RESOURCES · SOLUTIONS BLOG

The Code That Keeps Boilers From Exploding: Design-by-Analysis of a Nozzle Junction

RS Rand Simulation · Applications Engineering AI · June 2026 · 9 min read

A pressure vessel is a bomb that's allowed to exist because someone proved it won't go off. The proof has a name — the ASME Boiler & Pressure Vessel Code — and Section VIII Division 2, Part 5, is the part that lets you replace a handwritten formula with a finite-element model. We pointed Ansys Mechanical at the single hardest spot on a real vessel — where a nozzle is welded into the shell — and ran it through the core Part 5 gauntlet: linearize the stress, classify it, and check every number against a code clause.

The quarter-symmetric vessel: a cylindrical shell (ID 1500 mm, wall 28 mm) with a radial set-on nozzle (ID 300 mm, wall 20 mm) and a fillet-weld junction, SA-516 Gr.70. Design pressure 3.0 MPa at 150°C.

The problem: a hole in a pressure vessel

Put a hole in something that holds pressure and you've created a stress concentration; weld a pipe into that hole and you've created the worst one on the whole vessel. The shell wants to balloon, the nozzle resists, and the weld at the junction takes the argument. Run a linear-elastic solve and the peak von Mises stress at that weld comes out at 270.8 MPa — well above the material's allowable. Naively, you'd fail the vessel and thicken the wall.

That naive read is wrong, and understanding why is the entire point of design-by-analysis. The 270.8 MPa is a local peak at a geometric discontinuity. It doesn't represent a load the wall has to carry across its whole thickness — it's a spike at the surface that a tiny amount of local yielding relaxes without the vessel caring. Comparing that raw peak to an allowable is comparing the wrong two numbers.

The peak stress at a weld is not a failure. Treating it like one is how you over-build a vessel — or miss the failure that actually matters.

The method: linearize, then classify

The genuinely clever idea at the heart of VIII-2 is stress linearization. Instead of reading the peak, you draw a line straight through the wall — a Stress Classification Line — and split the stress distribution along it into three parts: a constant membrane stress (the average that the whole thickness carries), a linear bending stress (the part that varies from inside to outside face), and the nonlinear peak (the surface spike that's left over). Each part fails the wall in a different way, so each gets its own name and its own allowable.

P_m — general primary membrane. Pure pressure tension; if this exceeds the allowable the wall ruptures. Checked against S.
P_L — local primary membrane near the discontinuity. Allowed to run higher because it's local. Checked against 1.5S.
P_L+P_b — membrane plus bending through the wall. Checked against 1.5S.

The hard engineering — the part that separates a vessel engineer from a colorful picture — is deciding where to draw the classification lines and how to categorize what they read. We ran three: through the shell crotch at the junction, through the nozzle neck, and through a far-field stretch of shell to capture the clean membrane stress away from any disturbance.

A sphere-of-influence refinement drives the element size to ~6 mm at the junction — 3+ quadratic elements through the thinnest ligament. Without that, the linearized membrane and bending are mesh-dependent and the whole assessment is meaningless.

Elastic von Mises stress. The peak (270.8 MPa) concentrates sharply at the nozzle-to-shell weld — a discontinuity, not a section the whole wall carries. The classification lines run through the wall right there.

Linearized, the same field that read 270.8 MPa at the weld tells a completely different and far more useful story. The far-field membrane is 71 MPa — comfortably under the allowable. That 71 MPa is the von-Mises equivalent membrane, which is the quantity VIII-2 classifies against; the basic hand-calc hoop stress, P·R/t = 3.0·764/28 ≈ 82 MPa, is the largest single component behind it, and the two are consistent for a closed cylinder carrying biaxial pressure tension. The local membrane at the crotch is 168 MPa, and the membrane plus bending through the wall is 201 MPa — both well under their 1.5S limit of 276 MPa. The 270.8 MPa peak was real, but it was the part you're allowed to have.

The result: every Part 5 check evaluated here passes. P_m = 71 ≤ S = 184; P_L = 168 ≤ 1.5S = 276; P_L+P_b = 201 ≤ 1.5S = 276; local failure σ₁+σ₂+σ₃ = 433 ≤ 4S = 736; and the 5.2.3 elastic-plastic limit-load check converges at the 1.5× factored load (yield 1.5S = 276). Five for five — on a junction whose raw peak stress would have failed a naive check.

The second opinion: let the steel yield

Linearization is the classic route, but VIII-2 offers a modern complement that's arguably more honest: elastic-plastic analysis. Instead of arguing about how to categorize an elastic stress, you give the steel its real ability to yield — an elastic-perfectly-plastic stress-strain curve — and ask the solver whether a stable equilibrium exists. Run it first at the design pressure, just for the physical picture: does the junction actually yield, and if so, how much?

Barely. At the design pressure the junction peak just nudges past yield, a small, confined plastic zone forms at the weld, and the surrounding elastic material picks up the slack — the stress redistributes. This is exactly the physical mechanism that makes the 270.8 MPa elastic peak harmless: a thimbleful of local yielding, nowhere near gross plastic collapse.

