John A. Krembs, James M. Connolly
M&M Protection Consultants
Chicago
An analysis of the 150 largest losses caused by accidents and natural phenomena in the hydrocarbon processing and chemical industries during a period of 30 years ending Jan. 1, 1989, shows that the cost and number of losses is increasing.
The analysis is taken from data gathered for, "Large Property Damage Losses in the Hydrocarbon-Chemical Industries: A Thirty-Year Review," published by M&M Protection Consultants.
While losses of catastrophic proportion are relatively rare, the very nature of operations in the hydrocarbon processing and chemical industries represent the potential for events that can severely affect corporations, shareholders, employees, and the public. A combined property damage and business interruption loss exceeding a billion U.S. dollars was unheard of in the industry until recently.
The catastrophic losses analyzed were used to develop statistical trends from the losses in M&M's data base. The trended data offer valuable information on where efforts to prevent and mitigate future incidents should be directed.
Some hazard-control professionals would argue that high-frequency, low-consequence events should be analyzed to determine the adequacy of operating methods or protective systems. While not discounting the benefit of analyzing the more common incidents, there are unique and significant lessons to be learned from low-frequency, high-severity events. Conclusions reached from analyzing these different types of events can vary significantly.
The 150 losses analyzed represent more than $5 billion in property damage when adjusted for inflation. The magnitude of the trended losses has increased in each of the 5-year periods analyzed in the 30-year study (Fig. 1). Individual losses varied from $9.5 million to $309 million.
The average loss was $34 million, and the median loss was $20 million. These figures represent physical damage only. Financial loss from business interruption, liability, worker's compensation, or other consequential losses are not included in the figures.
The availability of more and better information probably serves as partial explanation for the increase in the number and cost of the losses over the 30-year period. A more significant force behind the trend to higher losses would be the ever greater concentration of assets to achieve economies of scale in this most capital-intensive industry.
Large, single-train units have replaced multiple, smaller units. In a modern refinery fluid catalytic cracking (FCC) unit, for example, a single centrifugal wet-gas compressor does the job that 10 reciprocating units did in an older plant.
Process unit production capacities have increased dramatically during the last 30 years. Ethylene production has risen from about 20 million lb/year to more than 1.5 billion lb/year in a single-unit operation.
The size of polyvinyl chloride batch reactors has grown from 1,000 gal to 50,000 gal. Attempts to achieve economies of scale have had an undeniable effect on the size of potential losses.
Reduced spacing of processing equipment to reduce land use, and to minimize energy, piping, and instrumentation requirements, has resulted in a higher concentration of assets exposed to a single adverse event.
Finally, there is growing evidence that available inflation indices may not provide a sufficient adjustment of older loss data. That is, the inflation-trended amount for a loss that occurred 25 years ago is more likely to understate, rather than overstate, the financial effect of a recurrence of the same event today.
It should be recognized that the data base is not perfect and that information is easier to obtain in some parts of the world than others.
The 150 losses have been grouped in one of eight "type-of-complex" categories: refining, petrochemical, terminal and bulk plants, chemical plants, plastic and rubber manufacturing, natural-gas processing, pipeline, and miscellaneous. Excluded are offshore production accidents, such as loss of the Piper Alpha platform in the North Sea, and losses involving marine vessels unless the loss occurred while the vessel was at a plant dock.
The financial loss data which have been trended to January 1989 include property damage, debris removal, and cleanup costs only. The cost of business interruption, employee injuries, environmental damage, liability claims, and fines were excluded.
While the main purpose of the study was to gather data relating to property damage losses, records were kept of the resulting injuries and fatalities in each incident.
Given the large quantities of hazardous materials that are handled at the facilities included in the study, it is significant that only three of the accidents caused fatalities to people outside of the plants. Of the 99 incidents occurring in the U.S. and Canada, there were no deaths of neighbors or bystanders.
In addition to data on events during the 30-year period ending Jan. 1, 1989, large losses in 1989 were analyzed. These losses shattered all previous records for hydrocarbon-chemical industry loss experience. It also appears that eight of the 1989 losses will qualify among the 100 largest in the 30-year period ending Jan.1, 1990. The largest of the 1989 losses, at $725 million, contributed to the high average loss of $133.5 million for the year. None-the-less, the median figure of $83.5 million indicates there were other very large losses in the context of prior experience.
Although the 1989 data are not included in the statistics and figures, the trends have been projected through 1989 by illustrating some of the larger losses that occurred during 1989.
