1,124 Corroded Sprinkler Heads

Every account of this one lands in the same place. A company let its sprinklers rot, water hit fourteen million pounds of pool chemicals, and a Georgia county spent seventeen days indoors. Run to failure. Cheap outfit, predictable ending.

Bio-Lab did replace its corroded sprinkler heads. It bought the right ones, engineered for exactly this chemistry, and installed them two years before the fire.

It installed them in the room where the chemicals used to be.

Bio-Lab Conyers made the chlorine tablets people drop into swimming pools. The active ingredients are chlorinated isocyanurates, mostly trichloroisocyanuric acid and sodium dichloroisocyanurate. They are dry solids that carry a large amount of chlorine, and they give it up when they get wet. That's the entire point of the product. Drop one in a pool and it releases chlorine into the water.

The hazard is that a warehouse isn't a pool. Wet these solids with a small amount of water and they don't dissolve politely. They decompose, and decomposition generates heat, and the heat drives more decomposition. What comes off is chlorine gas and hydrogen chloride, and later, from a related compound, bromine.

Those gases matter to a person because they go after wet tissue. Your eyes, your throat, the lining of your lungs. Chlorine reacts with the moisture already in your airway and turns into acid there. A small dose burns and makes you cough. A larger one can flood the lungs hours after you walked away feeling fine.

So the whole storage philosophy for this product is: keep it dry, keep it cool, keep it ventilated.

Plant 12 was the storage warehouse, built in 2019. Inside it was a fire-rated concrete room the crews called the bunker, where most of the chlorinated isocyanurates lived. Everything else sat in the open warehouse, in racks and stacked on the floor. Above all of it ran an ordinary wet-pipe sprinkler system, the same kind over your own warehouse, full of water, waiting for a fire.

The product decomposes when it gets wet. The safety system directly above it is a grid of pipes full of water.

Hurricane Helene came through on September 26 and 27, 2024. Georgia declared an emergency, Bio-Lab shut the site down, and the plant ran on a fire watch: two people walking the buildings around the clock, looking for exactly the thing that was about to happen. The site had made that fire watch permanent two or three months earlier, after people kept smelling oxidizer in the warehouse.

At about 5:00 a.m. on Sunday, September 29, the employee on fire watch in Plant 12 stepped out of the breakroom and heard popping. They first thought it was the ice maker. It wasn't, and they knew it. They'd been trained on what wet product sounds like. They called the second fire watch employee, then site management.

By 5:04 the second employee was there and had a full-face respirator on. Neither of them saw fire. What they saw was dense white fumes coming off super sacks stacked against the wall just outside the bunker. At 5:10 the first employee called 911 and reported, accurately, that a sprinkler head had burst and chemical smoke was everywhere.

Then they tried to get the reacting sacks out of the building. Each sack held between 2,205 and 2,805 pounds. The off-gassing ones sat behind three rows of double-stacked sacks. One employee got four sacks clear before the fumes took their visibility and the sprinkler water made the floor too slick to run a forklift. They left.

Rockdale County Fire Rescue arrived by 5:30 and found water already pouring out of every dock door except the bunker's. Decomposition makes heat long before it makes flame, and the heat shorted the alarm wiring at 5:32. The fire protection system failed completely at 5:37, a full hour before anyone saw fire. Firefighters saw the first small flames at 6:30, put the roof fire out around 8:10 using what they estimated as a couple of gallons, deliberately, because they knew from a 2020 event at this same site what water does here.

Crews spent the next few hours ventilating, and KIK employees went into the bunker, the one part of the building still uninvolved, to pull product out before the fire could reach it. The bunker held about 4,000 super sacks. They got 59 of them out before responders spotted a second fire on the roof and ordered everyone out of the building. At noon the roof opened up in a hole one responder put at 100 feet by 100 feet, throwing flames forty to fifty feet in the air. Walls began collapsing at 12:40. The fire was out by 4:00 p.m. and the rest of the building was down by 8:00.

Seventeen thousand people were evacuated. Ninety thousand across metro Atlanta were told to shelter in place. The emergency response ran seventeen days, with nightly shelter-in-place orders inside a two-mile radius and schools closed for nearly two weeks. Property damage came to $50.5 million and the site never reopened.