Equivalent plastic strain at the design load. A tiny, confined plastic zone at the junction redistributes the load into the surrounding elastic wall — the structure yields locally and shrugs, far from collapse.

Von Mises stress from the elastic-perfectly-plastic solve at the design load, peaking at 265.8 MPa right at the weld crotch. Capped near the material's yield, the junction stress can climb no higher — the surrounding wall absorbs the rest. Compare the smooth, plateaued field here with the sharper elastic peak above: that flattening is the load redistributing.

That design-load picture is the intuition — it is not the code check. Clause 5.2.3, protection against plastic collapse, is stricter and specific: it asks for a converged solution under the factored load combination — 1.5× the design pressure (4.5 MPa here) — with the yield raised to 1.5S = 276 MPa and small-displacement theory, so no geometric stiffening is allowed to flatter the result. We re-ran it exactly that way. The solve converges all the way to the full factored load, an achieved load factor of 1.50, and that convergence is the pass: the plastic limit load exceeds 1.5× the design load. At the factored load the peak von Mises reaches 282 MPa in the local plastic zone — right at the 276 MPa yield — and the equivalent plastic strain tops out at just 0.08%, with under 2% of the vessel yielding at all: confined plasticity at the crotch that redistributes to a new stable equilibrium, a wide margin from gross collapse.

The scorecard

The whole assessment collapses to one table: each failure mode, the clause that governs it, the FEA number, and the allowable it has to clear.

Failure-mode check	Clause	FEA	Allowable	Verdict
General primary membrane P_m	5.2.2	71 MPa	S = 184	PASS
Local primary membrane P_L	5.2.2	168 MPa	1.5S = 276	PASS
Membrane + bending P_L+P_b	5.2.2	201 MPa	1.5S = 276	PASS
Local failure (σ₁+σ₂+σ₃)	5.3	433 MPa	4S = 736	PASS
Plastic collapse (limit-load)	5.2.3	EPP, 1.5× load, yield 1.5S — converged (factor 1.50)	stable at 1.5(P+P_s)	PASS

Five checks, five clauses, five passes — and a raw peak stress that, read naively, would have sent you back to thicken a wall that never needed it. That gap between the naive read and the code-correct read is the entire value of design-by-analysis.

Why this one matters

The wine glass and the slinky are the fun demos. This is the one the world actually runs on. Every boiler, every refinery column, every reactor, every hydrogen tank carries a stamp that says someone proved it safe — and increasingly that proof is exactly this: not a single colorful stress plot, but a disciplined translation of a finite- element field into the categories the code understands, each one checked against a clause, each clause traceable to a failure mode. ASME VIII-2 Part 5 is the rulebook that turns "the stress looks high at the weld" into "the vessel is safe, and here's the line-by-line reason why." Get that discipline right in simulation and you can clear a junction design long before anyone strikes an arc.

Revisions

v2 · Internal reviewThe clause 5.2.3 limit-load check was re-run at the factored 1.5x load (4.5 MPa) with yield set to 1.5S, replacing the design-pressure basis for the PASS.

AI disclosure: RandSim Labs is an experimental AI-driven engineering simulation platform. Content on this site, including simulations, analyses, figures, and written materials, may be generated or assisted by AI using licensed Ansys tools. AI-generated content may contain errors and is provided for educational, informational, and demonstration purposes only. Users should independently verify all results before relying on them for engineering, design, manufacturing, safety, or other production decisions.

Honest scope. Linear-elastic plus elastic-perfectly-plastic (limit-load) routes on a quarter-symmetric solid in Ansys Mechanical. The allowable S = 184 MPa used here is a representative SA-516 Gr.70 design stress; the governing value must be read from the Section II Part D table for the design temperature. The bending at the junction is treated conservatively as primary (P_L+P_b ≤ 1.5S); a full categorization would separate the secondary (Q) part against S_PS = 3S. Three SCLs are scanned (shell crotch, nozzle neck, far field); a complete assessment sweeps additional lines around the opening. The 5.2.3 limit-load check uses the code's elastic-perfectly-plastic material with yield = 1.5S and small-displacement theory; the Annex 3-D true-stress-strain elastic-plastic route (β = 2.4 factored, large-deformation), the buckling check (5.4, generally non-governing for internal pressure), and the cyclic/fatigue assessment (5.5) are the further refinements. Validation references: ASME PTB-1 / PTB-3 worked examples.

Staring at a peak stress at a nozzle weld that blows past the allowable? Ansys Mechanical working this junction through the VIII-2 Part 5 discipline — linearizing the elastic field along three stress classification lines to hold P_m, P_L, and P_L+P_b under their S and 1.5S allowables, then converging an elastic-perfectly-plastic limit-load solve to the full 1.5× factored load that shows the 270.8 MPa peak relaxing into a small confined plastic zone, five code checks passed out of five — is how simulation separates a harmless surface spike from a real failure before you thicken a wall that never needed it. That's innovation through insight.