| APPLICATION OF PHYSICAL MODEL TEST-BASED LONG
SHORT-TERM MEMORY ALGORITHM AS A VIRTUAL SENSOR FOR
NITROGEN OXIDE PREDICTION IN DIESEL ENGINES |
| Dalho Shin 1, Seongin Jo 2, Hyung Jun Kim 3, Suhan Park 4 |
1Department of Mechanical Engineering, Konkuk University
2Department of Mechanical Engineering, Chonnam National University
3Transportation Pollution Research Center, National Institute of Environmental Research
4School of Mechanical and Aerospace Engineering, Konkuk University |
|
|
|
|
|
|
|
| ABSTRACT |
|
In this study, exhaust gas emissions are predicted using long short-term memory (LSTM) algorithm and
minimum engine data, such as intake air temperature, emission gas temperature, and injection timing. Unlike existing
modeling analysis methods, deep learning does not require various vehicle specifications and data, and the correlation
between the measured data is derived by itself; therefore, it can serve as a virtual emission sensor. As it is difficult to analyze
the correlation between the deep learning and test data from actual road cars because of the complex environment, an
experimental single-cylinder diesel engine is used in this study. The intake air temperature is varied from 0 °C to 100 °C, and
the injection timing is varied for nitrogen oxide measurement. Consequently, nitrogen oxide is successfully predicted with a
high correlation R2 of 0.994 using minimal engine data. |
| Key Words:
Deep learning, Long short-term memory algorithm, Intake temperature, Injection timing, Indicated
specific nitrogen oxide (ISNOX), Indicated mean effective pressure (IMEP) |
|
Corresponding Author.
Suhan Park , Email. 
Preview 
Full text via DOI 
Download Citation
Print
