Process safety is on my mind for several reasons, including the recent explosion at a Texas fertilizer plant and this month’s On the Horizon article, “Improve Process Safety with Near-Miss Analysis,” by Ulku Oktem, Warren Seider, Masoud Soroush, and Ankur Pariyani (pp. 20–27).
The February Spotlight on Safety column focused on black swan events — i.e., highly improbable events that are unpredictable, carry massive impacts, and in retrospect can be explained in a way that makes them appear less random and more predictable than previously thought. In my editorial in that issue, I referred to work by Elisabeth Paté-Cornell of Stanford Univ., who suggested that the term black swan is used too liberally after a disaster as an excuse for poor planning.
Often, that poor planning stems from the failure to learn lessons from previous incidents and near-misses. Near-misses are high-probability, low-consequence abnormal events that, if allowed to progress (because layers of protection are ineffective), would likely result in a major accident with severe consequences, such as economic losses, environmental damage, injuries, and even fatalities.
Most accidents are preceded by several observable near-misses. The Center for Chemical Process Safety (CCPS) book Incidents that Deﬁne Process Safety has a chapter devoted to near-misses that begins with the statement: “It is said that there is nothing new in safety — it has all happened before.” The book points out several clear indications that the events leading up to the fatal explosion at BP’s Texas City reﬁnery in March 2005 had occurred in part on many occasions in the preceding ﬁve years. The blowdown drum that was involved had experienced 19 releases from its atmospheric vent (although those were all much smaller). The column that was being started up had experienced pressure spikes during previous startups; in two instances, the pressure came close to the relief valve setpoint, and in two others the pressure reached the setpoint, probably causing the valve to lift and discharge hydrocarbons into the blowdown drum. Investigators found that some previous incidents had not been reported, fully investigated, or documented with corrective actions to prevent a recurrence.
Oktem et al. introduce the concept of dynamic risk analysis (DRA), which makes use of the often-overlooked information contained in the large alarm databases of distributed control systems and emergency shutdown systems. It involves: tracking near-misses (abnormal events) using raw data from alarm databases; creating event trees that show all of the possible paths an abnormal event can take when propagating through the safety systems; using a set-theoretic framework to compact the data; and applying Bayesian analysis to estimate the failure probabilities of each safety system, the probability of trips, and the probability of accidents.
Unless process near-misses are analyzed thoroughly and regularly, plant personnel are likely to overlook the development of risky conditions, and thus, in time, trajectories of accident opportunities develop. DRA using alarm data provides a way to perform such analyses, and although it is still in its infancy, its application is expected to grow rapidly.
Over the coming months, more information will come out about the circumstances surrounding the Texas fertilizer plant explosion. Will investigators ﬁnd that the contributing causes had happened in the past? Would near-miss analysis have allowed the plant to predict and avoid such a failure?
Would you like to reuse content from CEP Magazine? It’s easy to request permission to reuse content. Simply click here to connect instantly to licensing services, where you can choose from a list of options regarding how you would like to reuse the desired content and complete the transaction.