Saddles, Skirts and Differential Expansion — One Holds, One Lets Go

The exchanger that tore its own foundation

A 1.2 m diameter crude preheat exchanger, 6 m between saddle centres, runs at 220 °C. It was installed on a cold Sunday by a crew who were told to "bolt it down properly", and they did — both saddles, all eight M24 bolts, torqued to the same value off the same chart.

Eighteen months later the routine inspection finds a hairline crack in the grout at the east pier, two anchor bolts bent like hairpins, paint flaked off the shell at the saddle horn, and one slotted hole visibly elongated into a bright steel smear.

Guess before you read on: the vessel only moved about 15 mm. What size of force does 15 mm of blocked movement generate at that saddle — same order as the 15 kN of friction the design allowed for, or something else?

878 kN. Fifty-nine times the design number. Not because anything was overloaded in the usual sense, but because a displacement that had nowhere to go got converted into a force by the stiffness of a very good concrete pier.

The misconception, stated plainly: "Both saddles just hold the vessel up. Bolting both of them tight is the safe choice."

It sounds like belt and braces. It is the opposite. A horizontal vessel that changes temperature must change length, and the only question a designer gets to answer is where the growth is allowed to happen. Deny it and the growth does not disappear — it becomes force, and force goes looking for the weakest thing in the chain: the anchor bolts, the base plate, the grout, the shell at the saddle horn.

The correct arrangement is asymmetric on purpose. One saddle is fixed — round holes, bolts torqued, this is the datum from which everything is measured. The other slides — slotted holes, a low-friction plate, nuts snugged then backed off and locked so they retain without clamping. That asymmetry is not a detail left to the fabricator. It is the load path.

Symbol key — every symbol on this sheet

Read the subscripts as words: f = friction, s = seized, slot = the hole, sad = the saddle. The one letter that fights you is μ — here it is only the friction coefficient, never ductility.

• L — distance from the fixed saddle to the sliding saddle · m, mm
• ΔT — metal temperature minus installation temperature (signed: negative in cold service) · °C
• α — mean coefficient of thermal expansion over that range · µm/m·°C (i.e. ×10⁻⁶ /°C)
• ΔL — growth at the sliding end = α·L·ΔT · mm
• μ — friction coefficient of the sliding interface · dimensionless
• W — vertical load carried by the sliding saddle (shell + contents + insulation) · kN
• F_f — friction force when the slide works = μ·W · kN
• k — combined lateral stiffness of saddle + base plate + pier + soil · kN/mm
• F_s — force when the slide seizes ≈ k·ΔL, ceiling E·A·α·ΔT · kN
• A — shell cross-sectional metal area = π·(D − t)·t · mm²
• E — Young's modulus at temperature (≈ 190 GPa for steel hot, not 200) · MPa
• L_slot — slotted-hole length · mm
• A_zick — distance from the head tangent line to the saddle centreline · mm
• θ — saddle included (contact) angle, typically 120°–150° · degrees
• H — skirt height, tangent line to base ring · m

Fixed, sliding, and the arithmetic of 15 millimetres

Everything starts at the fixed saddle, because that is the origin of the coordinate system the vessel actually uses:

ΔL      = α · L · ΔT                      growth at the sliding end
F_f     = μ · W                           price of letting it happen
F_s     ≈ k · ΔL   (ceiling E·A·α·ΔT)     price of refusing
L_slot  ≥ 1.25·|ΔL| + d_bolt + erection tolerance

Three consequences fall straight out and are worth holding on to:

• Growth is measured from the fixed point, not from the middle. Put the fixed saddle at the east end and the west end moves the full ΔL while the east end moves nothing. Move the fixed saddle to the middle of a three-support arrangement and both ends move ΔL/2 — which is sometimes exactly what you want, if the piping is symmetrical.
• The piping group needs three displacements at every nozzle, not one: axial growth from the fixed saddle, radial growth of the shell (α·R·ΔT — small, but it lifts top nozzles), and vertical growth of the support itself. Hand them the fixed-point location, not just "the vessel gets hot".
• Friction is reacted at the fixed saddle. The μ·W that appears at the sliding end does not vanish into the ground there; the vessel is one rigid body, so the fixed saddle's anchor bolts see it too, plus wind and seismic. Size them for the sum, in both directions — the vessel cools down as well as heating up.

