironment while the system interacts with it alone [48].
As shown in Eq. 5.7 the sum of the exergy input ( �Ein) is equal to the exergyoutput ( �Eout) plus the destructed exergy ( �Ed). When dealing with second lawanalysis, some amount of exergy is always destroyed in the process.
∑in
�Ein =∑out
�Eout + �Ed (5.7)
The exergy balance may alternatively be formulated for an energy system,as seen in Eq. 5.8, wherein the exergy product ( �Ep) is the useful power,exergy. The exergy product is equal to the exergy of the fuel ( �Ef), the exergydestruction ( �Ed) and exergy losses ( �El) which are released into the environmentin the form of exhaust gases or cooling water.
�Ep = �Ef − �Ed − �El (5.8)
The exergy product corresponds to the desired output of the system in termsof exergy and could, for instance, be the useful power produced by an engine.
40
In Paper VI, the exergy flows were calculated for one full year of operation, asshown in Figure 5.11. As discussed earlier in this section 5.4, and relatingto the Sankey diagram Figure 5.3 p. 29, an energy analysis does not containinformation on the usability of the energy. When looking at the energy systemand accounting for the amount of work which can theoretically be used,however, the picture changes. As seen in the Grassman diagram, Figure 5.11,the green is exergy for electricity and propulsion, and the yellow representsthe exergy in the exhaust gases. Notably, that this diagram significantly differsfrom the Sankey diagram as the exergy entering the diagram on the left isnot the same as that which leaves on the right due to the fact that exergyis destroyed in the process. What can be derived from this is that thereare significant exergy losses from the cooling water as well as the exhaustgases, which is an indicator that the system can be further optimised. Whencomparing the energy and exergy diagrams (Figures 5.3 and 5.11), the energylosses for cooling water are 19 GW h, while the exergy losses are 5.5 GW h.The exergy analysis provides a more concise picture of what can be used,and also where the irreversible losses occur. The reference state for theexergy analysis was set to the measured sea water temperature at an standardatmospheric pressure.
41
In an isolated system, such as a room filled with air or a fuel container, noenergy or mass can be transferred outside the boundaries of the room, andaccording to the first law of thermodynamics (energy conservation), the totalenergy is always constant. In Case A, nothing occurs, with the fuel containingchemically stored energy. In Case B, the fuel is ignited, combusting withthe surrounding air and the temperature rising as the chemical energy istransformed into heat. In both Cases A and B, the amounts of energy areidentical, but comparing the actual economic or useful value of the two casesilluminates differences which do not reflect in the measured energy content.The fuel in Case A could be used to power an engine, producing mechanicalwork which in turn can be transformed into electricity. The hot combustedgasses in Case B can, of course, be used for heating, but if we want to power acomputer or drive a car, they are much less useful. Case A thus has a moresignificant potential for use compared to case B.
Exergy is the property that quantifies the potential for mechanical work, whichis in most cases the purpose of a machine. Exergy can be defined as themaximum theoretical useful work (shaft work or electrical work) as the systemis brought into complete thermodynamic equilibrium with the thermodynamicenvironment while the system interacts with it alone [48].
As shown in Eq. 5.7 the sum of the exergy input ( �Ein) is equal to the exergyoutput ( �Eout) plus the destructed exergy ( �Ed). When dealing with second lawanalysis, some amount of exergy is always destroyed in the process.
∑in
�Ein =∑out
�Eout + �Ed (5.7)
The exergy balance may alternatively be formulated for an energy system,as seen in Eq. 5.8, wherein the exergy product ( �Ep) is the useful power,exergy. The exergy product is equal to the exergy of the fuel ( �Ef), the exergydestruction ( �Ed) and exergy losses ( �El) which are released into the environmentin the form of exhaust gases or cooling water.
�Ep = �Ef − �Ed − �El (5.8)
The exergy product corresponds to the desired output of the system in termsof exergy and could, for instance, be the useful power produced by an engine.
40
Chapter 6
Learning from data
Big Data is like teenage sex:everyone talks about it, nobodyreally knows how to do it, everyonethinks everyone else is doing it, soeveryone claims they are doing it.
Dan Ariely
There are estimates that the amount of data (digitally stored) created in the pastdecade accounts for over 90 % of all data that we have stored today, and thisfigure is expected to rise exponentially with even cheaper and more availablesensors [49]. Due to this exponential expansion of data in conjunction withmore powerful processors, both new challenges and new possibilities emerge.Data alone does not provide any meaning unless it is analysed and usedpurposively. Big Data is the broad concept of large amounts of data, whichleaves opportunities to construct computer algorithms which can learn fromthis data and make accurate predictions. These predictions can pertain toclassification, translating languages, showing the optimal advertisement fora customer in a web-shop or finding the best treatment for patients and asdemonstrated in this thesis, predicting the fuel consumption of a ship.
Information (not in the sense of knowledge) is something which exists every-where, and it becomes data as soon as we measure and store that informationfrom sensors. This is done today to a much larger extent than at any point
43
Propeller shaft
Shaft losses
Losses
Losses
Losses
AG losses
Losses
Losses
ThrustersHVAC
HRSG
HRSG
Cylinder
Turbine
Cylinder
LOC
JWC
HTC
LTC
Compressor
Bypass valve
CAC-HT
CAC-LT
Compressor
CAC-HT
CAC-LT
LOC
JWC
Turbine
Boiler
Auxiliary boiler
Environment
Environment
0.8
25
0.4
Cooling water5.5
Exhausts2.8
Exhausts2.2
1.3
64
3.2
0.21.0
HRHT
Exhausts3.5
2.3
0.5
0.8
Fuel67
Fuel43
Fuel5
Switchboard Others
Propeller
Preheater
Reheater
Hot water
Machinery space heaters
HFO tank heating
Tank heatingGalley
Other tanks
1.0
Figure 5.11: Grassman diagram of M/S Birka, Paper VI. Flow values are in GWh/year.
42
Chapter 6
Learning from data
Big Data is like teenage sex:everyone talks about it, nobodyreally knows how to do it, everyonethinks everyone else is doing it, soeveryone claims they are doing it.
Dan Ariely
There are estimates that the amount of data (digitally stored) created in the pastdecade accounts for over 90 % of all data that we have stored today, and thisfigure is expected to rise exponentially with even cheaper and more availablesensors [49]. Due to this exponential expansion of data in conjunction withmore powerful processors, both new challenges and new possibilities emerge.Data alone does not provide any meaning unless it is analysed and usedpurposively. Big Data is the broad concept of large amounts of data, whichleaves opportunities to construct computer algorithms which can learn fromthis data and make accurate predictions. These predictions can pertain toclassification, translating languages, showing the optimal advertisement fora customer in a web-shop or finding the best treatment for patients and asdemonstrated in this thesis, predicting the fuel consumption of a ship.
Information (not in the sense of knowledge) is something which exists every-where, and it becomes data as soon as we measure and store that informationfrom sensors. This is done today to a much larger extent than at any point
43
Propeller shaft
Shaft losses
Losses
Losses
Losses
AG losses
Losses
Losses
ThrustersHVAC
HRSG
HRSG
Cylinder
Turbine
Cylinder
LOC
JWC
HTC
LTC
Compressor
Bypass valve
CAC-HT
CAC-LT
Compressor
CAC-HT
CAC-LT
LOC
JWC
Turbine
Boiler
Auxiliary boiler
Environment
Environment
0.8
25
0.4
Cooling water5.5
Exhausts2.8
Exhausts2.2
1.3
64
3.2
0.21.0
HRHT
Exhausts3.5
2.3
0.5
0.8
Fuel67
Fuel43
Fuel5
Switchboard Others
Propeller
Preheater
Reheater
Hot water
Machinery space heaters
HFO tank heating
Tank heatingGalley
Other tanks
1.0
Figure 5.11: Grassman diagram of M/S Birka, Paper VI. Flow values are in GWh/year.
42
The dynamics of fuel consumption are also important for an accurate andhighly precise analysis. For instance, if the fuel measurements are logged at15-minute intervals but the ship is manoeuvring, the high dynamic engineloads are averaged out for that time period. Further, if the navigational officerwho operates the ship does not know how much fuel is being consumed,adjustments and behavioural changes are more difficult [51].
6.3 Basic machine learning concepts
6.3.1 Features and labels
Machine learning (ML) involves learning from data, which means that theprocess is not dependent on knowing anything of how the system works.When learning an algorithm, the inputs (what the machine can see) to thealgorithm are called features, and the output is called the label or target. InPapers II and III, the training data are features consisting of engine RPM,exhaust temperatures, turbo RPM and fuel rack positions, while the label ortarget is the fuel consumption.
6.3.2 Train and test split
Depending on the size of the quantities of data available, the train and test splitit somewhere between 50 % and 90 % (larger for big data sets). All modelsrequire some form of parameter input, which is defined as hyperparameters.Hyperparameters for an ML-model can, for instance, be the number oflayers in a neural network, loss factor for training or number of clusters in aK-means [52].
6.3.3 Scaling
When feeding an ML-algorithm, some kind of feature scaling of the data isoften conducted [52]. The scaling can be done in several ways depending onhow the data is distributed, with the classical ML-approach to finding out howto best pre-process the data being either by previous experience or by tryingout which provides the best results. Rescaling, shown in Eq. 6.1, where x ′ is
45
previously, as sensors are decreasingly expensive and much easier to deploy.One challenge with Big Data is that it is not always clear what to do with theavailable data, as the data itself says nothing until it is analysed. There is thena question of how much data is needed for added value.
6.1 Machine learning and artificial intelligence
Artificial intelligence (AI) is a research field which has existed for over 50years. The word artificial means created by man, and intelligence refers toan agent doing something intelligently. The agent can be a human, robot,thermostat, algorithm and so on. Moreover, depending on how the agent acts,it is seen as intelligent according to the following criteria [50]:
• What it does is appropriate for its circumstances and goals.• It is flexible to changing environments and changing goals.• It learns from experience.• It makes appropriate choices given its perceptual and computational
limitations. An agent typically cannot observe the state of the worlddirectly; it has only a finite memory and it does not have unlimited timeto act.
AI and ML can be used to complement and replacing sensors, providing betterdecision support for the operators and in the long run replacing much humaneffort.
