Data-driven Methods for Industrial Alarm Flood Analysis

• Author / Creator

  Lai, Shiqi

• The effectiveness of industrial process monitoring depends heavily on alarm systems. If alarm configurations are not rationally designed, the problem of excessive alarm messages would impact negatively the efficiency or even the safety of plant operations due to distracted information provided to operators. An alarm flood is an extreme case of this problem, during which the operator efficiency of handling important alarms is usually reduced significantly because of the overwhelming workload created by the numerous alarm messages. Consequently, techniques are needed to reduce the number and severity of alarm floods, as well as facilitate operators for proper operations during alarm floods. Motivated by this, this thesis focuses on the development of the data-driven techniques for alarm flood analysis. Two research topics are considered. The first topic is pattern mining in multiple alarm flood sequences. Incorporating time information into the evaluation of an alignment and dealing with large scale datasets are the main challenges. Two methods have been developed. The first one conducts a traversal search based on dynamic programming to obtain the optimal alignment of multiple alarm flood sequences. The second one achieves significant improvement on computational efficiency by applying approximations, but at the cost of a small amount of alignment accuracy. The second topic is online pattern matching and prediction of incoming alarm floods. The objective is to match the online alarm sequence with the patterns in the database and identify whether the online sequence is similar to any of the patterns in the database; predict the oncoming alarm flood if a matching is found. A method has been developed for this topic, which utilizes a proposed incremental dynamic programming procedure to break the whole computational burden of matching two sequences into small pieces that can be finished quickly in each individual step. The effectiveness of the proposed methods and their parameter robustness are tested by case studies based on datasets from a real chemical plant. Furthermore, a causality analysis for alarm floods is conducted. Process data associated with the alarms raised during an alarm flood is acquired and causality analysis is applied on the process data to generate causal maps, which are useful for root cause analysis and early predictions of future similar alarm floods.

• Subjects / Keywords

  • Sequence pattern analysis
  • Causality analysis
  • Process monitoring
  • Alarm flood analysis

• Graduation date

  Fall 2017

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/R30P0X492

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.