Direct soft identification for truss structure using noise-contaminated macro-strain

The increasing development of distributed strain monitoring technologies such as optic fiber Bragg grating and embedded piezoelectric sensors necessitates the development of strain-based identification methodologies. For the identification of member stiffness and damping parameters of a truss structure from free vibration-induced noise-free strain measurement, a three-step damage identification strategy using a strain-based emulator neural network (SENN) and a parametric evaluation neural network (PENN) has been reviewed. In civil engineering applications, the measurement of dynamic responses in field condition always contains noise components from environmental factors. To make the proposed direct soft identification strategy practical, the effect of noise on the performance of the strategy must be considered. The identification accuracy of the PENN for strain measurements contaminated with different level noise is investigated using a truss structure with a known mass distribution. The performance of noise injection learning for identification accuracy improvement is investigated.