High-Order Nonstationary Feature Extraction for Industrial Process Monitoring Based on Multicointegration Analysis
Document Type
Article
Publication Date
5-29-2024
Abstract
The statistical characteristics of industrial process data may change over time due to equipment aging, working condition adjustment, etc. It may lead to the failure of traditional Multivariate Statistical Process Monitoring (MSPM) methods. Long-term equilibrium relationship analysis (LERA) is considered to be effective for nonstationary processes monitoring. However, in industrial practice, certain variables may exhibit different orders of nonstationarity, which may lead to insufficient feature extraction by LERA. To address this issue, a nonstationary process monitoring strategy based on multicointegration analysis is proposed in this work. The variables are first classified according to their integrated orders. Multicointegration analysis is then performed to extract the long-term equilibrium and multicointegration relationships, based on which the monitoring statistics are constructed to determine the process operating status. The proposed method is validated through two cases and compared with other methods. Validation results demonstrate its effectiveness in monitoring nonstationary process including high-order nonstationary variables.
Publication Source (Journal or Book title)
Industrial and Engineering Chemistry Research
First Page
9489
Last Page
9503
Recommended Citation
Rao, J., Ji, C., Wang, J., Sun, W., & Romagnoli, J. (2024). High-Order Nonstationary Feature Extraction for Industrial Process Monitoring Based on Multicointegration Analysis. Industrial and Engineering Chemistry Research, 63 (21), 9489-9503. https://doi.org/10.1021/acs.iecr.4c00423