No injuries were reported, at the site or across the seventeen days of response that followed. That one came down to luck. The plant was shut down for a hurricane, it was 5:00 a.m., and it was a Sunday. A Tuesday day shift is a different report.

Trace it the way a hurried investigation would.

Were the sprinkler heads corroded? Yes. Documented. Was it documented for years? Yes, in 2021, in 2022, and again in December 2023. Did anyone count them? Yes, 1,124 of them. Did the company replace them proactively? No. It replaced them as they failed and leaked.

Run to failure. The CSB names it that way itself, as the first of five safety issues. Every box checks. The trade headlines wrote themselves: not replacing corroded sprinklers caused this fire.

That version is accurate and it stops one question early. It says Bio-Lab wouldn't spend the money. The record says Bio-Lab spent the money, on the right hardware, and still ended up here. The question worth asking is how a company that correctly diagnosed this hazard and correctly fixed it still burned its warehouse down.

The corrosion showed up almost immediately. TCCA went into the bunker in 2019 and within two months the fire protection piping was corroding. By December 30, 2019, a coupling had corroded through its bolts and leaked.

Nobody had to guess why: this wasn't exotic chemistry. The sacks off-gas chlorine as a matter of routine, the chlorine meets humidity in the air, and it condenses on cold metal as hydrochloric acid. The warehouse had no air conditioning and a ventilation system too small to clear the vapor. The building was quietly manufacturing acid on its own ceiling..

The site read this correctly. Annual inspections in 2021 and 2022 found hundreds of corroded heads, concentrated in the bunker where the chemicals were. Insurance risk assessments in 2021 and 2022 found extensive corrosion on piping and sprinklers across multiple buildings, and one building's fire protection system was completely out of service from corrosion damage. So in late 2022 Bio-Lab replaced the bunker's brass-finish heads with wax-coated corrosion-resistant heads, and in 2023 upgraded the bunker's bolts and couplings too.

That's a correct fix. It worked. On the morning of the fire, responders looked into the bunker and reported no fumes, no sprinkler activation, nothing involved.

During 2023 the business grew the inventory, and the bunker ran out of room, so chlorinated isocyanurates started being stored outside the bunker as well. The chemistry moved into the part of the building that still had ordinary brass heads. And the standing instruction out there didn't change: when a corroded head failed and leaked, replace it in kind, with another brass one.

The December 2023 inspection reads like a diagram of the mistake. Inside the bunker, where the corrosion-resistant heads were: no deficiencies. Outside the bunker, where the chemicals now lived: 1,124 corroded heads across four of the system's seven zones. December 2023 sprinkler deficiencies, shown as vertical bands, over the chemical storage layout. Every one was outside the bunker. (Credit: CSB)

Everything after that is the building telling them, in writing, monthly. Between January 1, 2023 and September 29, 2024, the fire protection system sat in an abnormal state on at least 60 of 637 days. About one day in ten. In July 2024 crews made emergency repairs to a hole in the fire protection piping. The five-year inspection on September 12, 2024 recorded no corrosion at all, and then corroded heads were replaced on an emergency basis on September 13, and again from September 17 through 20. Nine days before the fire.

The corrosion-resistant heads stayed in the bunker. The chemicals didn't.

Between January 1, 2023 and September 29, 2024, the fire protection system was logged abnormal on at least 60 of 637 days. Across those twenty-one months, the system meant to catch this had about two months of not being right. In July 2024 crews made emergency repairs to a hole in the fire protection piping. The five-year inspection on September 12, 2024 recorded no corrosion at all, and then corroded heads were replaced on an emergency basis on September 13, and again from September 17 through 20. Nine days before the fire.

Under all of it sat a classification problem, and it is worse than the number everyone is going to quote.

The number everyone is going to quote is the inventory. Before construction, Bio-Lab's design-build contractor told local authorities the warehouse would hold an average inventory of about 6.2 million pounds. On the morning of the fire it held close to 14 million, more than double what the county was told.

The CSB found a bigger problem underneath it. The building as designed in 2019 did not comply with NFPA standards, and for its construction type the limit on Class 2 oxidizers was 500 pounds per control area. The warehouse held roughly 1.2 million pounds of Class 2 material. Had it been built to the high-hazard standard instead, it could have held four million pounds of Class 2 legitimately, outside the bunker, with the ventilation and the detection that classification demands.