6.2 Making predictions from data
In this thesis, machine learning is used as a tool to better predict fuelconsumption, an idea formed during work on Paper V, ‘Energy and ExergyAnalysis in a Cruise Ship’ [29]. Because knowing fuel consumption isessential, and as the dataset lacked the measurements of all the individualcomponents of fuel consumption, many assumptions needed to be made.For example, fuel boiler consumption was not measured and needed to beapproximated by outside temperature and crew interviews.
44
The dynamics of fuel consumption are also important for an accurate andhighly precise analysis. For instance, if the fuel measurements are logged at15-minute intervals but the ship is manoeuvring, the high dynamic engineloads are averaged out for that time period. Further, if the navigational officerwho operates the ship does not know how much fuel is being consumed,adjustments and behavioural changes are more difficult [51].
6.3 Basic machine learning concepts
6.3.1 Features and labels
Machine learning (ML) involves learning from data, which means that theprocess is not dependent on knowing anything of how the system works.When learning an algorithm, the inputs (what the machine can see) to thealgorithm are called features, and the output is called the label or target. InPapers II and III, the training data are features consisting of engine RPM,exhaust temperatures, turbo RPM and fuel rack positions, while the label ortarget is the fuel consumption.
6.3.2 Train and test split
Depending on the size of the quantities of data available, the train and test splitit somewhere between 50 % and 90 % (larger for big data sets). All modelsrequire some form of parameter input, which is defined as hyperparameters.Hyperparameters for an ML-model can, for instance, be the number oflayers in a neural network, loss factor for training or number of clusters in aK-means [52].
6.3.3 Scaling
When feeding an ML-algorithm, some kind of feature scaling of the data isoften conducted [52]. The scaling can be done in several ways depending onhow the data is distributed, with the classical ML-approach to finding out howto best pre-process the data being either by previous experience or by tryingout which provides the best results. Rescaling, shown in Eq. 6.1, where x ′ is
45
previously, as sensors are decreasingly expensive and much easier to deploy.One challenge with Big Data is that it is not always clear what to do with theavailable data, as the data itself says nothing until it is analysed. There is thena question of how much data is needed for added value.
6.1 Machine learning and artificial intelligence
Artificial intelligence (AI) is a research field which has existed for over 50years. The word artificial means created by man, and intelligence refers toan agent doing something intelligently. The agent can be a human, robot,thermostat, algorithm and so on. Moreover, depending on how the agent acts,it is seen as intelligent according to the following criteria [50]:
• What it does is appropriate for its circumstances and goals.• It is flexible to changing environments and changing goals.• It learns from experience.• It makes appropriate choices given its perceptual and computational
limitations. An agent typically cannot observe the state of the worlddirectly; it has only a finite memory and it does not have unlimited timeto act.
AI and ML can be used to complement and replacing sensors, providing betterdecision support for the operators and in the long run replacing much humaneffort.
6.2 Making predictions from data
In this thesis, machine learning is used as a tool to better predict fuelconsumption, an idea formed during work on Paper V, ‘Energy and ExergyAnalysis in a Cruise Ship’ [29]. Because knowing fuel consumption isessential, and as the dataset lacked the measurements of all the individualcomponents of fuel consumption, many assumptions needed to be made.For example, fuel boiler consumption was not measured and needed to beapproximated by outside temperature and crew interviews.
44
of data, the optimal ML-algorithm and by also tuning the hyper-parameters.All this is combined in a pipeline, which is a Python object containing severalML-classes, where often a pre-processor and one or several ML-models aretrained in series. The advantage of using pipelines of models is that severalsimple models can perform well together if integrated in a series. To derive arobust algorithm and a suitable prediction, the data must be pre-processed,the appropriate algorithm chosen and the data divided into a random train andtest set.
Several studies have been conducted in the area of AutoML, including opensource tools based on the Python Sci-Kit learn library [53]. The Tree-basedPipeline Optimization Tool for Automating Machine Learning (TPOT) usesgenetic algorithms to optimise the best pipeline, both pre-processors (scalers)and algorithms as well as hyperparameters [55, 56, 57]. The Auto-sklearn toolis also based upon the scikit-learn library and works similarly to the TPOTbut also factors in neural networks [58]. Both the TPOT and Auto-sklearnwere evaluated in Paper II, with similar results.
6.6 Predicting the energy consumption
In Papers II and III, an ML-algorithm using auto machine learning was appliedto fuel measurements from the dataset. As it is an added cost of installingadditional mass flow meters an ML-model can be useful of complementingthe existing ones, or creating virtual sensors. If the ship has several engines,the cost goes up accordingly with each added installed fuel meter. The shipM/S Birka Stockholm, for instance, has eight engines and two fuel oil boilers.If the ship has only one fuel meter for all consumers, a machine learningalgorithm trained on the engine and boiler data, which could be temperatures,revolutions per minute, pressures or other sensor data, can identify individualcharacteristics of each consumer. Even when adding a higher time interval,such as only using noon reports for each day, an algorithm can be trained topredict the dynamics of the consumers. In Paper III, ‘Predicting DynamicFuel Oil Consumption on Ships with Automated Machine Learning’, the sumsfrom each day were used to train an ML-pipeline on the average values oflabels from the engines. This proved feasible for an average of fuel sums forup to 96h intervals. The predicted dynamic fuel consumption was plottedagainst the real measurements, as demonstrated in Figure 6.2. The model fit(how well it performs) for the entire year of data in the specific example is
47
the rescaled value, occurs when the features are simply scaled to a new scale,often [−1,1] or [0,1].
x ′ =x − min(x)
max(x) − min(x) (6.1)
Machine learning often uses data from different sensors which are notnecessarily linear or evenly distributed, and in the case of predicting fuel oil asconducted in Papers II and III, the features were scaled with a standardisationaccounting for standard deviation. The standard scaler is shown in Eq. 6.2,where σ is the standard deviation.
x ′ =x − x̄σ
(6.2)
Choosing the best performing feature scaler is dependent on the type of dataand on which algorithm is chosen, the latter of which is done automaticallyby AutoML-tools.
6.4 Choosing the right algorithm
The available ML algorithms and pre-processors in the Scikit-learn packageare extensive and comprise of many regressors, classifiers and clusteringalgorithms [53]. The vast number of algorithms render model selection a taskfor an experienced ML-practitioner, and as can be seen in the Scikit-learncheat-sheet, Figure 6.1, many paths can be chosen to provide the optimalmodel for the task. Depending on what data is available, as well as theamount and quality of data, a model is often created based on what will be asuitable predictor as well as being generalisable (meaning it can be applied toother similar cases). In Section 6.5, a framework for automating this task isexplained which was the basis of Studies II and III.
6.5 Auto machine learning
Auto machine learning (AutoML) is the concept of automating the work ofthe data scientist by running optimisation routines on the best pre-processing
46
of data, the optimal ML-algorithm and by also tuning the hyper-parameters.All this is combined in a pipeline, which is a Python object containing severalML-classes, where often a pre-processor and one or several ML-models aretrained in series. The advantage of using pipelines of models is that severalsimple models can perform well together if integrated in a series. To derive arobust algorithm and a suitable prediction, the data must be pre-processed,the appropriate algorithm chosen and the data divided into a random train andtest set.
Several studies have been conducted in the area of AutoML, including opensource tools based on the Python Sci-Kit learn library [53]. The Tree-basedPipeline Optimization Tool for Automating Machine Learning (TPOT) usesgenetic algorithms to optimise the best pipeline, both pre-processors (scalers)and algorithms as well as hyperparameters [55, 56, 57]. The Auto-sklearn toolis also based upon the scikit-learn library and works similarly to the TPOTbut also factors in neural networks [58]. Both the TPOT and Auto-sklearnwere evaluated in Paper II, with similar results.
6.6 Predicting the energy consumption
In Papers II and III, an ML-algorithm using auto machine learning was appliedto fuel measurements from the dataset. As it is an added cost of installingadditional mass flow meters an ML-model can be useful of complementingthe existing ones, or creating virtual sensors. If the ship has several engines,the cost goes up accordingly with each added installed fuel meter. The shipM/S Birka Stockholm, for instance, has eight engines and two fuel oil boilers.If the ship has only one fuel meter for all consumers, a machine learningalgorithm trained on the engine and boiler data, which could be temperatures,revolutions per minute, pressures or other sensor data, can identify individualcharacteristics of each consumer. Even when adding a higher time interval,such as only using noon reports for each day, an algorithm can be trained topredict the dynamics of the consumers. In Paper III, ‘Predicting DynamicFuel Oil Consumption on Ships with Automated Machine Learning’, the sumsfrom each day were used to train an ML-pipeline on the average values oflabels from the engines. This proved feasible for an average of fuel sums forup to 96h intervals. The predicted dynamic fuel consumption was plottedagainst the real measurements, as demonstrated in Figure 6.2. The model fit(how well it performs) for the entire year of data in the specific example is
47
the rescaled value, occurs when the features are simply scaled to a new scale,often [−1,1] or [0,1].
x ′ =x − min(x)
max(x) − min(x) (6.1)
Machine learning often uses data from different sensors which are notnecessarily linear or evenly distributed, and in the case of predicting fuel oil asconducted in Papers II and III, the features were scaled with a standardisationaccounting for standard deviation. The standard scaler is shown in Eq. 6.2,where σ is the standard deviation.
x ′ =x − x̄σ
(6.2)
Choosing the best performing feature scaler is dependent on the type of dataand on which algorithm is chosen, the latter of which is done automaticallyby AutoML-tools.
6.4 Choosing the right algorithm
The available ML algorithms and pre-processors in the Scikit-learn packageare extensive and comprise of many regressors, classifiers and clusteringalgorithms [53]. The vast number of algorithms render model selection a taskfor an experienced ML-practitioner, and as can be seen in the Scikit-learncheat-sheet, Figure 6.1, many paths can be chosen to provide the optimalmodel for the task. Depending on what data is available, as well as theamount and quality of data, a model is often created based on what will be asuitable predictor as well as being generalisable (meaning it can be applied toother similar cases). In Section 6.5, a framework for automating this task isexplained which was the basis of Studies II and III.
6.5 Auto machine learning
Auto machine learning (AutoML) is the concept of automating the work ofthe data scientist by running optimisation routines on the best pre-processing
46
Figure 6.1: Scikit-learn algorithm cheat-sheet [54].49
an R2-score of 0.9889. Some of the higher dynamics are missing, but thealgorithm does well at identifying the dynamics, considering that it has onlyseen the sums for each 96h-period. This was proven to be feasible in Papers IIand III, and this is only the beginning as there are many applications still tobe explored with this technology.
