Got the call at 10:47 on a Tuesday. Baisheng’s production manager — a guy named Chen who I’d worked with once before on a pump seal issue — sounded more confused than panicked. Their reactor temperature had been swinging for three days. Not the usual drift. Wild swings. -15°C to -28°C in under two minutes, then back up. The operators were convinced the TCU controller was bad. They’d already swapped it out once. No change.

I flew down to Shenzhen the next morning. The plant is in one of those industrial parks out near the airport, the kind where every building looks identical and the taxi driver circles twice before admitting he’s lost. Chen met me at the gate and walked me to the reactor bay. They were running LM-11D, water-free, rated to -60°C. Good fluid. We’ve sold hundreds of tons of it.

I spent the first two hours checking everything except the coolant. TCU calibration. Valve positions. Pump curves. Expansion tank level. All normal. The operators were watching me like I was supposed to pull a rabbit out of a hat. I remember Chen leaning against the control panel, arms crossed, saying, “It’s got to be the fluid, right? What else is left?”

He was right. But I didn’t want to believe it. LM-11D doesn’t freeze. That’s the whole point.

I pulled a sample from the low point of the loop. The liquid looked fine. Clear, the right shade of pale yellow. Smelled normal — that faint glycol-like sweetness that LM-11D has. But when I held the sample bottle up to the light and tilted it, I saw something. A faint haze. Not sediment. More like a shimmer. The kind of thing you see when there’s a tiny amount of immiscible water suspended in a non-aqueous fluid.

Water content read 0.3% by Karl Fischer. That’s 3,000 ppm. The spec for LM-11D is under 500 ppm in service. Someone had either contaminated the system during top-up or the original fill had been sitting too long with the expansion tank vent open to humid Shenzhen air. June in Shenzhen, the humidity is brutal. We’re talking 90% RH day after day. An open vent on an expansion tank is basically a dehumidifier running in reverse — pulling moisture out of the air and condensing it into your coolant.

The ice blockage mechanism in LM-11D is weird, by the way. It’s not like water-based fluids where you get a solid plug. With water-free fluids, the water doesn’t dissolve. It forms micro-droplets. At low temperatures, those droplets freeze into tiny ice crystals — maybe 5 to 20 microns. They don’t block the pipe outright. They aggregate at restrictions: valve seats, heat exchanger inlets, pump suction strainers. Then they melt. Then they re-freeze. The result is exactly what Chen was seeing: temperature swings that look like a control problem.

We drained the system. Flushed it with dry nitrogen for six hours. Recharged with fresh LM-11D. The temperature stabilized within 45 minutes. Chen offered to buy me dinner. I was too tired. Fell asleep in the hotel with my boots still on.

This one still bothers me.

Guangzhou Hengxing called in mid-June. Their LM-11D system was showing intermittent flow restrictions. Same symptoms as Baisheng — temperature oscillations, reduced heat transfer, occasional pump cavitation. The local distributor had already sent a tech out. He pulled a sample. Water content: 47 ppm. That’s practically nothing. Well within spec. The distributor told them the fluid was fine and the problem must be mechanical.

It wasn’t mechanical.

I got involved because the distributor was stuck. Two pump replacements later, the problem was unchanged. The plant manager — a woman named Liang who had zero patience for consultants telling her to “monitor the situation” — demanded someone from Glacier fly down.

I was in the middle of another job in Dongguan. Drove over. The system was a 500-liter loop feeding four jacketed reactors. All LM-11D. The fluid had been in service for about 11 months. The water content was genuinely low — 47 ppm by KF titration. I ran it twice to be sure. Same result.

So I started looking at other things. Viscosity at low temperature. Normal. Particulate count. Normal. Acidity. Slightly elevated, but nothing that would explain the symptoms. I was about to give up and recommend a full system teardown when I noticed something on the sample bottle. A thin film on the glass. Oily. Slightly iridescent under the lab light.

