(595b) Computer-Aided Emergency Operation of Process Plants

With the exception of those sudden events provoked by external factors in poorly monitored areas or when important malfunctions of the control system occur, a great amount of the incidents and accidents in a chemical plant are first detected by the firing of the monitoring system alarms, leading after some time to the activation of the safety instrumented system if the cause remains unsolved. Considering this successive triggering at the protection layers, it is possible to distinguish three states with regard to the plant: 1) A normal state, in which current measurements are within the alarm limits. 2) An alarm state, produced when any of those variables goes beyond the limits but the safety system is not activated. 3) An emergency state, which occurs in the case that any of the values reaches the limit of the emergency system. In the first case, the control system maintains the process between what has been considered as acceptable bounds and so, the operator action goes no further than routinely supervising the monitoring system, changing setpoints or outputs and initiating sequences. At the emergency state, the safety system takes control of the plant. The alarm state is the only time in which the operator could actively intervene and when it usually becomes clear that the control system has been primarily designed for the regulatory function and only subsidiarily for the operation. Under these circumstances, the possibilities to avoid damage to people, materials or environment as well as economic losses are extremely reduced. The typical inertia of chemical plants makes usable our current control systems, but in likely more situations than needed an emergency shut down occurs. Taking a car with a primitive safety system as an analogy, the airbag is the only option for any abnormal situation. But modern vehicles include anti-lock braking systems (ABS) and electronic stability programs (ESP) which allow safe driving under difficult circumstances. Can we think in something similar for process plants? That is the question we try to answer. What the operator requires at this point is a fast identification and evaluation of the problem, proper decision making and its prompt and safe execution. Many papers have been published about the monitoring and fault detection issues. Regarding operation, the possibility of generating automatically operating procedures from rigourous plant models was demonstrated several years ago [1,2]. A different approach is taken here, where it has to be considered that the plant may have changed as a consequence of the accident and that the previous models can be of dubious utility. Emergency aided operation in a chemical plant essentially copes with the task of avoiding to exceed safety limits during abnormal situations modifying where possible the operators' response while approaching a hazardous point. What we present in this paper is an implementation as a constrained control problem, using ARX models continuosly estimated with current plant data. The results of the simulated examples are promising but much more has to be investigated regarding stability as well as control, alarm and safety systems design and efficient operator interfaces. This research has been supported by Fundación Repsol - ETSII/UPM. [1] S. Galán and P. I. Barton. Dynamic optimization formulations for operating procedure synthesis. AIChE Annual Meeting, November 1997. [2] S. Galán, J. R. Banga, and P. I. Barton. Automatic generation of operating procedures. AIChE Annual Meeting, November 1998.