Damage detection in fluid-structure systems: A case study on cylindrical steel tank with varying fluid levels

Traditional damage detection methods often struggle to accurately identify structural damage in systems involving structure–fluid interaction, particularly in cylindrical tanks, which necessitate high-fidelity numerical models and present localization challenges complicated by structural symmetry and varying fluid levels. Dataset-based damage detection (DBDD) approaches offer promise but encounter challenges in accurate numerical modeling, effective pattern recognition, efficient handling of input and output vector, and managing computational and storage costs. This study introduces a practical procedure incorporating novel techniques for damage detection in a cylindrical tank with varying fluid levels. Initially, dynamic equations governing the structure–fluid interaction are developed and validated against experimental results. The finite elements (FE) method is then employed to model the tank under different fluid conditions. A new method is proposed to enhance pattern recognition algorithms by decomposition output vector followed by the application of a stacking ensemble for damage detection. To address computational and storage challenges, a hybrid feature engineering framework incorporating a novel methodology is proposed and implemented. The approach is validated on a 3D steel tank case study with varying fluid levels, demonstrating a significant reduction in FEM computational cost (by 27%) and storage requirements (by 75%), while achieving damage detection accuracies of 90% for clean datasets and 86.5% for noisy datasets.