 This study tests a multi-temporal facade damage detection approach using convolutional neural networks and compares it with monotemporal approaches. The results show that multi-temporal approaches outperform monotemporal ones by up to 25% in F1 score, with the best performing approach taking 6 tuples of 3 views per epoch as input for late fusion image classification. However, detecting small damages such as smaller cracks or areas of spalling remains challenging due to low resolution of the dataset used. This article was authored by Diogo Duarte, Francesco Nex, Norman Kerl, and others.