LPS 2009 ____________________________________________________________ Paper 96a

Lessons Learned from the 2006 Facility Explosion in Danvers, MA

Scott G. Davis*, Ph.D., P.E. and Olav R. Hansen, M.Sc. GexCon US

3 Bethesda Metro Center, Suite 700 Bethesda, MD 20814

sgdavis@gexcon.com, olav@gexcon.com http://www.gexconus.com

Prepared for Presentation at American Institute of Chemical Engineers

2009 Spring National Meeting 43rd Annual Loss Prevention Symposium

Tampa, Florida April 26-30, 2009

UNPUBLISHED

AIChE shall not be responsible for statements or opinions contained in papers or printed in its publications

LPS 2009 ____________________________________________________________ Paper 96a

Lessons Learned from the 2006 Facility Explosion in Danvers, MA

Scott G. Davis, Ph.D., P.E. and Olav R. Hansen, M.Sc. GexCon US

3 Bethesda Metro Center, Suite 700 Bethesda, MD 20814 sgdavis@gexcon.com

Abstract

On November 22, 2006 the largest explosion in the history of Massachusetts occurred in Danvers, MA at approximately 2:46 am. This paper presents a detailed analysis into the potential causes and lessons learned from the Danvers explosion. Other investigative groups concluded that the cause of the explosion was an overheated production tank. However, the analyses presented here demonstrate that their proposed scenario could not have occurred and that other potential causes are more likely. Using the computational fluid dynamics tool FLACS, it was possible to investigate the chain of events leading to the explosion, including: (1) evaluating various leak scenarios by modeling the dispersion and mixing of gases and vapors within the facility, (2) evaluating potential ignition sources within the facility of the flammable fuel-air mixture, and (3) evaluating the explosion itself by comparing the resulting overpressures of the exploding fuel-air cloud with the structural response of the facility and the observed near-field and far-field blast damage. These results, along with key witness statements and other analyses, provide valuable insight into the likely cause of this incident. Based on the results of our detailed analysis, lessons learned regarding the investigative procedure and methods for mitigating this and future explosions are discussed. 1. Introduction An explosion and ensuing fire occurred in Danvers, MA at approximately 2:46 am on November 22, 2006. The explosion and resulting fire occurred at an ink and paint manufacturing facility that was jointly operated by Coatings, Adhesives and Inks, Inc. (CAI) and Arnel Company, Inc (Arnel). The facility was unoccupied when the explosion occurred. The blast destroyed the facility and caused significant damage to the surrounding property and structures. It was reported that approximately 17 to 19 structures were damaged beyond repair as a result of the blast wave. Several boats located in a neighboring marina suffered damage from projectiles and the blast wave. Approximately a dozen individuals sustained minor injuries and no fatalities were reported as a result of the incident. CAI and Arnel manufactured separate product lines at the Danvers facility. CAI manufactured solvent-based inks at the Danvers facility and used various types of solvents (alcohols, aliphatic hydrocarbons, glycols, and esters), pigments and resins within the building. Arnel manufactured solvent- and water-based paints and coatings at the Danvers facility and therefore also stored and used various solvents (alcohols, ketones, aromatic hydrocarbons, and esters), pigments and resins

LPS 2009 ____________________________________________________________ Paper 96a

within the building. Both companies stored solvents in three underground storage tanks and industrial grade nitrocellulose in trailers located outside of the facility adjacent to the building. This paper presents a detailed analysis into the potential causes and lessons learned from the November 22, 2006 incident. The release and dispersion of flammable gases and vapors, as well as their subsequent explosions, are complex phenomena. Using simplified models and approximations can result in incorrect determinations of cause and origin. Detailed inspections, including relevant data collection, along with computational fluid dynamics (CFD) tools such as FLACS allow complex scenarios to be modeled and reconciled with the available evidence. The paper will present analyses that contradict proposed scenarios1,2 into the cause and origin of the explosion and present other potential causes not considered by other groups. A brief description of the facility is first provided, followed by the sequence of events leading up to the explosion. Next, a summary of the inspection findings is provided. Finally, analyses of the various possible scenarios using various tools including the CFD-tool FLACS are presented. Lessons learned regarding the investigation are discussed as well as methods for mitigating this and future explosions. 2. Background Information 2.1 Facility

2.1.1 General Description The CAI/Arnel manufacturing facility was located at 126 Rear Water St., Danvers, MA on the west end of the Danversport peninsula. The facility was bounded by a large swamp on its east side, an access road to the Danversport Marina on the south side, and a few small businesses and Water St. (Route 35) on the west side. The backyards of several residential properties on Bates St. bordered the north side of the facility. A bulk propane distribution facility and a large warehouse were located about 500 feet away on the west side of the building, across Water St. Figure 1 shows an aerial view of the CAI/Arnel manufacturing facility and its surroundings prior to the explosion of November 22, 2006. The south side of the CAI/Arnel facility (Figure 2) housed the employee offices and the laboratory areas for the two companies. The west and north side of the facility housed the warehouse and the manufacturing areas, which contain the production and storage areas for both CAI and Arnel. Most of the facility’s exterior and interior walls were constructed of unreinforced concrete masonry units (CMUs or ‘cinder blocks’). The manufacturing section of the facility was comprised of two rooms: (1) Room E and (2) Arnel’s storage and packing (Figure 3). All process equipment in the warehouse and manufacturing areas was bonded and grounded. According to interviews and agency reports, equipment within the production part of the facility was properly rated for use in flammable environments.

LPS 2009 ____________________________________________________________ Paper 96a

Figure 1: Aerial view of the Danversport area prior to the accident on November 22, 2006.

Figure 2: Layout of the CAI/Arnel facility.

Three 3,000-gallon underground storage tanks (USTs) used by CAI and Arnel to store solvents were located on the north side of the facility (Figure 3). The storage trailers containing the industrial grade nitrocellulose drums used by both companies were located on the east side of the facility. Two oil-fired boilers provided low-pressure steam to the various heaters throughout the facility and to certain production tanks. Only one boiler was in use on the day of the incident.