The benefit of this technology is providing not only a means of creating newand better predictions but also a system which can adapt itself and learn fromexperiences. Algorithms can be trained as classifiers to predict failures, whichin turn becomes a tool for predictive maintenance.
In the studies in this thesis, all data for ML has been collected from an alreadyexisting logging system, which serves as an example of what is available for aship. The logged data was seldom used in any analysis. the ML data is not alarge dataset, as it comprises only one year of data, but given the number ofsensors available, it still demonstrated to be useful for training algorithms andmaking good predictions.
48
Figure 6.1: Scikit-learn algorithm cheat-sheet [54].49
an R2-score of 0.9889. Some of the higher dynamics are missing, but thealgorithm does well at identifying the dynamics, considering that it has onlyseen the sums for each 96h-period. This was proven to be feasible in Papers IIand III, and this is only the beginning as there are many applications still tobe explored with this technology.
The benefit of this technology is providing not only a means of creating newand better predictions but also a system which can adapt itself and learn fromexperiences. Algorithms can be trained as classifiers to predict failures, whichin turn becomes a tool for predictive maintenance.
In the studies in this thesis, all data for ML has been collected from an alreadyexisting logging system, which serves as an example of what is available for aship. The logged data was seldom used in any analysis. the ML data is not alarge dataset, as it comprises only one year of data, but given the number ofsensors available, it still demonstrated to be useful for training algorithms andmaking good predictions.
48
Chapter 7
The impact and context
7.1 Waste heat recovery feasibility
The results of Papers I and VI indicate a significant potential for producingelectricity from exhaust gases with an ORC, though the results differ between16 % and 22 % electricity production due to the different engineering ap-proaches and choices made. In the first study, the ORC was optimised inrelation to efficiency from a design condition of 12 to 14 knots to identifythe best working fluid and set-up, which demonstrated the possibility ofmaximising the savings with benzene in the design condition [23]. In thesecond ORC study, the same set-up was used, but with another approachconsidering the simulated power production in terms of the running profile,which resulted in smaller percentage of savings [31]. Fuel savings could beestimated to approximately 1372 kg/day. This represents a 5.9 % savings infuel use considering the total energy use of a one-year average fuel consump-tion of 23 436 kg each day. Studies on the waste heat recovery potential forships indicates fuel savings between 5 % to 15 % can be realistic [59, 60].
51
Figure 6.2: Dynamic fuel oil consumption, SVR-algorithm 96h sum average
50
Chapter 7
The impact and context
7.1 Waste heat recovery feasibility
The results of Papers I and VI indicate a significant potential for producingelectricity from exhaust gases with an ORC, though the results differ between16 % and 22 % electricity production due to the different engineering ap-proaches and choices made. In the first study, the ORC was optimised inrelation to efficiency from a design condition of 12 to 14 knots to identifythe best working fluid and set-up, which demonstrated the possibility ofmaximising the savings with benzene in the design condition [23]. In thesecond ORC study, the same set-up was used, but with another approachconsidering the simulated power production in terms of the running profile,which resulted in smaller percentage of savings [31]. Fuel savings could beestimated to approximately 1372 kg/day. This represents a 5.9 % savings infuel use considering the total energy use of a one-year average fuel consump-tion of 23 436 kg each day. Studies on the waste heat recovery potential forships indicates fuel savings between 5 % to 15 % can be realistic [59, 60].
51
Figure 6.2: Dynamic fuel oil consumption, SVR-algorithm 96h sum average
50
7.3 Measuring energy with machine learning
In Papers II and III, we presented results to obtain better predictions ofthe energy flow using the existing fuel flow meters in conjunction with theexisting machine learning models. Operational efficiency is crucial, and oneof the keys to knowing how to operate the ship efficiently is knowing the fuelconsumption not only after a trip is made but as instant feedback during atrip. Given the effects of weather (waves and wind) and the effects of the trimand speed, providing the operator a means optimise energy consumption isessential. Machine learning can be used to provide this measurement whilemaking use of existing flow meters without installing additional ones.
This method provides the advantage not only by creating new data points forthe dynamics but also giving a prediction of the individual engines even ifseveral share a common fuel meter. In Paper II, a model was created whichdemonstrated an ML-model predicting the fuel with high accuracy, and inPaper III, the same method was applied to data from fuel sums of up to fourdays.
The approach in both Papers II and III use an AutoML framework for tooptimise the best performing ML-model. With this approach, many newpossibilities emerge for creating ML-models which can predict more than fuelconsumption. The combination of mass flow meters and machine learningalgorithms, coupled with already available data, creates the opportunity topredict individual consumers without the need to install a flow meter on eachpoint.
Arguably, the use of ML makes it more difficult to see what is really happeningand to understand the basics of how the system functions. The ML tools canproduce accurate predictions based on the data provided for training, but withthe risk of making erroneous predictions if fed with data for which it has notbeen trained. It is not possible to dissect an ML model in the same way as aphysical model built upon equations, which means that the model must betrusted as a black box.
53
If an ORC produces additional electrical power and the electrical demand isconsidered to be the same, the electrical generators must run at lower power.The ORC is powered by the exhaust gases from all engines, including theAE, which means that a decreased mass flow of exhaust gasses results in adecreased power production from the ORC. This feedback loop has not beenconsidered in any of the papers and implies that the real results are likely tobe marginally lower than those presented.
7.2 The exergy destruction
Energy efficiency actually refers to the amount of useful energy derived froma process. The amount of energy must always be considered in relation towhat kind of energy it is and what we want to do with it. To illustrate this,the amount of heat energy in a bathtub (400 l) filled with 40 ◦C water, heatedfrom 5 ◦C, is 61 200 kJ. This amount of energy is roughly equivalent to 4,080standard fully charged alkaline AA-batteries (14 kJ each), that is, 100 kg ofbatteries. It is also equivalent to the chemical energy stored in only 0.69 kgof diesel (42 700 kJ/kg). The same unit, Joule, is used for all these examples,but if we were to light a house, power a computer or drive a car or ship, theexamples of batteries or diesel fuel are more useful. This is where the conceptof exergy and the second law thermodynamics becomes useful.
Significant exergy is demonstrated to have been lost into the atmosphere, inaddition, also the destruction of exergy. Destruction of exergy is irreversible,and in Paper VI, the greatest exergy destruction during the full year ofoperation occurred in the engine turbochargers at 28.9 %, and after that, theheat recovery steam generators (HRSG) and oil-fired boilers accounted forexergy losses of 10.7 % and 10.5 %, respectively. The design and efficiencyof the engines are important, but these elements fall outside the scope of thisthesis. In the analysis, the engines were evidently often operating at low loadsrather than at the optimal efficiency of about 85 %. Electrifying the systemand making use of battery storage could, however, lead the engines to greaterefficiency [61]. The analysis also illuminates that steam generation, both fromthe heat recovered by the exhaust gasses as well as the oil-fired boilers, is amajor contributor to the exergy losses onboard.
52
7.3 Measuring energy with machine learning
In Papers II and III, we presented results to obtain better predictions ofthe energy flow using the existing fuel flow meters in conjunction with theexisting machine learning models. Operational efficiency is crucial, and oneof the keys to knowing how to operate the ship efficiently is knowing the fuelconsumption not only after a trip is made but as instant feedback during atrip. Given the effects of weather (waves and wind) and the effects of the trimand speed, providing the operator a means optimise energy consumption isessential. Machine learning can be used to provide this measurement whilemaking use of existing flow meters without installing additional ones.
This method provides the advantage not only by creating new data points forthe dynamics but also giving a prediction of the individual engines even ifseveral share a common fuel meter. In Paper II, a model was created whichdemonstrated an ML-model predicting the fuel with high accuracy, and inPaper III, the same method was applied to data from fuel sums of up to fourdays.
The approach in both Papers II and III use an AutoML framework for tooptimise the best performing ML-model. With this approach, many newpossibilities emerge for creating ML-models which can predict more than fuelconsumption. The combination of mass flow meters and machine learningalgorithms, coupled with already available data, creates the opportunity topredict individual consumers without the need to install a flow meter on eachpoint.
Arguably, the use of ML makes it more difficult to see what is really happeningand to understand the basics of how the system functions. The ML tools canproduce accurate predictions based on the data provided for training, but withthe risk of making erroneous predictions if fed with data for which it has notbeen trained. It is not possible to dissect an ML model in the same way as aphysical model built upon equations, which means that the model must betrusted as a black box.
53
If an ORC produces additional electrical power and the electrical demand isconsidered to be the same, the electrical generators must run at lower power.The ORC is powered by the exhaust gases from all engines, including theAE, which means that a decreased mass flow of exhaust gasses results in adecreased power production from the ORC. This feedback loop has not beenconsidered in any of the papers and implies that the real results are likely tobe marginally lower than those presented.
7.2 The exergy destruction
Energy efficiency actually refers to the amount of useful energy derived froma process. The amount of energy must always be considered in relation towhat kind of energy it is and what we want to do with it. To illustrate this,the amount of heat energy in a bathtub (400 l) filled with 40 ◦C water, heatedfrom 5 ◦C, is 61 200 kJ. This amount of energy is roughly equivalent to 4,080standard fully charged alkaline AA-batteries (14 kJ each), that is, 100 kg ofbatteries. It is also equivalent to the chemical energy stored in only 0.69 kgof diesel (42 700 kJ/kg). The same unit, Joule, is used for all these examples,but if we were to light a house, power a computer or drive a car or ship, theexamples of batteries or diesel fuel are more useful. This is where the conceptof exergy and the second law thermodynamics becomes useful.
Significant exergy is demonstrated to have been lost into the atmosphere, inaddition, also the destruction of exergy. Destruction of exergy is irreversible,and in Paper VI, the greatest exergy destruction during the full year ofoperation occurred in the engine turbochargers at 28.9 %, and after that, theheat recovery steam generators (HRSG) and oil-fired boilers accounted forexergy losses of 10.7 % and 10.5 %, respectively. The design and efficiencyof the engines are important, but these elements fall outside the scope of thisthesis. In the analysis, the engines were evidently often operating at low loadsrather than at the optimal efficiency of about 85 %. Electrifying the systemand making use of battery storage could, however, lead the engines to greaterefficiency [61]. The analysis also illuminates that steam generation, both fromthe heat recovered by the exhaust gasses as well as the oil-fired boilers, is amajor contributor to the exergy losses onboard.
