Enhancing the accuracy of forecasting impact of accidents in chemical process industry by the application of cellular automata technique

Abstract

Nearly all the techniques, tools and strategies for preventing accidents in chemical process industry revolve round the question: what would happen if a vessel, a conduit, a controller, or a process suddenly fails leading to an accident?' To answer this question accidental scenarios are developed and, based on the wisdom of hindsight derived from the post-mortem of past accidents, the damage likely in each scenario is assessed. This information provides the key feedback for quantifying and 'scaling' hazards: higher the probability-cum- severity of the likely accident, higher the risk posed by the unit whose failure may cause the accident. In other words, assessment of the likely consequences of a likely accident is one of most common and important steps associated with loss prevention and safety promotion in process industry. In the hitherto used approach it is assumed that the blast wave and heat load generated by an explosion or explosion-cum-fire would travel radically outward with equal intensity in all directions from the accident epicentre. It is also assumed that the intensity of the impact felt by a receptor would only be a function of the receptor's distance from the epicentre but shall be independent of the receptor's direction. In real-life situations, however, the flux of energy, mass and momentum ensuing from an accident would not travel with equal intensity in all directions but would be absorbed or attenuated by different objects lying in its path in different directions. In order to foresee a more plausible picture of the 'risk contours' of a likely accident it is necessary to develop a methodology which takes cognizance of the site characteristics likely to influence the accident's impact zone. In this paper the technique of cellular automata (CA) has been used, in all probability for the first time ever, as a basis to develop a new methodology for assessing the consequences of an accident occurring in a chemical process industry. CA was originally developed by von Neumann to model life-like processes but until recently its use was very limited due to its need for expensive computer time. The technique is now experiencing a resurgence as the falling cost-capability ratio of computers has made CA increasingly affordable. In the CA-based methodology presented here, we have treated an industrial accident as an event occurring in a hypothetical square-shaped 'cell', surrounded on all sides by other square 'cells' of identical area. The grid simulates the accidental site and the model correlates all the cells of the grid with each other on the basis of 'influencing' or 'transition' functions; these functions, in turn, are governed by the cell characteristics - in other words the composition of the fragment of the site represented by that cell. The accident is modeled to proceed in small but discrete time-steps; in each step the impact of the 'seed cell' (the one in which the accident occurs) causes the state of the other cells to change in accordance with the appropriate transition function. It is, therefore, possible to take into account the nature of space that lies between the accident epicentre and the receptor in each direction. As a consequence 'the impact zone' of any accident can be forecasted much more accurately than possible hitherto. The applicability of the methodology has been demonstrated with the help of an illustrative case study. © 2006 Institution of Chemical Engineers.