Transfer Learning for Verification and Validation

One of the key challenges in Structural Health Monitoring (SHM) and Performance Assessment is the difficulty associated with gathering experimental data from a structure in its damaged state. Numerical modelling offers the potential to overcome this problem by making physically informed predictions of how the structure will behave once damaged. However, numerical modelling raises challenges of its own, with a major question being how one goes about establishing the credibility of the model predictions.

The aim of this project is to develop best practice for developing and validating predictive models for structural damage, and thus demonstrate the feasibility of model-augmented SHM approaches for use on practical systems. This will involve:
(1) Developing FE modelling techniques for predicting dynamic behaviour of structures subject to realistic damage;
(2) Developing and conducting associated test protocols specifically targeted at the validation of such models.

The aim will be to seek 'effective' models that contain sufficient physical insight to train SHM systems guided by ideas from Decision Theory on the notion of model sufficiency. As part of the work the effects of both model and experimental uncertainty and confounding influences (e.g. environmental variation) will be considered. The project will build from considering models of small-scale systems at ambient conditions to the testing of real-world structures under realistic environmental conditions at the Laboratory for Verification and Validation (www.lvv.ac.uk).

This project will fit into the EPSRC's Performance and Inspection of Mechanical Structures and Systems research area. The work is primarily concerned with experimental and modelling techniques, in this case those relevant, SHM, which is by its nature closely related to the fields of dynamics and acoustics.