Data-driven Analysis of Engineering Drawings Using Component-based Graphs

Engineering technical drawings serve as a universal medium for information exchange between part designers and manufacturers. Despite the ease of direct digital access when using vector drawings, raster drawings are still preferred for sharing in the current industrial settings, which results in a major impediment in the automation of the extensive part indexing, cost estimation, and process planning due to the need of human involvement in interpreting these drawings. This work designs a data-driven pipeline to systematically address two major challenges when automatically interpreting raster part drawings: (1) The shortage of labeled data for training, (2) The difficulty in interpreting semantic information from sparse man-made images. 

To generate more training data, a data augmentation method based on dimension set randomization is introduced to synthetically generate an arbitrarily large dataset given only a corpus of existing examples. Two major constraints for synthesis are proposed to ensure the validity of the synthetic drawings. An evaluation of several image segmentation methods trained on our synthetic dataset shows that our approach to new data generation can boost the segmentation accuracy and the generalizability of the machine learning models to unseen drawings. Additionally, a data-driven framework is proposed to effectively vectorize the drawings and interpret the semantic meaning at the component level. Taking a raster drawing as input, vectorized components can be obtained through thinning, stroke tracing, and curve fitting, which then forms a graph structure embedded with topological and contextual information. Finally, a deep-learning graph convolutional network is introduced to learn such graph data and predict the semantic type of each component. Results show that our method yields the best performance compared to the recent image-based, and graph-based segmentation methods.