Model Updating Based on Physics Informed Machine Learning on Welded Stiffened Structure

A stiffened structure is integral to complex structures such as ships. An advantage of the stiffened plate is that it has a greater load-carrying capacity. Therefore, structural health monitoring of stiffened structures is essential in a ship structure. The stiffened structure is modeled as a coupled system with a plate and stiffener forming the subsystems coupled with welded joints. The complex coupled system is modeled using finite element analysis. The coupling springs between a plate and beam could replicate the modal characteristics of welded joints in a stiffened structure. Since the welding characteristics could change depending on the operating conditions, the coupling spring was varied to account for the uncertainty in the weld strength. A hyperspace of coupling springs, one longitudinal and two torsional, was spanned using the Latin hypercube sampling technique following a uniform distribution. The eigenvalue problem for the stiffened structure was solved, and the modal characteristics of the system were determined. Dynamic features of the system, such as natural frequency, mode shape and frequency response function (FRF), were extracted. A metamodel for the system was developed using a Gaussian process emulator (GPE). The dataset was generated for a driving point response since this location's response magnitude was high. A validation study carried out on the metamodel indicated a good prediction of the weld strength. Future work would include a study to span more sensing locations and damping.