LPS 2009 ____________________________________________________________ Paper 96a

Figure 3: Arnel/CAI facility layout

2.1.2 Heating and Ventilation

It has been reported that the heaters in the facility were steam coil air heaters with enclosed explosion proof fan motors. On the production side of the facility, there was one Modine heater in the warehouse, and two other steam-coil heaters in the manufacturing section of the facility. A 6000-CFM (~2.8 m3/s) supply fan provided fresh make-up air near ceiling level to the production part of the facility. There were four exhaust fans installed in the production area of the facility. The facility also contained two dust collectors. 2.1.3 Warehouse

The warehouse section of the facility contains both CAI’s production area and Arnel’s storage area (shown in Figure 4). CAI had five closed-top, atmospheric pressure mixing tanks in the manufacturing section as shown in Figure 4 and Figure 5. The first mixing tank (V1) was not in use, while the other mixing tanks were used to manufacture ink vehicles. Each mixing tank was equipped with a top-mounted agitator and a steam heating jacket. Operators controlled the temperature of the tanks by opening and closing valves on the steam piping to the respective heating jacket. Arnel stored hundreds of 55-gallon drums and thousands of pounds of powdered raw material in the south area of the warehouse section of the facility. Arnel also stored solvents in four to six 350-gallon portable totes stacked on the south wall. It was reported that the HC-258 Modine heater (258,000 Btu/hr ~ 75 kW), which was located between the totes as shown in Figure 4, was the only heater in operation in the production areas on the night of the incident.

LPS 2009 ____________________________________________________________ Paper 96a

Figure 4: Perspective view of warehouse section of the Danver’s facility

Figure 5: CAI mixing tanks as viewed from mezzanine (prior to explosion)

2.1.4 Manufacturing

The manufacturing section of the facility contained Arnel’s storage and packing area and Room E as shown in Figure 6. There were two fire doors in Room E, one on the west wall leading to the warehouse section and one on the east wall leading to Arnel’s storage and packing area. Room E had a roof with two different ceiling heights and housed CAI’s T-1250 (V12) mixing tank, eight 500-gallon totes used by CAI for solvent storage, and Arnel’s 1000-gallon mixing tank. The USTs were piped directly into each company’s respective pump and distribution piping manifold. Arnel’s 1000-gallon mixing tank did not have an explosion proof motor and the motor was located just outside the facility on the lower roof section east of the tank in a motor housing. Employees indicated that Arnel used the 1000-gallon tank for nitrocellulose-based products. Solvents would first be pumped into the tank, followed by the addition of nitrocellulose and

LPS 2009 ____________________________________________________________ Paper 96a

resins, and then constituents were allowed to mix overnight. It was reported that the tank mixer was not energized on the night of the incident, but this could not be confirmed.

Figure 6: Perspective view of manufacturing section of the CAI/Arnel facility

3. Accident Description 3.1 Events Prior to the Incident 3.1.1 Facility Related Events On Tuesday, November 21, 2006, no abnormal operations at the facility were reported. A batch for the 01-1038 (V4) was the only CAI batch being manufactured. In the late morning, solvents (alcohols, heptane and water) were added to the 01-1038 (V4) to start the new batch. After lunch, the agitator and the steam heater on the tank were turned on and solids (resins) were added to the mixture. A CAI employee recalled starting to add the resins at approximately 12:30 pm and the production manager recalled starting at 1:00 or 1:30 pm. The loading of the solids was completed by approximately 3:00 pm. Around 5:00 pm, the CAI production manager verified that the temperature of tank 01-1038 (V4) had reached the 90°F (32°C) set point, and turned off the steam valves. Therefore, the steam heat was applied to the 01-1038 (V4) tank for approximately 3.5 to 4.5 hours (from 12:30-1:30 pm to 5:00 pm). That evening the lights, the dust collectors, the make-up air and exhaust fans in the production area were turned off. CAI employees indicated that only one heater was in operation in the production area, the fire door to the office and laboratory space was closed, and the two fire doors in Room E were left open.

LPS 2009 ____________________________________________________________ Paper 96a

3.1.2 Non-facility related incidents A history of natural gas leaks have been reported in the neighborhood and many residents reported smelling natural gas near the facility the day prior to the explosion. In the days prior to the incident, demolition and excavation using a backhoe occurred at the Abbey Fence building and around the Danversport Bottled Gas building, which are both located on Water Street in front of this facility. Eastern Propane, which was located across Water Street, was excavating the day before the incident and ruptured a 1¼-inch propane vapor pipe. On Nov 22, 2006 at approximately 2:00 am, an individual arrived at Harvey Industries (across Water St. from the facility) in his sleeper-cab tractor-trailer truck. Slightly before 2:30 am, he noticed a dark-colored pickup truck enter the driveway to the CAI/Arnel plant, and the dome light became illuminated. His view of the CAI/Arnel facility driveway was obscured, so he could not confirm if anyone exited the vehicle or entered the facility. Just after 2:30 am, the witness retired to the back of his truck and was awoken by the explosion at approximately 2:46 am. Both CAI and Arnel employees had access to the facility, and investigative reports indicated two Arnel employees were recently fired in the months preceding the incident. 3.1.3 Incident

At approximately 2:46 am, multiple witnesses reported that they were woken up by a loud explosion. Many witnesses also reported hearing a second, slightly less severe explosion within one minute after the initial blast. At the time of the incident, outside temperatures were between 27°F (–3°C) and 29°F (–2°C) and the wind was from the north at approximately 6 mph (2.7 m/s). Video footage from the Fox 25 News 5:00 am program showed significant burning in various areas of the facility. The contents of some tanks were shown burning, but not the 01-1038 (V4) tank, which had been used by CAI the day prior for batch production. The blast destroyed the CAI/Arnel facility and caused significant damage to the surrounding property and structures. 4. Inspections It was reported that the gas company had several leaks in the natural gas underground distribution piping surrounding the facility footprint on both Water and Bates Streets. The underground distribution piping on Water and Bates Streets consisted of 4- and 6-inch cast iron pipes connected by bell and spigot joints located every 10 feet. The gas company had repaired several joints on the underground distribution piping after the explosion and ultimately abandoned the system in place on the streets surrounding the facility. 4.1 Site Inspection 4.1.1 Near Field Blast Effects

Figure 7 is an overhead image of the remains of the CAI/Arnel facility taken two days after the explosion. It is clear from the image that the facility was destroyed and many objects were blown further from the building. Significant damage was observed to the surrounding structures (see Figure 7).

