| 학회 |
한국화학공학회 |
| 학술대회 |
2017년 봄 (04/26 ~ 04/28, ICC 제주) |
| 권호 |
23권 1호, p.180 |
| 발표분야 |
공정시스템 |
| 제목 |
Multivariable Statistical Monitoring Using Independent Component Analysis for LDPE Production Process |
| 초록 |
Low-density polyethylene (LDPE) production process is a high pressure process. Therefore, it is important to detect the failure prognostics for safety and maintenance reasons. Principal component analysis (PCA) is the most popular multivariate statistical monitoring method. However, the PCA cannot extract the most effective features when the observations have a non-Gaussian distribution so that it cannot guarantee acceptable detectability in the LDPE production process. To overcome the limitation, independent component analysis (ICA) is developed for non-Gaussian distribution data. So, in this research, the ICA is applied to the observations from the LDPE production process to detect the failure prognostics. After applying the ICA, the independent components (ICs) are sorted by their non-Gaussianity to determine the most dominant ICs. Then, I2 statistics are calculated using the most dominant ICs. The I2 statistics reflect an abnormal behavior of the process better than the T2 statistics. The result shows that the ICA-based method has better detectability for the failure prognostics before the shutdown occurred from the hyper compressor failure than the PCA-based method. |
| 저자 |
박병언1, 이정근1, 이창송2, 이규황2, 이호경2, 이인범1
|
| 소속 |
1포항공과대, 2LG화학 |
| 키워드 |
이상진단 |
| E-Mail |
|
| VOD |
VOD 보기 |
| 원문파일 |
초록 보기 |