So the doubling isn't the story. At 14 million pounds this was the wrong building. At 6.2 million pounds it was the wrong building. At 500 pounds it was the wrong building. The compliance failure was there the day the doors opened, at any inventory anyone would realistically have put inside, because the design had answered a question about storing combustible goods rather than a question about storing oxidizers.

What the inventory growth actually decided was location. The bunker filled, product moved out under uncoated brass, and no process existed to make anyone re-ask the sprinkler question when it did.

And none of this was the first time.

On May 25, 2004, a 260,000 square foot warehouse at Bio-Lab Conyers was destroyed the same way, holding about 12.5 million pounds of pool chemicals and oxidizers. It released hydrogen chloride, shut down Interstate 20, sent 28 people to hospitals, and threw a plume that people reported feeling more than fifty miles away. Plant 12 was later built on that ground.

Then September 14, 2020, in Plant 6 on the same site. A three-eighths-inch water line off a water heater developed a leak, floodwater reached TCCA sitting on the floor, and the material decomposed. Employees and nine firefighters were exposed, and the interstate closed for about six hours. Bio-Lab moved the surviving inventory out of Plant 6 and into trailers. That inventory had already been wetted. Four days later, one of the trailers went up.

That one is the same failure, four years early and one building over. They moved the material to get it away from the problem and nobody re-asked what the material had already been through. A Rockdale County responder gave the CSB the summary: "we've been here many times."

The corporate layer above the site had every reason to catch it. Beyond Conyers' own history, KIK had watched this chemistry burn at Bio-Lab Lake Charles in Westlake, Louisiana in 2020, which the CSB also investigated. Rockdale County firefighters remembered all of it well enough to ration their water on the roof in 2024. KIK also had the paperwork for exactly this problem, including a corporate procedure it called PSM Lite, a requirement to run hazard analyses, and a fire protection impairment process. The CSB found none of them fully implemented or enforced, and it found the company treating off-gassing events as routine business as usual.

Another Bio-Lab site had already solved part of it. Bio-Lab On-Cal used preventive and mitigation practices that Conyers didn't, including thermal cameras that spot a super sack heating up before a person could smell it. Nobody required Conyers to adopt them. A fix existing somewhere inside your company isn't the same as a fix reaching the building that needs it.

The investigators used a frame worth stealing, and they took it from the CCPS book Recognizing Catastrophic Incident Warning Signs in the Process Industries, which opens with the claim that most if not all incidents were preceded by warning signs. Three of the named asset-integrity warning signs are run to failure, corrosion, and frequent leaks. Bio-Lab had all three, for five years, in writing.

The move that makes this transferable is one sentence in Section 4.1. Sprinkler systems are not process equipment, the CSB says, but they meet the definition of asset integrity, because asset integrity covers equipment designed to contain, prevent, or mitigate a hazardous release. So the sprinkler grid gets judged by process safety standards rather than facilities standards, and once you do that, 1,124 corroded heads stop being a maintenance backlog and become a degraded safeguard on a live hazard.

Then they did the boring, decisive work. They asked for every sprinkler inspection report and insurance assessment for five years and got eight documents, and they read them against each other. They pulled 637 days of fire alarm monitoring logs and turned them into a calendar. They mapped the December 2023 deficiencies onto the seven fire protection zones and laid the chemical storage locations over the top, which is the picture that proves the whole argument.

And they killed the competing explanation instead of ignoring it. Roof leaks were documented during Helene, and a roof leak is the obvious suspect. So they pulled the rain gauge nearest the site: ten inches on September 26 and 27, then nothing for the 36 hours before the fire. Rain had stopped a day and a half earlier. Combined with the employee's own account and the water observed running out the dock doors before any hose was charged, the sprinkler system was the only source left standing.

None of that needed a technique with a name. It needed somebody to read eight documents against each other, build a calendar out of alarm logs, lay two maps on top of one another, and go get the rain data before assuming.

Your fire protection system is process safety equipment, and almost no mechanical integrity program in this industry says so.

