(672f) Identification of Repeating Sequential Alarms in Historical Operation Data of Chemical Plants

The advance of distributed control systems (DCS) in the chemical industry has made it possible to inexpensively and easily install numerous alarms in DCS. While most alarms help operators detect an abnormality and identify its cause, some are unnecessary. A poor alarm system might cause sequential alarms, which are a collection of numerous alarms that almost always occur simultaneously with specific time lags within a short period of time. Sequential alarms reduce the ability of operators to cope with plant abnormalities because critical alarms are buried under many unnecessary ones. It is very important to identify sequential alarms in a plant operation data for safe plant operations.

Nishiguchi and Takai (2010) proposed an identifying method of sequential alarms in plant operation data using event correlation analysis. Event correlation analysis is a method of extracting knowledge that detects statistical similarities between discrete events. The method can identify sequential alarms from the operation data of chemical plant, but it occasionally failed to detect similarities between two physically related sequential alarms when deletions, substitutions, and/or transpositions occurred in the alarm sequence. A modified Smith-Waterman algorithm was proposed to calculate similarities of alarm flood sequences in plant operation data (Cheng et al., 2013). Experts can make a thoroughly analysis, such as root cause, based on the clustered patterns of alarm floods. However it is not the method that directly finds out sequential alarms hidden in the plant operation data. Wang and Noda (2017) proposed an identification method of sequential alarms by applying the dot matrix method to a plant operation data. The dot matrix method is one of sequence alignment methods for identifying similar regions in DNA or RNA, which may be a consequence of functional, structural, or evolutionary relationships between the sequences (Mount, 2004). The method can identify sequential alarms from the operation data of chemical plant, but it occasionally failed to detect similarities between two physically related sequential alarms because the time information when alarms occurred is not for evaluating similarities between them.

In this research, we propose a new identification method of repeating sequential alarms in historical operation data of a chemical plants. In the method, time stamped alarm data is converted into a set of windows containing adjacent alarms. All combinations of windows are compared, and repeating similar windows are identified on the basis of similarities between them. The alarms in each repeating window comprise a sequential alarm. Application of this method to simulated operation data for an azeotropic distillation column demonstrated that it can identify sequential alarms in noisy plant-operation data. By classifying such alarms into small numbers of sequential alarms, an engineer is able to effectively reduce the number of unnecessary alarms.

References

1. Cheng, Y., Izadi, I., and Chen, T., Pattern matching of alarm flood sequences by a modified Smith-Waterman algorithm, Chemical Engineering Research and Design, 91, 1085-1094 (2013)
2. Nishiguchi, J., and Takai, T., IPL2 and 3 Performance Improvement Method for Process Safety using Event Correlation Analysis, Computers & Chemical Engineering, 34(12), 2007-2013 (2010)
3. Wang, Z., and Noda, M., Identification of Repeated Sequential Alarms in Noisy Plant Operation Data Using Dot Matrix Method with Sliding Window, Journal of Chemical Engineering of Japan, 50(6), 445-449 (2017)
4. Mount, D., W., Bioinformatics Sequence and Genome Analysis Second Edition, Cold Spring Harbor Laboratory Press, Cold Spring Harbor, New York (2004)