LPS 2009 ____________________________________________________________ Paper 96a

Figure 7: Blast damage at the CAI/Arnel facility (taken Nov. 24, 2006)

4.1.2 Manufacturing Section Figure 8 shows a representative image of the original position and post blast movement of the major components that were in Room E as viewed from the north. CAI’s T-1250 and the eight 500-gallon totes were still standing after the blast. The fire door to the west was found blown into the warehouse area 20-feet away. The east fire door blew further east into the manufacturing area and was found 18-feet away covered with 8-inch thick cinder blocks. The non-explosion proof motor was blown over the totes to the east (see Figure 8). Inspection of Room E further revealed that the cinder block walls (12-inch thick) supporting the high ceiling section were blown west into the manufacturing area, and the cinder block walls supporting the lower section ceiling (8-inch thick) were blown to the east. Inspection also revealed that there were practically no cinder blocks remaining in Room E after the explosion.

Figure 8: Directional analysis of blast in Room E

LPS 2009 ____________________________________________________________ Paper 96a

Inspection of Room E also revealed that the Arnel 1000-gallon mixing tank was severely deformed as shown in Figure 9. The damage to the tank suggests that it was loaded with a force such that the bottom of the tank, which was originally domed outward, was deformed inward to the extent that the shaft of the mixer had punctured through the bottom tank shell. The tank walls were deformed inward and the legs buckled outward. The scale below the 1000-gallon tank also showed significant damage.

Figure 9: Arnel 1000-gallon mixing tank (Dec. 4, 2006)

A manhole cover, just north of the facility, was blown-off during the explosion and found 75 feet from its original location. While the mechanism responsible for the displaced manhole cover has not been analyzed, this could possibly be the result of a flammable mixture accumulating within the sewer or the result of the dynamics associated with the facility explosion. It was reported that approximately 7450 gallons of liquid material were pumped from the 9000-gallon capacity USTs and therefore, these tanks cannot be ruled out as a possible fuel source. The nitrocellulose drums stored in the trailers all appeared to have burned as a result of the ensuing fire. 4.1.2 Warehouse Section

Significant heat and fire damage was observed in the south section of the warehouse due to the storage of large quantities of raw materials in this area. In contrast, the northern wall had minimal heat and fire damage. The CAI mixing tanks were still standing and uncompromised after the blast. All four mixing tanks had significant thermal damage on the sides facing the Arnel storage racks due to the ensuing fire. The steam valves to the 01-1038 (V4) were never found. Extensive thermal and mechanical damage was found in the Arnel storage area. Inspections further revealed that permanent lighting fixtures, mixer motors, electrical conduit and other equipment in the production areas appeared to be properly rated for use in a flammable environment. Directional damage from east to the west was observed in the warehouse, and consisted of lifted and deformed grating on the mezzanine (Figure 10), displaced steel trusses, displaced railing and piping. This damage was likely due to drag effects associated with a blast wind originating from the direction of Room E.

LPS 2009 ____________________________________________________________ Paper 96a

Figure 10: Lifted and bent grating on mezzanine as viewed from the North

4.2 Blast Damage Blast wave interaction with structures and the resulting damage is very complex to analyze and is dependent on many factors. Due to this complexity, the relationship between overpressures and resulting blast damage serve only as an approximation of the energy released in an explosion. From the energy released, one can further estimate the amount of fuel necessary to cause such an event. Residences and buildings immediately surrounding the facility were severely damaged by the explosion, suggesting a strong explosion (overpressures > 10 psi) in the near field. At the time of our initial inspection (Dec. 2, 2006), much of the resulting blast damage had been altered or repaired. Some sporadic window glazing (glass) failures were identified in the far field. The glass analysis software SafeVue3, developed by the Naval Facilities Engineering Service Center to determine the dynamic blast pressure capacity of windows, was used to estimate the peak blast pressure distribution. When considering estimates of glass failure probabilities and wave reflection, the lower bound peak free field pressure at about 0.5 miles (805m) was approximated as 0.2 psi (1.4kPa) and rather insensitive to pulse duration. The CSB, who arrived at the scene earlier (November 26, 2006), reported blast overpressures of approximately 1.2 psi at 580 feet (8.3kPa at 177m) and 2.3 psi at 365 feet (15.9kPa at 111m) from the facility. 4.3 Soil Gas Sampling, Ground Penetrating Radar, 3-D Laser Scan/Model of Facility Authority to install permanent soil gas probes was not obtained until December 26, 2006, over a month after the incident occurred. Soil gas samples were taken on a regular basis in the weeks and months (up to a year) following the installation. Methane concentrations around the facility were generally below 100 ppm except for one location designated as SG-101 (see Figure 3). This soil gas was located just north of Room E in the area of the sewer piping and piping from the USTs. Methane concentration readings at this location were as high 18.5% (185,000 ppm). Elevated levels of methane (18.5%, 4.2% and 6.2%) were measured in SG-101 on three occasions from July to December of 2007.

