[Intermediate → Advanced] — Turning a six-component stress tensor into one number and a verdict. The criteria, what each assumes, and why most real failures are fatigue.
The problem
You have a stress state — six independent components. Material data comes from a uniaxial tensile test — one number.
A failure criterion is a function that reduces the tensor to a scalar comparable with that one number:
Different criteria embody different assumptions about how the material fails, and choosing the wrong one gives a confidently wrong answer.
Ductile vs brittle
The distinction that determines which criterion applies.
| Ductile | Brittle | |
|---|---|---|
| Mechanism | shear — dislocation slip | tensile separation |
| Warning | large plastic deformation first | none — sudden |
| Fracture plane | 45° to max principal stress | ⊥ to max principal stress |
| Tension vs compression | roughly symmetric | much stronger in compression |
| Examples | mild steel, aluminium, copper | cast iron, concrete, glass, ceramics |
The fracture surface tells you the mechanism. Twist a piece of chalk (brittle) and it breaks along a 45° helix — following the principal tension. Twist mild steel and it fails on a transverse plane — following the maximum shear. Same loading, different criterion, visibly different result. → Stress
Von Mises — for ductile materials
The standard, and the one you’ll use most:
Yield when .
Equivalently, in terms of the deviatoric stress:
Why it’s built on the deviator: metals yield by dislocation slip, which is driven by shear. Hydrostatic pressure produces no shear, so it produces no yielding — a solid steel ball at the bottom of the ocean is unharmed, while a hollow one implodes (that’s buckling, a different failure). → the deviatoric split
The physical interpretation is distortion energy: yielding begins when the energy of shape change reaches a critical value, independent of the energy of volume change.
In principal stress space, the von Mises surface is a cylinder whose axis is the hydrostatic line . Move along that axis — pure pressure — and you never reach the surface.
Von Mises stress is what FE post-processors plot by default, and it’s usually the right thing to look at for metals. But it is a scalar with no sign — it cannot distinguish tension from compression. For a material with different tensile and compressive strengths (concrete, cast iron, most polymers) it is simply the wrong criterion.
Tresca — maximum shear stress
Simpler, and always conservative relative to von Mises — up to 15% more so in pure shear.
In principal space it’s a hexagonal prism inscribed inside the von Mises cylinder.
Used in pressure-vessel codes (ASME) precisely because it’s conservative and easy to hand-check. Experimentally, von Mises fits metal data better; Tresca errs safe. The corners of the hexagon make it awkward numerically, which is why FE plasticity uses von Mises.
Maximum principal stress (Rankine) — for brittle materials
Failure when the largest principal tension reaches the ultimate tensile strength.
Correct for brittle materials, which separate rather than slip. And it captures the asymmetry: concrete is roughly 10× stronger in compression than tension, which is the entire reason for reinforced concrete — steel carries the tension the concrete cannot.
Mohr–Coulomb generalises this for materials whose shear strength depends on normal stress:
with the cohesion and the friction angle. The standard criterion in soil and rock mechanics, where confining pressure genuinely does increase strength — which is why deep rock is stronger than surface rock, and why sand holds together under load but not loose.
Drucker–Prager is the smooth version, used in FE because it has no corners.
Choosing
| Material | Criterion |
|---|---|
| Ductile metal (steel, aluminium) | von Mises |
| Ductile, conservative / code work | Tresca |
| Brittle (cast iron, glass, ceramic) | max principal stress |
| Concrete, rock, soil | Mohr–Coulomb or Drucker–Prager |
| Composites | Tsai–Wu / Hashin — direction-dependent |
| Anything cyclic | none of these — see fatigue |
That last row matters more than the rest.
Fatigue
Most real structural failures are fatigue. A part that survives its static load indefinitely can fail after millions of cycles at a fraction of that load.
Cracks initiate at a stress concentration, grow a little each cycle, and eventually the remaining section fails suddenly. Static criteria say nothing about this.
