Ongoing advancements in the fields of 3D modelling and digital archiving have led to an outburst in the amount of data stored digitally. Consequently, several classification and retrieval systems (search engines) have been developed depending on the type of data stored in these databases. Considering that we are in the digital age with most information archived digitally, the problem of automatic classification and search becomes a predominant one. The advent of deep learning techniques calls for a large number of inputs. However, the datasets for Engineering CAD models have not grown to the extent of image / graphical model datasets. This is because the people involved in the classification of CAD models should have rich domain knowledge and experience. Also, many of the design data are proprietary in nature and hence not very suitable for making them public.

Therefore, in this work, we develop deep learning frameworks for the problems of classification and search of 3D CAD models. For the work on classification, it is proposed to collect Engineering CAD Models from the publicly available and well-annotated datasets, and a dataset termed ‘CADNET’ is prepared with many augmented CAD models. The developed CADNET dataset is released publicly to aid in deep learning research for CAD models. A Convolutional Neural Network (CNN) approach is then proposed for the classification of Engineering CAD models perhaps for the first time. A novel method of weighted viewing directions by using gradient boosting is also proposed. The proposed CNN obtains state-of-the-art classification accuracy.

In order to develop a search engine for 3D CAD models, it is essential to first create a dataset of query sketches that can be used for training. Using the CAD models from the available databases, a large-scale dataset of computer-generated sketch data is created. Additional hand-drawn sketches of a few CAD models are also incorporated into the dataset. This dataset, termed CADSketchNet, is made available publicly. Many experimental learning algorithms have been tested on the CADSketchNet and their search results are reported. Following the development of this search engine, a deep learning based sketch cleanup and enhancement methodology is also proposed to enhance the computer-generated query sketches.