Ask any structural engineer where they begin the analysis of a new steel frame. Nine times out of ten, you’ll hear one word: bending. Not shear, not torsion, not even buckling first. Beams and girders, before anything else, and inside them, the bending moment, before any other check.
This article explains why. What sits behind the deceptively simple σ = M·c/I, how standards like Eurocode 3 and AISC 360 turn that one formula into a chain of connected checks, where manual calculations systematically miss things, and what changes when automation runs on top of FEA.
Bending as the primary load in any frame
Almost every load-bearing element in the horizontal plane works in bending. Floor beams transferring slab loads, frame girders, roof purlins, crane booms, bridge main beams, and longitudinal hull stiffeners on a ship. Load through a lever arm every time. Always a moment.
The base formula traces back to Euler-Bernoulli beam theory from the mid-1700s: σ = M·c/I. Stress at the extreme fiber equals moment divided by section modulus S. Zero at the neutral axis, peak at the outer fibers. That’s where the steel is always working hardest, and that’s where you check it first.
From an engineer’s perspective, the thing to grasp is this: M is fixed by statics. You can’t shrink it without redistributing loads across the structure. The only knob the designer can turn is section geometry. Section modulus S, or Wel in European notation. Bending sets the section. Every other check has to live inside that constraint.
Why does bending go first
The logic is straightforward. Before you size a weld, you need the section it joins. Before you check the combined bending plus axial, you need Mc,Rd. Before you run fatigue, you need the stress field at the location of the peak moment. Bending sets the reference point.
And then there are deflections. SLS limits like L/360 for floors, or L/240 under full load, often drive the section even harder than ULS strength. A nine-meter beam under code live load can pass bending with a margin and still fail deflection. Same underlying parameter on both sides: moment of inertia I. So the smarter move is to check both in a single pass, not iterate one against the other.
This is the baseline sequence taught in the first year of structural engineering, and nobody outgrows it for the rest of their career.
What hides behind σ = M·c/I
The formula looks elementary. But the moment you apply it under any real code, it expands into a chain of connected checks. And that expansion is where mistakes creep in.
Take the choice of section modulus. Eurocode 3 sorts cross-sections into four classes based on local buckling risk. Class 1 and 2 use plastic Wpl, Class 3 uses elastic Wel, Class 4 uses an effective Weff. For standard I-shapes, the Wpl/Wel ratio sits in the 1.10–1.17 range. For rectangular plates, it climbs to 1.50. Plug Wel into a compact Class 1 section, and you’ve lost 10–17% of capacity and added unnecessary steel. Plug Wpl into a Class 3 section and you’ve overestimated capacity by the same margin. The first costs money. The second costs more than that.
Layered on top of section classification is lateral-torsional buckling. An I-beam with no restraint on its compression flange loses moment capacity. Textbook case from AISC 360 Chapter F: a W14×22 with an unbraced length of 10 ft gives φMn = 98.8 ft·kips, versus φMp = 124.5 ft·kips fully braced. That’s a 26% capacity hit from a single missing brace point. Closed sections like CHS or SHS barely feel this effect, which is the kind of detail that gets forgotten when the profile is swapped during optimization.
On top of all that sits the AISC Cb factor, which accounts for the shape of the moment diagram along the unbraced length. A default Cb = 1.0 for a midspan point load can underestimate capacity by close to a factor of 1.5. Each of these corrections lives on top of the basic formula. Which is why a real-world calculation of bending stress under code requirements doesn’t reduce to plugging M into σ = M·c/I. It becomes a chain of four or five connected checks, and getting any one of them wrong throws the whole chain off.
And that’s just the baseline. Beyond it sit shear-bending interaction, biaxial bending, combined bending plus axial, and fatigue at peak-moment locations.
Standards: shared logic across national codes
Open Eurocode 3 EN 1993-1-1 clauses 6.2.5 and 6.3.2, or AISC 360 Chapter F, or DNV-OS-C101 for offshore steel structures, or API 2A-WSD for fixed platforms. The skeleton is identical. Classify the section first. Compute Mc,Rd, or Mn from the matching modulus. Run a stability check through χLT in Eurocode or the Lp/Lr zones in AISC. Then run the interaction if there’s axial force or shear in play.
The formulas and partial factors differ. The logic of the check stays the same. Worth knowing, because an engineer moving between European, offshore, and US projects is essentially learning a new vocabulary on top of the same grammar. This is also where automation pays back the most: the same FEA result run through multiple code sets without redoing the math by hand.
Where manual calculations break
Hand calculations for bending work cleanly only in clean cases. Single-span beam, uniform load, symmetric section, no interaction with neighboring elements. Once the model gets real, the problems show up.
It starts with the obvious. A hand calculation only looks at the cross-sections the engineer picked. Midspan peak moment gets checked almost every time. The section ten centimeters off the support, where moment combines with shear and a welded gusset adds a stress concentrator, often goes unchecked. An FEA model evaluates the whole structure, and the highest-utilized section gets flagged automatically.
Then there’s load combination cherry-picking. On a beam-column, the load case that maximizes midspan moment isn’t always the one that maximizes combined utilization. Manually iterating through dozens of combinations takes hours, and a few will slip through anyway.
And then there’s the rework after a geometry change. Swap HEA 280 for HEA 240, and every hand-calc spreadsheet needs to be redone. Usually, it gets done. Not always in time, though, and old revision numbers occasionally end up in the final report. Anyone who’s run a verification cycle on a deadline knows how this goes.
What automation on top of FEA actually changes
When bending checks become part of automated verification, the formula doesn’t change. The speed and coverage do. A modern FEA platform with built-in code checks recognizes element types automatically: what’s a beam, what’s a column, what’s a brace. Each one gets the correct check for the chosen standard. Local axes line up with geometry rather than manual entry, and the strong/weak axis mix-up disappears from the top of the error list.
ULS and SLS get evaluated in one pass. A beam that passes the moment but blows past L/360 is flagged in the same run, not the next iteration. Documentation is generated from templates, and the time to assemble a report drops by 50–70% compared to hand-built deliverables.
One more piece worth flagging: in shell models, the bending check pulls the linearized stress through the thickness, not the depth-averaged von Mises. That distinction matters. The code check needs the normal stress component aligned with the member axis, not an equivalent stress.
The takeaway
Bending isn’t first because of tradition. Every check downstream depends on it: welds, fatigue, combined loading and deflections. A section that passes the bending check cleanly, with the right class, real LTB treatment, and the actual Cb factor, becomes a solid anchor point for everything that follows. A section checked sloppily drags that error all the way to the final report. This is not the place to cut corners. Speed comes from automation, not from skipping work.
Image source: Unsplash
