Fluid-structure interaction (FSI) problems involving transient phenomena are crucial in numerous engineering applications. In nuclear facilities, these include submersion scenarios due to flooding or dam break, sloshing in spent fuel pools or reactor pools during seismic events, and concrete ablation by high-velocity corium jets during severe accidents. These events involve free surface flows with large deformations and potential structural damage. They require robust numerical approaches to be modelled, capable of capturing complex FSI phenomena. This work proposes a numerical method to address such FSI problems in high-performance computing (HPC) environments, targeting exascale capabilities. A partitioned coupling is developed, using the Finite Element Method (FEM) and Smoothed Particle Hydrodynamics (SPH) method. The objective of such an approach is to leverage the respective strengths of each solver while maintaining computational efficiency on massively parallel architectures. The structural dynamics is handled with Europlexus, capable of modelling structural behaviour from elastic response to failure. The fluid dynamics is dealt with exaSPH, built with a GPU-optimised kernel. Both solvers employ conditionally stable explicit time integration schemes (finite differences). The coupling is achieved through the ICoCo standardised API, with field and mesh exchanges managed via the MEDCoupling library, enabling flexible data transfer between solvers. This presentation focuses on the validation of the coupled approach through classical 2D and 3D benchmarks, including dam break scenarios and water impact on flexible cylinders. Scalability studies are conducted by systematically refining spatial discretisation while monitoring solution convergence. To demonstrate the modularity of the approach and its capability to efficiently exploit large-scale parallel computing resources, future work will focus on coupling exaSPH with a next-generation CEA structural solver, specifically designed for HPC environments.