It wasn’t water. It was a low-molecular-weight fraction of the coolant that had partially degraded. The system had a hot spot — a section of piping that ran too close to a steam line. The localized temperature was hitting close to 180°C, well above the fluid’s thermal stability limit. The fluid was thermally cracking. The degradation products had a higher freezing point than the base fluid. At -40°C, they were precipitating out as a waxy solid.

Not ice blockage. But it behaved exactly like ice blockage. The pump strainer had a coating of this waxy deposit that looked like frost. I scraped some off with a knife and it melted in my palm at about 15°C. Definitely not ice. Definitely not water.

We rerouted the piping away from the steam line. Insulated the hot section. The problem disappeared. But the experience shook me. I’d been so focused on water contamination that I almost missed a completely different failure mode. Thermal degradation masquerading as ice blockage. If I hadn’t noticed that film on the glass, I’d still be chasing my tail.

This case started with a WeChat message at 6:14 a.m. A photo of a pressure gauge pegged at zero. The caption, translated from Mongolian: “It won’t move.”

The customer was a chemical plant in Inner Mongolia. They’d commissioned a new reactor line in April. LM-11D coolant, 800-liter loop. The system had been running fine for two months, then suddenly — nothing. The pump was running. You could hear it. But the flow meter showed zero. The reactor jacket wasn’t getting any cooling. Production was down.

I flew to Hohhot. Then a four-hour drive through terrain that made me understand why people describe Inner Mongolia as “emptiness with grass.” The plant was isolated. Nearest town was 90 kilometers away. If they needed a spare part, it was a two-day wait.

The system looked fine on paper. Pump discharge pressure was normal. Suction pressure was normal. But the flow meter at the reactor inlet? Zero. I checked the meter. It wasn’t broken. There was genuinely no flow reaching the reactor.

I traced the piping. Pump to heat exchanger. Heat exchanger to manifold. Manifold to reactor. At the manifold, I found it. A ball valve that had been installed backwards. The valve seat was acting as a check valve in the wrong direction. At full flow, the pressure drop across the seat was enough to cause localized flashing. The LM-11D wasn’t boiling — it was cavitating. The vapor bubbles were collapsing downstream, creating a vapor lock that looked exactly like an ice blockage.

But here’s the part that matters: the cavitation was causing localized cooling through the Joule-Thomson effect. The fluid temperature at the valve seat was dropping 15-20°C below the bulk temperature. Enough to push it below the cloud point of the fluid. The dissolved components were precipitating out, forming a gel-like deposit on the valve seat. That deposit was what actually stopped the flow.

So the sequence was: backwards valve → cavitation → localized cooling → component precipitation → gel deposit → blocked flow. Five steps. Any one of them would have been obvious. All five together? It took me three days to figure out.

We flipped the valve. Flushed the manifold. The system started right up. The plant manager — a man of few words — shook my hand and said, “Next time, come faster.” I think that was a compliment.

Chongqing in September is like working inside a dishwasher. Hot, wet, and the air feels thick enough to chew. I was there for a customer who’d been running LM-11D in a pharmaceutical intermediate process. They’d installed a molecular sieve dryer on the coolant loop — a belt-and-suspenders approach to water management. The idea was sound: continuously circulate a side stream of coolant through the molecular sieve bed to scavenge any moisture that crept in.

The problem was that the molecular sieve was also scavenging something else.

The system had been running for about six months. Everything was stable. Then, over the course of about two weeks, the low-temperature performance started degrading. The system could reach -50°C, but the cooling capacity at that temperature was maybe 60% of what it should have been. The operators compensated by increasing flow rate. That helped for a few days. Then the flow rate started dropping too.

When I arrived, the system was barely holding -42°C. The molecular sieve bed had been replaced twice. The operators assumed the sieve was saturated with water and needed more frequent changeouts. But the water content in the fluid was only 85 ppm. The sieve should have been fine.

