Fire sprinkler systems have a deserved reputation of reliability and effectiveness. When properly designed, installed, inspected, and maintained, they are expected to lie dormant for decades, poised ready to respond to a potentially damaging fire. As a result, fire protection engineers have confidence that sprinkler systems will reliably function when called upon to do so. However, on rare occasions, sprinkler systems may fail to perform as intended or expected. When this occurs, fire protection engineers may be called upon to determine the cause of the system failure. This article summarizes a portion of an investigation and analysis of one such failure: the unexpected separation of a mechanical coupling that had been installed for nearly three years.

At approximately 11:00 p.m., a 2-inch (50 mm) diameter wet pipe sprinkler line installed in a large metropolitan hospital suddenly separated and began to immediately discharge water. The separated pipe was above the suspended ceiling of the high-rise hospital's eighth floor. The hospital building housed in-patient sleeping areas, diagnostic and treatment suites, operating rooms, support, and related patient care areas.

It was estimated that water discharged from the separated pipes at a flow rate in excess of 1,000 gallons per minute (0.63 m3/s). In total, more than 10,000 gallons (40 m3) of water flowed directly into the hospital from the separated pipes. On-duty hospital personnel responded immediately, investigated the separation, and shut down the flow of water. In the meantime, the water rapidly flooded the upper floor of the hospital and cascaded to all floors below, causing extensive damage to the building, equipment, and contents.

In response to the massive flow of water, the hospital instituted their formal Internal Disaster Procedures to assure patient care and safety, and to mitigate the impact of the water. Off-duty personnel were called in. Patients were relocated, and efforts were made to stem the flow of water. There were no injuries reported.

After the water was shut off, it was observed that the pipe separation had occurred at a mechanical coupling used to join a piece of feed main to a fitting on a wet pipe sprinkler system.

A mechanical coupling connects pipe ends and fittings together using a specially designed clamping collar and gasket. The coupling used two bolts that, when tightened, clamp the pipe ends and fittings together. Hospital staff took photographs of the coupling still in place on the pipe end as it was found after the water to the sprinkler system was shut off.

The separated coupling was a flexible, rubber-gasketed fitting, which was listed and approved for fire protection service. The joining pipe was manufactured of thin-wall steel with rolled-groove treated ends with the coupling still attached on one end to a rolled-groove coupling. The coupling was compatible with and listed for use with the installed fittings and pipe.

A material and dimensional analysis of the connecting pipe and fitting determined that their wall thicknesses and groove dimensions were within tolerances specified by the manufacturer. No material defects were observed in either component. However, upon inspection, striations were evident on the separated end of the pipe section. (See Figure 1.)

When originally constructed in the early 1970s, the hospital was not provided with automatic sprinklers throughout the upper floors. As a result, many sprinkler systems were retrofitted into areas of the hospital after the original construction. A review of available records showed that the system that experienced the coupling separation had only been installed for approximately three years at the time of the separation. The automatic fire pump supplying the sprinkler system and other areas of the hospital campus was rated for 1,000 gpm (0.63 m3/s) at 165 psi (1.13 MPa). The pump developed a maximum churn discharge pressure of more than 230 psi (1.6 MPa). The fire pump was preexisting when the subject sprinkler system was designed and installed.

Due to the extensive damage and disruption caused to the hospital and to guard against a possible reoccurrence, it was critically important to hospital representatives that the cause of the coupling separation be determined. As a result, an engineering investigation was undertaken to determine the underlying cause.

The initial phase of the investigation included gathering and documenting all available factual data surrounding the circumstances of the coupling separation. Data was collected through a detailed review and documentation of the existing sprinkler system installation, study of available construction records, nondestructive examination and testing of system components, and eyewitness interviews.

From an initial inductive analysis of the collected data, the following factors were identified as potential contributing factors to the coupling separation.