Play with the numbers: vessel length, ΔT, material, μ, weight and pier stiffness, with the force–travel plot showing the friction plateau against the seized ramp: ▶ open the interactive: static saddles expansion calc

What makes a slide actually slide. Four details, all of which get lost between the drawing and the field:

• Slotted holes in the base plate of the sliding saddle, running along the vessel axis, long enough for the growth plus the bolt diameter plus erection tolerance. Sized for growth alone, the slot is already too short before the vessel is hot.
• A low-friction plate: PTFE on stainless (μ ≈ 0.06–0.15), graphite or bronze (≈ 0.15–0.25). Bare steel on steel is 0.3–0.4 when clean and 0.6 or worse once it rusts, and rusty steel on concrete is worse again. Choosing "no slide plate" multiplies the design friction force by 3.5 (0.35 vs 0.10) and then degrades toward seizure over the plant's life.
• Nuts that retain but do not clamp. Snug, back off an eighth to a quarter turn, then lock with a jam nut or a tack. Use an oversize or spherical washer. A torqued nut over a slotted hole is a fixed saddle with extra steps — and it is the single most common way a sliding end dies.
• Nothing else holding it: no dowel, no grout bund around the base plate, no conduit or drain clamped hard between the vessel and the structure, no fireproofing poured across the slot. Site loves to fill a gap.

Watch the difference: heat the vessel and see the sliding end walk while the fixed end stays put, then toggle Seized slide and watch the force path light up from shell to horn to base plate to bolts to pier: ▶ open the interactive: static saddles expansion 3d

Where the load actually goes — Zick, horns, and skirts

Letting the vessel slide solves the longitudinal problem. It does not solve the local one, because a saddle is a hard point pressed into a thin shell — the same "a shell hates local loads" story as a nozzle, in a different costume.

Zick's logic (L. P. Zick, Welding Journal, 1951; now embedded in ASME VIII-2 Part 4.15 and PD 5500 Annex G) is the industry's answer, and its reasoning is worth knowing even when software does the arithmetic:

• Two saddles, never three. Two supports on a beam are statically determinate: the reactions are known whatever the foundation does. A third saddle makes the reactions depend on settlement you cannot predict, and the "extra support" can end up carrying almost everything — or nothing. Zick is a two-saddle method by deliberate choice.
• Saddle angle θ ≥ 120°. The shell is carried by tangential shear spreading up the saddle face. Too narrow an angle concentrates it; too wide and the saddle starts to fight the shell's own ovalisation. 120°–150° is the practical window.
• Put the saddle near the head, or near the quarter point. With A ≤ R/2 the dished head acts as a stiffening ring and carries the shell through. Further out (A up to ~0.2 L) the shell is on its own and you may need a stiffening ring — or a wear plate that does the job locally.
• The horn is the hot spot. Circumferential bending peaks at the saddle horn, the top edge of the saddle contact. That is what the wear plate exists for: it must be wider than the saddle by roughly 1.56√(R·t) in total (≈ 0.78√(R·t) each side) and extend above the horn, or it simply relocates the peak to its own edge. (Same √(R·t) boundary layer as the nozzle — the shell only has one yardstick.)
• Zick's four stresses are longitudinal bending at midspan and over the saddles, tangential shear, circumferential stress at the horn, and ring compression at the bottom. It is the horn one that surprises people, and the one a seized slide makes worse: the friction force arrives as a longitudinal shear dragged through that same contact patch.

Vertical vessels take the other road entirely. A tall column sits on a skirt — a cylinder welded to the bottom head, carrying the vessel down to a base ring, anchor bolts and foundation. The skirt solves differential expansion by sharing it: the skirt is at roughly vessel temperature, grows with the vessel, and there is no relative movement to accommodate. There is no sliding end because there is no second support. Legs and lugs, by contrast, reintroduce exactly the problem saddles have — different metal at a different temperature bolted to a fixed structure.

The skirt's load path is a couple, not a shear:

wind / seismic overturning moment  →  skirt shell (tension one side, compression the other)
                                   →  base ring bearing + anchor bolts in tension
                                   →  foundation and soil
weight                             →  skirt shell in compression  →  base ring  →  concrete
thermal growth                     →  the whole column rises by α·H·ΔT — a nozzle displacement

Three skirt details that cost money when missed: the skirt-to-head weld is the thermal and fatigue hot spot (a steep temperature gradient sits right at a structural discontinuity), which is why hot columns get a hot box — an insulated annulus that stretches the gradient out over a length of skirt instead of concentrating it at the weld. The access opening and vent holes remove material exactly where the overturning couple is highest, so they get reinforced and kept off the neutral-axis-perpendicular. And the fireproofing and its stand-off must not bridge the base ring into a rigid, water-trapping detail.

Worked example — crude preheat exchanger

D = 1200 mm OD, t = 14 mm, carbon steel A516 (α = 12.2 µm/m·°C), L = 6.0 m between saddle centres, operating metal temperature 220 °C above an installation temperature of 20 °C so ΔT = +200 °C. PTFE slide plate, μ = 0.10. Sliding saddle carries W = 150 kN. Saddle + pier lateral stiffness k = 60 kN/mm. Anchor bolts M24.

