Modeling and classification of non-linear systems using neural networks--II. A preliminary experiment

Abstract In the first part of this study, a method for classifying non-linear systems using neural networks was proposed and validated using data from numerical simulation. In order to extend this validation to experimental data, a system was required with a repeatable non-linearity of controllable severity, which could be simply relocated within the system. A preliminary study is presented here of a free-free beam containing a non-linear gap element which induces a bilinear stiffness characteristic. It is shown that the dynamic behaviour is characteristic of a beam with an actual fatigue crack. A neural network is trained to distinguish between cracked and uncracked states of the beam when presented with measured time data.