Fault Propagation Analysis by Implementing Nearest Neighbors Method Using Process Connectivity

Rinat Landman, Sirkka-Liisa Jämsä-Jounela

Research output: Contribution to journalArticleScientificpeer-review

11 Citations (Scopus)
134 Downloads (Pure)


Industrial systems often encounter abnormal conditions due to various faults or external disturbances which deteriorate the process performance. In such cases, it is essential to detect and eliminate the root cause of the faulty condition as early as possible in order to minimize its adverse effect on the entire process performance. Capturing the process causality plays a key role in identifying the propagation path of faults and their root cause. In recent times, several data-based methods have been developed in order to capture causality from the measured process data. However, each of the methods suffers from several limitations and deficiencies which might compromise their ability to provide an adequate causal model, especially in multivariate (MV) systems. This paper proposes a new methodology for retracing the propagation path of oscillation using a nearest neighbors method by utilizing the information on process connectivity. The two-phase methodology yields a directionality measure based on the type of connectivity in the process using a unique search algorithm. In phase I, the bivariate directionality measure is calculated to include only the interactions that are considered as direct based on the plant topology. In phase II, a new MV directionality measure based on the nearest neighbors method is introduced in order to exclude indirect interactions. The methodology is successfully demonstrated on industrial board machine exhibiting oscillations in its drying section.
Original languageEnglish
Pages (from-to)2058-2067
Number of pages10
JournalIEEE Transactions on Control Systems Technology
Issue number5
Publication statusPublished - Sept 2019
MoE publication typeA1 Journal article-refereed


Dive into the research topics of 'Fault Propagation Analysis by Implementing Nearest Neighbors Method Using Process Connectivity'. Together they form a unique fingerprint.

Cite this