서브메뉴
검색
Modeling and Predicting Heavy-Duty Vehicle Nitrogen Oxide Emissions Using Deep Learning- [electronic resource]
Modeling and Predicting Heavy-Duty Vehicle Nitrogen Oxide Emissions Using Deep Learning- [electronic resource]
상세정보
- 자료유형
- 학위논문
- Control Number
- 0016933666
- International Standard Book Number
- 9798379565473
- Dewey Decimal Classification Number
- 621
- Main Entry-Personal Name
- Pillai, Rinav Raveendran.
- Publication, Distribution, etc. (Imprint
- [S.l.] : University of Michigan., 2023
- Publication, Distribution, etc. (Imprint
- Ann Arbor : ProQuest Dissertations & Theses, 2023
- Physical Description
- 1 online resource(213 p.)
- General Note
- Source: Dissertations Abstracts International, Volume: 84-12, Section: B.
- General Note
- Advisor: Boehman, Andre L.
- Dissertation Note
- Thesis (Ph.D.)--University of Michigan, 2023.
- Restrictions on Access Note
- This item must not be sold to any third party vendors.
- Restrictions on Access Note
- This item must not be added to any third party search indexes.
- Summary, Etc.
- 요약Heavy-duty vehicles are powered by diesel engines that emit significant amounts of NOx emissions which are detrimental to human health and the environment. As emissions regulations for transportation become stricter, it has become increasingly important to develop accurate NOx emissions models for heavy-duty vehicles. However, estimation of transient NOx emissions is challenging due to its highly dynamic nature. Traditional thermophysical, chemical and semi-empirical NOx models require large number of assumptions and consume high quantities of computational time and cost. Therefore, this research investigates a multi-dimensional data driven approach to estimate NOx emissions in heavy-duty diesel vehicles using machine learning techniques.With increasing quantities of data being collected through laboratory and on-road compliance tests, a deep learning neural network (DNN) method to predicting NOx emissions at representative engine and aftertreatment operating conditions was explored in this thesis. DNN models were developed for estimation of engine-out and tailpipe NOx emissions using physics inspired engine and exhaust aftertreatment variables readily available from On-Board Diagnostics (OBD) of heavy-duty vehicles. A 6.7L engine (different model years) from a heavy-duty bus was tested on an engine and chassis dynamometer using test cycles representative of in-use operation, to develop a comprehensive dataset for training the DNN models. Results from testing on unseen dynamometer test data showed high prediction accuracy for engine-out (R2 = 0.99 and 0.97) and tailpipe (R2 = 0.99 and 0.93) NOx emissions on engine and chassis dynamometer datasets respectively.Variable importance studies conducted to determine significant engine and aftertreatment inputs that affect DNN NOx prediction demonstrated high prediction accuracy (R2 = 0.92 to 0.95), while utilizing minimal significant inputs. Additionally, the model was shown to successfully detect 60% higher tailpipe NOx emissions on a similar 6.7L engine fitted with a faulty selective catalytic reduction (SCR) system. On-road testing was conducted on a conventional heavy-duty diesel bus operating on winter and summer biodiesel fuels using an OBD logger and a Horiba Smart Emissions Measurement System (SEMS) to develop a dataset that included variables such as road grade and ambient conditions, the effects of which are not necessarily captured in the controlled environment of laboratory testing.DNN tailpipe NOx models subsequently developed using on-road datasets showed high accuracy (R2 = 0.97) on unseen on-road test data collected using SEMS. In addition, the effect of different fuel types on DNN tailpipe NOx predictions was explored utilizing fuel type as a categorical input to the DNN model. The results from the study showed good estimation of NOx emissions using multiple fuels (R2 = 0.97). A Feed-Forward DNN (FF-DNN) model with a single time shifted NOx feature was proposed which demonstrated comparable accuracy with time series forecasting DNN models in the literature, with reduced computational time for model training. A potential macro-scale anomaly detection tool was also developed by training a DNN classification model using on-road datasets.This dissertation demonstrates that accurate data-driven NOx emissions models can be developed using DNNs without the need for physical or chemical equations. Moreover, the potential application of DNN models to NOx control and aftertreatment operation, fault detection in SCR systems and supplementing in-use heavy-duty vehicle NOx emissions inventories in understanding the potential for future NOx emissions regulations has been explored through these detailed studies.
- Subject Added Entry-Topical Term
- Mechanical engineering.
- Index Term-Uncontrolled
- Modelling nitrogen oxide emissions
- Index Term-Uncontrolled
- Deep learning
- Index Term-Uncontrolled
- Nitrogen oxide emissions
- Index Term-Uncontrolled
- Data-driven modelling
- Index Term-Uncontrolled
- On-road vehicle testing
- Index Term-Uncontrolled
- Deep learning neural networks
- Index Term-Uncontrolled
- Emissions modelling
- Added Entry-Corporate Name
- University of Michigan Mechanical Engineering
- Host Item Entry
- Dissertations Abstracts International. 84-12B.