52
7.5 Trends
About half of the world’s oil supply is transported on ships, and the useof fossil fuels must decrease soon if we are going to limit the temperatureincrease to below 2 ◦C [67]. This fact implies that the shipping sector willbe impacted, not only because of the shift from fossil fuel for propulsion butalso because a large share of the transported goods are oil, and a decrease ofoil use in the road sector will decrease the need for the oil transport as well.Thus, not only are ships going to need to consume less fuel, but the sector alsoneeds to adapt to lower transport volumes of oil, which implies a decreasedoverall transport volume.
In 2013, there was a significant focus on the upcoming scrubber directive,which was enforced on 1 January 2015 [68]. Much discussion centred aroundthe expense of installing scrubbers, but at the same time during the years 2014and 2015, fuel prices dropped significantly, as shown in Figure 7.1, meaningthat ship owners could switch fuel to MDO instead of undertaking a costlyscrubber installation. This because the price of MDO dropped to previousprice of IFO380 in the same period. In the next years even stricter sulphurregulations will take place, as it will be a global cap of 0.1 % in 2020, whichmeans it is an incentive to invest in a scrubber if the investment cost for ascrubber is lower than the added price of MGO compared to IFO.
Battery powered ships are already in operation, and for shorter distances, theycan compete economically with fossil fuel even though that fuel is tax-free.Local emissions and sound will drastically decrease, and these ships benefitfrom better manoeuvrability (a flatter torque curve compared to a dieselengine) and decreased maintenance [69, 70]. However, battery propelled shipsare not yet able to undertake a transatlantic voyage, as the batteries require asubstantial amount of space and weight in comparison with fuel oil. A batterywith Li-ion technology has a substantially lower energy content in relation toboth mass and volume compared to a liquid fuel. The order of magnitude isapproximately 2 % compared to the volumetric energy density of diesel fuel[71]. However, for shorter distances in short sea shipping, and where it also ispossible to charge at each stop in port, battery power is currently already apossibility, so long as the ports have decent charging capacities installed andthe ships’ time-schedules allow for re-charging time.
55
7.4 Machine learning applications
In the era of the internet, machine learning is rapidly evolving. Gathering thedata has been a priority in recent years, and the data is essential if we wishto learn from it, but without the appropriate algorithms and tools, the data isuseless.
The main task ahead is to develop better applications for the onboard energysystem using the machine learning tools which have been developed incomputer science. The key is to cross-breed the areas. Recent developmentsin deep learning have led to programmes which can beat humans in areaswhich were previously thought to be impossible for a computer as theywere thought to require intuition. The programme Alpha-Go beat the besthuman players in Go during 2015 and 2016 used neural networks which weretrained and supervised by human interaction, but only a year later, the nextversion, Alpha-Go Zero, was released. This programme trained on itself usingreinforced learning, eventually beating the old Alpha-Go programme 100 to 0[62]. A machine learning method that proves to be accurate for predictingonline shopping behaviours, predicts age from pictures of humans, or beatsthe top human player in Go, can also be used for energy systems [63].
With even more powerful, affordable, energy efficient and smaller computers,the applications are not limited to stationary centres but can be put on board.For instance, the computing power in a modern Tesla car with the NVIDIA’sDrive PX 2 delivers ten teraflops (floating operating per second), equivalent tothe capabilities of the supercomputer IBM ASCII White, which was the world’sfastest supercomputer in 2000 [64, 65, 66]. That computer weighed 106 tonsand consumed 3 MW of electricity. The NVIDIA Drive PX 2 consumesapproximately 250 W and is smaller than an ordinary laptop computer.
54
7.5 Trends
About half of the world’s oil supply is transported on ships, and the useof fossil fuels must decrease soon if we are going to limit the temperatureincrease to below 2 ◦C [67]. This fact implies that the shipping sector willbe impacted, not only because of the shift from fossil fuel for propulsion butalso because a large share of the transported goods are oil, and a decrease ofoil use in the road sector will decrease the need for the oil transport as well.Thus, not only are ships going to need to consume less fuel, but the sector alsoneeds to adapt to lower transport volumes of oil, which implies a decreasedoverall transport volume.
In 2013, there was a significant focus on the upcoming scrubber directive,which was enforced on 1 January 2015 [68]. Much discussion centred aroundthe expense of installing scrubbers, but at the same time during the years 2014and 2015, fuel prices dropped significantly, as shown in Figure 7.1, meaningthat ship owners could switch fuel to MDO instead of undertaking a costlyscrubber installation. This because the price of MDO dropped to previousprice of IFO380 in the same period. In the next years even stricter sulphurregulations will take place, as it will be a global cap of 0.1 % in 2020, whichmeans it is an incentive to invest in a scrubber if the investment cost for ascrubber is lower than the added price of MGO compared to IFO.
Battery powered ships are already in operation, and for shorter distances, theycan compete economically with fossil fuel even though that fuel is tax-free.Local emissions and sound will drastically decrease, and these ships benefitfrom better manoeuvrability (a flatter torque curve compared to a dieselengine) and decreased maintenance [69, 70]. However, battery propelled shipsare not yet able to undertake a transatlantic voyage, as the batteries require asubstantial amount of space and weight in comparison with fuel oil. A batterywith Li-ion technology has a substantially lower energy content in relation toboth mass and volume compared to a liquid fuel. The order of magnitude isapproximately 2 % compared to the volumetric energy density of diesel fuel[71]. However, for shorter distances in short sea shipping, and where it also ispossible to charge at each stop in port, battery power is currently already apossibility, so long as the ports have decent charging capacities installed andthe ships’ time-schedules allow for re-charging time.
55
7.4 Machine learning applications
In the era of the internet, machine learning is rapidly evolving. Gathering thedata has been a priority in recent years, and the data is essential if we wishto learn from it, but without the appropriate algorithms and tools, the data isuseless.
The main task ahead is to develop better applications for the onboard energysystem using the machine learning tools which have been developed incomputer science. The key is to cross-breed the areas. Recent developmentsin deep learning have led to programmes which can beat humans in areaswhich were previously thought to be impossible for a computer as theywere thought to require intuition. The programme Alpha-Go beat the besthuman players in Go during 2015 and 2016 used neural networks which weretrained and supervised by human interaction, but only a year later, the nextversion, Alpha-Go Zero, was released. This programme trained on itself usingreinforced learning, eventually beating the old Alpha-Go programme 100 to 0[62]. A machine learning method that proves to be accurate for predictingonline shopping behaviours, predicts age from pictures of humans, or beatsthe top human player in Go, can also be used for energy systems [63].
With even more powerful, affordable, energy efficient and smaller computers,the applications are not limited to stationary centres but can be put on board.For instance, the computing power in a modern Tesla car with the NVIDIA’sDrive PX 2 delivers ten teraflops (floating operating per second), equivalent tothe capabilities of the supercomputer IBM ASCII White, which was the world’sfastest supercomputer in 2000 [64, 65, 66]. That computer weighed 106 tonsand consumed 3 MW of electricity. The NVIDIA Drive PX 2 consumesapproximately 250 W and is smaller than an ordinary laptop computer.
54
Chapter 8
Concluding remarks
The thesis has presented results that address the aim enhancing energyefficiency in ships, summarised in the following items:
• Gathering and analysing data from a cruise-ship, in collaboration withmany researchers, to advance knowledge of ship operation.
• Simulating an organic Rankine cycle on operational data from the ship.
• Simulating an organic Rankine cycle integrated into a two-stroke dieselengine.
• Simulating the energy system for a cruise ship for a year’s worth of datato provide increased knowledge of the energy flows.
• Developing machine learning methods to better predict fuel consump-tion.
The main conclusion of this thesis is that the data from the ship’s operationsare usable not only for maintenance and performance monitoring but also toprovide a solid base from which to build energy models for the ship. Theoperational data together with physical modelling tools and machine learningthus creates new possibilities. My recommendations to both academia andthe shipping industry are that the data on board should be better used. Onebarrier that we need to address is the way to standardise both the collectionand sharing of data. Much research has so far focussed on optimising the
57
0
200
400
600
800
1000
1200
2013-11-22
2014-05-11
2014-10-28
2015-04-16
2015-10-03
2016-03-21
2016-09-07
2017-02-24
2017-08-13
2018-01-30
2018-07-19
USD
/ m
etric
tonn
e
IFO380 MDO
Figure 7.1: IFO380 and MDO prices, adopted from Ship and Bunker [72]
56
Chapter 8
Concluding remarks
The thesis has presented results that address the aim enhancing energyefficiency in ships, summarised in the following items:
• Gathering and analysing data from a cruise-ship, in collaboration withmany researchers, to advance knowledge of ship operation.
• Simulating an organic Rankine cycle on operational data from the ship.
• Simulating an organic Rankine cycle integrated into a two-stroke dieselengine.
• Simulating the energy system for a cruise ship for a year’s worth of datato provide increased knowledge of the energy flows.
• Developing machine learning methods to better predict fuel consump-tion.
The main conclusion of this thesis is that the data from the ship’s operationsare usable not only for maintenance and performance monitoring but also toprovide a solid base from which to build energy models for the ship. Theoperational data together with physical modelling tools and machine learningthus creates new possibilities. My recommendations to both academia andthe shipping industry are that the data on board should be better used. Onebarrier that we need to address is the way to standardise both the collectionand sharing of data. Much research has so far focussed on optimising the
57
0
200
400
600
800
1000
1200
2013-11-22
2014-05-11
2014-10-28
2015-04-16
2015-10-03
2016-03-21
2016-09-07
2017-02-24
2017-08-13
2018-01-30
2018-07-19
USD
/ m
etric
tonn
e
IFO380 MDO
Figure 7.1: IFO380 and MDO prices, adopted from Ship and Bunker [72]
56
Chapter 9
Summary of the papers
9.1 Waste Heat Recovery in a Cruise Vessel in theBaltic Sea by Using and Organic Rankine Cycle:A Case Study
This paper investigates the feasibility of an organic Rankine cycle on board acruise ship. The simulations were based on data which was collected fromthe logging system on board. The ORC was optimised for the best performingworking media, and it demonstrated that the ORC could produce 22% of theelectrical power on board. I collected all the data and was responsible for thedata analysis. I wrote half of the paper, and Maria E Mondejar the other half.Maria was responsible for the ORC simulations and the theoretical part aboutorganic fluids and thermodynamics.
