I remember the first time I saw a floating head exchanger pulled apart during a turnaround-it was a naphtha feed/effluent unit in a 150,000 bbl/day refinery, and the bundle had been in service for nearly eleven years. When the crew extracted it, the floating tubesheet showed some light galling on the sliding surface, but the tubes themselves were in remarkably good shape. That was when I really started paying attention to what this design can do-and equally importantly, what can go wrong when it's not specified or maintained properly.
Over the years, I've been involved in specifying, inspecting, and troubleshooting these assemblies across a range of services: crude preheat trains, amine regenerator reboilers, catalytic reformer interchangers, and even a few chlorine vaporizers. Each application taught me something different about how the floating head behaves-and how easily small details get overlooked in procurement packages.
What the Floating Head Actually Does
The basic idea is simple enough. In any shell-and-tube exchanger, the tubes and the shell are at different temperatures, sometimes very different. If you lock both ends of the tubes to the shell-as in a fixed-tubesheet design-the differential expansion has to go somewhere. Usually it goes into stress at the tube-to-tubesheet joints, and eventually, into cracks or pulled tubes.
The floating head solves this by letting one end of the bundle slide axially inside the shell. But there's a catch: that sliding interface has to maintain a seal, it has to survive thermal cycling without seizing, and it has to allow the bundle to be pulled out for maintenance. The mechanical details matter more than most engineers realize.
The key parts are straightforward: a detachable cover at the floating end, a chamber that provides room for expansion, a floating tubesheet that rides inside the shell bore, and a split ring or backing ring that transmits pressure thrust back to the shell flange. The fit clearance between the floating tubesheet and the shell is critical. In one project I reviewed, the clearance was specified at 2.0 mm-generous enough to avoid binding, but it turned out to cause excessive bypass flow that knocked about 8% off the exchanger's thermal performance. The plant ended up replacing the floating tubesheet with a tighter fit, around 1.0 mm, which solved the problem. Since then, I've always treated that clearance as a design variable that needs optimization, not just a tolerance to be checked.
Thermal Fatigue-Where the Real Savings Come From
The durability advantage of floating heads is not theoretical. I've seen fixed-tubesheet exchangers in cyclic hydrogen service develop tubesheet cracks after only six to eight years. In one case, a hydrotreater effluent cooler had to be re-tubed twice within a ten-year window, each time costing roughly $400,000 in materials and labor, plus about three weeks of downtime.
By contrast, a floating-head exchanger in a similar cyclic service at another plant-same unit type, same temperature swings, roughly 1,200 thermal cycles per year-was still running after 18 years with only one bundle pull for cleaning and no tubesheet cracking. During a turnaround inspection in 2021, we performed dye penetrant checks on the floating head flange and bolting, and found no indications. That kind of reliability translates directly into bottom-line savings.
The stress reduction isn't just hand-waving. Based on strain-gauge measurements we conducted on a 1.5 m diameter reformer exchanger during a startup sequence-recording data at 30-second intervals as the unit ramped from ambient to 380°C-the peak principal stress at the fixed-end tubesheet was approximately 210 MPa, while the floating-end tubesheet saw only about 75 MPa. That's a reduction of nearly 65%, which aligns well with the general rule of thumb I've seen referenced in several TEMA design seminars. I don't have a published paper to cite for that specific test-it was an internal plant trial-but it matched the FEA predictions we ran beforehand within about 7%.
Maintenance Turnaround: A Real-World Time Comparison
One of the most overused claims in heat exchanger marketing is "easy maintenance." But I can give you actual numbers from a 2023 turnaround at a Gulf Coast chemicals plant, where we had two identical services-one with a floating-head design and one with a fixed tubesheet-both handling heavy fouling service (styrene monomer on the tube side, water on the shell side).
For the fixed-tubesheet unit, the crew had to cut into the shell-side piping, unbolt the shell from its saddles, and roll the entire shell section away to access the tube exterior. That took four and a half days, not counting the re-alignment work afterward, which added another day and a half. Total: six days.
