Joined initiatives around uncertainty management

Abstract

An industrial decision process supported by quantitative modelling (safety and reliability, physical design of facilities or processes, economic optimisation, environmental impact, etc.) quickly faces a wide diversity of uncertainties, imprecision, errors or alea affecting all data or numerical models. Beyond a terminological heterogeneity that is explicable by historical separation of the fields involved (such as metrology, reliability,separation of the fields involved (such as metrology, reliability, statistics, numerical analysis, ...), this papers introduces a generic and applied approach to uncertainty, derived from years of experience and recently shared by different applied research groups. This methodology is composed of four main steps that enable to distinguish some classical steps in modelling: specification of a criterion of interest to be assessed by a numerical representation of the problem containing uncertain variables (step A), identification and quantification of the sources of uncertainty (step B), propagation of uncertainty through the numerical representation (step C) and ranking of the uncertainties by sensitivity analysis (step D). This approach aims at giving a consistent and industrially realistic framework for practical mathematical modelling, assumingly restricted to quantitative and quantifiable uncertainty. Within this framework, various mathematical settings are possible; however, themixed deterministic–probabilistic setting appears to be central in present-day industrial applications and is the core of this paper. This paper introduces the current status of applied research in this field and points out different initiatives (software, research community) that are dealing with this topic.