Mitigating high dimensionality in damage identification for plate-like structures through substructuring with interacting filtering-based approaches

Plate-like structures form the essential components of diverse engineering systems, such as building floors, aircraft wings, wind turbine blades, etc. Detection of early onset of damage/s and prevention of sudden catastrophic failures during operation are necessary for maintaining and enhancing the safety standards of such high-dimensional and intricate systems. To detect anomalies in high resolution within the structure, the employment of a high-dimensional support model parameterized with damage attributes becomes imperative within a model-based assessment framework. Further, to support the estimation of consequent high- dimensional parameter sets, the traditional structural health monitoring methods require dense instrumentation over the entire structure for accurate damage location and severity assessment. Targeting an overall cost reduction in the implementation and subsequently circumventing the augmented cost of computation, this article employs a robust Interacting Particle Kalman filtering (IPEnKF)-based approach for health estimation of plate-like structures that is capable of monitoring only a manageable subdomain of interest while being independent of the rest of the domain. Eventually, this approach allows stage-wise health estimation of complex plate-like structures under the limitation of instrumentation. T he proposed approach has been validated through both numerical simulations and real experiments on a scaled model of the plate roughly representing the NASA CRM wing, idealized as a trapezoidal Mindlin plate divided into independent substructures. The proposed method employs the IPEnKF algorithm to estimate each substructure independently by modifying the process model by injecting output towards achieving robustness against unknown substructure boundary forces. By using this subdomain estimation approach, the sensor requirement has been reduced by 60% while maintaining 95% accuracy in damage detection and assessment. Additionally, computational costs have been significantly reduced due to the algorithm's ability to independently monitor the system at its substructure level without depending on the rest of the structure. The method also exhibits excellent reliability, with minimal false alarms. The results emphasize the algorithm's robustness and the essential role of sensor placement in improving the accuracy of damage identification.