This paper investigates damage quantification through deep learning algorithms. A bidirectional long short-term memory (BiLSTM) network and a convolutional neural network (CNN) are trained based on synthetic vibration accelerations of a reinforced single-span concrete beam. The damage extent is modeled by crack patterns, which differ in crack lengths and number of cracks. Different levels of accuracy in the damage quantification are analyzed by investigating various classes of damage extents. High classification accuracies are obtained for both networks, which shows the benefit of deep learning algorithms for Structural Health Monitoring (SHM).