- Wenkai Hu
- Sirish L. Shah
- Tongwen Chen
- Jiandong Wang
- Fan Yang
- Masaru Noda
- Martin Hollender
- Tongwen Chen, University of Alberta, Canada
- Sirish L. Shah, University of Alberta, Canada
- Martin Hollender, ABB Corporate Research Germany, Germany
- Masaru Noda, Fukuoka University, Japan
- Jiandong Wang, Shandong University of Science and Technology, China
- Fan Yang, Tsinghua University, China
- Wenkai Hu, China University of Geosciences, China
Reliable alarm monitoring is of critical importance to the safety and efficiency of process operations in complex industrial facilities. However, the presence of nuisance alarms and alarm floods severely impair the performance of alarm systems and compromise the safety of system operations. The objective of this workshop is to introduce participants to ideas, solutions, and recent advances of advanced industrial alarm monitoring based on advanced data analytics and machine learning techniques. This workshop will cover a variety of interesting topics, including the pattern analysis of alarm floods, design of alarm systems, association rule mining, causality inference, root cause analysis, and data visualization, which are based on seamless integration of information from process and alarm databases complemented with process connectivity.
The expected goals are to present recent advances on industrial alarm monitoring, to demonstrate the applicability and practicality of developed techniques, and to show how alarm management problems can be effectively solved using new tools. More specifically, the emphasis will be on how to improve industrial alarm management with machine learning techniques, how to design better alarm systems and evaluate alarm management performance, how to extract interesting patterns from historical alarm data and provide decision supports, how to track the propagation of abnormalities and find out root causes of alarms by causality inference and process topology, and how to identify sequential alarms and evaluate performance of alarm systems. The talks will be accompanied by industrial case studies to convey the practical utility of advanced industrial alarm monitoring tools.
Alarm Systems, Past, Present and the Future: A Vendor's Perspective, Martin Hollender
Milford Haven, EEMUA 191, ISA 18.2 and IEC 62682 have been important milestones on the way towards better alarm management. Today, well-established best practice is available, but still only applied in a subset of industrial automation projects. Many operators still run their plants without really using alarms. Modern data analytics tools can lower the prohibitive high cost of alarm system engineering. Alarm analytics can also help to optimize plant operation.
Machine Learning Tools for Advanced Alarm Management, Sirish L. Shah
The process industry is awash with all types of data archived over many years: sensor data, alarm data with operator actions to ‘navigate’ the process to operate safely at desired conditions and process models that are used for advanced control. The fusion of information from such disparate sources of process data is the key step in devising strategies for a smart analytics platform for safe and autonomous process operation. The purpose of this talk is to present results and strategies that will ultimately lead to safe and optimal autonomous or semi-autonomous process operation using smart sensor and alarm data analytics.
Optimal Design of Univariate and Multivariate Industrial Alarm Systems, Jiandong Wang
Industrial alarm systems have to be designed in a systematic manner in order to remove nuisance alarms and preserve true alarms. This talk will focus on the optimal design of univariate alarm systems based on single process variable, including alarm delay times and deadbands, as well as multivariate alarm systems based on multiple-correlated process variables, including operating-zone-based multivariate alarm systems, and qualitative-trend-based multivariate alarm systems.
Advanced Alarm Analytics Tools Developed at the University of Alberta, Tongwen Chen
This talk will summarize some recent results on advanced alarm analytics and present a new set of tools for design of alarm systems and improvement of alarm management. The essential functionalities of the tools include alarm visualization, alarm performance evaluation and analysis, and rationalization design, thereby to help industrial processes to comply with the new standards. The tools have been tested with real industrial data and used by process engineers in Canada and elsewhere.
Evaluation Methods of Plant Alarm Systems, Masaru Noda
This presentation will introduce two evaluation methods for plant alarm systems. The first one is for identifying sequential alarms hidden in plant operational data using dot matrix analysis, which is one of the sequence alignment methods for identifying similar regions in DNA or RNA sequences. The second method uses an operator model that mimics humans’ fault detection and identification (FDI) behavior. The model automatically produces an FDI track in an emergency after a malfunction occurs. By analyzing the FDI tracks after causing all the assumed malfunctions in the plant, we can evaluate the performance of the alarm system.
Causality Analysis Based on Data Analytics, Fan Yang
This presentation will introduce advanced alarm strategy and abnormal situation monitoring based on process data analytics and, in particular, correlation/causality analysis based on mining of process and alarm data in combination with process connectivity knowledge, with applications to root cause analysis of propagated or even plant-wide abnormalities. The methods of Granger causality, transfer entropy, and Bayesian networks will be demonstrated.
Pattern Mining from Historical Alarm & Event Database, Wenkai Hu
This presentation will introduce advanced data analytics that mine interesting patterns from historical alarm & event database and demonstrate their effectiveness and practicality with real industrial case studies. The involved techniques include the detection of mode dependent alarms, determination of first-out alarms, extraction of frequent alarm patterns, similarity analysis of alarm floods, and process discovery of alarm response workflows.