A causal mixture model decomposition for root cause identification. Issue 1 (2021)

• Record Type: Journal Article
• Authors: Atoui, M. Amine; Cocquempot, Vincent
• Is Part Of: IFAC-PapersOnLine. Volume 54:Issue 1(2021)
• Page Start: 1241
• Page End: 1247
• Publication Date: 2021
• Subjects: statistical process monitoring -- root cause identification -- fault path propagation -- Gaussian mixture models; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2021.08.148
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: Multivariate statistical process monitoring methods usually assume the Gaussianity of data. However, in practice, data are multi-modal. Therefore, it is not always reasonable and enough to use methods that only deal with the data overall covariance matrix. As the latter may wrap less information compared to the data distribution. Also, such prior assumption is prejudicial to the estimation of the data' structure and the causal direction of variables. An interesting challenge would then be the development of relevant metrics to monitor variables and address their causal nature in the context of the non-Gaussianity of the data. Therefore, adequate parametric tests are required to ensure an acceptable and adjustable compromise between false positives and false negatives. In this paper, a new statistical approach is introduced to root cause and fault path propagation analysis. The obtained results demonstrate that the proposed method performs better than the existing methods.