A probabilistic reliability analysis technique that estimates the probability of failure by locating the design (most probable failure) point on the limit‑state surface in a transformed standard-normal space and approximating the limit-state function by its first‑order (linear) Taylor expansion about that point. The reliability index (beta) is the shortest distance from the origin to the linearized failure surface; failure probability is approximated by Φ(−beta), where Φ is the standard normal cumulative distribution.

Principle

Local linearization of the limit-state surface at the most probable failure point reduces an arbitrary failure region to an equivalent half‑space in standard normal space; the distance (beta) to this tangent hyperplane quantifies reliability and maps to a Gaussian tail probability.

Demonstration

Illustrative scenario → A simply supported beam has uncertain uniform load and uncertain flexural capacity defined by random variables with known distributions. Recognition → Define the limit‑state g(X)=capacity−demand and transform variables to standard normal space. Action → Use an optimization algorithm to find the design point that minimizes distance to the origin subject to g=0; linearize g at that point to compute beta. Consequence → Compute Pf ≈ Φ(−beta). The result gives a quick estimate of failure probability and identifies sensitivities via the design‑point direction cosines.

Misapplication

Applying FORM without verifying single dominant failure region or weak nonlinearities (large curvature) of the limit‑state surface. The semantic error is assuming the local linear approximation is globally representative; in multimodal or highly curved cases the identified design point may be local and Pf estimated by Φ(−beta) can be significantly inaccurate.

Consequence

When its assumptions hold, FORM yields an efficient estimate of failure probability and provides sensitivity information for design optimization; when misapplied it can underpredict or overpredict risk, potentially producing nonconservative designs or unnecessary over‑design depending on the curvature and multiple failure modes present.

Reversal

If the limit‑state exhibits significant curvature near the design point or multiple comparable failure regions exist, second‑order methods (SORM) or direct simulation (Monte Carlo, importance sampling) should be used; likewise, for heavy‑tailed distributions a transformation to Gaussian space may not render FORM accurate.

Boundary

Clearly within: smooth, monotonic limit‑state functions with a single dominant failure point and variables that can be transformed to near‑Gaussian space. Boundary case: moderate nonlinearity or weakly non‑Gaussian inputs where FORM may provide approximate results but should be validated. Clearly outside: discontinuous limit states, multiple widely separated failure regions, or strongly non‑Gaussian/heavy‑tailed variables.

Semantic Tension

Trade‑off between computational efficiency (FORM is fast and provides gradients) and fidelity (accuracy suffers with curvature or multiple failure regions); tension between approximating reliability analytically and using costly but more accurate sampling methods.

Synthesis

FORM converts a probabilistic failure problem into a geometric distance problem in standard normal space via local linearization at the most probable failure point; it is a first‑order expedient that yields useful sensitivity information but requires checks for curvature and multimodality to ensure reliability of results.