First-Order Reliability Method — FORM

Why FORM is efficient

The First-Order Reliability Method estimates a small failure probability without sampling the entire input space. It transforms uncertain variables to a standard-normal space, identifies the point on the failure boundary with the highest probability density — equivalently the shortest distance from the origin — and replaces the limit-state surface locally with a tangent hyperplane. The distance is the reliability index β.

Step 1 — transform variables

Physical variables may be Normal, Lognormal, Weibull or otherwise distributed and can be correlated. FORM operates most naturally in a space of independent standard-normal variables u. Probability-preserving transformations such as Rosenblatt or Nataf-based approaches are used depending on the joint model. The transformation itself can materially affect the result when dependence is complex, so it is part of the method, not merely preprocessing.

Step 2 — find the design point

FORM solves a constrained optimisation problem: minimise ||u|| subject to g(u)=0. The solution u* is the most probable failure point under the transformed density. Algorithms based on Hasofer–Lind–Rackwitz–Fiessler concepts are common, but convergence depends on scaling, gradient quality and smoothness of the limit state. Multiple starting points are prudent when several failure regions may exist.

β = min ||u||
subject to g(u) = 0

Step 3 — linearise the boundary

At the design point, FORM replaces the actual limit-state surface with its tangent hyperplane. Failure probability is then approximated by Φ(−β). If the surface is nearly linear over the important probability mass, this can be highly accurate. Strong curvature, discontinuity or several comparable design points can make the first-order approximation inadequate.

Gradient calculation

The search requires derivatives of g with respect to the transformed variables. Analytical derivatives are ideal but rarely available for black-box FEA. Finite differences are straightforward but require extra solves and can be noisy if perturbations are too small. Adjoint derivatives can be efficient for many design variables when the solver supports them. Always perform a step-size sensitivity check on finite-difference gradients.

FORM with FEA

1. Define a scalar, verified limit state from the FE results.
2. Choose transformations and dependence model for the uncertain inputs.
3. Evaluate g at the current point and obtain gradients.
4. Iterate toward the design point until both boundary and location convergence criteria are met.
5. Re-run the high-fidelity model at the final design point and nearby perturbations.
6. Compare the FORM result with targeted Monte Carlo or importance sampling where practical.

Sensitivity information

The design-point direction indicates which variables control the rare failure event locally. This is often more useful for design improvement than the Pf number alone. A variable with modest effect on ordinary response variance can dominate reliability if its tail moves the system toward the limit state. Use these local importance measures alongside global sensitivity analysis rather than as substitutes.

Failure modes and multiple design points

A non-convex failure region may contain several local design points. A single converged FORM run can find only one. Try physically distinct starting points or formulate separate limit states for distinct mechanisms. If several points have similar β, system-reliability or multi-point approximations may be needed.

Transformation quality for non-Normal variables

The transformation from physical variables to standard-normal space is central to FORM. For independent variables, marginal probability transforms are straightforward. Correlated non-Normal variables require a consistent joint model, and the transformed correlation may differ from the physical Pearson correlation. Verify that simulated samples transformed back to physical space reproduce the intended marginals and dependence before relying on the FORM design point.

Practical convergence criteria

A robust implementation should monitor several quantities: change in u, distance to the limit-state surface, change in β and change in design-point direction. Convergence of one scalar criterion alone can hide oscillation or a poor gradient. If the iteration repeatedly crosses a discontinuity or alternates between mechanisms, reformulate the limit state or use a simulation-based method rather than forcing numerical convergence.

FORM convergence is not proof of accuracy. Always inspect the final design point physically and verify that the local linear approximation represents the real limit state in the region controlling failure.

Engineering judgement — governing sensitivities

For First-Order Reliability Method — FORM, the most useful review question is not simply whether the solver has produced a plausible contour or scalar result, but whether the model preserves the geometry of the limit-state surface in transformed probability space. Gradient quality, nonlinearity and multiple design points determine whether a local approximation is trustworthy. This is where apparently small modelling choices can change the engineering conclusion. The analyst should identify the variables that can move the governing response, separate physical uncertainty from deliberate conservatism, and show that the selected modelling fidelity is proportionate to the decision being supported. Where the response is close to an acceptance boundary, sensitivity cases should bracket credible changes rather than apply arbitrary percentage perturbations.

Verification evidence for the engineering record

A defensible First-Order Reliability Method — FORM assessment should leave an evidence trail that another engineer can independently interrogate. At minimum, review design-point convergence, gradient verification, transformation/dependence assumptions, curvature diagnostics, comparison with simulation or importance sampling and sensitivity to alternative starting points. Numerical convergence should be demonstrated on the response quantity that drives the decision, not only on generic mesh or solver metrics. The report should distinguish verified numerical behaviour from validation against test or service evidence, record any extrapolation beyond the supporting data, and state which assumption would most likely change the conclusion. This turns the analysis from a plausible calculation into an auditable engineering substantiation.

Key takeaways

• FORM converts a rare-event integration problem into a design-point search.
• Transformation, gradient quality and limit-state smoothness govern robustness.
• Verify the design point with direct model evaluations and an independent probabilistic method where feasible.