In a fertilizer plant, an ammonia leak (NH3, toxic gas) was discovered at 7:30 a.m. on a non-return valve (DN 50) installed on the purge line of a pipe (DN 200, Design pressure: 2.8 bar) connecting the NH3 storage tanks to a manufacturing unit that was in the process of being shut down for maintenance. The NH3 detection system automatically shut down the NH3 compression system. At 7:45 a.m., the first responders shut several manual valves to isolate the piping and deployed water curtains near the valve to abate the NH3 vapours. Staff at a neighbouring company took cover at 8 a.m. The onsite firefighting personnel intervened, and at 8:30 a.m., the NH3 concentration was measured at 50 ppm downwind, near the leak. The confinement of the personnel in the neighbouring company was lifted. Several NH3 compressors were restarted to avoid damaging the site’s storage facilities (owing to the risk of pressure build-up), and additional water curtains were set up downwind. At 11 a.m., the valve was removed, and blind flanges were installed in its place.
The operator responded to requests from the local media at 8:15 a.m., and then issued a press release at 10:30 a.m. The prefecture, neighbouring municipalities and the Classified Facilities Inspection authorities were informed. The valve flange had broken in two at the clamping bolts; 2 rods still had their nuts in place but were twisted. Another nut had been torn off, and a threaded rod was outside the valve body. The quantity of leaked NH3 was estimated at 1.5 m³, of which 505 kg evaporated.
The brittle failure of the valve cap was determined to be the cause of the leak. This cap had been flapping when the leak was noticed (caused by the sporadic release of NH3). Cross-threading of the bolts (improper tightening, improper mounting of studs), corrosion or a defect in the cap were determined to be the cause of the failure. The unit’s shutdown phase, during which the leak occurred, involved injecting hot gas into the circuits to flush out the liquid-phase NH3. The thermal expansion caused an estimated overpressure of 19 bar compared to 1.5 bar during normal operation. The installations were designed to withstand this pressure. Protective valves, calibrated at 20 bar, were installed but had not been activated on the day of the accident. The 4 other valves of the same design had been replaced by stainless steel caps, less sensitive to such failure, and secured by stainless steel bolts. A pressure recorder was installed on the affected line to better understand pressure variations during transient phases such as shutdowns.