The WHA Dataset is a composed by SEM images of the fracture surface of WHA samples with different tungsten composition. This dataset is used to evaluate the efficiency of two Unsupervised Machine Learning algorithms that aim to cluster and classify the fracture images according to the tungsten composition. Additionally, the Activation Maps of the last convolution layer of the neural network (VGG16) that is used in an intermediate stage of the algorithms are computed, according to their weights in the next stages of the clustering algorithm.