Robust malfunction diagnosis in process industry time series
Authors: T. Vafeiadis, S. Krinidis, C. Ziogou, D. Ioannidis, S. Voutetakis, D. Tzovaras
In this work, a modified version of a Slope Statistic Profile (SSP) method is proposed, capable to detect real-time incidents that occur in two interdependent time series. The estimation of incident time point is based on the combination of their linear trend profiles test statistics, computed on a consecutive overlapping data window. Furthermore, the proposed method uses a self-adaptive sliding data window. The adaptation of the size of the sliding data window is based on real-time classification of the linear trend profiles in constant and equal time intervals, according to two different linear trend scenarios, suitably adjusted to the conditions of the problem we face. The proposed method is used for the robust identification of a malfunction and it is demonstrated to real datasets from a chemical process pilot plant that is situated at the premises of CERTH / CPERI during the evolution of the performed experiments at the process unit.
T. Vafeiadis, S. Krinidis, C. Ziogou, D. Ioannidis, S. Voutetakis, D. Tzovaras, “Robust malfunction diagnosis in process industry time series”, 14th IEEE International Conference on Industrial Informatics (INDIN’16), Futurescope, Poitiers, France, 18-21 July 2016