I took a sample of the spent molecular sieve and sent it to our lab in Shanghai. The results came back two days later — the sieve wasn’t adsorbing water. It was adsorbing the corrosion inhibitor package from the LM-11D. The inhibitor molecules were small enough to fit into the sieve pores. Over time, the sieve had stripped about 40% of the inhibitor out of the system.

Without the inhibitor, the fluid was still chemically stable. But the inhibitor wasn’t just for corrosion — it also functioned as a pour point depressant. With the inhibitor concentration reduced, the fluid’s low-temperature flow properties degraded. The viscosity at -50°C had increased by a factor of nearly three. The pump couldn’t push it through the heat exchanger at the design flow rate. The system looked like it was freezing up. It wasn’t. It was just too thick to pump.

We bypassed the molecular sieve. Re-inhibited the fluid. The system recovered within 48 hours. The customer was furious — not at us, but at the engineering firm that had specified the molecular sieve without checking compatibility. I spent an hour on the phone with the firm’s technical director, who kept saying “but the datasheet says it’s compatible with glycols.” LM-11D isn’t a glycol. It’s a water-free synthetic fluid. The datasheet doesn’t say it’s compatible with everything. It says what it is compatible with. There’s a difference.

This was the case that made me write this diary.

Shanghai Nuoleng is a contract manufacturer. They run multiple reactor lines with different coolants — some LM-4, some LM-11D, some LM-8. Their LM-11D system was a 1,200-liter loop running at -55°C. The water content was 104 ppm. Very low. The fluid should have been in perfect condition.

But the system was freezing. Not partially. Not intermittently. Completely. The entire loop had turned into a solid block.

I got there on a Wednesday evening. The plant was shut down. The maintenance team had already tried thawing the system with external heaters. They’d gotten the piping warm enough to flow, but as soon as they turned the heaters off, it froze again. They’d been through this cycle three times and were starting to damage the pipe insulation.

The water content was 104 ppm. That’s 0.0104%. At that concentration, the freezing point depression should be negligible. The fluid should still flow at -60°C. So why was it freezing?

I pulled a sample from the low point. Then another from the high point. Then one from the pump discharge. The water content was consistent — 104, 107, 102 ppm. No stratification. No localized pockets of high moisture. The fluid was homogenous.

I put the sample in the freezer at -55°C. The whole thing turned solid.

That’s when it hit me. I wasn’t dealing with LM-11D. I was dealing with LM-11D that had been cross-contaminated with something else. The plant used LM-4 in another loop. LM-4 is water-based. If someone had accidentally topped up the LM-11D system with LM-4 — even a small amount — the water in the LM-4 would have been enough to push the mixture past its cold tolerance.

I asked the maintenance supervisor to check the topping-up logs. He pulled out a clipboard. Sure enough, three weeks earlier, a night-shift operator had added 40 liters of “coolant” to the LM-11D expansion tank. The drum was labeled in Chinese. The operator had grabbed an LM-4 drum by mistake. Same color label. Same size drum. Same storage rack.

40 liters of LM-4 into 1,200 liters of LM-11D. That’s about 3.2% contamination. LM-4 contains roughly 40% water. So we’d introduced about 16 liters of water into the system. The water content should have been through the roof. But here’s the thing: the water from the LM-4 had partially phase-separated. It wasn’t showing up in the KF titration because the water was trapped in a separate phase that the sample didn’t capture. The 104 ppm reading was just the dissolved water. The free water — the stuff that actually froze — was hiding in the low points of the loop, pooled in the expansion tank, and coating the inside of the heat exchanger tubes.

We drained the entire system. Pressure-washed the heat exchanger. Dried everything with nitrogen for 12 hours. Recharged with fresh LM-11D. The system came back up and ran perfectly. The plant manager implemented a color-coded drum labeling system and a two-person verification for all coolant top-ups. Twenty thousand yuan in lost production, all because of a label that two people read differently at 3 a.m.