An Augmented Risk-Based Paradigm for Structural Health Monitoring

Abstract

The notion of risk is comprised of two components: the likelihood of an adverse event occurring, and the severity of the consequences. Probabilistic risk assessment (PRA) is an established methodology for quantifying risks, used by engineers in a range of industries to inform decisions regarding the design and operation of safety-critical or high-value structures and systems.

In comparison, a salient motivation for implementing a structural health monitoring (SHM) system is to facilitate the decision-making process throughout the lifetime of a structure. Oftentimes, there is uncertainty when assessing the damage state of a structure. As such, a statistical pattern recognition (SPR) approach to structural health monitoring is employed in which data acquired from a structure of interest are processed to yield features indicative of the damage state. The current paper details how decision-making under uncertainty can be aided by augmenting the established structural health monitoring paradigm to incorporate risk, utilising a framework based on probabilistic graphical models.

The modelling of failure events as fault trees is a core process in conducting a PRA and, by modelling key failure modes of interest for a given structure in this way, provides a convenient and rigorous basis for formulating risk-based SHM problems. As statements in Boolean logic, fault trees are limited to representing binary damage states. Fortunately, it is possible to map fault trees into Bayesian networks which are capable of representing multi-state variables whilst also affording other benefits.

Risk is incorporated into the framework by introducing utility nodes into the probabilistic graphical model, thereby attributing costs to failure events. Decision nodes are also included, enabling the evaluation of potential courses of action such that a strategy that maximises utility may be determined.