On a simple scheme for systems modeling and identification using big data techniques

In the field of reliability engineering and systems safety, it is a common challenge to identify the state of a system with basis in a limited set of observations of system performances. Often, there are a multitude of different possible system states, including states of damages, which compete in explaining the observations. To account for these in the context of risk-informed management of systems, the probabilities of the relevant possible different states are needed. In the present contribution, an idea on how this might be supported through big data techniques is presented. Here, systems are considered more holistically and not only as the relationship between input and output. The starting point is to establish a knowledge-consistent probabilistic representation of the system, its key performance characteristics, and the observations (exposures, condition state and performances) that may be collected from the system in reality. Monte Carlo simulations are then employed to establish the relevant scenarios of realizations of the random variables describing possible system states, system performance characteristics, and observations. Using big data classification on the simulated scenarios, the probabilities of the system being in a given state, given particular outcomes of observations, may then be straightforwardly evaluated. The application of the presented idea is illustrated through two examples considering damage identification in structural systems subject to extreme loading.