Glossary term
Uncertainty Analysis
The systematic evaluation of how input uncertainty affects model, measurement, or design results.
Definition
methodUncertainty analysis evaluates how imperfect knowledge of inputs, measurements, assumptions, or models affects a calculated result or decision.
Uncertainty analysis quantifies the credibility of engineering results by identifying uncertain inputs, assigning defensible ranges or probability distributions, propagating them through calculations, and interpreting the output in relation to risk, requirements, confidence, and validation evidence.
Uncertainty analysis starts by asking what could make the result wrong or incomplete. Sources include measurement error, calibration limits, material scatter, environmental variation, manufacturing tolerances, numerical approximation, model-form error, boundary-condition uncertainty, incomplete data, and human judgement.
For simple independent inputs, uncertainty can sometimes be propagated with sensitivity coefficients:
For nonlinear models, correlated variables, discontinuities, or non-normal distributions, Monte Carlo simulation, interval methods, polynomial approximations, or scenario analysis may be more appropriate.
Engineering use
Uncertainty analysis supports measurement reporting, simulation credibility, reliability estimates, tolerance allocation, safety margins, calibration, model validation, and risk decisions. It is useful only when tied to a decision: passing a requirement, selecting a design margin, ranking failure modes, or deciding whether more testing is worth the cost.
Uncertainty is not the same as variability. Variability is real variation in the population or process; uncertainty is lack of knowledge about the value or model. A robust analysis states which is which, because the mitigation differs.
Common mistakes
A common mistake is adding conservative assumptions one by one without tracking whether the final result has a known confidence level. Another is assuming inputs are independent when they share a calibration source, supplier lot, environment, or model parameter. A strong uncertainty review states input sources, distributions or bounds, correlations, propagation method, confidence level, sensitivity ranking, validation evidence, and the decision threshold affected by the result.