9.2 Auto Machine Learning for predicting Ship FuelConsumption
This paper uses machine learning to predict the fuel consumption of theengines. By using the existing logged machinery data from the logging systemand learning an algorithm with the volume flow metres, it was possible totrain a model which proved to be very accurate. The model was trained by
59
route, dependent on outside conditions such as weather, wind, waves androute planning. However, it is easy to miss the internal energy system, whichlies within the boundaries of the ship’s hull. By utilising existing waste heatrecovery technologies such as an ORC, fuel consumption can be reducedsignificantly on an existing ship. This fact alone is perhaps self-evident, butthis thesis has shown the potential for doing so in real operating conditions, notjust by designing from manufacturers data. This thesis has also contributed toa better understanding of how energy flows in a cruise ship are distributedand has led to laying additional pieces to the puzzle on how the energy isdistributed internally and where the low hanging fruits are. Part of the workhas also considered calculating exergy efficiency, which has not been donepreviously.
8.1 Future research
Many ways exist of further optimising the onboard energy system, and with newmachine learning algorithms, many possibilities have not yet been explored.We now stand before a revolution in the energy industry, when sensors andcomputer power are getting not only more powerful but also inexpensive andsmall. Cars and ships that can drive autonomously, artificial intelligent voiceassistants that help us in our home, automatic vacuum robots and other robotsall ease our tasks. Technology is enabling new ways of thinking and doingthings which were previously impossible.
In the next phase of this research, energy efficiency will be explored withmachine learning tools. This research will investigate questions of how tobetter predict the energy efficiency of ships using machine learning tools, andhow far can these tools enable the existing technology to be optimised. Byusing the extensive sensor data that is available on many more ship types,ML-models might be a key technology to further optimise efficiency andthereby reducing environmental impact.
58
Chapter 9
Summary of the papers
9.1 Waste Heat Recovery in a Cruise Vessel in theBaltic Sea by Using and Organic Rankine Cycle:A Case Study
This paper investigates the feasibility of an organic Rankine cycle on board acruise ship. The simulations were based on data which was collected fromthe logging system on board. The ORC was optimised for the best performingworking media, and it demonstrated that the ORC could produce 22% of theelectrical power on board. I collected all the data and was responsible for thedata analysis. I wrote half of the paper, and Maria E Mondejar the other half.Maria was responsible for the ORC simulations and the theoretical part aboutorganic fluids and thermodynamics.
9.2 Auto Machine Learning for predicting Ship FuelConsumption
This paper uses machine learning to predict the fuel consumption of theengines. By using the existing logged machinery data from the logging systemand learning an algorithm with the volume flow metres, it was possible totrain a model which proved to be very accurate. The model was trained by
59
route, dependent on outside conditions such as weather, wind, waves androute planning. However, it is easy to miss the internal energy system, whichlies within the boundaries of the ship’s hull. By utilising existing waste heatrecovery technologies such as an ORC, fuel consumption can be reducedsignificantly on an existing ship. This fact alone is perhaps self-evident, butthis thesis has shown the potential for doing so in real operating conditions, notjust by designing from manufacturers data. This thesis has also contributed toa better understanding of how energy flows in a cruise ship are distributedand has led to laying additional pieces to the puzzle on how the energy isdistributed internally and where the low hanging fruits are. Part of the workhas also considered calculating exergy efficiency, which has not been donepreviously.
8.1 Future research
Many ways exist of further optimising the onboard energy system, and with newmachine learning algorithms, many possibilities have not yet been explored.We now stand before a revolution in the energy industry, when sensors andcomputer power are getting not only more powerful but also inexpensive andsmall. Cars and ships that can drive autonomously, artificial intelligent voiceassistants that help us in our home, automatic vacuum robots and other robotsall ease our tasks. Technology is enabling new ways of thinking and doingthings which were previously impossible.
In the next phase of this research, energy efficiency will be explored withmachine learning tools. This research will investigate questions of how tobetter predict the energy efficiency of ships using machine learning tools, andhow far can these tools enable the existing technology to be optimised. Byusing the extensive sensor data that is available on many more ship types,ML-models might be a key technology to further optimise efficiency andthereby reducing environmental impact.
58
9.4 Energy integration of Organic Rankine Cycle, Ex-haust Gas recirculation and Scrubber
The study investigated the potential of integrating an organic Rankine cyclewith a two-stroke 10 MW marine diesel engine equipped with exhaust gasrecirculation and a seawater scrubber. The study was based on engine datafrom a de-rated MAN 6G50ME-C95 engine; MAN Diesel & Turbo providedthe data. We identified four possible heat sources: the charge air cooler, theexhaust gas after the turbo, exhaust gas in the EGR-loop and the fresh waterjacket cooler. An organic Rankine cycle was simulated for three different heatsource combinations and with a total of six different organic fluids.
The study demonstrated that an ORC integrated with the engine could producea significant amount of electrical power, and for this specific installation, anexcess power of 184 kW was obtained. The ORC integration can cover andprovide an excess power of the scrubber installation, making the system bothmore efficient and fulfilling Tier III standard for the NOxand sulphur directive.
I performed the simulations and analysis as well as writing the paper. MariaE Mondejar supported with the simulation setup and fluid selection.
9.5 Quasi-steady state simulation of an organic Rank-ine cycle for waste heat recovery in a passengervessel
In this paper, we studied the performance of a regenerated ORC integratedinto a passenger-vessel for waste heat recovery of the engines exhaust gasesover a regular trip. Experimental temperature data from the engine’s exhaustwas logged during the vessel operation for four weeks. The exhaust massflow rates were estimated as a function of the engine’s load and speed and theturbocharger air cooler pressures and temperatures. The exhaust conditionsover the regular round trip were used as input values for an off-design model,which consisted of a regenerated ORC using benzene as working fluid. Thedesign conditions for the off-design model were selected to maximise netpower production over the round trip.
I was responsible for data collection and filtering and calculating the model
61
using optimisation algorithms to select the best performing machine learningalgorithm. These tools are called auto machine learning, which attempts avast set of algorithms as well as tuning of the hyperparameters. The modelscan estimate the performance of each engine without installing flow metresfor each individual engine.
I collected the data, developed the method, built the ML-models and performedthe analysis, as well as writing the paper.
9.3 Predicting Dynamic Fuel Oil Consumption usingAutomated Machine Learning from Large TimeInterval Fuel sums
This paper tests the hypothesis of using fuel readings from larger time intervals,such as noon-reports or even intervals of up to four days, to predict the instantfuel consumption onboard a ship. The data used is the engine data, whichis recorded with high timer intervals, and this is then used as an input forthe model, training on fuel readings for larger time intervals of up to fourdays. The model is then tested for performance at smaller time intervals.Using experiences from the previous study on using auto machine learningto predict the fuel consumption, an optimisation of the best performingML-algorithm was conducted on a very small subset of the ship’s data. Thestudy demonstrated that it is possible to make predictions on dynamic fuelconsumption by training a model with only a few data points.
This can be used as a complement to the installation of additional mass flowmetres and a better base for the decision supports systems on board. Themethod demonstrates that operation data already available on many ships canbe used with noon-reports, or at more seldom bunker intervals, to predictinstant fuel consumption without the additional cost of installing more sensors.The method can be applied directly to existing ships and is cheaper thaninstalling a mass flow metre for each consumer.
I collected the data, developed the method, built the ML-models, and conductedthe analysis, as well as writing the paper.
60
9.4 Energy integration of Organic Rankine Cycle, Ex-haust Gas recirculation and Scrubber
The study investigated the potential of integrating an organic Rankine cyclewith a two-stroke 10 MW marine diesel engine equipped with exhaust gasrecirculation and a seawater scrubber. The study was based on engine datafrom a de-rated MAN 6G50ME-C95 engine; MAN Diesel & Turbo providedthe data. We identified four possible heat sources: the charge air cooler, theexhaust gas after the turbo, exhaust gas in the EGR-loop and the fresh waterjacket cooler. An organic Rankine cycle was simulated for three different heatsource combinations and with a total of six different organic fluids.
The study demonstrated that an ORC integrated with the engine could producea significant amount of electrical power, and for this specific installation, anexcess power of 184 kW was obtained. The ORC integration can cover andprovide an excess power of the scrubber installation, making the system bothmore efficient and fulfilling Tier III standard for the NOxand sulphur directive.
I performed the simulations and analysis as well as writing the paper. MariaE Mondejar supported with the simulation setup and fluid selection.
9.5 Quasi-steady state simulation of an organic Rank-ine cycle for waste heat recovery in a passengervessel
In this paper, we studied the performance of a regenerated ORC integratedinto a passenger-vessel for waste heat recovery of the engines exhaust gasesover a regular trip. Experimental temperature data from the engine’s exhaustwas logged during the vessel operation for four weeks. The exhaust massflow rates were estimated as a function of the engine’s load and speed and theturbocharger air cooler pressures and temperatures. The exhaust conditionsover the regular round trip were used as input values for an off-design model,which consisted of a regenerated ORC using benzene as working fluid. Thedesign conditions for the off-design model were selected to maximise netpower production over the round trip.
I was responsible for data collection and filtering and calculating the model
61
using optimisation algorithms to select the best performing machine learningalgorithm. These tools are called auto machine learning, which attempts avast set of algorithms as well as tuning of the hyperparameters. The modelscan estimate the performance of each engine without installing flow metresfor each individual engine.
I collected the data, developed the method, built the ML-models and performedthe analysis, as well as writing the paper.
9.3 Predicting Dynamic Fuel Oil Consumption usingAutomated Machine Learning from Large TimeInterval Fuel sums
This paper tests the hypothesis of using fuel readings from larger time intervals,such as noon-reports or even intervals of up to four days, to predict the instantfuel consumption onboard a ship. The data used is the engine data, whichis recorded with high timer intervals, and this is then used as an input forthe model, training on fuel readings for larger time intervals of up to fourdays. The model is then tested for performance at smaller time intervals.Using experiences from the previous study on using auto machine learningto predict the fuel consumption, an optimisation of the best performingML-algorithm was conducted on a very small subset of the ship’s data. Thestudy demonstrated that it is possible to make predictions on dynamic fuelconsumption by training a model with only a few data points.
This can be used as a complement to the installation of additional mass flowmetres and a better base for the decision supports systems on board. Themethod demonstrates that operation data already available on many ships canbe used with noon-reports, or at more seldom bunker intervals, to predictinstant fuel consumption without the additional cost of installing more sensors.The method can be applied directly to existing ships and is cheaper thaninstalling a mass flow metre for each consumer.