LPS 2009 ____________________________________________________________ Paper 96a

Ground-penetrating radar (GPR) was performed at the site to evaluate possible pathways and conduits into the building (i.e., voids, utilities, piping). The GPR data revealed multiple utilities and voids leading to the building footprint from the north as well as from the west as far as Water Street. A detailed 3-D laser scan was performed of the building footprint and surrounding area, and provided accurate three-dimensional information regarding the building layout as well as the surrounding area. These data, along with drawings, interviews, photographs, and measurements were used to construct a model of the facility prior to the explosion. From this model, the total available volume for vapor to accumulate was calculated to be 114,600 ft3 (~3245 m3) 5. Analysis NFPA 921 Guide for Fire and Explosion Investigations states that all available fuel sources should be considered and eliminated until one fuel can be identified as meeting all of the physical damage criteria.4 Possible fuel sources that could accumulate within the facility to significant levels include n-heptane, acetone, other solvents, natural gas, methane, and propane. Three general scenarios have been identified as possible causes and include: (1) steam valves to the 01-1038 (V4) being left open, causing the contents to boil and release significant solvent vapors, (2) solvent spill or intentional solvent release from the totes (or pumped from the USTs), and (3) gas migration into the facility. The following analyses will evaluate the consequence and likelihood of the three hypothesized scenarios. To understand which mechanisms could lead to the accident and observed damage, a number of dispersion and explosion CFD simulations were performed using the consequence tool FLACS5. FLACS is a leading CFD tool for dispersion and explosion predictions and is extensively used for safety studies in the petrochemical industry. FLACS is capable of modeling releases, dispersion of vapors, ventilation in structures, and ignition of flammable fuel-gas mixtures to evaluate the flame progression and overpressures due to explosions. Sensor points can be placed throughout the computational domain to monitor the time history of relevant variables such as static or dynamic pressure, velocities, fuel concentration, etc. FLACS has been extensively validated against numerous experiments, including large-scale realistic release and explosion tests performed at GexCon and full-scale experiments performed in semi-confined model of an offshore module. Many hydrocarbons, including n-heptane, have similar burning characteristics as propane. Therefore, explosion simulations were conducted with propane and it is expected that the blast results for heptane or similar hydrocarbons will be comparable in strength6. Methane and natural gas have very similar burning characteristics yet are distinctly different from propane. As such, these fuels were modeled independently as methane. 5.1 Blast Overpressure Analysis Energy is released during a chemical explosion causing an overpressure and a blast (pressure) wave to propagate away from the source or epicenter of the explosion. After all the energy is released, this pressure wave decreases in strength as it propagates away from the epicenter. The

LPS 2009 ____________________________________________________________ Paper 96a

resulting blast wave can cause damage to surrounding structures. Evaluating the damage caused by the blast wave allows one to approximate the energy released in an explosion, and hence the amount of fuel involved. In addition, computational tools, which are capable of modeling flame propagation and resulting overpressures, can be used to evaluate the likely fuel-air mixtures as well as ignition source location by comparing the predicted and observed blast damage. The CFD tool FLACS was used in the present study to model the explosion itself, and compare the predicted overpressures and dynamics of the burning fuel mixture with the observed structural response and blast damage within the facility. While the near-field blast damage can provide valuable information in determining the location of the ignition source and the dynamics of the intiating event, the far-field overpressure damage is less sensitive to the local dynamics of the explosion. The far-field damage can be used to estimate amount of energy released (or amount of fuel involved) in the explosion. Therefore, explosion simulations were conducted first in order to reconcile the near- and far-field blast damage with (1) the type of fuel involved, (2) the quantity of fuel required, and (3) the likely location of the ignition source. Some key observations regarding the blast damage include:

• Observed far field blast damage estimates • Walls of Room E blown outward by the explosion (Figure 8) • Directional damage in the warehouse including the lifted and bent grating (Figure 10)

The first step in modeling the explosion is to construct a geometry model of the facility. Flame acceleration and pressure buildup within the facility are very sensitive to both confinement and congestion of obstacles within the facility (i.e., railing, support structures, various types of piping, hoses and ducts, small containers, etc.). An exact replica of the facility was not possible because it was completely destroyed by the blast, and there were only a limited number of photographs showing the interior of the building prior to the incident. Therefore, the goal of the present study was to create a moderately accurate geometry model, capturing the details of the major elements. Figure 11 shows the geometry model used for FLACS simulations. Based on photographs and the 3-D laser scan, the position of walls, doors and windows are modeled with reasonable accuracy. The congestion of objects within the model, which creates turbulence and flame accelerations, was significantly less than observed in the few available photographs within the facility taken prior to the incident. In order to evaluate the effect of the actual congestion within the facility, objects were added to the model at given intervals (to be discussed later).

Figure 11: Geometry model of the CAI/Arnel facility in FLACS. The surrounding houses were modeled as simple boxes

LPS 2009 ____________________________________________________________ Paper 96a

5.1.1 Fuel Reactivity and Vapor Cloud Size The observed far-field damage from the blast wave can be used to estimate the blast overpressures. Using explosion models to predict this damage can only provide a rough estimate of the strength of the explosion and cannot be used to precisely determine the amount of fuel. For example, the estimated overpressures based on observed blast damage in Danvers are above those predicted by the strong explosion TNO Multi-Energy (TNO) curves (#6 to #10 as shown in Figure 12), assuming the entire volume of the warehouse and manufacturing areas (114,600 ft3) are filled with a fuel to just above stoichiometric (ideal) conditions.

Figure 12: Blast overpressures and predicted overpressures using the TNO method

Based on this result, certain investigative groups have concluded that 50% additional energy or fuel is required beyond stoichiometric conditions in order to account for observed damage. In addition, these groups further concluded that a fuel-rich mixture of heptane is required (equivalence ratio = 1.49) and that the explosion could not be caused by methane or natural gas. We disagree with these conclusions for the following reasons:

• Flame speeds fall off markedly for increasing large equivalence ratios for heptane/air mixtures (see Figure 13) and cannot generate the necessary overpressures to cause the observed damage.

