ABSTRACT
This paper presents a new method for assessment of semi-rigid connections based on dynamic data processed by neural networks. The study was carried out on two kinds of a steel joints for which natural frequencies were obtained from dynamic tests. Two approaches was used: one-stage and two-stage. In the onestage procedure the forces is calculated directly from the experimental data. The second approach consists of two steps: i) the neural simulators for predicting frequencies depended on prestressed forces in bolts are created, ii) next the neural network trained by the patterns generated by the first step neural networks is used to assess condition of joints. The results of the paper show that the neural network trained with experimental dynamic data was capable for detecting actual state of the joint.