I collected the data, developed the method, built the ML-models, and conductedthe analysis, as well as writing the paper.
60
Postface
If you would have asked a younger version of me 10 years ago, that personwould never have thought this would have happened. Great things can happenby always trying to see the opportunities which lie ahead of you and chasingthem when they appear, and by maintaining a positive and optimistic attitude.The last five years has been a journey which has transformed me into a differentperson; I would say a better version. At the beginning of this process, ortransformation, I could not imagine that this day would come. It always feltso distant and far away. But here I am, writing the postface of my thesis.
The scientific contribution of this thesis is more about showing what can bedone with the operational data based on what we already have. The storybetween the lines is that everything is possible if you have a positive attitudeand are open to collaboration.
63
simulation input parameters and operation intervals, and I contributed to thewriting. Maria E. Mondejar wrote significant parts of the paper and performedthe simulations and the study results.
9.6 Energy and exergy analysis of a cruise ship
In this paper, we analysed the energy flows in the ship’s energy system.By analysing both the energy as well as exergy flows, we provided a betterunderstanding of how to further optimise new designs, as well as optimisingexisting energy systems. The study used data from the cruise ship M/S BirkaStockholm from a total of one year of operation. The study demonstratedthat thermal energy storage could be feasible. Most of the exergy destructionoccurred in the engines, and the analysis also showed that the engines wereoperating at a low load condition, indicating that hybridisation could be moreoptimal.
I performed the data collection as well as the filtering of data. I contributedto the manuscript, as well as outlining and contributing to the methodologyin discussion with Francesco Baldi and Tuong-Van Nguyen. Tuong-Vandid the clustering of ship operations and also contributed to the manuscript.Francesco was responsible for all energy and exergy flow calculations, theresults, and major parts of the manuscript.
62
Postface
If you would have asked a younger version of me 10 years ago, that personwould never have thought this would have happened. Great things can happenby always trying to see the opportunities which lie ahead of you and chasingthem when they appear, and by maintaining a positive and optimistic attitude.The last five years has been a journey which has transformed me into a differentperson; I would say a better version. At the beginning of this process, ortransformation, I could not imagine that this day would come. It always feltso distant and far away. But here I am, writing the postface of my thesis.
The scientific contribution of this thesis is more about showing what can bedone with the operational data based on what we already have. The storybetween the lines is that everything is possible if you have a positive attitudeand are open to collaboration.
63
simulation input parameters and operation intervals, and I contributed to thewriting. Maria E. Mondejar wrote significant parts of the paper and performedthe simulations and the study results.
9.6 Energy and exergy analysis of a cruise ship
In this paper, we analysed the energy flows in the ship’s energy system.By analysing both the energy as well as exergy flows, we provided a betterunderstanding of how to further optimise new designs, as well as optimisingexisting energy systems. The study used data from the cruise ship M/S BirkaStockholm from a total of one year of operation. The study demonstratedthat thermal energy storage could be feasible. Most of the exergy destructionoccurred in the engines, and the analysis also showed that the engines wereoperating at a low load condition, indicating that hybridisation could be moreoptimal.
I performed the data collection as well as the filtering of data. I contributedto the manuscript, as well as outlining and contributing to the methodologyin discussion with Francesco Baldi and Tuong-Van Nguyen. Tuong-Vandid the clustering of ship operations and also contributed to the manuscript.Francesco was responsible for all energy and exergy flow calculations, theresults, and major parts of the manuscript.
62
References
[1] Thomas Stocker, ed. Climate Change 2013: The Physical Science Basis:Working Group I Contribution to the Fifth Assessment Report of theIntergovernmental Panel on Climate Change. New York: CambridgeUniversity Press, 2014. 1535 pp.
[2] CO2 Emissions from Fuel Combustion 2017 - Highlights. IEA, 2017,p. 162.
[3] World Energy Outlook-2017. International Energy Agency, 2017, p. 32.[4] G Turner. ‘A Comparison of The Limits to Growth with 30 Years of
Reality’. In: Global Environmental Change 18.3 (Aug. 2008), pp. 397–411.
[5] Christophe McGlade and Paul Ekins. ‘The Geographical Distributionof Fossil Fuels Unused When Limiting Global Warming to 2 °C’. In:Nature 517.7533 (Jan. 2015), pp. 187–190.
[6] United Nations Conference on Trade and Development. Review ofMaritime Transport 2018. 2018.
[7] V. Eyring. ‘Emissions from International Shipping: 1. The Last 50Years’. In: Journal of Geophysical Research 110.D17 (2005).
[8] Jun Yuan, Szu Hui Ng and Weng Sut Sou. ‘Uncertainty Quantificationof CO2 Emission Reduction for Maritime Shipping’. In: Energy Policy88 (Jan. 2016), pp. 113–130.
[9] Jana Moldanová et al. ‘Characterisation of Particulate Matter andGaseous Emissions from a Large Ship Diesel Engine’. In: AtmosphericEnvironment 43.16 (May 2009), pp. 2632–2641.
[10] Smith, T. W. P. et al. Third IMO GHG Study 2014 – Final Report.London: International Maritime Organization (IMO), 2014.
65
References
[1] Thomas Stocker, ed. Climate Change 2013: The Physical Science Basis:Working Group I Contribution to the Fifth Assessment Report of theIntergovernmental Panel on Climate Change. New York: CambridgeUniversity Press, 2014. 1535 pp.
[2] CO2 Emissions from Fuel Combustion 2017 - Highlights. IEA, 2017,p. 162.
[3] World Energy Outlook-2017. International Energy Agency, 2017, p. 32.[4] G Turner. ‘A Comparison of The Limits to Growth with 30 Years of
Reality’. In: Global Environmental Change 18.3 (Aug. 2008), pp. 397–411.
[5] Christophe McGlade and Paul Ekins. ‘The Geographical Distributionof Fossil Fuels Unused When Limiting Global Warming to 2 °C’. In:Nature 517.7533 (Jan. 2015), pp. 187–190.
[6] United Nations Conference on Trade and Development. Review ofMaritime Transport 2018. 2018.
[7] V. Eyring. ‘Emissions from International Shipping: 1. The Last 50Years’. In: Journal of Geophysical Research 110.D17 (2005).
[8] Jun Yuan, Szu Hui Ng and Weng Sut Sou. ‘Uncertainty Quantificationof CO2 Emission Reduction for Maritime Shipping’. In: Energy Policy88 (Jan. 2016), pp. 113–130.
[9] Jana Moldanová et al. ‘Characterisation of Particulate Matter andGaseous Emissions from a Large Ship Diesel Engine’. In: AtmosphericEnvironment 43.16 (May 2009), pp. 2632–2641.
[10] Smith, T. W. P. et al. Third IMO GHG Study 2014 – Final Report.London: International Maritime Organization (IMO), 2014.
65
[23] Fredrik Ahlgren et al. ‘Waste Heat Recovery in a Cruise Vessel in theBaltic Sea by Using an Organic Rankine Cycle: A Case Study’. In:Journal of Engineering for Gas Turbines and Power 138.1 (12th Aug.2015), p. 011702.
[24] Francesco Baldi et al. ‘Energy and Exergy Analysis of a Cruise Ship’.In: Energies 11.10 (20th Sept. 2018), p. 2508.
[25] Wes McKinney. ‘Data Structures for Statistical Computing in Python’.In: (2010), p. 6.
[26] Maria E Mondejar and Marcus Thern. ‘A New IPSEpro® Libraryfor the Simulation of Binary Mixtures of Real Fluids in Power CycleAnalysis’. In: Postdoc journal (19th Mar. 2014).
[27] Fredrik Ahlgren et al. ‘Energy Integration of Organic Rankine Cycle,Exhaust Gas Recirculation and Scrubber’. In: Trends and Challengesin Maritime Energy Management. Ed. by Aykut I. Ölçer et al. Vol. 6.Cham: Springer International Publishing, 2018, pp. 157–168.
[28] Baldi Francesco. Github Project Repository for the Analysis of CruiseShip Operational Data for Energy and Exergy Analysis. url: https://github.com/francescobaldi86/Ecos2015PaperExtension.
[29] Francesco Baldi et al. ‘Energy and Exergy Analysis of a Cruise Ship’.In: Proceedings of ECOS 2015. The 28th International Conference onEfficiency, Cost, Optimization, Simulation and Environmental Impactof Energy Systems. Pau, France, 2015, p. 17.
[30] Maria E. Mondejar et al. ‘Study of the On-Route Operation of a WasteHeat Recovery System in a Passenger Vessel’. In: Energy Procedia.Vol. 75. Aug. 2015, pp. 1646–1653.
[31] Maria E. Mondejar et al. ‘Quasi-Steady State Simulation of an OrganicRankine Cycle for Waste Heat Recovery in a Passenger Vessel’. In:Applied Energy 185 (Jan. 2017), pp. 1324–1335.
[32] Anaconda Software Distribution. Computer Software. Version 5.2.30th May 2018.
[33] Kluyver Thomas et al. ‘Jupyter Notebooks - a Publishing Formatfor Reproducible Computational Workflows’. In: Stand Alone (2016),pp. 87–90.
[34] Peter Atkins. The Laws of Thermodynamics: A Very Short Introduction.Oxford University Press, 1st Mar. 2010.
67
[11] Maritime Executive. ‘IMO Agrees to CO2 Emissions Target’. In: TheMaritime Executive, LLC (13th Apr. 2018).
[12] Daria Gritsenko. ‘Regulating GHG Emissions from Shipping: Local,Global, or Polycentric Approach?’ In: Marine Policy 84 (Oct. 2017),pp. 130–133.
[13] Stephen Horvath, Mahdi Fasihi and Christian Breyer. ‘Techno-EconomicAnalysis of a Decarbonized Shipping Sector: Technology Suggestionsfor a Fleet in 2030 and 2040’. In: Energy Conversion and Management164 (May 2018), pp. 230–241.
[14] Martin Stopford. MARITIME ECONOMICS, Third Edition. Routledge,Taylor & Francis, 2008.
[15] Jukka-Pekka Jalkanen, Lasse Johansson and Jaakko Kukkonen. ‘AComprehensive Inventory of the Ship Traffic Exhaust Emissions in theBaltic Sea from 2006 to 2009’. In: AMBIO 43.3 (Apr. 2014), pp. 311–324.
[16] V. Eyring. ‘Emissions from International Shipping: 1. The Last 50Years’. In: Journal of Geophysical Research 110.D17 (2005).
