Structures can be subjected to damage, leading to catastrophic failures and significant financial losses. Thus, researchers have been studying several tools to ensure reliability and safety. Thus, structural health monitoring has drawn attention, mainly by using tools such as vibration-based model and artificial neural networks. So, this work aims to develop a methodology to identify and classify damage in glass fiber-reinforced plastic composite beams through vibration data and artificial neural network. For this, healthy and damaged beams were manufactured considering different delamination sizes. Then, dynamic tests were performed to obtain both time and frequency domain data. As the large dimension of the data obtained by the vibrational tests, hinders its direct use to feed the neural networks, a strategy called dislocated series is used to reduce the raw signal size in mini batches, without losing important information to detect the damage. Results show that the artificial neural network topology and the parameters of the dislocated time series are crucial to the success of the proposed methodology. When these parameters are properly selected, it is possible to successfully detect and classify damage with less computational cost when compared to the direct use of the vibration-based model data.