Vibration-based substructural inverse analysis approach for large-scale discrete structural systems

Computationally effective inverse analysis algorithms are crucial for damage detection and identification, performance evaluation and control design of real dynamic systems. From the computational point of view, parametric identification for large-scale dynamic structures presents a challenging problem because of the convergence of the inverse analysis and the limitation of available measurements. A non-classical sub-structural inverse analysis approach with two neural networks by the direct use of acceleration measurements is proposed. The accuracy and robustness of the approach for parameter identification are examined by numerical simulations with a substructure within a fairly large-scale discrete shear structure model with 50 degree-of-freedom involving all stiffness and damping coefficient values unknown. Based on the two neural networks, the inter-storey stiffness and damping coefficients of the substructure are identified. Different from most of the existing inverse analysis methodologies, the proposed methodology does not require any structural modes or frequencies extraction from dynamic measurements.