  • Improperly installed coupling Photographs of the separated coupling showed a gap between the mating surfaces of the coupling's bolt shoulders. (See Figure 2.) A coupling installed per the manufacturer's installation instructions for the coupling exhibited no such gap. (See Figure 3.)
  • Lack of pressure reducing valve (PRV) Hydraulic calculation errors were made during the design of the subject sprinkler system that resulted in the omission of a PRV for the subject sprinkler system. A PRV was required by applicable standards based on the operating pressures of the existing fire pump serving the hospital campus, the sprinkler system's relative location to the fire pump, and the rating of sprinkler system components.
  • Improper hangers Certain pipe hangers and supports that were required by applicable standards were omitted near the location of the coupling separation.
  • Improper pump start pressures An hydraulic calculation error resulted in the fire pump start pressure being established well below the setting recommended by relevant standards (NFPA 20).
  • Jockey pump cycling On the afternoon of the separation, the hospital's jockey pump was noted by hospital staff to be cycling on and off very frequently.
  • Air in the piping system The subject sprinkler system piping that separated served the uppermost level of the hospital, providing a greater opportunity for the accumulation of trapped air.

To assess the contribution of the above potential factors, tests were conducted at the Maryland Fire and Rescue Institute (MFRI) in College Park, Maryland, under the direction of Dr. Fred Mowrer, Associate Professor of Fire Protection Engineering at the University of Maryland. A substantially similar replica of the fire sprinkler system was constructed using exemplar fittings, couplings, and piping. The layout of the system mirrored detailed measurements and technical data from the failed system. The replicated system addressed factors such as downstream piping volume, specific hanger configuration, and fire/jockey pump arrangement and settings.

A series of initial tests was conducted to validate the testing concept for the replica system and to investigate the general relationship between the identified potential factors and their tendency to cause or contribute to a coupling separation. Factors including coupling bolt torque, the presence of a pressure reducing valve, jockey pump cycling, fire pump pressure surges and air entrainment in the piping system were varied to determine which, if any, may have led to the coupling separation (alone or in combination with other potential factors).

One of the initial test series focused on the potential impact of jockey pump cycling on the integrity of the installed coupling. The purpose of the testing was to investigate if the cycling of the jockey pump observed by the hospital staff on the afternoon of the separation would generate enough pressure, pipe movement, etc., to loosen an installed coupling and/or contribute in any way to the coupling separation.

Jockey pump cycling tests were run on both a properly installed coupling (tightened to the manufacturer's specified bolt torque) as well as on exemplar couplings with loosened and fingertightened bolts. Conservative pump pressure ranges and cycle frequencies (i.e., start/stop duration) were used. The cycling tests were conducted for durations as long as 15 hours. In each test, the sprinkler system piping simply swayed gently with each start/stop cycle. No separations occurred, and no leaks were observed anywhere in the system including during tests with loosened and finger-tight couplings.

Further jockey pump cycling tests were conducted using the actual jockey pump that was removed from the hospital. The regenerative turbine jockey pump was physically replaced and reinstalled in a test set up to evaluate a theory developed that the repetitive jockey pump cycling observed by the hospital staff would result in a cumulative increase in downstream system pressure due to the specific configuration of the hospital system's check valves and backflow preventer. Reportedly, an independent surge analysis/computer model predicted that extraordinary pressures well in excess of the rated system pressure could develop due to the cumulative effects of jockey pump cycling. While the model was not reviewed, it was theorized that excessive forces from these predicted excessive pressures could be related to the separation. The additional testing clearly demonstrated that excessive pressures predicted by the computer model would not develop from the observed jockey pump cycling. As expected, the maximum pressure achieved by the hospital's jockey pump system was equivalent to the jockey pump shut-off pressure plus the maximum available static pressure from the pump source of supply. The maximum potential pressure from the hospital's jockey pump system was well within the working pressure of the sprinkler system and coupling.

Another important portion of the investigation was to conduct hydrostatic tests on the replica system to the requirements of NFPA 13, Standard for Installation of Sprinkler Systems. The testing was performed to assess the hydrostatic performance of the separated coupling under varying bolt tightness conditions to the acceptance testing criteria contained in NFPA 13. To accomplish this, the replicated sprinkler system was hydrostatically tested several times under two configurations:

  • Each coupling on the system was installed in strict accordance with the manufacturer's installation instructions. Specifically, all coupling bolts were tightened to the required torque.
  • The exemplar coupling at the location of the separation was installed with its bolts loosened and only "finger-tight." That is, the final tightening step specified by the manufacturer's instructions was not accomplished. The "finger-tight" and loosened bolts resulted in a gap between the mating surfaces of the bolt shoulders similar to that observed in the photographs taken of the actual separated coupling.