[17] Øyvind Endresen et al. ‘Improved Modelling of Ship SO2 Emissions—aFuel-Based Approach’. In: Atmospheric Environment 39.20 (June2005), pp. 3621–3628.
[18] International Maritime Organization. IMO Sets 2020 Date for Ships toComply with Low Sulphur Fuel Oil Requirement. 28th Oct. 2016. url:http://www.imo.org/en/mediacentre/pressbriefings/pages/mepc-70-2020sulphur.aspx.
[19] Victor N. Armstrong and Charlotte Banks. ‘Integrated Approach toVessel Energy Efficiency’. In: Ocean Engineering 110 (Dec. 2015),pp. 39–48.
[20] Evert A. Bouman et al. ‘State-of-the-Art Technologies, Measures, andPotential for Reducing GHG Emissions from Shipping – A Review’.In: Transportation Research Part D: Transport and Environment 52(May 2017), pp. 408–421.
[21] Nishatabbas Rehmatulla and Tristan Smith. ‘Barriers to Energy Efficientand Low Carbon Shipping’. In: Ocean Engineering 110 (Dec. 2015),pp. 102–112.
[22] Wärtsilä. Project Guide Wärtsilä 46 - 3/2007. 2007.
66
[23] Fredrik Ahlgren et al. ‘Waste Heat Recovery in a Cruise Vessel in theBaltic Sea by Using an Organic Rankine Cycle: A Case Study’. In:Journal of Engineering for Gas Turbines and Power 138.1 (12th Aug.2015), p. 011702.
[24] Francesco Baldi et al. ‘Energy and Exergy Analysis of a Cruise Ship’.In: Energies 11.10 (20th Sept. 2018), p. 2508.
[25] Wes McKinney. ‘Data Structures for Statistical Computing in Python’.In: (2010), p. 6.
[26] Maria E Mondejar and Marcus Thern. ‘A New IPSEpro® Libraryfor the Simulation of Binary Mixtures of Real Fluids in Power CycleAnalysis’. In: Postdoc journal (19th Mar. 2014).
[27] Fredrik Ahlgren et al. ‘Energy Integration of Organic Rankine Cycle,Exhaust Gas Recirculation and Scrubber’. In: Trends and Challengesin Maritime Energy Management. Ed. by Aykut I. Ölçer et al. Vol. 6.Cham: Springer International Publishing, 2018, pp. 157–168.
[28] Baldi Francesco. Github Project Repository for the Analysis of CruiseShip Operational Data for Energy and Exergy Analysis. url: https://github.com/francescobaldi86/Ecos2015PaperExtension.
[29] Francesco Baldi et al. ‘Energy and Exergy Analysis of a Cruise Ship’.In: Proceedings of ECOS 2015. The 28th International Conference onEfficiency, Cost, Optimization, Simulation and Environmental Impactof Energy Systems. Pau, France, 2015, p. 17.
[30] Maria E. Mondejar et al. ‘Study of the On-Route Operation of a WasteHeat Recovery System in a Passenger Vessel’. In: Energy Procedia.Vol. 75. Aug. 2015, pp. 1646–1653.
[31] Maria E. Mondejar et al. ‘Quasi-Steady State Simulation of an OrganicRankine Cycle for Waste Heat Recovery in a Passenger Vessel’. In:Applied Energy 185 (Jan. 2017), pp. 1324–1335.
[32] Anaconda Software Distribution. Computer Software. Version 5.2.30th May 2018.
[33] Kluyver Thomas et al. ‘Jupyter Notebooks - a Publishing Formatfor Reproducible Computational Workflows’. In: Stand Alone (2016),pp. 87–90.
[34] Peter Atkins. The Laws of Thermodynamics: A Very Short Introduction.Oxford University Press, 1st Mar. 2010.
67
[11] Maritime Executive. ‘IMO Agrees to CO2 Emissions Target’. In: TheMaritime Executive, LLC (13th Apr. 2018).
[12] Daria Gritsenko. ‘Regulating GHG Emissions from Shipping: Local,Global, or Polycentric Approach?’ In: Marine Policy 84 (Oct. 2017),pp. 130–133.
[13] Stephen Horvath, Mahdi Fasihi and Christian Breyer. ‘Techno-EconomicAnalysis of a Decarbonized Shipping Sector: Technology Suggestionsfor a Fleet in 2030 and 2040’. In: Energy Conversion and Management164 (May 2018), pp. 230–241.
[14] Martin Stopford. MARITIME ECONOMICS, Third Edition. Routledge,Taylor & Francis, 2008.
[15] Jukka-Pekka Jalkanen, Lasse Johansson and Jaakko Kukkonen. ‘AComprehensive Inventory of the Ship Traffic Exhaust Emissions in theBaltic Sea from 2006 to 2009’. In: AMBIO 43.3 (Apr. 2014), pp. 311–324.
[16] V. Eyring. ‘Emissions from International Shipping: 1. The Last 50Years’. In: Journal of Geophysical Research 110.D17 (2005).
[17] Øyvind Endresen et al. ‘Improved Modelling of Ship SO2 Emissions—aFuel-Based Approach’. In: Atmospheric Environment 39.20 (June2005), pp. 3621–3628.
[18] International Maritime Organization. IMO Sets 2020 Date for Ships toComply with Low Sulphur Fuel Oil Requirement. 28th Oct. 2016. url:http://www.imo.org/en/mediacentre/pressbriefings/pages/mepc-70-2020sulphur.aspx.
[19] Victor N. Armstrong and Charlotte Banks. ‘Integrated Approach toVessel Energy Efficiency’. In: Ocean Engineering 110 (Dec. 2015),pp. 39–48.
[20] Evert A. Bouman et al. ‘State-of-the-Art Technologies, Measures, andPotential for Reducing GHG Emissions from Shipping – A Review’.In: Transportation Research Part D: Transport and Environment 52(May 2017), pp. 408–421.
[21] Nishatabbas Rehmatulla and Tristan Smith. ‘Barriers to Energy Efficientand Low Carbon Shipping’. In: Ocean Engineering 110 (Dec. 2015),pp. 102–112.
[22] Wärtsilä. Project Guide Wärtsilä 46 - 3/2007. 2007.
66
[48] George Tsatsaronis. ‘Definitions and Nomenclature in Exergy Analysisand Exergoeconomics’. In: Energy 32.4 (Apr. 2007), pp. 249–253.
[49] IBM. 10 Key Marketing Trends for 2017 and Ideas for ExceedingCustomer Expectations. 9th May 2017.
[50] David L. Poole and Alan K. Mackworth. Artificial Intelligence: Found-ations of Computational Agents. The Edinburgh Building, CambridgeCB2 8RU, UK: Cambridge University Press, 2010.
[51] E. Bal Beşikçi et al. ‘An Artificial Neural Network Based DecisionSupport System for Energy Efficient Ship Operations’. In: Computers& Operations Research 66 (Feb. 2016), pp. 393–401.
[52] Sebastian Raschka. Python Machine Learning: Unlock Deeper Insightsinto Machine Learning with This Vital Guide to Cutting-Edge Predict-ive Analytics. Community experience distilled. OCLC: 927507196.Birmingham Mumbai: Packt Publishing open source, 2016. 425 pp.
[53] Fabian Pedregosa et al. ‘Scikit-Learn: Machine Learning in Python’.In: Journal of Machine Learning Research 12 (), p. 6.
[54] scikit-learn developers. Choosing the Right Estimator. url: http://scikit-learn.org/stable/tutorial/machine_learning_map/index.html.
[55] Randal S. Olson et al. ‘Evaluation of a Tree-Based Pipeline OptimizationTool for Automating Data Science’. In: ACM Press, 2016, pp. 485–492.
[56] Randal S. Olson et al. ‘A System for Accessible Artificial Intelligence’.In: arXiv preprint arXiv:1705.00594 (2017).
[57] Randal S Olson. ‘TPOT: A Tree-Based Pipeline Optimization Tool forAutomating Machine Learning’. In: JMLR: Workshop and ConferenceProceedings 64:66–74, 2016. ICML 2016 AutoML Workshop, p. 9.
[58] Matthias Feurer et al. ‘Efficient and Robust Automated Machine Learn-ing’. In: Advances in Neural Information Processing Systems. 2015,pp. 2962–2970.
[59] Francesco Baldi and Cecilia Gabrielii. ‘A Feasibility Analysis of WasteHeat Recovery Systems for Marine Applications’. In: Energy 80 (Feb.2015), pp. 654–665.
[60] Gequn Shu et al. ‘A Review of Waste Heat Recovery on Two-Stroke ICEngine Aboard Ships’. In: Renewable and Sustainable Energy Reviews19 (Mar. 2013), pp. 385–401.
69
[35] Moran Michael J. et al. Principles of Engineering Thermodynamics(7th Ed.) 7th. John Wiley & Sons, Inc., 2015.
[36] V. Eyring. ‘Emissions from International Shipping: 2. Impact of FutureTechnologies on Scenarios until 2050’. In: Journal of GeophysicalResearch 110.D17 (2005).
[37] Wärtsilä Finland Oy. New Wärtsilä 31 Engine Achieves Guinness WorldRecords Title. 2nd June 2015. url: https://www.wartsila.com/media/news/02-06-2015-new-wartsila-31-engine-achieves-guinness-world-records-title.
[38] Douglas F. Woodyard, ed. Pounder’s Marine Diesel Engines and GasTurbines. 9. ed. OCLC: 836807832. Amsterdam: Elsevier, 2009. 896 pp.
[39] Fredrik Ahlgren et al. ‘Waste Heat Recovery in a Cruise Vessel in theBaltic Sea by Using an Organic Rankine Cycle: A Case Study’. In:Journal of Engineering for Gas Turbines and Power 138.1 (12th Aug.2015), p. 011702.
[40] Monthly Electricity Statistics. February 2018. International EnergyAgency, 15th May 2018.
[41] Tomei Gregory L, ed. Steam Its Generation and Use. 42nd edition.Charlotte, North Carolina, U.S.A: The Babcock & Wilcox Company,2015.
[42] Thomas Tartière and Marco Astolfi. ‘A World Overview of the OrganicRankine Cycle Market’. In: Energy Procedia 129 (Sept. 2017), pp. 2–9.
[43] Piero Colonna et al. ‘Organic Rankine Cycle Power Systems: Fromthe Concept to Current Technology, Applications, and an Outlook tothe Future’. In: Journal of Engineering for Gas Turbines and Power137.10 (1st Oct. 2015), p. 100801.