Figure 13: Dependence of laminar burning velocity on equivalence ratio7,8

LPS 2009 ____________________________________________________________ Paper 96a

• For strong explosions (> 10 psi), there are very small differences in the far-field blast

pressure for different fuels. Large-scale, unconfined explosion experiments performed by British Gas (MERGE)9, showed comparable far-field pressures for strong methane and propane explosions, even if source pressures were 2 times higher for propane.

• Differences in the explosion model predictions and observed blast damage can be the result of large uncertainties associated with estimating blast overpressures from resulting damage. This method cannot be used to precisely determine the amount of fuel involved in the explosion.

• Partially confined, natural gas explosion experiments (BFETS10 Test 7 and HSE11 Phase 3a Test 4) have shown that experimental overpressures can be as high as those predicted by the strong TNO explosion curves (see Figure 12). Furthermore, these same experiments demonstrated that certain overpressures closer to the source were even higher than those predicted by the TNO method due to non-symmetrical explosion development (Figure 12).

Based on previous experimental and modeling data, methane or natural gas can only be ruled out as a cause if a strong explosion cannot occur within the facility (source pressures > 10 psi or 70kPa). Initial simulations using the simplified geometry resulted in somewhat lower pressures than those required to cause the observed damage for methane. However, this is to be expected as this simplified model contains considerably less congestion than what existed within the facility on the day of the incident (see Figure 14). Therefore, an anticipated congestion method12, which is regularly used in safety risk analyses, was employed in the present study to evaluate the results with the expected object density and distribution within the facility. A coarse grid of approximately 6-inch (15 cm) beams every 6.6 feet (2 m) was added to the model to more accurately replicate the actual obstruction density within the facility on the day of the incident. With the additional anticipated congestion, the far-field overpressures from a methane explosion were comparable to that of propane and confirm that a methane explosion cannot be ruled out.

Figure 14: Simplified geometry model (left), actual congestion within facility (right)

Initial FLACS simulations also showed that in order to cause the observed blast damage, a fuel must fill almost the entire available volume (warehouse and manufacturing) to levels very near stoichiometric concentrations.

LPS 2009 ____________________________________________________________ Paper 96a

5.1.1 Explosion Dynamics and Ignition Source In order to determine the location of the ignition source and resulting building damage during the explosion, it is necessary to understand the dynamics of an explosion. FLACS was used to evaluate the ignition of a fuel-air cloud, the propagation and acceleration of the flame front around obstacles and congestion, and the resulting overpressures. Figure 15 shows an example of the flame and pressure development for an exploding stoichiometric fuel-air mixture, which was ignited in Arnel’s storage and packing area as viewed from the north.

Figure 15: Flame and overpressure front progression (as viewed from North)

For this ignition, the flame accelerates away from the ignition source and vents out of the building and into Room E, causing a significant overpressure in Room E. The flame front then continues to the warehouse and significant overpressures are observed in this area of the facility. As the flame vents from the facility a blast wave is seen propagating away from the building.

Figure 16: Example of peak blast pressure distribution around the facility

LPS 2009 ____________________________________________________________ Paper 96a

Figure 16 shows an example of the calculated peak blast pressures measured at various locations within the facility and at the surrounding structures (residences and buildings). These peak blast pressures can be used to evaluate the likelihood of various explosion scenarios. In order to understand the explosion dynamics, GexCon modeled explosions for fuel-air clouds that filled the entire production area to just over stoichiometric conditions. More specifically, eight scenarios were simulated assuming different ignition locations shown in Figure 17. All ignition locations were chosen near the floor level to represent worst-case conditions, with locations #31-#34 within the warehouse, #21-#22 within Room E, and #11-#12 in Arnel’s storage and packing east of Room E. These simulations were performed with propane-air and methane-air mixtures using the assumed level of anticipated congestion within the facility. The results for methane were comparable to those for propane.

Figure 17: Ignition locations evaluated (left) and variation in side-on pressure at the severely

damaged bakery WNW of the facility (right). The explosion destroyed the entire facility, and a failure mechanism for the roof and walls needed to be considered. The detailed development of an explosion is very complicated and the goal of explosion modeling is not to understand the failure of each element within the facility, but to include enough failure mode details to understand the dynamics of the explosion and reconcile the observed damage. A simplified failure mechanism is modeled in FLACS by assigning a weight, a failure pressure, and a certain available venting area after failure, to various walls and ceiling elements. Windows have been assumed to fail at approximately 0.3-0.4 psi, doors at approximately 0.7 psi, and walls and ceilings at approximately 1-3 psi depending on the type of construction. When a roof (concrete or wooden) fails during the explosion, it was assumed that available effective venting area initially would be of the order 25-50%. Simulations were also performed assuming weaker roofs (50% lower opening pressure and twice as high effective vent area) to confirm that observed trends do not change. The failure of internal walls and some of the external walls has been ignored in the modeling, as these were not assumed to have significant impact on the initial explosion development. These explosion simulations provided interesting results regarding the dynamics of the explosion. One finding is that the blast waves emanating from the facility were somewhat directional and depended on the location of the ignition source (Figure 17 and Figure 18). It became immediately evident from the 8 explosion scenarios that significant blast damage to the west (bakery) and north (houses) was the result of most scenarios except the two most western ignition locations (#31, #33).

LPS 2009 ____________________________________________________________ Paper 96a

Figure 18: Illustrations of peak blast pressure distribution (psig) for the eight ignition scenarios. The walls of Room E were observed to have been blown outward during the explosion and must have been due to a significant overpressure in Room E early in the explosion. In order to achieve significant overpressures in Room E, the 8 explosion scenarios revealed that:

• Ignitions outside of Room E resulted in strong flame accelerations venting into Room E, causing significant overpressures in Room E sufficient to blow out its walls. Hence, the ignition must occur within a limited distance from the doorway to Room E in order for the flame to reach Room E prior to any significant pressure buildup in the room of ignition (#11, #12, #32, #34).