Think about where the sprinkler ITM report lands at your site. Plenty of us do see it. Fire protection sits on the EHS desk at a lot of plants, and if it sits on yours you have signed one of these off in the last year. So the problem isn't that nobody looks. The problem is what the document is understood to be when it arrives. It comes from a third-party vendor, it is written to NFPA 25, and it reports deficiencies to be scheduled and closed. It gets read as a compliance obligation with a work-order queue attached. Almost nowhere does it get read as the current condition of a safeguard sitting over one specific hazard, which is the only reading that turns 1,124 corroded heads into a countdown instead of a backlog. I have signed those reports. I checked the deficiencies had work orders against them. I never once asked what was underneath the heads.

The standards didn't help either. The CSB found the corrosion hazard split across three of them, each looking at a different victim. NFPA 400 treats corrosion as a human health question. NFPA 13 treats it as sprinkler degradation. NFPA 70 treats it as an electrical atmosphere problem. Nothing tells a warehouse storing pool chemicals to read the other two.

And the regulatory line is sharper than it looks. If Bio-Lab had stored chlorine, it would have been covered by OSHA's process safety standard. It stored something that makes chlorine, so it wasn't. The CSB found the release very likely exceeded PSM threshold quantities for chlorine, hydrogen chloride, and bromine, all of which are covered chemicals, but they arrived as reaction products rather than inventory. No PSM, no RMP, therefore no required process hazard analysis and no required management of change on the day somebody decided the product could start living outside the bunker.

That's the question I keep turning over, and I don't think it has a clean answer. If a regulation covers what you store but not what your storage can produce, whose job is the gap? I'd like to hear where you land on that, because I suspect the honest answer at most sites is nobody's.

Screenshot the next three checks, or save the post. They are built to run in about ten minutes at a toolbox meeting.

1. Go stand under your worst chemical and look up. Not at the product, at the steel above it. Pipe, conduit, hangers, sprinkler heads, roof deck. Ask what the material below is doing to that metal during normal storage, on an ordinary day when nothing is going wrong. Bio-Lab's warehouse was corroding its own fire protection within two months of the product arriving, and the mechanism was just the product sitting there doing what it does.

2. Pull the design basis for your most loaded building and check the hazard class before you check the quantity. Find the original permit or design submittal, find the occupancy classification and the hazard class it was built for, then walk the building with the current inventory report. Quantity is the easy comparison and it is the less important one. Bio-Lab's warehouse was described to the county at 6.2 million pounds and held 14 million, but the deeper problem was that the building had been classified for combustible storage rather than for oxidizers, which made it the wrong building at any tonnage. Nobody smuggled the inventory in. It arrived one approved purchase order at a time, into a building that was never right for it.

3. Decide, in a number, when a degraded safeguard stops the job. This is the one I expect argument on, and it should get argument. At Bio-Lab, 1,124 documented corroded sprinkler heads did not trigger a stop, and neither did a fire protection system logged abnormal on 60 days out of twenty-one months, nor emergency head replacements nine days out. Every one of those was handled as a work order. If you can't say what your number is, then in practice your number is whatever number you happen to survive.

Bio-Lab knew the chemistry, diagnosed the corrosion, and bought the correct hardware. What it never built was anything that noticed the hazard had moved.

Knowing the fix isn't the hard part. Keeping the fix pointed at the hazard while the plant changes underneath it, that's the part nobody writes a procedure for.

Where has a fix at your site quietly stopped covering the thing it was bought for? I read every reply.

Before Human Error is a weekly incident teardown for industrial EHS leaders who know "the operator screwed up" is where a lazy investigation stops, not where a real one starts. This issue is built on the U.S. Chemical Safety Board's final report on the Bio-Lab Conyers incident, released July 21, 2026. Worth noting while you read it: the CSB used this report to reiterate a recommendation it first made to the EPA in 2002, asking for reactive hazards to be covered by federal rule. That recommendation remains formally open, twenty-four years later. Subscribe here on LinkedIn to get each issue. The inbox edition, plus the Process-Safety Teardown Checklist, is linked in the first comment.

/

/

/

/

/

Written in a personal capacity. The views are mine, not my employer's, and nothing here is written on their behalf.