On-Line Monitoring, Control, and Reliability of Structural Dynamical Systems

Abstract

Ideas for improving the efficiency of Monte Carlo simulation (MCS) is the subject of the final section. Determining the low failure probabilities of typical engineering systems is quite difficult without using millions of MCS realizations to characterize the probability distribution. Several links between some MCS variance reduction techniques and Genetic Algorithms are discussed. A simple example, incorporating Genetic Algorithm operators into MCS, is shown to estimate probabilities a couple orders of magnitude smaller than standard MCS. Several concepts for characterizing the sense of a realization's "importance", such as discrepancy sensitivity and phase space velocity, are examined and found to successfully quantify realization importance.