Deep learning-based analysis to identify fluid-structure interaction effects during the response of blast-loaded plates

Blast events within urban areas in recent decades necessitate that protective design is no longer reserved for military installations. Modern civil infrastructure composed of light-weight, flexible materials has introduced the consideration of fluid-structure interaction (FSI) effects in blast-resistant design. While the action of blast loading on massive, rigid structures in military fortifications is well established, assessment of FSI effects is, at present, only possible through computationally expensive coupled simulations. In this study, a data-driven approach is proposed to assist in the identification of the blast-loading scenarios for which FSI effects play a significant role. A series of feed-forward deep neural networks (DNNs) were designed to learn weighted associations between characteristics of uncoupled simulations and a correction factor determined by the out-of-plane displacement arising from FSI effects in corresponding coupled simulations. The DNNs were trained, validated and tested on simulation results of various blast-loading conditions and material parameters for metallic target plates. DNNs exposed to mass-per-unit-area, identified as an influential factor in quantifying FSI effects, generalised well across a range of unseen data. The explainability approach was used to highlight the driving parameters of FSI effect predictions which further evidenced the findings. The ability to provide quick assessments of FSI influence may serve to identify opportunities to exploit FSI effects for improved structural integrity of light-weight protective structures where the use of uncoupled numerical models is currently limited.