• Some of the ignition locations modeled in the warehouse (#31, #33) resulted in significant overpressures in the warehouse before the flame enters Room E. This would blow the wall that separates the warehouse and Room E in the opposite direction as that observed after the explosion.

• Ideal stoichiometric mixtures ignited in Room E (#21, #22) resulted in moderately high overpressures (3-10 psi) in Room E. However, if the mixture deviated from ideal conditions, significant overpressures occurred outside of Room E prior to significant overpressures in Room E, causing the walls in Room E to fail inward.

Based on the above analyses, it appears that ignition locations well within the warehouse can be ruled out, whereas ignitions just inside the warehouse close to Room E (#32, #34) and in Arnel’s storage and packing (#11, #12) yield the highest overpressures in Room E. While ignition in Room E (#21, #22) cannot be ruled out, these scenarios appear less likely as near ideal conditions are required. A more detailed assessment of observed damage could possibly have ruled out an ignition location in Room E. As mentioned earlier, directional damage from east to the west was observed in the warehouse, and consisted of lifted and deformed grating on the mezzanine (Figure 10), displaced steel trusses, and displaced railing and piping. Diamond plating originally covered the mezzanine grating. This damage appears to be consistent with significant forces associated with a directional blast wind from east to west, as well as an initial pressure difference across the mezzanine (higher pressure from below). Sensor points were placed within the model to monitor the simulated dynamic pressure (½ρu2), or the forces associated with drag loads, and pressure both above and below the grated mezzanine for the 6 remaining scenarios.

LPS 2009 ____________________________________________________________ Paper 96a

The simulations suggest that the dynamic pressure (from east to west) for the ignition scenarios within the warehouse (#32, #34) were too low to be consistent with the observed damage (see Figure 19). Simulations also revealed that the dynamic pressure associated with ignition locations in Arnel’s storage and packing (#11, #12) resulted in larger dynamic pressures on the mezzanine grating in the location where they were lifted and bent towards the west (see Figure 19) as compared to ignitions within Room E (#21, #22). Furthermore, these dynamic pressures were accompanied by pressure differences of approximately 4 to 5 psi across the mezzanine surface for ignition positions #11 and #12, which are consistent with the lifting of the grating. Lower pressure differences across the mezzanine were observed for ignition positions #21 and #22.

Figure 19: Dynamic pressure (upper) from east to west and pressure difference (lower) measured

across the mezzanine (Pbelow-Pabove). While ignition in Room E cannot be entirely ruled out, these results for near stoichiometric gas clouds throughout the production area indicate that ignition in Arnel’s storage and packing area is the most likely cause. Ignition locations in Arnel’s storage and packing were capable of reconciling the observed blast damage in the facility, including: (1) significant overpressure in Room E causing the walls to blow outward, (2) the blast venting into the warehouse causing the observed drag effects associated with a blast wind in this section of the buildings, and (3) directional damage to surrounding structures outside the facility. This suggests that this area of the facility likely contained a piece of equipment that was not rated for use in flammable environments, however, this was not verified during the inspections. 5.2 Dispersion Analysis GexCon next evaluated the likelihood of various scenarios creating a flammable mixture throughout the facility. The CFD tool FLACS was used to evaluate the release and dispersion of: (1) heptane vapor generation from the 01-1038 (V4) tank, (2) solvent spills or intentional solvent release, and (3) gas migration into the facility. A scenario was only considered possible if the simulated release resulted in a flammable mixture throughout the production area and was capable of producing the observed overpressure damage when ignited.

LPS 2009 ____________________________________________________________ Paper 96a

In order to model the flow and temperature distribution within the facility, it was necessary to account for natural and forced convection sources in the facility. The only heater in operation in the production areas was the Modine heater located on the south wall in the warehouse. The flow rate and heating capacity of this heater were modeled in all FLACS dispersion calculations and had a significant effect on the mixing and homogeneity of the vapors within the facility. More specifically, the heater was responsible for mixing the contents within the facility, regardless of the leak scenario. Had the heater been off, many of the leak scenarios involving the heavier components would have likely stratified to the ground level of the facility. Figure 20 shows the flow patterns and temperatures associated with the heater.

Figure 20: Flow and temperature profiles associated with the Modine heater in the warehouse

Furthermore, the air change rate of the facility, given as air changes per hour (ACH), was also modeled in all FLACS dispersion calculations. The facility was old construction with some broken and boarded up windows, large overhead doors, louvered openings for fans, and was not considered “tight” construction. Moreover, infiltration would be further enhanced by the wind (up to 6 mph ~ 2.7 m/s) reported from the north on the night of the incident and due to stack effects from the outside temperature (27 to 29°F, or –2 to –3°C) being colder than indoor temperature (~60°F or 15°C). While the ACH for the facility was likely greater than 1.0, we performed calculations using 1.0 ACH. Airflow into and out of the facility due to wind and stack effects was first evaluated with passive porous openings at the windows, doors and other openings. Next, to more accurately control the ACH low momentum sources of outdoor air were implemented on the north side windows/doors as well as other low level openings. Mechanical exhausts were implemented at the 2nd story windows and openings, and passive exhausts were modeled at ceiling ducts in the warehouse. 5.2.1 01-1038 (V4) Steam Valves Contrary to witness statements, one identified scenario is that the steam valves were left open on the 01-1038 (V4) tank and caused the product in the tank to boil, releasing significant quantities of flammable vapors into the facility. The following heat transfer calculation and analysis evaluate this scenario. We selected conservative estimates for various parameters that overestimated the steam input and underestimated heat losses from the tank.