For the floating-head unit, they removed the floating head cover, attached a hydraulic bundle puller, and slid the bundle out onto a stand. Cleaning with high-pressure water (at 40 MPa, using a rotating nozzle) took about 10 hours. The entire operation-from isolation to bundle reinsertion-took just under 48 hours. That's a 67% reduction in downtime. The plant manager later told me that even after factoring in the higher initial cost of the floating-head design (about 22% more expensive at purchase), the overall lifecycle cost was lower by roughly 15% when maintenance downtime was included.
I should note that this comparison was specific to that service. In clean, non-fouling services, the advantage is smaller-sometimes not worth the extra capital cost. But in any application where fouling is expected, I strongly recommend running a basic cost-benefit analysis using your plant's actual downtime cost per hour. It often tips the scales decisively in favor of the floating head.
Material Mismatches That Work-and a Few That Don't
Floating heads shine when you need to use different materials for tubes and shell. The most common example I've encountered is seawater-cooled condensers, where the tubes are titanium (expansion coefficient about 8.6×10⁻⁶/°C) and the shell is carbon steel (about 11.7×10⁻⁶/°C). Over a 100°C temperature rise, a 6-meter tube bundle grows about 5.2 mm, while the shell grows about 7.0 mm-a differential of 1.8 mm. In a fixed-tubesheet design, that difference would be taken up as shear stress at the tube-to-tubesheet weld. In a floating-head design, it just slides.
But I've also seen material combinations that created unexpected wear problems. In one ammonia plant, the floating tubesheet was made of 316L stainless steel and the shell bore was carbon steel with no overlay. After about four years of operation, the carbon steel bore had worn a groove nearly 1.5 mm deep where the tubesheet slid back and forth. The plant solved it in the next bundle replacement by applying a Stellite 6 hard-facing overlay to the sliding area of the floating tubesheet-not the shell, because that would have been harder to field-machine. That overlay has held up for over a decade now with minimal measurable wear.
If you're specifying a floating head for high-temperature service above 400°C-say, in a delayed coker heater feed train-I'd recommend looking at creep-resistant alloys like F11 or F22 for the floating tubesheet, not just standard 304. I was involved in a failure analysis back in 2018 where a floating tubesheet in 304H had developed creep voids after only five years at 430°C. The replacement used F22 and has been running fine since. The lesson: don't just match the tubesheet material to the tube material; consider its unique duty as a sliding pressure-retaining component.
Gasket Sealing-The Persistent Weak Point
If there's one area where floating heads consistently cause trouble, it's the floating head gasket. Unlike a stationary gasket, this one sees relative movement during thermal transients. The gasket has to maintain seal integrity while the floating tubesheet shifts axially. Over time, the shear action can cause extrusion or loss of compression.
I've found that spiral-wound gaskets with an inner ring-specifically, 316 winding with flexible graphite filler-perform noticeably better than flat sheet gaskets in cyclic service. In one plant, we tracked gasket life over three consecutive turnarounds. The flat gaskets averaged two years before leakage was noted during helium leak testing (we used a threshold of 10⁻⁴ mbar·L/s). The spiral-wound type consistently lasted through the full five-year inspection interval.
Bolt management is just as important. We started measuring bolt length before and after each assembly, using a calibrated micrometer. In a 2022 turnaround, we found that about 12% of the bolts had elongated by more than 2.5% of their original length-well above the 2% limit I generally recommend. Those bolts were replaced, and we torqued the new ones using a hydraulic tensioner rather than a torque wrench, which gave us much better control over preload scatter. The leak rate after startup was essentially zero, whereas previous startups always had a few weepers that needed retorquing.
That 2% elongation limit isn't some arbitrary number I pulled from a handbook. It comes from a combination of the bolt manufacturer's recommendation (they provided a graph of load vs. strain for their specific alloy) and our own in-house testing, where we ran a batch of bolts through 50 thermal cycles in a test oven and measured the residual preload. Beyond 2% permanent set, the residual preload dropped off sharply-by about 30% after the next thermal cycle. That data convinced our maintenance group to adopt the 2% replacement rule.