In each hydrostatic test, the replicated system successfully passed the acceptance requirements of NFPA 13. No leaks developed or occurred during any hydrostatic test. Surprisingly, the hydrostatic tests required by NFPA 13 did not uncover the improperly installed couplings, including those with their bolts installed only "finger-tight."

Following the initial exploratory testing, a series of preliminary surge tests was conducted on the replicated system. For each preliminary surge test, water was introduced into the piping system by rapidly opening a valve installed in the sprinkler system's supply. The character and magnitude of the induced surge was established from a study of the flow and pressure potential of the hospital's fire pump and the pump "start pressure" settings. and recorded using pressure transducers and data recording equipment. A similar capacity mobile pump was used to simulate the hospital's fire pump.

The following identified potential factors were considered in the preliminary surge testing. Each potential factor was tested independently and in combination.

  • Coupling installation (i.e., varied bolt tightness)
  • Installation of pressure reducing valve (set per NFPA 13 requirements)
  • Air in the piping system

During each of the preliminary surge tests, the piping system moved dramatically in response to the rapid introduction of water flow and increase in pressure. However, as expected, the magnitude of the piping system's response to the induced surge was clearly dampened for tests that included a pressure reducing valve and those when the piping system was "water solid" (i.e., with no introduced air).

For all but one of the preliminary surge test sequences, the piping system (including with loosened couplings) maintained the system pressure boundary and did not separate or even leak. However, during one of the surge tests, the test coupling catastrophically separated. (See Figure 4.) It was noted that striations similar to those observed on the actual separated pipe were also created on the pipe from the test.

The factors present at the time of the coupling separation during the preliminary surge testing included: 1. "finger-tight" coupling bolts, 2. no pressure reducing valve, and 3. air introduced into the piping system. When only one or two of these factors were present, the coupling did not separate.

Based on the results of the preliminary surge test series, subsequent surge testing was performed varying coupling bolt torque, installing a compliant pressure reducing valve, and introducing air in the sprinkler system piping. Again, the coupling separated only when all of the above three conditions were present.

Sprinkler systems have a deserved history of reliable and durable performance. This investigation supports this reputation. However, when failures do occur, the lessons learned from related investigations can provide useful knowledge to fire protection professionals.

The fundamental cause of this coupling separation was the improper installation of a mechanical coupling. Specifically, the bolts of the coupling were left finger tight when originally installed. While the lack of a PRV and accumulated air in the system also contributed to this particular occurrence, these factors would have been inconsequential but for the loose coupling bolts.

The use of a torque wrench by sprinkler system installers is unusual, and traditional rolled-groove couplings do not have a visual tell-tale, such as break-away bolt heads, to assure minimum torque is reached. Fire protection engineers should recognize that the absence of a gap between mating coupling surfaces is not a reliable indicator that the specified bolt tightness has been achieved. (For example, the gap on this coupling tested completely closed when less than 10 percent of the required bolt torque was applied.)

Further, while counterintuitive, the hydrostatic testing requirements of NFPA 13 may not reveal improperly installed couplings, including those that are only "finger-tight."

Notwithstanding the above, rolledgroove mechanical couplings are well accepted and reliably used in the fire protection industry. Overall, the testing conducted in this investigation supports this comfortable reliance. Even when the coupling bolts were installed with significantly less torque than specified, the couplings performed well and did not separate or leak when subjected to this surge test protocol.

The sole purpose of this investigation was to review a discrete and specific set of potential factors on a particular coupling configuration and separation, and the related information should not be considered a general review of coupling performance. However, it is somewhat instructive on a practical level that there were more than 350 opportunities for a coupling to separate during the surge testing, yet the only separations or leaks that occurred were with couplings that were "finger-tight." This outstanding performance, as well as the rarity of coupling separations generally in fire protection systems, is clearly due to the robust design of the coupling and diligent installers. Nevertheless, installation oversights can occur. Fire protection engineers in conjunction with the installing contractor should give appropriate consideration to specific methods to assure the tightness of coupling bolts in accordance with manufacturer's requirements.

Mr. Duane Johnson, Fire Protection Engineering student at the University of Maryland, is specially acknowledged for his assistance in constructing the replicated system and conducting of the testing. Also, thanks are extended to the Maryland Fire & Rescue Institute and College Park Volunteer Fire Department for use of equipment and facilities.

Daniel Arnold is with Seneca Fire Engineering, LLC.