Framework and models for an event-based early warning system design

点击翻译按钮获取中文摘要

I have proposed a five-year research program that aims to develop an event-based approach to the design and management of warning systems used in process industries. The special features and novelty of the proposed approach are in the allocation of warnings to specific events, the use of risk as a basis for alarm annunciation and categorization, and the use of prediction for warnings. The challenges in the design of such a warning system are tied to identification of events, selection of their associated variable sets, and continuous estimation of the risk of corresponding events. Moreover, the use of prediction will require the development of identification schemes to estimate models with the ability to predict variables around safety limits. My experience with the development of novel system identification algorithms and our recent works in the field of alarm system design will be valuable tools to overcome the above challenges. My record of contributions, previous and current HQP training performance, and detailed plan for HQP training will ensure the success of the program. The outcome of the program will be a novel warning system design framework and models, corresponding intellectual properties, and a group of highly qualified personnel. This program will also help to establish the applicant as a leading contributor to this field of significant industrial importance. Warning systems for industrial operations, including alerts and both process and safety alarms, play a key role in monitoring process safety. These systems warn operators of abnormal operating conditions, and seek the attention of an operator when intervention is required. Timely action, based on real warnings, is needed to help enhance plant safety and minimize costs through the effective prevention, control and mitigation of abnormal situations. In industrial plants, however, misleading warnings are a regular occurrence. Moreover, when flooded with spurious alarms, the warning system may itself contribute to abnormal situations and to plant accidents. A root cause of the ineffectiveness of existing warning systems is the large number of warnings that result from the use of single-variable settings. The number of warnings in a plant needs to be significantly reduced. Warnings should be made informative, actionable and clear indicators of the risk associated with the current state of process/system variables. Plant operators should not need to monitor the trend of individual variables; instead, the risk profiles of a plant should be monitored, with warnings that have predictive capabilities. To achieve adoption of this novel approach, a paradigm shift in the overall design of the warning systems is needed. To overcome the problems associated with existing warning systems (use of single variables, alarm flooding), the proposed research program will develop an event-based approach to warning system design and management that will improve the ability of plant operators to identify potential problems, make timely decisions and take any necessary actions. By assigning warnings to events, the number of warnings will be significantly reduced, as will the number of false warnings. Incorporation of risk in the methodology for annunciation will make it possible to accommodate the importance of the process variables and the criticality of an abnormal situation. Thus it will be possible to categorize warnings using a well-defined metric. Finally, the use of prediction will provide operators with extra time to take necessary measures. The new warning system will be a valuable addition to the arsenal of plant operators in their management of abnormal situations, the prevention of plant accidents, and improve process safety.