Similarity Analysis of Industrial Alarm Flood Data

Abstract

Flooding of alarms is a very crucial problem in process industries. An alarm flood makes an operator ineffective of taking necessary actions, and often risking an emergency shutdown or a major upset. In this work, the flooding of alarms is discussed based on the standards presented in ISA 18.2. A new analysis method is proposed to investigate similar alarm floods from the historic alarm data and group them on the basis of the patterns of alarm occurrences. A case study on real industrial alarm data is also presented to demonstrate the utility of the proposed analysis.Note to Practitioners-In an industrial plant, alarms are designed to inform the operator about any fault or abnormality in the operation. However, alarm systems are not always designed, implemented, and maintained properly. This usually results in an excessive number of enunciated alarms, most of which are false or nuisance. Specifically, during a plant upset, operators receive far more alarms that they can handle (hundreds or even thousands of alarms in a short period of time). This is known as an alarm flood or alarm shower. Our study shows that in some cases, alarm floods follow similar patterns. As a result by studying and classifying flood patterns, it is possible to find the root cause of an abnormality based on the history of the plant and previous similar floods. This allows the operator to react to a plant upset at the early stages of or even before an alarm flood.

Validation of similarity measures for industrial alarm flood analysis?

“…Our study shows that the flood distance measure in [1] behaves significantly different from the other analysed distance measures; in particular, the other measures produce a more stable clustering in the presence of noise in the data.…”

“…Analysis of the behaviour of the distance measure can then help choose the most suitable distance measure. We also reproduce and extend the alarm flood detection and clustering approach by Ahmed et al [1] with additional similarity measures based on Term frequency-inverse document frequency (TF-IDF) representation [12] and Levenshtein distance [14], apply these measures to a large real industrial alarm log and evaluate them using our validation method.…”

A Hybrid Failure Diagnosis and Prediction using Natural Language-based Process Map and Rule-based Expert System

“…In the research of [3], an evaluation of the alarm filters which are suggested in [1] to find the optimal ones. In the research of [1] clustering of alarm is done by evaluating the consecutive occurrence pattern of alarms.…”

“…In the research of [1] clustering of alarm is done by evaluating the consecutive occurrence pattern of alarms. Also, various alarm management techniques including frameworks, data filtering, alarm delay and alarm deadlines are suggested, in order to design an optimal alarm system [5].…”

“…They focused on reducing the number of alarms to an affordable level, by extracting unusual alarms with a categorization of alarm types and analysis of the alarm patterns using statistical models. The limitation of such projects is that even if they succeed and the alarms are provided in manageable level, there are many additional works which require enormous time and efforts such as analysis of alarms, diagnosis, and failure prediction [1].…”