Assessment of Weak Segments in Cantilever Beams Using an Artificial Neural Network

Damage detection is a fundamental task in the safety assessment of engineering structures. Various nondestructive damage detection techniques have been proposed to accomplish this task. The research presented herein aims to prove the reliability of an original method for detecting damage in beams using analytically generated data. The method consists of dividing the beam in several segments, among which one or two are considered as damaged. The damage is induced by decreasing the modulus of elasticity of the segment. For each damage scenario we derive the natural frequencies of a few numbers of vibration modes and the resulted frequency shifts (RFS). Taking the RFS values as input data and the damaged segments’ number and the modulus of elasticity alteration as output data, we train an Artificial Neural Network (ANN), which can locate the damaged segment and assess the stiffness degradation. The accuracy of the ANN model is validated using analytical data and Finite Element Method (FEM) simulations. Since we succeeded in assessing damage accurately, we demonstrated that the proposed analytical method provides the means for generating large databases that can be used to train ANN’s that can effectively detect damage.