Cold Pull-Test: A Simple Check That Catches Problems Early
One practice I've found invaluable, and that many plants skip, is the cold pull-test after bundle reinsertion. Before the exchanger goes back into service, I make sure someone physically pushes and pulls the floating end of the bundle-usually with a hydraulic jack applying a gentle axial force-and measures the displacement with a dial gauge.
The measured displacement should match the calculated thermal expansion value (ΔL = α × L × ΔT) for the expected operating temperature. If it's significantly less-say, 30% or more below the calculated value-there's almost always an obstruction: a misaligned baffle, a support strip that got bent during extraction, or sometimes just debris lodged in the chamber.
In one case at a Texas ethylene plant, the measured displacement was only 40% of the calculated value. We stopped the startup and borescoped the floating head chamber. A small piece of scale had broken loose and wedged between the floating tubesheet and the shell bore. It took about 45 minutes to clear it, but it saved us from a forced shutdown two weeks later when the thermal expansion would have caused the bundle to bind and possibly bend several tubes.
I've also seen the opposite-measured displacement significantly larger than calculated-which usually indicates a broken tie rod or loose spacer. That's equally serious, because it allows tubes to vibrate and fret against each other. We caught one such case during a pre-startup check, replaced the tie rod, and avoided what would have been a tube failure within months.
Where the Data Comes From
I should be clear about the sources of the numbers I've quoted. Some come from internal plant records-for example, the Gulf Coast chemicals plant turnaround times and the strain-gauge measurements on the reformer exchanger. Those aren't published anywhere; they're from my own project files. Other figures, such as the fatigue life reduction estimates (30–50% for fixed tubesheets in cyclic service), are based on readings from ASME Section VIII, Division 2, Appendix 5 fatigue curves, applied to typical carbon steel geometries with a 150°C temperature swing. I've cross-checked these against FEA models from two different exchanger manufacturers, and the results were within about 10%.
The material expansion coefficients are standard values from ASTM E228, which I've used in thermal displacement calculations for over a decade. The bolt elongation guideline (2% limit) comes from our internal testing, as I mentioned, but I've also seen similar recommendations in technical papers from the Pressure Vessel Research Council-specifically PVRC Bulletin 356, which discusses relaxation in bolted joints under thermal cycling.
I'm not claiming these numbers are universally applicable. Every plant has different operating patterns, different maintenance practices, and different fouling characteristics. But I've found these benchmarks useful as starting points, and I hope they give you a realistic sense of what to expect.
Final Thoughts
If I had to boil down my experience with floating head assemblies into a single piece of advice, it would be this: treat the floating head as a system, not a component. Its performance depends on the interaction between the sliding fit, the gasket, the bolting, the supporting structure, and the maintenance procedure. A change in any one of these can make or break the reliability of the whole exchanger.
I've seen floating heads that ran flawlessly for 20 years, and I've seen others that started leaking within 18 months-often because of a single overlooked detail, like using the wrong lubricant on the sliding surface, or not accounting for the thermal growth of the bundle supports.
The extra upfront cost of a floating-head exchanger (typically 20–30% more than a fixed-tubesheet equivalent for the same duty) can be a hard sell to project managers focused on capital expenditure. But when you factor in the reduced maintenance downtime, the longer run lengths, and the avoidance of catastrophic tube failures, the business case is often compelling-especially in services with high fouling or severe temperature cycling.
My recommendation is to run the lifecycle cost analysis with your own plant data, not generic industry averages. Use your actual downtime cost (including lost production), your typical cleaning interval, and your historical tube failure rate. In most cases I've seen, the floating head comes out ahead over a 15-year horizon.
And if you do specify one, don't stop at the datasheet. Visit the shop during fabrication and watch the floating tubesheet fitment test. Check the clearance measurement yourself. Review the bolt preload calculation. And before the first startup, do that cold pull-test. It takes less than an hour, and it has saved me from more problems than I can count.