The S–N curve plots stress amplitude against cycles to failure:
- Steel has an endurance limit — below roughly it survives indefinitely ( cycles)
- Aluminium does not. It fails eventually at any stress amplitude, which is why aircraft have finite fatigue lives and mandated inspection intervals
Mean stress matters. A tensile mean stress reduces allowable amplitude, captured by the Goodman or Gerber relations. This is why compressive residual stress improves fatigue life — and why shot peening, cold-rolling threads, and autofrettage of gun barrels all work.
The practical levers:
- Eliminate stress concentrations. Fillets, generous radii, no sharp internal corners. This dominates everything else → stress concentration
- Surface finish matters enormously — cracks initiate at the surface, so a machined finish outperforms an as-cast one substantially
- Compressive residual stress via peening
- Avoid fretting at joints and press fits
Miner’s rule sums damage over variable amplitude: predicts failure. It’s crude — it ignores load sequence — and it’s still what’s used.
Historical note worth carrying: the de Havilland Comet crashes of 1954 were fatigue cracks starting at square window corners. The fix — rounded windows — is why every aircraft window you’ve looked through has a radius.
Fracture mechanics
For a body that already contains a crack, stress-based criteria fail: linear elasticity predicts infinite stress at a crack tip.
Replace stress with the stress intensity factor:
where is crack length and a geometry factor. Fracture when , the fracture toughness — a measured material property.
The is the important part. Doubling crack length raises by only 41% — cracks grow slowly, then accelerate. Combined with Paris’ law (, with for steel), this is the basis of damage-tolerant design: assume a crack exists, predict its growth, and inspect before it reaches critical length. That’s how aircraft are certified.
Toughness and strength trade off. High-strength steels are generally less tough — they tolerate smaller cracks before fracturing. Choosing the strongest available alloy can reduce structural safety, which is genuinely counterintuitive and a real design trap.
Ductile-to-brittle transition: many steels become brittle below a transition temperature. The Liberty ship fractures of WWII — hulls splitting in cold North Atlantic water — established this, and it’s why toughness is specified at service temperature rather than at room temperature.
Other failure modes
Stress-based criteria miss several important ones entirely:
Buckling — instability, not strength. A slender column fails at a stress far below yield, with the material undamaged. Strength doesn’t help; only stiffness and geometry do. → Beams and Structures
Creep — slow deformation under sustained load at high temperature (above ~0.4 of the melting point in kelvin). The design limit for turbine blades, boiler tubes, and lead pipes.
Corrosion and stress-corrosion cracking — chemistry, not mechanics, and often the actual initiator of a fatigue crack.
Wear and fretting — surface degradation at contacts.
Impact — high strain rate changes material behaviour; many materials embrittle. A static toughness value can be badly optimistic.
Safety factors
Typical values: 1.5–2 for well-characterised static loads; 2–4 where loads are uncertain; 1.5 for aircraft structure (with extremely well-known loads and rigorous inspection); 3–5 for pressure vessels.
A safety factor covers ignorance, not physics. Material variability, load uncertainty, manufacturing tolerance, modelling error, and degradation over life.
Which is why a large safety factor does not compensate for the wrong criterion. Applying von Mises to a brittle material with is still wrong — you’ve scaled a prediction that was never valid.
Limit-state design (Eurocode, LRFD) is the modern replacement: separate partial factors on loads and on resistances, calibrated so the probability of failure meets a target. More honest than a single lumped number, because it distinguishes “we don’t know the load” from “we don’t know the material”.
The practical checklist
- Ductile or brittle? Choose von Mises or max principal accordingly
- Static or cyclic? If cyclic, fatigue governs and static criteria are irrelevant
- Stress concentrations? Check fillets and holes; that’s where it will fail
- Slender in compression? Check buckling before strength
- Hot, or loaded for years? Check creep
- Cracks assumed present? Fracture mechanics, not stress
- Cold service? Check the ductile–brittle transition
- Did the analysis stay inside its assumptions? If a linear-elastic run predicts stress above yield, the result is invalid — rerun it plastic → Constitutive Models
That last one is the most commonly ignored, and the most consequential.
Related
- Stress — principal stresses and the deviator
- Linear Elasticity — stress concentration and the crack-tip singularity
- Beams and Structures — buckling
- Continuum mechanics map
Software engineering
Infrastructure & security
Computer science
Hardware & engineering