Monitoring transitions in chemical plants using enhanced trend analysis

Chemical plants operate in a multitude of steady states called modes and frequently undergo transitions between them. Transitions usually require extensive operator involvement and hence monitoring the safe execution of transitions is crucial. Existing techniques for process monitoring are currently configured assuming a single mode of operation and hence cannot be directly applied to transitions. In this paper, we use a trend analysis-based technique for monitoring transitions in continuous chemical plants where the real-time evolution of the process variables is abstracted into semi-quantitative trends. These real-time trends are then compared with the dictionary trends, which correspond to those during normal operation of that transition. One challenge is to determine the location in the dictionary with which the real-time trend has to be compared. For this, a dynamic feature synchronization algorithm is proposed which helps in tracking the process during transitions. The details of the algorithm, its implementation as an expert system, and the application to two case studies-a pilot-scale distillation column and a simulation of a fluidized catalytic cracking unit-are presented. (C) 2003 Elsevier Science Ltd. All rights reserved.