Vacuum insulated glazing (VIG) defines an energy efficient glazing unit. Their thermal performance units can reach values as good as quadruple insulating glass, with thinner dimensions and less material. Owing to their internal vacuum, VIGs are under the permanent influence of the atmospheric pressure acting on them throughout their entire service life. To withstand the pressure and to ensure sufficient distance between the glass panes, small pillars are positioned in-between. Especially the area around the pillars is prone to damage during the manufacturing of a VIG and during its lifetime. In order to assess this damage efficiently, automatic damage detection is necessary. For this purpose, we use a convolutional neural network. The binary classification model achieves an accuracy of 100 % for clearly recognisable damage and is also able to visualize the detected damage. Through our object recognition model, the input resolution can effectively be increased by cropping the image before the classification. The proposed methods can therefore be used to detect systematic defects even without large amounts of training data. Damage detection and classification can be used for quality control and enables the application of fracture mechanical models for assessing the stability of initial cracks during lifetime.