BROWSE
Detail
Conference Paper
Data-Driven Multiplicative Fault Detection Using Hybrid of Multivariate Statistical Techniques
Cited - time in 
Share 
Share
- Authors
- Jungwon Yu, Jinhong Kim, Youngjae Lee, Soyoung Yang, Kil-Taek Lim
- Issue Date
- 2019-08
- Citation
- International Workshop on Big Data for Ceramics and Smart Manufacturing (BDCSM) 2019, pp.1-8
- Language
- English
- Type
- Conference Paper
- Abstract
- Accurate and timely detection of possible faults is indispensable for safe and cost-effective operations of industrial plants. In this paper, we present the results of applying an integration of auto-associative kernel regression (AAKR) and dynamic independent component analysis (which is proposed by Yu et al. [1]) to multiplicative fault detection (FD); in this method, after extracting several latent variables (i.e., independent components) from residual vectors generated by AAKR, detection indices are calculated based on the latent variables and FD is then performed via statistical hypothesis tests. The FD performance of the hybrid method is evaluated with a benchmark example relevant with multiplicative fault type, and compared with various popular FD methods. The experimental results show that the hybrid method achieves the lowest type II error (i.e., miss detection rate) and, at the same time, acceptable type I error (i.e., false alarm rate).
- KSP Keywords
- Auto-Associative Kernel Regression, Data-Driven, Dynamic independent component analysis, False Alarm Rate(FAR), Fault Type, Industrial Plants, Latent variables, Miss detection, Multivariate statistical techniques, Statistical hypothesis test, Type I error