442475 Process Accident Prediction Using Transition-Path Sampling

Wednesday, April 13, 2016: 8:00 AM
370 (George R. Brown )
Ian Moskowitz, Chem. and Biomolec. Eng, Univ. of Pennsylvania, Philadelphia, PA, Warren D. Seider, Department of Chemical and Biomolecular Engineering, University of Pennsylvania, Philadelphia, PA, Amish Patel, University of Pennsylvania, Philadelphia, PA and Jeffrey E. Arbogast, Applied Mathematics R&D, American Air Liquide, Newark, DE

Chemical manufacturing processes have the potential for catastrophic accidents – those involving loss of human life or major environmental impacts.  The numerous ways through which a complex chemical process can fail are limited neither by our experience, nor by our imagination; consequently, so-called unpostulated accident scenarios pose a major concern.  By combining extensive chemical process data, process modeling, and state-of-the-art path-sampling algorithms, rare and unpostulated process accidents are uncovered, secondary process variables that portend plant trips and accidents are identified (to be incorporated in alarm systems), and their rates of occurrence are characterized.

            Transition-path sampling (TPS) has been widely used in molecular modeling.  It identifies rare-event trajectories that are of investigative interest; such as the dissociation of a weak acid in aqueous solution or the flipping of phospholipids in a lipid bilayer.  The types of rare-events which are best handled by TPS are those whose time-scale is far shorter than the time-scale over which the events may take place.  This ratio of time-scales is similar to an accident in a chemical manufacturing process; the accident may occur once in five years of operation, while the accident  duration is just five hours; and to simulate 100 accidents, typical process simulations in response to expected disturbances would be carried out over 500 years!  This paper applies TPS to focus on accident trajectories, excluding nearly all of the vastly dominant safe operating trajectories. 

            The details of the Monte-Carlo approach are described, and its ability to discover unpostulated rare events is shown.  The accident trajectories investigated with TPS are used to create event-specific alarms, whose simulated false-positive and false-negative rates are explored.  The use of extensive process, control, and alarm data is essential is highlighted.  An Air Liquide Steam-Methane Reformer (SMR) is the test-bed for the TPS technique.  This process provides interesting dynamics – a major heat recycle loop and cyclical operation of its pressure-swing adsorption (PSA) process to recover hydrogen.  These attributes lead to rare expensive plant shut-downs and accidents.  In this paper, TPS is applied to study its embedded steam-drum, which is simpler to analyze and experiences occasional rapid pressure buildups resulting in plant shut-downs.

Extended Abstract: File Uploaded
See more of this Session: Advanced Prevention and Mitigation Techniques
See more of this Group/Topical: Global Congress on Process Safety