Dynamic risk analysis of chemical plants to improve process safety, reliability, and product quality

In most industrial plants, hundreds of variables monitor the dynamics of their processes. When the variables move out of their normal operating ranges, alarms are triggered to notify the operators. Often, vast amounts of alarm data are recorded in databases associated with their distributed control and emergency shutdown systems that often remain under-utilized . In this work, these large databases are utilized efficiently to: (a) project probabilities of occurrence of shutdowns and accidents, (b) assess the real-time safety and operability performances of processes and improve their alarm systems, and (c) develop leading indicators to forewarn operators of approaching incidents several hours in advance. Abnormal events, which occur when process and quality variables depart from and return to their normal operating ranges, are recognized as near-misses—as they could have propagated into incidents. Consequently, vast amounts of near-miss data are available for analyzing the performances of the regulating (process control) and protection (emergency shutdown) systems. A novel dynamic risk assessment methodology is introduced that uses the alarm databases to calculate the failure probabilities of regulating and protection systems, and the probabilities of the occurrence of incidents. The methodology consists of three steps: (i) tracking of abnormal events over an extended period of time, (ii) event-tree and set-theoretic formulations, and (iii) Bayesian analysis. New event-trees and set-theoretic formulations are introduced to compact the massive numbers (millions) of abnormal events, over several months of operation, leading to significant improvements in the efficiency of the probabilistic calculations and permitting Bayesian analysis of large databases in real-time. The Bayesian model accounts for the interdependences among the regulating and protection systems using copulas, which occur due to the nonlinear relationships between the variables and behavior-based factors involving human operators. As a case study, the methodology is applied to an industrial, fluidized-catalytic-cracking unit (FCCU) at a major petroleum refinery. New performance indices, which utilize the abundant alarm data, are introduced to conduct quantitative and qualitative (absolute and relative) assessment of the real-time safety and operability performances. Also, techniques for improving the alarm system performance are introduced and it is shown how they can be used to suggest improvements in the alarm-system structures for the FCCU. A new leading indicators methodology using the alarm data is introduced to help identify an escalation in the probability of the occurrence of incidents—permitting the system to alert the operator(s) of a major problem likely to occur in the near future. Using historical data and errors before past incidents, these techniques have the great potential to foresee the occurrence of safety and quality incidents by identifying and utilizing near-misses in real-time