Operational reliability — Analysis and reliability databases

This article presents the various steps, methods and data required to carry out a predictive operational safety analysis of complex industrial systems. Given the importance of the consequences of failures on availability, people and the environment, manufacturers are increasingly concerned with carrying out studies to predict the level of operational safety for existing installations or those under design. A predictive operational safety analysis of a complex system can be broken down into several main stages: structural and functional analysis of the system, followed by qualitative and/or quantitative analyses based on available feedback data. The first section describes the flowchart of tasks and analyses to be carried out in order to draw up a predictive operational safety analysis file. It is used to check that the results obtained from the various analyses comply with the specifications defined in the user's specifications. The second section, after a reminder of the aims of value analysis, presents the main functional analysis methods used to identify the functions of a complex system based on structural models. Knowledge of these functions is essential to identify the failure modes which, through their effects, will lead to failure. The principles of the FAST, RELIASEP ^® , SADT ^® , IDEF0 and APTE ^® functional analysis methods are briefly described. The third section is dedicated to a typology of classical predictive analysis methods. It defines deductive and inductive methods, and describes the characteristics of qualitative and quantitative methods. In particular, quantitative methods assess the parameters associated with dependability. They use human reliability data and reliability databases, which will be presented in this article. Qualitative methods include FMECA, PRA, MCPR and HAZOP. Quantitative methods under development include: reliability diagrams, truth tables, consequence trees, cause-consequence diagrams, bow ties, state spaces and Markov processes. To take account of developments over the last two decades in the field of predictive analysis in operational safety under the names of "condition-based-monitoring-CBM", "diagnosis, prognostics and health management (PHM)" and "RUL - remaining useful life", the fourth section will present these advanced methods. They have been developed specifically to diagnose and prognose failures. Their ultimate purpose is to predict the service life before failure (DEFAD) and its probability (noted as "DEFAD-RUL" in this article). Numerous methods have been developed for estimating DEFAD-RUL, and will be briefly described here. They include model-driven, data-driven, experimental and hybrid methods. These call on classical statistical...