ΔL      = 12.2e−6 × 6000 × 200                       = 14.6 mm
L_slot  ≥ 1.25 × 14.6 + 24 + 10                      = 52 mm      (so: 55 mm slot, M24)
F_f     = 0.10 × 150                                 = 15 kN      slide working
A       = π × (1200 − 14) × 14                       = 52 162 mm²
ceiling = E·A·α·ΔT = 190 000 × 52 162 × 12.2e−6 × 200 = 24 182 kN  (full restraint — never reached)
F_s     = k·ΔL = 60 × 14.6                           = 878 kN     slide seized
ratio   = 878 / 15                                   = ×58.6
σ_long  = 878 000 / 52 162                           = 16.8 MPa   in the shell — the shell survives

Read the last two lines together, because they are the whole story. The shell shrugs: 17 MPa of longitudinal membrane is nothing. What does not shrug is everything the 878 kN passes through — four M24 bolts in shear, a 34 mm base plate in bending, a grouted pier designed for 15 kN horizontal, and the saddle horn already carrying its Zick circumferential stress. The failure is never in the vessel. It is always in the connection.

Swap the PTFE for bare steel-on-steel (μ = 0.35) and the working force becomes 53 kN — still survivable, but now 3.5× what the foundation drawing says, and on a trajectory toward seizure as it rusts. That is the quiet version of this failure, and it is far more common than the dramatic one.

Common pitfalls

• Bolting both saddles tight. The headline error. Redundancy in a support is not redundancy in a load path.
• Slot length sized for growth only. The slot must cover growth + bolt diameter + erection tolerance, and both directions if the vessel is ever steamed out or runs cold.
• Nuts torqued on a slotted hole. Slots plus clamping equals no slide. Snug, back off, lock.
• No slide plate. Steel on steel starts at μ ≈ 0.35 and rusts to 0.6+. The "saving" is a few hundred rupees of PTFE against a foundation designed for a quarter of the real load.
• Forgetting the friction is reacted at the fixed saddle, in both directions, on top of wind and seismic.
• Not giving the piping group the fixed-point location. Nozzle displacements are measured from the fixed saddle; assuming the vessel grows symmetrically about its centre halves the number at one end and doubles it at the other.
• Adding a third saddle "for support". Indeterminate reactions, no reliable load share, and outside Zick's basis.
• Saddle angle below 120°, or a wear plate the same width as the saddle. The peak just moves to the wear plate edge — extend it by ~1.56√(R·t) and above the horn.
• Treating cold service as the easy case. ΔT is negative, the vessel shrinks toward the fixed saddle, the slot must allow it, and the contraction of an austenitic (α ≈ 17) is 40 % larger than carbon steel for the same ΔT.
• On skirts: forgetting the column rises. α·H·ΔT at the top of a 40 m column at 250 °C is over 100 mm of vertical nozzle movement. Also: no hot box on a hot skirt, and reinforcement missed at the access opening.
• Site fills the gap. Fireproofing, grout, a cable tray bracket or a drain clamp across the sliding end will seize it just as effectively as rust. Walk the sliding saddle at commissioning.

Outcome

• A horizontal vessel gets one fixed saddle and one sliding saddle — round holes vs slotted holes, clamped vs retained. The asymmetry is the design; both-tight is a fault, not caution.
• Growth is ΔL = α·L·ΔT measured from the fixed point. A 6 m carbon-steel exchanger at ΔT = 200 °C grows 14.6 mm and needs a ~52 mm slot for M24.
• A working slide costs μ·W — a flat force, independent of travel, ~15 kN with PTFE here. A seized slide costs k·ΔL — 878 kN, ×59 — and a stiffer pier makes it worse. The theoretical ceiling E·A·α·ΔT (24 000 kN) is never reached because bolts, grout and base plate fail first.
• The shell barely notices the longitudinal force (17 MPa); the connection — anchor bolts, base plate, grout and the Zick horn stress — is what fails.
• Zick's logic: two saddles only (determinate), θ ≥ 120°, saddle near the head or within ~0.2 L, wear plate extended ~1.56√(R·t) past the saddle and above the horn, where circumferential bending peaks.
• Skirts avoid differential expansion by growing with the vessel; their job is to convert wind and seismic overturning into a bolt-tension/bearing couple. Watch the skirt-to-head weld, the hot box, and the access-opening reinforcement — and remember the whole column rises α·H·ΔT.
• Interactive: ▶ open the interactive: static saddles expansion calc — growth, slot, friction vs seized force, and the force–travel plot with a warning band.
• 3D: ▶ open the interactive: static saddles expansion 3d — heat it, watch one end walk, seize it and follow the load path into the foundation.

Open items

• Add the Zick stress set (S1–S4) to the calculator so the saddle-horn check sits next to the friction check instead of being described only in words
• Add a skirt mode: wind/seismic base moment → anchor-bolt tension and base-ring bearing
• Field-measured friction coefficients for aged PTFE and graphite plates, to replace the textbook range with something defensible after ten years in service
• A slotted-hole detail sheet (washer, jam nut, keeper bar, travel witness mark) for the construction handover pack

Educational teaching material — simplified and illustrative. Not for engineering design use.
© EPiC Circle · provided by PipeAxis