(66f) A Quantitative Approach for Optimization of Alarm Identification and Rationalization

Efficient monitoring of process variables and timely corrective measures form the crux of Abnormal Event Management (AEM). A prevalent monitoring technique in current practice involves the use of alarms to alert the operator in case of an abnormal event. This is achieved by configuring a subset of all measured variables such that alarms are triggered when they are not within predefined operating limits. The operator is then required to take corrective actions to restore normal operations. Poor alarm design was noted as the contributing factor for several disastrous incidents.

The process of choosing the potential subset of measurements that are to be configured to the alarm system is known as alarm identification. The rationalization of alarms involves the documentation of the possible causes for each of the alarm configured [1]. Alarm rationalization together with identification constitute important aspects of alarm system design, and they are the focus of this research. Most of the previous works on alarm identification and rationalization involved qualitative methods [1, 2]. Due to high interaction among the variables in most chemical plants arising from the complex nature of the process, the dynamics are not fully captured by qualitative approaches. In the present work, a quantitative approach that incorporates the temporal aspects such as the time taken by measurements for deviation under abnormal scenarios is proposed.

A linear multi-objective optimization formulation was proposed in [3] to identify the set of measurements that maximizes the response time available for the operator for the listed faults, keeping the number of alarms below a threshold. The optimization formulation was represented as a mixed-integer linear programming (MILP) problem. It accounted for additional criteria, such as order of priority of potential faults to be able to provide ample time to respond to high risk scenarios. Although the optimal alarm configuration could detect the fault and provide enough time for the operator to respond, it could not identify the fault that triggered the alarms. In this paper, the proposed optimization formulation, which is an extension of the work in [3], guarantees the optimal alarm configuration provides enough information to identify the cause of the alarm. The optimization formulation is represented as a mixed-integer nonlinear programming (MINLP) problem.

It is assumed that a closed loop dynamic simulator for the entire plant is available and all pertinent safety constraints have been identified. An exhaustive list of possible hazardous scenarios based on a safety review process [4] is available. Finally, it will be assumed that through the plant simulator time response information for all possible faults and all measurements are available and (ii) the time after which a fault can create a hazardous situation has been computed. Some terminology required for the formulation are as follows: (i). Alarm Configuration refers to the subset of measured variables that will be configured to the alarm system; (ii). Available Response Time (ART) of a fault is the time available for the operator to take response action after the fault is detected to avoid safety critical constraint violation.

A multi-objective optimization formulation is proposed, which is represented as a mixed-integer nonlinear programming problem. The primary objective of the optimization formulation is to identify a configuration that maximizes the minimum available response time (ART) of the process such that all faults are detected, and the number of alarms triggered for a fault is less than a ceiling number (N). The formulation ensures that each fault has a unique subset of alarms from the optimal configuration that is sufficient to identify the fault when it occurs. The secondary goals incorporate additional optimality criteria that arise from understanding the severity of faults to further refine the alarm identification process. It is achieved by maximizing the ART for faults sequentially in order of their priority with the highest priority being assigned to the fault with most severe consequence.

The formulation is then applied to the industrial case study of Vinyl acetate Monomer (VAM) plant. A closed-loop simulator was used for fault propagation. GAMS was used to solve the MIP problem. The design algorithm provides the opportunity to weigh the trade-offs and choose a reasonable ceiling number, such that the operator has sufficient time and information to respond to faults. For the identified optimal design, dynamic responses of the fault propagation, alarm generation and information provided to the operator with online diagnosis will be shown to illustrate the effectiveness of the approach.

References

1. Hollifield, B.R., E. Habibi, and J. Pinto, Alarm Management: A Comprehensive Guide. Research Traingle Park, NC: ISA, 2011.
2. Takeda, K., et al., A design method of a plant alarm system for first alarm alternative signals using a modularized CE model. Process Safety and Environmental Protection, 2014. 92(5): p. 406-411.
3. Venkidasalapathy, J.A., M.S. Mannan, and C. Kravaris, A quantitative approach for optimal alarm identification. Journal of Loss Prevention in the Process Industries, 2018. 55: p. 213-222.
4. Mannan, S. and F.P. Lees, Lee's loss prevention in the process industries. 2012, Butterworth-Heinemann;.