Hybrid Uncertainty and Sensitivity Analysis of the Model of a Twin-Jet Aircraft

International audienceThe mathematical models employed in the risk assessment of complex, safety-critical engineering systems cannot capture all the characteristics of the system under analysis, due to: (i) the intrinsically random nature of several of the phenomena occurring during system operation (aleatory uncertainty); (ii) the incomplete knowledge about some of the phenomena (epistemic uncertainty). In this work, we consider the model of a twin-jet aircraft, which includes twenty-one inputs and eight outputs. The inputs are affected by mixed aleatory and epistemic uncertainties represented by probability distributions and intervals, respectively. Within this context, we address the following issues: (A) improvement of the input uncertainty models (i.e., reduction of the corresponding epistemic uncertainties) based on experimental data; (B) sensitivity analysis to rank the importance of the inputs in contributing to output uncertainties; (C) propagation of the input uncertainties to the outputs; (D) extreme case analysis to identify those system configurations that prescribe extreme values of some system performance metrics of interest (e.g., the failure probability). All the tasks are tackled and solved by means of an efficient combination of: (i) Monte Carlo Simulation (MCS) to propagate the aleatory uncertainty described by probability distributions; (ii) Genetic Algorithms (GAs) to solve the numerous optimization problems related to the propagation of epistemic uncertainty by interval analysis, and (iii) fast-running Artificial Neural Network (ANN) regression models to reduce the computational time related to the repeated model evaluations required by uncertainty and sensitivity analyses