Application of artificial neural network to forecast engine performance and emissions of a spark ignition engine
Fu, Jiahong LU ; Yang, Ruomiao ; Li, Xin ; Sun, Xiaoxia ; Li, Yong LU
; Liu, Zhentao
; Zhang, Yu
and Sunden, Bengt
LU
(2022)
In Applied Thermal Engineering
201.
- Abstract
Increasing the application of machine learning algorithms in engine development has the potential to reduce the number of experimental runs and the computation cost of computational fluid dynamics simulations. The objective of this study is to assess if such a statistical modelling approach can predict engine efficiency and emissions at any given condition for an already calibrated spark ignition (SI) engine. Engine responses at various engine speeds and load are recorded and used for correlative modelling. The artificial neural network (ANN) algorithm is utilized in this study, with engine speed and load as the model inputs, and fuel consumption and emission as the model outputs. The comparisons between experimentally measured data and... (More)
Increasing the application of machine learning algorithms in engine development has the potential to reduce the number of experimental runs and the computation cost of computational fluid dynamics simulations. The objective of this study is to assess if such a statistical modelling approach can predict engine efficiency and emissions at any given condition for an already calibrated spark ignition (SI) engine. Engine responses at various engine speeds and load are recorded and used for correlative modelling. The artificial neural network (ANN) algorithm is utilized in this study, with engine speed and load as the model inputs, and fuel consumption and emission as the model outputs. The comparisons between experimentally measured data and model predictions indicate that the well-trained network is capable of forecasting engine efficiency, unburned hydrocarbons, carbon monoxide, and nitrogen oxide emissions with close-to-zero root mean squared error performance metric. In addition, the relatively small errors do not affect the relations between model inputs and outputs, as evidenced by the close-to-unity coefficient of determination. Overall, all these results indicate ANN model is appropriate for the application investigated in this study. Moreover, this study also suggests that the “black-box” modelling approach has the potential to effectively predict engine-related variables. And the predicted engine map can be used as a reference to accelerate the motor development in the hybrid vehicles. Also, the ANN model forecast the fuel consumption and emissions under transient operating conditions, while the literature is scarce to date on the investigation of the prediction of engine responses for transient conditions.
(Less)
- author
- Fu, Jiahong
LU
; Yang, Ruomiao
; Li, Xin
; Sun, Xiaoxia
; Li, Yong
LU
; Liu, Zhentao
; Zhang, Yu
and Sunden, Bengt
LU
- organization
- publishing date
- 2022
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- Artificial neural network, Engine response prediction, Machine learning, Spark ignition engine
- in
- Applied Thermal Engineering
- volume
- 201
- article number
- 117749
- publisher
- Elsevier
- external identifiers
-
- scopus:85120324908
- ISSN
- 1359-4311
- DOI
- 10.1016/j.applthermaleng.2021.117749
- language
- English
- LU publication?
- yes
- id
- 7b40d854-3e8b-4ed7-bdde-00c84fc7e63f
- date added to LUP
- 2021-12-15 10:43:30
- date last changed
- 2023-11-09 01:48:58
@article{7b40d854-3e8b-4ed7-bdde-00c84fc7e63f,
abstract = {{<p>Increasing the application of machine learning algorithms in engine development has the potential to reduce the number of experimental runs and the computation cost of computational fluid dynamics simulations. The objective of this study is to assess if such a statistical modelling approach can predict engine efficiency and emissions at any given condition for an already calibrated spark ignition (SI) engine. Engine responses at various engine speeds and load are recorded and used for correlative modelling. The artificial neural network (ANN) algorithm is utilized in this study, with engine speed and load as the model inputs, and fuel consumption and emission as the model outputs. The comparisons between experimentally measured data and model predictions indicate that the well-trained network is capable of forecasting engine efficiency, unburned hydrocarbons, carbon monoxide, and nitrogen oxide emissions with close-to-zero root mean squared error performance metric. In addition, the relatively small errors do not affect the relations between model inputs and outputs, as evidenced by the close-to-unity coefficient of determination. Overall, all these results indicate ANN model is appropriate for the application investigated in this study. Moreover, this study also suggests that the “black-box” modelling approach has the potential to effectively predict engine-related variables. And the predicted engine map can be used as a reference to accelerate the motor development in the hybrid vehicles. Also, the ANN model forecast the fuel consumption and emissions under transient operating conditions, while the literature is scarce to date on the investigation of the prediction of engine responses for transient conditions.</p>}},
author = {{Fu, Jiahong and Yang, Ruomiao and Li, Xin and Sun, Xiaoxia and Li, Yong and Liu, Zhentao and Zhang, Yu and Sunden, Bengt}},
issn = {{1359-4311}},
keywords = {{Artificial neural network; Engine response prediction; Machine learning; Spark ignition engine}},
language = {{eng}},
publisher = {{Elsevier}},
series = {{Applied Thermal Engineering}},
title = {{Application of artificial neural network to forecast engine performance and emissions of a spark ignition engine}},
url = {{http://dx.doi.org/10.1016/j.applthermaleng.2021.117749}},
doi = {{10.1016/j.applthermaleng.2021.117749}},
volume = {{201}},
year = {{2022}},
}