[44] M.E. Mondejar et al. ‘A Review of the Use of Organic RankineCycle Power Systems for Maritime Applications’. In: Renewable andSustainable Energy Reviews 91 (Aug. 2018), pp. 126–151.
[45] D.A. Snow. Plant Engineer’s Reference Book. Oxford, UK.: Butterworth-Heinemann, 2001.
[46] Veronika Eyring et al. ‘Transport Impacts on Atmosphere and Climate:Shipping’. In: Atmospheric Environment 44.37 (Dec. 2010), pp. 4735–4771.
[47] Ennio Macchi and Marco Astolfi. Organic Rankine Cycle (ORC) PowerSystems - Technologies and Applications. Woodhead Publishing, 2016.
68
[48] George Tsatsaronis. ‘Definitions and Nomenclature in Exergy Analysisand Exergoeconomics’. In: Energy 32.4 (Apr. 2007), pp. 249–253.
[49] IBM. 10 Key Marketing Trends for 2017 and Ideas for ExceedingCustomer Expectations. 9th May 2017.
[50] David L. Poole and Alan K. Mackworth. Artificial Intelligence: Found-ations of Computational Agents. The Edinburgh Building, CambridgeCB2 8RU, UK: Cambridge University Press, 2010.
[51] E. Bal Beşikçi et al. ‘An Artificial Neural Network Based DecisionSupport System for Energy Efficient Ship Operations’. In: Computers& Operations Research 66 (Feb. 2016), pp. 393–401.
[52] Sebastian Raschka. Python Machine Learning: Unlock Deeper Insightsinto Machine Learning with This Vital Guide to Cutting-Edge Predict-ive Analytics. Community experience distilled. OCLC: 927507196.Birmingham Mumbai: Packt Publishing open source, 2016. 425 pp.
[53] Fabian Pedregosa et al. ‘Scikit-Learn: Machine Learning in Python’.In: Journal of Machine Learning Research 12 (), p. 6.
[54] scikit-learn developers. Choosing the Right Estimator. url: http://scikit-learn.org/stable/tutorial/machine_learning_map/index.html.
[55] Randal S. Olson et al. ‘Evaluation of a Tree-Based Pipeline OptimizationTool for Automating Data Science’. In: ACM Press, 2016, pp. 485–492.
[56] Randal S. Olson et al. ‘A System for Accessible Artificial Intelligence’.In: arXiv preprint arXiv:1705.00594 (2017).
[57] Randal S Olson. ‘TPOT: A Tree-Based Pipeline Optimization Tool forAutomating Machine Learning’. In: JMLR: Workshop and ConferenceProceedings 64:66–74, 2016. ICML 2016 AutoML Workshop, p. 9.
[58] Matthias Feurer et al. ‘Efficient and Robust Automated Machine Learn-ing’. In: Advances in Neural Information Processing Systems. 2015,pp. 2962–2970.
[59] Francesco Baldi and Cecilia Gabrielii. ‘A Feasibility Analysis of WasteHeat Recovery Systems for Marine Applications’. In: Energy 80 (Feb.2015), pp. 654–665.
[60] Gequn Shu et al. ‘A Review of Waste Heat Recovery on Two-Stroke ICEngine Aboard Ships’. In: Renewable and Sustainable Energy Reviews19 (Mar. 2013), pp. 385–401.
69
[35] Moran Michael J. et al. Principles of Engineering Thermodynamics(7th Ed.) 7th. John Wiley & Sons, Inc., 2015.
[36] V. Eyring. ‘Emissions from International Shipping: 2. Impact of FutureTechnologies on Scenarios until 2050’. In: Journal of GeophysicalResearch 110.D17 (2005).
[37] Wärtsilä Finland Oy. New Wärtsilä 31 Engine Achieves Guinness WorldRecords Title. 2nd June 2015. url: https://www.wartsila.com/media/news/02-06-2015-new-wartsila-31-engine-achieves-guinness-world-records-title.
[38] Douglas F. Woodyard, ed. Pounder’s Marine Diesel Engines and GasTurbines. 9. ed. OCLC: 836807832. Amsterdam: Elsevier, 2009. 896 pp.
[39] Fredrik Ahlgren et al. ‘Waste Heat Recovery in a Cruise Vessel in theBaltic Sea by Using an Organic Rankine Cycle: A Case Study’. In:Journal of Engineering for Gas Turbines and Power 138.1 (12th Aug.2015), p. 011702.
[40] Monthly Electricity Statistics. February 2018. International EnergyAgency, 15th May 2018.
[41] Tomei Gregory L, ed. Steam Its Generation and Use. 42nd edition.Charlotte, North Carolina, U.S.A: The Babcock & Wilcox Company,2015.
[42] Thomas Tartière and Marco Astolfi. ‘A World Overview of the OrganicRankine Cycle Market’. In: Energy Procedia 129 (Sept. 2017), pp. 2–9.
[43] Piero Colonna et al. ‘Organic Rankine Cycle Power Systems: Fromthe Concept to Current Technology, Applications, and an Outlook tothe Future’. In: Journal of Engineering for Gas Turbines and Power137.10 (1st Oct. 2015), p. 100801.
[44] M.E. Mondejar et al. ‘A Review of the Use of Organic RankineCycle Power Systems for Maritime Applications’. In: Renewable andSustainable Energy Reviews 91 (Aug. 2018), pp. 126–151.
[45] D.A. Snow. Plant Engineer’s Reference Book. Oxford, UK.: Butterworth-Heinemann, 2001.
[46] Veronika Eyring et al. ‘Transport Impacts on Atmosphere and Climate:Shipping’. In: Atmospheric Environment 44.37 (Dec. 2010), pp. 4735–4771.
[47] Ennio Macchi and Marco Astolfi. Organic Rankine Cycle (ORC) PowerSystems - Technologies and Applications. Woodhead Publishing, 2016.
68
I
62
support. Perspectives on Psychological Science, 4(3), 236-255. doi: 10.1111/j.1745-6924.2009.01122.x
Van Humbeeck, L., Piers, R. D., Van Camp, S., Dillen L, Verhaeghe, S. T., van den
Noortgate N. J. (2013). Aged parents’ experiences during a critical illness trajectory and after the death of an adult child: A review of the literature. Palliative Medicine, 27(7):583-95. doi: 10.1177/0269216313483662
Wei, M., Liao, K. Y. H., Ku, T. Y., & Shaffer, P. A. (2011). Attachment, self‐compassion,
empathy, and subjective well‐being among college students and community adults. Journal of Personality, 79(1), 191-221. doi: 10.1111/j.1467-6494.2010.00677.x
Weiss, A., & Costa, P. T. (2005). Domain and facet personality predictors of all-cause mortality among medicare patients aged 65 to 100. Psychosomatic medicine, 67(5), 724-733. doi: 10.1097/01.psy.0000181272.58103.18
Wijngaards-de Meij, L., Stroebe, M. S., Schut, H. A. W., Stroebe, W., van den Bout, J., van der Heijden, P. G. M. (2005). Couples at risk following the death of their child: Predictors of grief versus depression. Journal of Consulting and Clinical Psychology, 73(4), 617-623. doi: 10.1037/0022-006x.73.4.617
Wijngaards-De Meij, L., Stroebe, M., Stroebe, W., Schut, H., Van den Bout, J., Van Der
Heijden, P.G., et al. (2008). The impact of circumstances surrounding the death of a child on parents’ grief. Death Studies, 32(3), 237-252. doi: 10.1080/07481180701881263
World Health Organization (WHO). (2006). Constitution of the World Health
Organization. Available at http://www.who.int/governance/eb/who_constitution_en.pdf Zimet, G. D., Dahlem, N. W., Zimet, S. G., & Farley, G. K. (1988). The multidimensional
scale of perceived social support. Journal of Personality Assessment, 52(1), 30-41. doi: 10.1207/s15327752jpa5201_2
[61] Francesco Baldi et al. ‘Optimal Load Allocation of Complex ShipPower Plants’. In: Energy Conversion and Management 124 (Sept.2016), pp. 344–356.
[62] David Silver et al. ‘Mastering the Game of Go without Human Know-ledge’. In: Nature 550.7676 (18th Oct. 2017), pp. 354–359.
[63] Evans Richard and Gao Jim. DeepMind AI Reduces Google Data CentreCooling Bill by 40%. 20th June 2016. url: https://deepmind.com/blog/deepmind-ai-reduces-google-data-centre-cooling-bill-40/.
[64] Tesla. All Tesla Cars Being Produced Now Have Full Self-DrivingHardware. 16th Sept. 2016. url: https://www.tesla.com/sv_SE/blog/all-tesla- cars-being-produced-now-have- full- self- driving-hardware(visited on 19/09/2018).
[65] Umar Zakir Abdul Hamid et al. ‘Current Collision Mitigation Tech-nologies for Advanced Driver Assistance Systems – A Survey’. In:PERINTIS eJournal (2016).
[66] Lawrence Livermore National Laboratory. ASCI White: The World’sFastest Computer - Meeting the Challenge of Stockpile Stewardship.Nov. 2014. url: https://web.archive.org/web/20141111101755/http:/ / computation . llnl .gov / casc / sc2001_fliers /ASCI_White /ASCI_White01.html.
[67] Maria Sharmina et al. ‘Global Energy Scenarios and Their Implicationsfor Future Shipped Trade’. In: Marine Policy 84 (Oct. 2017), pp. 12–21.
[68] IMO - 2010 - RESOLUTION MEPC.176(58) (Revised MARPOL AnnexVI). International Maritime Organization, July 2010, pp. 1–45.
[69] Andrea Vicenzutti et al. ‘The Role of Voltage Controls in Modern All-Electric Ships: Toward the All Electric Ship.’ In: IEEE ElectrificationMagazine 3.2 (June 2015), pp. 49–65.
[70] George S. Misyris et al. ‘On Battery State Estimation Algorithms forElectric Ship Applications’. In: Electric Power Systems Research 151(Oct. 2017), pp. 115–124.
[71] Michael Fischer, Mathew Werber and Peter V. Schwartz. ‘Batteries:Higher Energy Density than Gasoline?’ In: Energy Policy 37.7 (July2009), pp. 2639–2641.
[72] Ship & Bunker. Ship & Bunker Prices, 2013-2018. 19th Sept. 2018.url: https://shipandbunker.com/prices (visited on 19/09/2018).
70
Recommended