DISTRIBUTION BY COMPLEX TYPE
When losses were analyzed by type of complex, refineries were involved most frequently, accounting for 40% of the losses as shown in Fig. 2. Refinery losses accounted for three of the eight candidate losses in 1989.
The largest of the three refinery incidents was caused by hurricane Hugo, which struck a refinery at St. Croix, Virgin Islands, and resulted in an estimated $150 million property loss. Several 500,000 and 600,000-bbl storage tanks were destroyed or severely damaged along with company housing.
Process units, which had been shut down before the hurricane hit on Sept. 18, 1989, sustained moderate damage. To the best of our knowledge, this incident is the largest single plant loss caused by an event other than fire or explosion.
The natural gas processing category had higher average trended losses than did the refining category. Five of the ten natural gas processing losses occurred in extremely large plants in the Middle East. The $52.2 million average loss was no surprise given the size and concentration of assets characteristics of these plants.
Although pipeline incidents accounted for the smallest percentage of losses, they resulted in a high average trended loss of $45.6 million.
On June 3, 1989, a 28-in. natural gas liquids pipeline carrying 13,500 tons/day from Western Siberia to petrochemical centers in European Russia ruptured, releasing light hydrocarbons. A dense vapor cloud formed in the valley of the Ural Mountains and was apparently ignited by sparks from one of two electrically powered trains which were passing through the area almost 1 mile from the pipeline.
It is estimated that the explosion developed blast pressures equivalent to 10,000 metric tons of TNT. The resulting explosion felled trees within 2.5 miles of the pipeline and shattered windows for miles.
Western Siberian natural gas liquid production had to be reduced by about one third because of this incident, and many Soviet petrochemical plants were forced to shut down or operate at reduced capacity.
In spite of the rural location, the total property loss was estimated at $150 million. The loss amount is not directly comparable to other loss figures in the study because it includes damage to parties other than the pipeline organization, and accurate conversion of the Ruble is difficult.
CAUSE OF LOSS
Mechanical failure of equipment was, by far, the leading cause of loss, accounting for 41% of the incidents, including the largest single-owner loss prior to 1989 (Fig. 3).
This loss, which exceeded $300 million, occurred in Norco, La., on May 5, 1988, when an 8-in. pipe in a large fluid catalytic cracking unit failed, releasing at least 20,000 lb of light hydrocarbons. Analysis of damaged equipment after the loss indicated blast overpressures as high as 10 psi. Minor damage was recorded at distances up to 6 miles from the refinery.
On Apr. 10, 1989, mechanical failure caused an estimated $90 million damage to a hydrocracking unit at a refinery in Richmond, Calif. A line carrying hydrogen gas at approximately 2,800 psi failed at a weld, resulting in a high-pressure fire that impinged on the supports of a 100-ft high reactor.
The reactor support skirt, which was 10-12 ft in diameter and had a wall thickness of 7 in., subsequently failed. The reactor fell on other process equipment, greatly increasing the size of the loss.
It was understood that the reactor was in a hydrogen purge cycle in preparation for a maintenance shutdown. Approximately 25% of the refinery throughput capacity was lost as a result of the incident.
The second most frequent cause of loss was operational error, accounting for 19% of the losses. Recognizing that all loss events except natural disasters can ultimately be attributed to human error in some way, the definition of operational error is limited to errors relating to operations or maintenance on site, leading directly to the loss.
This category, then, deals with errors that are the proximate rather than the ultimate cause of the loss.
While design error accounted for a very small 4% of the losses, six incidents produced the highest average loss at $60.5 million, Most of these events involved the failure of piping due to improper metallurgy or faulty design.
The Soviet gas pipeline loss discussed previously is an example of possible design error followed by operational error after the pipe failed. The pipeline, designed to operate at about 735 psi, was not equipped with remote or automatic controls.
Upon detecting a decrease in pipeline pressure, pipeline operators tried to increase the pressure by turning on more pumps instead of initiating an emergency shutdown.
Sabotage or arson fires accounted for only 4% of the 150 large losses. Four of six fires in this class occurred outside of the U.S.
EQUIPMENT INVOLVED
Piping systems are the most frequent type of equipment involved in the cause of large losses (Fig. 4). This category includes pipe, hose, and tubing, as well as flanges, gauges, strainers, expansion joints, and other pipe fittings.
Piping system failures caused 31% of the losses at an average cost of $42 million per loss.
