Classification of 3D models is an open research problem. Many architectures have been proposed to classify 3D models. However, most architectures have tackled the ModelNet (Wu et al. 2015) or ShapeNet (Chang et al. 2015) datasets which offer everyday objects. To test the limitations of existing architectures and motivate new architectures, we present a challenging dataset of 3D models of flanges. There are two variants of flanges in this dataset: 1) 4 holes vs 8 holes and 2) 1 to 9 hole flanges. These holes are placed randomly on the flange faces and also in a regular fashion. Not only does this dataset stress test classification ability of existing architectures, but also promotes visualisation and explanation of learned concepts to match with human reasoning. These models are for demonstration purpose only and do not reflect actual products.