A robust clustering method for detection of abnormal situations in a process with multiple steady-state operation modes

Abstract

Many classical multivariate statistical process monitoring (MSPM) techniques assume normal distribution of the data and independence of the samples. Very often, these assumptions do not hold for real industrial chemical processes, where multiple plant operating modes lead to multiple nominal operation regions. MSPM techniques that do not take account of this fact show increased false alarm and missing alarm rates. In this work, a simple fault detection tool based on a robust clustering technique is implemented to detect abnormal situations in an industrial installation with multiple operation modes. The tool is applied to three case studies: (i) a two-dimensional toy example, (ii) a realistic simulation usually used as a benchmark example, known as the Tennessee–Eastman Process, and (iii) real data from a methanol plant. The clustering technique on which the tool relies assumes that the observations come from multiple populations with a common covariance matrix (i.e., the same underlying physical relations). The clustering technique is also capable of coping with a certain percentage of outliers, thus avoiding the need of extensive preprocessing of the data. Moreover, improvements in detection capacity are found when comparing the results to those obtained with standard methodologies. Hence, the feasibility of implementing fault detection tools based on this technique in the field of chemical industrial processes is discussed.