During 1989, the trend of losses involving piping systems continued. For example, there was a $40 million loss at Morris, Ill., on June 7, 1989, in which a piping system failed and caused the release of an ethylene vapor cloud that exploded. The result was sustained fires that damaged several acres of the plant.
At 17% of the total, the tank and other storage category had the second highest frequency. The greatly increased value of petroleum and chemicals over the 30-year period, and the move toward larger storage tanks, accounted for this observation.
Of the 23 incidents in this category, 14 tank losses occurred in refineries and 9 occurred in storage terminals. The most likely reason chemical plants did not contribute to these statistics is the generally smaller tank size at chemical plants.
Process towers originated only 3% of the losses but were responsible for the highest average trended loss of $53.8 million. Reactors were also an important category with 13% of the losses, at an average loss of $28.09 million.
The average loss for reactor-related losses will increase markedly in any further analysis because the largest, single-owner loss in history, involving a reactor, occurred on Oct. 23, 1989. This $725 million loss will be discussed later. Pumps and compressors are often thought to be responsible for more process industry fires than any other equipment. In this study, however, only eight such losses appear, five involving pumps and three involving compressors.
This finding may be misleading in that the majority of losses originating from pumps and compressors may result in damage too low for consideration in the study. Other factors might include improved materials, improved maintenance procedures, such as vibration monitoring, and the increased use of double seals on pumps in flammable liquid service.
OPERATING STATUS
A significant 36 losses, or 24% of the total, occurred while plants were not in normal operating modes (Fig. 5). This category included losses during start-up, shutdown, online maintenance, or turnaround.
The finding is noteworthy because normal operations prevail most of the time. For example, a process unit may run for 3 years before it is shut down for a turnaround period of only a few weeks.
In some of the losses, operators were aware of process upsets, leaks, or mechanical problems, and were shutting down the process or attempting to correct the problem when the problem escalated into a major disaster.
Most of these incidents occurred in the chemical and petrochemical industry where batch operations account for a significant proportion of the total accidents.
TYPE OF LOSS
As would be expected, fires and explosions account for nearly all of the losses. In Table 1, vapor-cloud explosions are differentiated from other types of explosions.
The miscellaneous losses included a windstorm, a tank collapse, and a steam pressure rupture.
As these data show, vapor-cloud explosions involving large releases of flammable vapor with delayed ignition, produced the largest losses in the study.
These explosions may be detonations or rapid deflagrations, or they may involve both phenomena.
Destructive overpressures of a blast often produce flying debris that breaks piping and causes additional fires remote from the point of failure. The average trended property damage loss produced by 54 vapor-cloud explosions was $45.5 million, 60% higher than the average of the other 96 losses.
A significant example of a disastrous vapor-cloud explosion occurred on Oct. 23, 1989, in Pasadena, Tex. The vapor-cloud explosion was followed by explosions of nearby pressure storage tanks, and destroyed a multiunit polyethylene complex.
The accident killed 23 people and resulted in an estimated $725 million in property damage.
A very large business interruption loss will accrue during plant restoration, which may require 18-24 months, or more.
One of six product separation legs of the polyethylene loop reactor was undergoing a procedure to eliminate a blockage when a release of isobutane (the catalyst carrier) and ethylene occurred. With a pressure of about 700 psi in the reactor, the cloud expanded rapidly until it ignited about 1 min after the release.
Other process units at the plant suffered only slight damage and resumed operations shortly after the accident. No one outside the plant was hurt and off-premises property damage was minimal. Type of loss by complex shows that the most common types of loss vary by complex. Within chemical plants, 75% of the losses were initiated by explosions, compared with only 13% in refineries.
In refineries, 52% of the losses were caused by fires as compared with only 8% in chemical plants. In natural-gas processing operations, 60% of the losses were initiated by vapor-cloud explosions, a reasonable finding in view of the types of hydrocarbons handled by gas-processing facilities.
SOURCE OF IGNITION
Determination of the source of ignition in the fire and explosion losses is often difficult or impossible. In over half of the 150 incidents, the exact source of ignition remains unknown (Fig. 6).
Open flames, such as would be found in heaters or boilers, constitute the most common source of ignition where the source was determined. These units accounted for 12% of the losses.
It is noteworthy that only one incident in the study was attributed to cutting or welding.
Considering the amount of hot work that takes place in processing plants, it is remarkable that cutting and welding was not more prominent as a source of ignition.
This demonstrates the importance of administrative controls, such as hot-work permit systems generally used in the industry.
