A Method for Evaluating System Robustness in Stochastic, High-Dimensional, Multi-Failure Mode Models

Abstract

This paper introduces a method of predicting system robustness using engineering models with aleatory uncertainty. The Stochastic Robustness Evaluation and Categorization (SREC) method is useful for the design of systems where performance along some dimension is limited by several failure modes. SREC integrates and extends interaction plots and Monte Carlo methods to complex engineering models. These complex models are often difficult to evaluate and visualize due to the curse of dimensionality. SREC is effective for non-linear, non-convex, non-monotonic, and discontinuous models due to its basis in Monte Carlo methods. The method is based on the identification of low-performing solutions, the construction of probability density functions and intervals from these solutions, and the categorization of the input space based on the likelihood of low-performing solutions occurring. Using SREC in the late stages of the design process provides insight to the designer about possible improvements in the system’s robustness. A ground vehicle model based on a US Army test procedure is used to demonstrate the effectiveness of SREC on high-dimensional, multi-failure mode models.