A multi-level uncertainty integration strategy for forward model-driven SHM

Conducting full-system testing can be costly and infeasible in many structural health monitoring (SHM) applications. In addition, many SHM methodologies require data from all likely damage scenarios in order to produce robust classifications. Forward model-driven SHM is an approach in which validated numerical models simulate full-system outputs under different damage scenarios, which are subsequently used as training data for machine learning methods, improving the problem of a lack of available data. Here a framework is proposed in which uncertainties characterised at a sub-system level, using sub-system experimental data, are integrated into full-system predictions. The proposed methodology for multi-level uncertainty integration seeks to propagate sub-system parameter uncertainties whilst inferring any model discrepancies at each level. This paper demonstrates the effectiveness of multi-level uncertainty integration for forward model-driven SHM on a numerical case study.