Chemical reactions were the second largest sources of ignition, accounting for 9% of the total.
Of the 14 chemical reaction or decomposition ignitions, seven occurred in chemical plants and five in petrochemical plants.
However, the largest accident resulted in $81.5 million damage to a refinery. Each remaining ignition accounted for less than 5% of the total.
FIRE-WATER FLOW
Fire-water flow associated with the suppression of large fires was also analyzed, along with the effect of impairments of fire-water systems. A significant impairment of fire-water flow was documented in nearly 10% of the losses. It is expected, however, that the actual figure is higher because difficulties with fire protection systems are not always well documented.
Recent examples of significant impairments to the fire protection system include a Nov. 14, 1987, unconfined vapor-cloud explosion at a petrochemical plant in Pampa, Tex. The blast severed piping in many of the sprinkler systems and caused the rupture of an underground fire-water main.
The fire house collapsed with fire equipment inside. The last fire was extinguished 12 hr after the initial blast, resulting in approximately $185 million damage.
The losses that resulted in fire pump failure suggest the need for more careful location and design of fire pump installations. Generally, it is suggested that fire pumps be separated by at least 250 ft from hydrocarbon or chemical process or storage areas.
This distance should be increased to 500 ft where fire pumps are exposed to extremely high hazards or large concentrations of liquefied petroleum gases. Even if the suggested spacing criteria are met, masonry buildings with noncombustible roofs should be provided to protect pumps. Open-sided shelters should be avoided except for remote locations.
Loss experience has shown that diesel engine-driven pumps are the most reliable in severe-loss incidents. Electric, steam, and spark-ignition engines are more likely to be put out of action by the initial fire or explosion.
In one incident, a flash fire from an LPG vapor release burned up the electrical system of a natural gas-fueled fire pump engine located in an open-sided shelter.
But even diesel-powered pumps are not immune from impairment. The same flash fire that impaired the gas engine broke a gauge glass on the fuel tank of the diesel pump and drained the fuel supply. It should be noted that National Fire Protection Association (NFPA) Standard 20, "Centrifugal Fire Pumps," prohibits the use of gauge glasses on fire pump fuel tanks.2
A recently completed analysis of more than 400 fire pump tests, of which about half were in hydrocarbon or chemical facilities, indicated a higher-than-expected failure rate. The failure rate of fire pumps in hydrocarbon or chemical facilities was 38% compared to 17% in other industries.
Failure was defined as water delivery less than 90% of the pump's rated capacity.
Table 2 summarizes the estimated fire-water flows used at the peak of fire-fighting activity for 20 large fire losses.
The range of flows is from 6,500 gpm to 21,500 gpm, with an average of approximately 12,000 gpm.
It can be concluded from the analysis that all available water flow capacity will usually be used in large or catastrophic fires whether or not the available flow is actually needed.
Also, flow capability of 4,000-8,000 gpm at a minimum of 100 psi is needed at a process unit or other fire risk area, such as a group of storage tanks. Additional water will likely be brought in by relay pumping or drawing from adjacent natural bodies of water. Emergency response planning should take this into account.
Drainage capacity must exceed fire-water flow capacity, including the additional water brought from outside to cool adjacent process units. Inadequate drainage has contributed to a number of large losses by spreading burning liquids on the surface of accumulated water.
Significant improvements to drainage systems can be very expensive in existing facilities. Therefore, adequate drainage must be an important part of the initial design of the facilities.
ACKNOWLEDGMENT
Thanks go to William G. Garrison of Marsh & McLennan, the author of the large loss publication from the first edition in 1977 to the twelfth edition in 1989. He retired in January 1989 after 37 years of service. Thanks also go to David G. Mahoney who will continue the large-loss project, and whose research into the 1989 losses provided valuable input to this article.
REFERENCES
- Garrison, W.G., Large Property Damage Losses in the Hydrocarbon-Chemical Industries-A Thirty-Year Review, Twelfth Edition and analysis, M&M Protection Consultants, Chicago, 1989.
- Garrison, W.G., 100 Large Losses-A Thirty-Year Review of Property Damage Losses in the Hydrocarbon-Chemical Indus. tries, Eleventh Edition, M&M Protection Consultants, Chicago, 1988.
- NFPA 20, "Standard for the Installation of Centrifugal Fire Pumps," National Fire Protection Association, 1987.
Copyright 1990 Oil & Gas Journal. All Rights Reserved.