(426h) Gas Detectors Layout Optimization Considering False Positive and False Negatives: A Stochastic Programming Approach and a Quantitative Assessment of the Process Industries

Gas detection constitutes an important layer of protection in process facilities. From a review of the 100 largest property damage losses in the hydrocarbon industry, around 70 are attributed to fires, explosions, and/or vapor cloud explosions. Incidents where the fire and gas detection system played or could have played an important role in preventing further damages after loss of containment. While effective technology exists for gas detection, several difficulties make the problem of gas detector placement challenging. First, dispersion scenario source location, size, and duration are generally unknown. Second, the gas leak dispersion development and transport depend on fluid properties and dispersion characteristics, environmental factors, and facility geometry. Third, formal quantification of the final risk of a given dispersion scenario is a complicated task. Finally, even if all this data is high quality and can be consolidated, due to the combinatorial aspects of the problem, exhaustive search is not an option.

Most regulations, standards, and operator-specific practices propose the use of prescriptive approaches supported by qualitative considerations and rules of thumb. These approaches are far from optimal. The acknowledgement of this fact by the industry has increased interest in formal quantitative approaches supplemented by rigorous dispersion simulations. In previous work, a stochastic Mixed-integer linear programming (MILP) formulation (SP) was proposed, developed and validated for optimal placement of gas detectors using information from leak scenario simulations. Simulations were performed on a real, medium-scale, proprietary offshore facility geometry capturing the full process features, i.e. equipment, piping, support structures, etc. Data generation was performed via a consortium between the Mary Kay O’Connor Process Safety Center at Texas A&M University and GexCon. Using FLACS, a validated tool for release modeling in the technical safety context, probabilistic scenario simulations were performed accordance to the latest industry standards. Results demonstrated the potential and suitability of numerical optimization to the gas detector placement problem while rigorously considering its inherent uncertainties.

This initial work assumed perfect gas detectors. In reality gas detectors are prone to several failure modes that can result in false negatives and false positive alarms. To address these issues in real settings, the potential detector unavailability should be considered when designing the system, and detection redundancy is often required (via a voting strategy) before emergency actions are triggered. A generalized MILP formulation, SP-UV, is proposed that considers detector unavailability and voting strategies in the optimization problem. Compared with the SP formulation, relaxing the perfect detector assumption results in changes to the optimal detector placement. Furthermore, significant deterioration in performance is observed if unavailability and voting are neglected when optimizing the gas detector layout for real settings. In addition to these observations, a number of common practices for gas detector placement in the process industries were also considered, and the detector layouts produced by SP-UV outperform these heuristic approaches in a number of metrics.