LPS 2009 ____________________________________________________________ Paper 96a

As stated earlier, the steam heat was applied to the 01-1038 (V4) tank for approximately 4.5 to 3.5 hours (from 12:30-1:30 pm to 5:00 pm). In order to release a significant quantity of vapors (necessary to cause the explosion), the contents in the 01-1038 (V4) needed to boil on the night of the incident. The mixture will boil only if net heat transfer to the tank and its contents was sufficient to cause the material in the tank to reach its boiling point prior to the explosion. The CSB1 determined the mixture was a positive azeotrope with a boiling point of 165.6 °F (74.2°C).

Figure 21: Schematic of Heat Transfer in the 01-1038 Tank

Heat transfer into the 01-1038 (V4) mixture is driven from the hot pressurized steam in the steam heater through the stainless steel tank shell and then transferred into the mixture (Figure 21). The total heat energy transfer rate from the steam to the mixture is the product of the total heat transfer coefficient for the heater, Uin, the cross sectional area through which the heat transfer occurs, ACS, and the temperature difference across the steam heater. The total heat transfer coefficient from the steam to the mixture, Uin, is representative of all the thermal resistances across the heater and is system dependent. The ambient air in the facility was reportedly 60°F and was assumed constant throughout the event. It was also assumed that the mixture temperature was 60°F when the steam valve was turned on. While steam at a pressure of 8 to 10 psig was reported, the higher temperature of 239.4° F (115°C)13 was used as a conservative estimate in the calculations by providing the most heat transfer into the vessel. The total heat transfer coefficient into the mixture (from the steam) was determined by matching the heating rate to the temperature of the 01-1038 mixture when it reached 90°F. The shorter heat time (3.5 hours) was conservatively chosen in order to maximize steam input and shorten the time to heat the tank contents. After running the calculation for the 13.25 hours between 1:30 pm November 21, 2006 and 2:46 am November 22, 2006, the final mixture temperature was determined to be 148.5°F (64.7°C). This temperature is more than 17°F lower than the boiling point reported by the CSB. These calculations confirm that, even with conservative assumptions that increase steam input to the tank contents, it is not possible to have an explosive mixture of heptane vapors that emanated from the 01-1038 (V4) tank as a result of the steam heater being left on in the facility on the night of the incident.

LPS 2009 ____________________________________________________________ Paper 96a

Another agency performed a calculation to simulate the steam valves being left open on the 01-1038 (V4) tank. Errors were noted in these calculations and we were unable to confirm their results. The major difference in the other agencies analysis is that they assumed a 2-hour heat up time, resulting in a significant over-prediction of the heat input into the mixture. This assumption is in direct conflict with:

• CAI employee statements indicating it took 3.5 to 4.5 hours to make the batch on the day of the incident.

• CAI’s official response to a request for information, stated that it would normally take a representative batch of 01-1038 product 4.0 hours to reach the target temperature of 90°F before the steam heat is turned off

• According to interviews of other CAI employees, the steam heat is applied to the 01-1038 after the solvents have been added and prior to adding the solids because “if you warm it [solvents in the tank], it helps the resin cut in.”

Therefore in order for the contents of the 01-1038 tank (V4) to reach boiling conditions prior to the explosion one must incorrectly assume that the tank heat up time to 90°F was 2 hours. Regardless, even if the emission rate provided by the other agency (4.4 lb/min ≈ 33 g/s) is used in a dispersion calculation in FLACS, the mixture in most of the production area when approaching steady-state conditions (after 1.5 hours) is below the lower explosion limit (LEL) for a conservative ACH of 1.0 (see Figure 22). This is not surprising as the maximum steady-state concentration (well mixed in the facility) of a continuous leak of 4.4 lb/min given an ACH of 1.0 is below the LEL.

Figure 22: N-heptane concentrations at steady state for an emission rate of 4.4 lb/min out of the

01-1038 (V4) tank with an ACH of 1.0

Furthermore, even if one employs a very low and unrealistic ACH of 0.5 for this geometry, the concentrations of vapor within the facility after 2.5 hours are far from stoichiometric and are fuel lean (see Figure 23). While some of this mixture may burn, the flame speeds and energy will not be adequate to cause the observed damage.

Figure 23: N-heptane concentrations after 2.5 hours for an emission rate of 4.4 lb/min out of the

01-1038 (V4) tank with an ACH of 0.5 5.2.2 Solvent Spill or Intentional Release

GexCon evaluated the possible contribution of both a solvent spill due to a container failure or from an intentional release. On the night of the incident, an individual recalled seeing a pick-up

LPS 2009 ____________________________________________________________ Paper 96a

truck in the parking lot of the CAI/Arnel facility prior to the incident. One agency did not include this statement in their final report and another agency concluded that no suspicious activities were documented. Furthermore, leaking totes were observed within the facility following the explosion. GexCon performed dispersion studies regarding the release rates for two different cases: (1) 350-gallon (1325 liters) tote release (intentional/accidental) in the warehouse, (2) 500-gallon (1893 liters) tote release (intentional/accidental) in Room E or an intentional pump release in Room E (≥ 500 gallons). The vapor necessary to fill the whole facility to stoichiometric conditions corresponds to approximately 130 gallons (490 liters) of acetone. Simulations were performed using a fluid mixture with similar vapor pressure and heat of vaporization per unit mole as acetone, a solvent used in significant quantities in the facility. Due to its significantly higher vapor pressure than heptane, this solvent appeared to be the most likely fuel source during a release. Releases were simulated along the southern wall of the warehouse (350 gallons total) in the vicinity of the Modine heater and in Room E (500 gallons total). The release in Room E was performed towards the center and west of Room E, to model the slight grade down into the warehouse. Initial simulations were conducted at 60°F (15°C) in the facility. The extent of evaporation was driven by the pool spread and dominated by the heat and mixing induced by the Modine heater. Figure 24 shows the release and spread of the pool initiated in Room E.

