Stochastic Analysis & Reliability

When loads and materials are uncertain — what's the probability of failure?


1. What is it?

Stochastic analysis quantifies how uncertainty in inputs (loads, material properties, geometry) affects the output (stress, displacement, life):

"This bracket has a 5% probability of failure under the load spectrum — is that acceptable?"


2. When do I use it?

Use whenDon't use for
Loads are random or variableDeterministic loads (use Static Structural)
Material properties varySingle value known (use Static Structural)
Safety factor is insufficientSimple pass/fail check (use Static Structural)
Risk-based designCode-based design (use Static Structural)

3. The flow — every action described

Step 1: Design (geometry)

What you do: Draw your part or import a CAD file.

How:

Why: The solver needs to know the shape.

Step 2: Boundary (material + supports + loads + uncertainty)

What you do: Define deterministic and random variables.

How:

Why: The solver needs to know which variables are uncertain and how.

Step 3: Mesh

What you do: Cut the part into small cells.

How:

Why: The solver works on tiny triangles/quads.

Step 4: Simulate → Stochastic

What you do: Run the reliability analysis.

How:

Why: The solver computes the probability of failure.

Step 5: Results

What you do: Read the reliability index and sensitivity.

How:

Why: The reliability index tells you how safe the part is.


4. Reading your results

Reliability index β

ValueMeaning
β < 2.0High risk — Pf > 2.3%
β = 2.0–3.5Moderate risk — Pf = 0.23%–2.3%
β = 3.5–5.0Low risk — Pf = 0.02%–0.23%
β > 5.0Very low risk — Pf < 0.02%

Good: β > required (typically 3.5 for structural, 4.5 for aerospace). Bad: β < 2.0 — redesign needed.

Probability of failure Pf

ValueMeaning
Pf < 0.001Acceptable for most applications
Pf > 0.01Unacceptable — redesign

Sensitivity

Shows which uncertain variables contribute most to failure probability.


5. Boundaries of truth

What's validated

CaseReferenceError
FORM on linear-normal limit stateHasofer-Lind exact< 0.01%
Determinism across repeated callsBit-identical0
Honest-failure path (insensitive limit state)Returns nullCorrect

What's NOT covered


Common mistakes

MistakeFix
Too few Monte Carlo samplesIncrease samples for better accuracy
Wrong distribution typeChoose normal, lognormal, or uniform
Limit state insensitive to variablesCheck that variables appear in limit state
No convergence toleranceSet tolerance for FORM iteration
Ignoring correlationDefine correlation between variables

See also

Open the interactive workspace — mesh, solve, validate and export in the browser.

Start a guided solve
FEA Lab
Loading analysis environment…