Eventos Anais de eventos
COBEM 2017
24th ABCM International Congress of Mechanical Engineering
RECONCILIATION OF PATTERNS IN THE CLUSTERING OF TIME SERIES
Submission Author:
Izete Silva , BA
Co-Authors:
Izete Silva, Pedro Moreira Arruti Aragão, CRISTIANO FONTES, Raony Maia Fontes, Marcelo Embiruçu
Presenter: Izete Silva
doi://10.26678/ABCM.COBEM2017.COB17-2900
Abstract
The significant increase in the relevance of issues related to reliability and safety in the production processes of the various segments of the production network has pressured organizations to seek efficient methods to diagnose and detect possible failures in their processes. Although there are many works related to Fault Detection and Diagnosis (FDD), few of them are based on clustering and pattern recognition in time series, especially the multivariate ones. In addition, there are no works related to the pattern recognition in time series that consider the process model as a constraint. This paper proposes a new method for the pattern recognition in uni and multivariate time series, based on Fuzzy C-Means (FCM), that inserts the process dynamics in the context of the clustering problem in order to ensure the feasibility of the recognized patterns. The proposed method is applied in a case study that comprises the clustering and pattern recognition of abnormal (failures) operation of a nonisothermal Continuous Stirred Tank Reactor (CSTR), a well-known benchmark system used to compare various monitoring solutions and also used for the assessment of FDD techniques. The results show that the proposed method (FCM coupled with a process model) is able to recognize patterns consistent with the behavior of the process without worsening the quality of clustering and classification.
Keywords
Clustering, Multivariate Time Series, Fault Diagnosis, Process Model
