This paper discusses a number of examples of real world QRA analysis from within the oil and gas industry as well as other industries, identifying how these have failed to approximate the actual risks. The paper then analyses the process and describes the types of information used in QRA. This is contrasted with the data that is not included in these analyses and how that has contributed to real world failures.
In order to understand risk, we may consider two routes. Qualitative information, which tells us about what risks exist and for which we can use expert knowledge to weigh the likelihood of (adverse) events occurring. The alternative is quantitative information, which looks at data sets of occurrences and from these defines the likelihood of occurrence of a specific event. QRA is a methodology that uses a systemic approach to the identification and evaluation of risks. It further uses mathematical modeling methods that estimate the risk and effects of hazardous events. It is also known as quantitative risk assessment.
The use of QRA is now widespread across a range of industries, including aviation, health care, rail, finance, and oil and gas.
The purpose of QRA is to establish a number that represents the likelihood of incidents and to undertake measures that mitigate risk to within an acceptable level. In a range of industries, we see what are presumed to be very unlikely events, based on the QRA reports, that, nevertheless, appear to happen with regularity. Major events, such as refinery explosions and rail fires, are designed to happen so rarely that a single instance should not be expected.
The risk acceptance criteria for transport of hazardous materials from the US Department of Transport show that a single fatality incident is considered an unacceptable risk if it has an expected frequency of 10−3 per year, with fatal accidents with 10 fatalities having an expected frequency of 10−4 per year and accidents with 1,000 fatalities having an expected frequency of 10−6 per year. Acceptable risk criteria for these outcomes start at 10−5, 10−6, and 10−8 respectively. Yet in 2002, 2004, and 2005, there were rail incidents involving the release of hazardous cargo that killed 1, 3, and 9 people, respectively, and, in 2023, a near fatal incident with chemical spill and fire that affected an Ohio town causing the residents to abandon their homes. Similarly, an international rail tunnel connecting the United Kingdom to France suffered a number of major fires in 1996, 2006, 2008, 2012, and 2015. This tunnel and its operations are subject to a QRA that should demonstrate that a fire is extremely unlikely. The evidence, however, demonstrates that this QRA does not reflect the actual magnitude of the risk.
In oil and gas, a similar pattern can be seen with the frequency with which refinery explosions and fires take place. Here, the expected rate of fires and explosions is around one incident per 10 years per refinery. This is orders of magnitude away from the intended level. Hyperbolically, it should be closer to once in the lifetime of the universe.
One example of this is the BP Texas City Refinery explosion on 23 March 2005 that resulted in the deaths of 15 people and 170 injuries. The shortcomings included the over-reliance in the QRA on historical data that did not reflect the unique conditions and actual operations at the refinery. The QRA underestimated the likelihood of organizational and human error deficiencies, which were significant contributors to the incident. The QRA used simplified assumptions that failed to capture the complexity of the refinery’s operations and the scale of catastrophic failure modes.