- Host Item Entry
- Dissertation Abstract International
- Electronic Location and Access
- 로그인을 한후 보실 수 있는 자료입니다.
- Control Number
- joongbu:641013
MARC
008240220s2023 ulk 00 kor■001000016933666
■00520240214101320
■006m o d
■007cr#unu||||||||
■020 ▼a9798379565473
■035 ▼a(MiAaPQ)AAI30548513
■035 ▼a(MiAaPQ)umichrackham004882
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a621
■1001 ▼aPillai, Rinav Raveendran.
■24510▼aModeling and Predicting Heavy-Duty Vehicle Nitrogen Oxide Emissions Using Deep Learning▼h[electronic resource]
■260 ▼a[S.l.]▼bUniversity of Michigan. ▼c2023
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2023
■300 ▼a1 online resource(213 p.)
■500 ▼aSource: Dissertations Abstracts International, Volume: 84-12, Section: B.
■500 ▼aAdvisor: Boehman, Andre L.
■5021 ▼aThesis (Ph.D.)--University of Michigan, 2023.
■506 ▼aThis item must not be sold to any third party vendors.
■506 ▼aThis item must not be added to any third party search indexes.
■520 ▼aHeavy-duty vehicles are powered by diesel engines that emit significant amounts of NOx emissions which are detrimental to human health and the environment. As emissions regulations for transportation become stricter, it has become increasingly important to develop accurate NOx emissions models for heavy-duty vehicles. However, estimation of transient NOx emissions is challenging due to its highly dynamic nature. Traditional thermophysical, chemical and semi-empirical NOx models require large number of assumptions and consume high quantities of computational time and cost. Therefore, this research investigates a multi-dimensional data driven approach to estimate NOx emissions in heavy-duty diesel vehicles using machine learning techniques.With increasing quantities of data being collected through laboratory and on-road compliance tests, a deep learning neural network (DNN) method to predicting NOx emissions at representative engine and aftertreatment operating conditions was explored in this thesis. DNN models were developed for estimation of engine-out and tailpipe NOx emissions using physics inspired engine and exhaust aftertreatment variables readily available from On-Board Diagnostics (OBD) of heavy-duty vehicles. A 6.7L engine (different model years) from a heavy-duty bus was tested on an engine and chassis dynamometer using test cycles representative of in-use operation, to develop a comprehensive dataset for training the DNN models. Results from testing on unseen dynamometer test data showed high prediction accuracy for engine-out (R2 = 0.99 and 0.97) and tailpipe (R2 = 0.99 and 0.93) NOx emissions on engine and chassis dynamometer datasets respectively.Variable importance studies conducted to determine significant engine and aftertreatment inputs that affect DNN NOx prediction demonstrated high prediction accuracy (R2 = 0.92 to 0.95), while utilizing minimal significant inputs. Additionally, the model was shown to successfully detect 60% higher tailpipe NOx emissions on a similar 6.7L engine fitted with a faulty selective catalytic reduction (SCR) system. On-road testing was conducted on a conventional heavy-duty diesel bus operating on winter and summer biodiesel fuels using an OBD logger and a Horiba Smart Emissions Measurement System (SEMS) to develop a dataset that included variables such as road grade and ambient conditions, the effects of which are not necessarily captured in the controlled environment of laboratory testing.DNN tailpipe NOx models subsequently developed using on-road datasets showed high accuracy (R2 = 0.97) on unseen on-road test data collected using SEMS. In addition, the effect of different fuel types on DNN tailpipe NOx predictions was explored utilizing fuel type as a categorical input to the DNN model. The results from the study showed good estimation of NOx emissions using multiple fuels (R2 = 0.97). A Feed-Forward DNN (FF-DNN) model with a single time shifted NOx feature was proposed which demonstrated comparable accuracy with time series forecasting DNN models in the literature, with reduced computational time for model training. A potential macro-scale anomaly detection tool was also developed by training a DNN classification model using on-road datasets.This dissertation demonstrates that accurate data-driven NOx emissions models can be developed using DNNs without the need for physical or chemical equations. Moreover, the potential application of DNN models to NOx control and aftertreatment operation, fault detection in SCR systems and supplementing in-use heavy-duty vehicle NOx emissions inventories in understanding the potential for future NOx emissions regulations has been explored through these detailed studies.
■590 ▼aSchool code: 0127.
■650 4▼aMechanical engineering.
■653 ▼aModelling nitrogen oxide emissions
■653 ▼aDeep learning
■653 ▼aNitrogen oxide emissions
■653 ▼aData-driven modelling
■653 ▼aOn-road vehicle testing
■653 ▼aDeep learning neural networks
■653 ▼aEmissions modelling
■690 ▼a0548
■71020▼aUniversity of Michigan▼bMechanical Engineering.
■7730 ▼tDissertations Abstracts International▼g84-12B.
■773 ▼tDissertation Abstract International
■790 ▼a0127
■791 ▼aPh.D.
■792 ▼a2023
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T16933666▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
■980 ▼a202402▼f2024