| Title |
Fault Detection Using Signal Reconstruction Model Based on Autoencoder in Thermal Power Plant |
| Authors |
김규한(Kyuhan Kim) ; 정희명(Heemyung Jeong) ; 이흥석(Heungseok Lee) ; 이화석(Hwaseok Lee) ; 김형수(Hyungsu Kim) ; 박준호(June Ho Park) |
| DOI |
https://doi.org/10.5370/KIEE.2020.69.6.800 |
| Keywords |
Autoencoder; Fault Detection; Kernel Density Estimation; Principal Component Analysis |
| Abstract |
Unexpected faults in major industrial facilities, such as thermal power plants, cause a lot of economic damage. Therefore, Fault detection and diagnosis systems at major industrial facilities are essential to prevent faults in advance and reduce financial losses from these damages. This paper is proposed a fault detection system applied autoencoder based on deep-learning. Autoencoder is very useful for data-driven multivariate signal reconstruction modeling. The Hotelling's T2 and SPE(Squared Prediction Error), which commonly fault detection indices used in PCA(Principal Component Analysis)-based fault detection systems, are also applied to the proposed fault detection system. And then, the threshold values of these indices for fault detection are calculated using the KDE(Kernel Density Estimation). Finally, we apply two real-world fault cases to compare the performance of the autoencoder-based fault detection system with the PCA-based fault detection system using a FAR(False Alarm Rate), which is a representative fault detection performance index. |