Neural Networks Based Structural Model Updating Methodology Using Spatially Incomplete Accelerations

Because it is difficult to obtain structural dynamic measurements of the whole structure in reality, it is critical to develop structural model updating methodologies using spatially incomplete dynamic response measurements. A general structural model updating methodology by the direct use of free vibration acceleration time histories without any eigenvalue extraction process that is required in many inverse analysis algorithms is proposed. An acceleration-based neural network(ANN) and a parametric evaluation neural network(PENN) are constructed to update the inter-storey stiffness and damping coefficients of the object structure using an evaluation index called root mean square of prediction difference vector(RMSPDV). The performance of the proposed methodology using spatially complete and incomplete acceleration measurements is examined by numerical simulations with a multi-degree-of-freedom(MDOF) shear structure involving all stiffness and damping coefficient values unknown. Numerical simulation results show that the proposed methodology is robust and may be a practical method for structural model updating and damage detection when structural dynamic responses measurements are incomplete.