Damage location in structures through a connectivistic use of FEM modal analyses

The presence of damage in a structure gives rise to alterations in its response which make it possible to determine the ''symptomatic functionals'' of damage. Frequency analysis may provide a summary characterization of the distortions in a structure's dynamic behavior (as revealed, for instance, by the variations in a number of modal frequencies). This paper illustrates an application involving the use of ''feedforward'' neural networks to locate damage in an FEM truss model on the basis of the modal analyses carried out on the structure in its undamaged and damaged conditions. The encouraging performance of the neural networks (exploited directly in the ability to realize continuous maps) emphasize the need to work out more suitable methods for experimental applications.