On crack identification using neural networks

Most structures suffer fatigue damage at some point during their operational life. This damage may lead to a structural failure. An early damage identification is needed to prevent such a structural failure. A technique which depends on the measurement of the changes in the vibration characteristics of the structure can be effective, since inspection can be performed while the structure is in normal operation. This work presents a methodology for using neural networks in identifying structural damage employing the vibration signature data.;An experimental study was carried out to measure the random response of undamaged and damaged beam models. The damage was simulated by introducing a hand-made saw cut at different points along the length of the beam. The depth of crack was also varied. Two beam models were used: one was simply supported, and the other was a fixed-fixed beam. The beam was excited using random excitation. The auto-correlation function was calculated and used as an approximation for the free vibration of the model. A neural network technique was performed to identify the crack occurrence and its extent. The results show that this technique is able to detect the occurrence of the crack.