Figure 24: Spill from Room E. Pool shown after 1 min (top-left), 2 min (top-right), 3 min

(bottom-left), and 5 min (bottom-right) Initial simulations with spill rates of approximately 80 gal/min (300 liters/min) revealed that evaporation rates up to 21 lb/min (~160 g/s) were observed for the spills in the area of the Modine heater. Actual evaporation rates may even be higher, but due to stability issues regarding the pool spread model, they could not be analyzed further. These rates are in excess of those required to fill the facility to stoichiometric conditions with an ACH of 1.0, and suggest the spill scenario of acetone as a viable cause to this explosion.

5.2.3 Gas Migration into the Facility The ground penetrating radar analysis revealed many pathways and conduits into the facility namely from the north (towards Bates Street) and from the west (towards Water Street). Numerous used and abandoned pipes were found penetrating the footprint of the facility,

LPS 2009 ____________________________________________________________ Paper 96a

including: (1) a sewer pipe running from the manhole north of the manufacturing section to within the northeast corner of manufacturing, (2) the six pipes from the USTs through the building wall into Room E, and (3) a 6-inch pipe in the southwest corner of the warehouse that would have been behind Arnel’s drum storage. Inspection of the 6-inch pipe revealed that it led to Water Street. While, the contribution of this pipe and other pathways has not been investigated at the time of this paper, numerous natural gas leaks have been reported both prior to and after the explosion. GexCon evaluated the possible migration of natural gas into the facility. Our study concluded that approximately 11,000 to 14,500 ft3/hr (311m3/hr to 411m3/hr) were required to fill the facility in order to achieve flammable mixtures around stoichiometric conditions. Figure 25 shows the buildup of natural gas within the facility had the point of infiltration been located in the southwestern corner of the warehouse (i.e., via the abandoned 6-inch open pipe). Rich concentrations are observed above the leak because the leak is in the corner and not readily mixed by the Modine heater.

Figure 25: Natural gas concentrations at steady-state for 14,500 ft3/hr natural gas leak with an

ACH of 1.0 The pressure in the underground distribution system on Water and Bates Streets, approximately 150 feet away from the building, was reported to be approximately 11 inches water column (2.74 kPa). Therefore, in order for natural gas to enter the facility a fairly unobstructed pathway is required (e.g., abandoned pipes such as the 6-inch pipe in the warehouse). We have not conducted further investigation into the multiple pathways identified from ground-penetrating radar. 6. Conclusion Investigation of the largest explosion in the history of Massachusetts revealed that the cause and origin of the explosion proposed by other agencies (i.e., steam valves of a production vessel being left open causing its contents to boil, and emitting significant quantities of n-heptane vapor) is not possible. More specifically, a detailed heat transfer analysis confirmed that in the event the steam valves were left open, the contents within the tank would not boil nor emit significant quantities of n-heptane vapor. Regardless, dispersion studies using FLACS demonstrated that even if the contents boiled, the emission rate would not result in a flammable mixture throughout the facility. Key witness statements and damage to solvent storage vessels provide the possibility of an intentional or accidental solvent release within the facility. In addition, multiple reported natural gas leaks, along with recent excavation work and various pathways/conduits identified by ground-penetrating radar, provide the possibility of natural gas migration into the facility. Using

LPS 2009 ____________________________________________________________ Paper 96a

the CFD tool FLACS, it was possible to investigate these scenarios along with the dynamics of the explosion itself. These results confirmed that the facility needed to be filled with a fuel-air mixture to near stoichiometric conditions and that the ignition source location was in the eastern section of the facility. While the exact cause remains undetermined, the results suggest that a solvent release or gas migration appear to be the most likely scenarios. Gas dispersion and explosions are very complex. Using simplified models and approximations can result in incorrect determinations of cause and origin. Detailed inspections, including relevant data collection, along with software tools such as FLACS allow complex scenarios to be modeled and reconciled with the available evidence. This investigation emphasizes that investigators should not reach conclusions too early in the study without having examined or considered all of the relevant data. In addition, this study emphasizes that software tools such as FLACS can be used to perform consequence analyses for these types of facilities to avoid similar incidents in the future. Using a CFD tool such as FLACS, facility owners can identify optimal locations to install gas detectors that activate the ventilation or shutdown systems when vapor levels in the facility reach a threshold limit (below the LEL) for various fuels and releases. 7. References [1] CSB Investigation Report 2007-03-I-MA, May 2008 [2] MA State Police Fire/Explosion Investigations Unit Final Report May, 7 2007 [3] Baldwin, D.B., “SafeVue (Safety Viewport Analysis Code),” version 2.0, Naval Facilities

Engineering Service Center, Port Hueneme, CA, 1995 [4] NFPA 912 “Guide for Fire and Explosion Investigations,” 2008 [5] http://www.flacs.com [6] Hansen, O.R., Wilkins, B., Oil mist explosions in a test channel, 37th Annual Loss

Prevention, Symposium, March 31 – April 2, 2003, New Orleans [7] Davis, S.G. and Law, C.K. Combustion Science and Technology, 140: 427, 1998. [8] Vagelopoulos, and Egolfopoulos, Proc. Combust. Inst. Vol. 27, 513, 1998. [9] W.P.M. Mercx et al., Modelling and experimental research into gas explosions, Overall

Report of the MERGE Project, CEC contract: STEP-CT-0111 (SSMA), 1994 [10] Selby, C., Burgan, B., 1998, Blast and fire engineering for topside structures, Phase 2, Final

summary report, Steel Construction Institute, UK, SCI Publication Number 253 [11] Al-Hassan, T., Johnson, D.M., 1998, Gas explosions in large-scale offshore module

geometries: Overpressures, mitigation and repeatability, presented at OMAE-98, Lisbon, Portugal

[12] Hoorelbeke, P., Izatt, C., Bakke , J.R., Renoult, J., and Brewerton, R.W., 2006, Vapor Cloud Explosion Analysis of Onshore Petrochemical Facilities: ASSE-MEC-0306-38

[13] Y. Cengel and M. Boles, Thermodynamics, An Engineering Approach, 2nd Edition.