Date of Award
Open Access Thesis
Feature recognition is a critical sub-discipline of CAD/CAM that focuses on the design and implementation of algorithms for automated identification of manufacturing features. The development of feature recognition methods has been active for more than two decades for academic research. However, in this domain, there are still many drawbacks that hinder its practical applications, such as lack of robustness, inability to learn, limited domain of features, and computational complexity. The most critical one is the difficulty of recognizing interacting features, which arises from the fact that feature interactions change the boundaries that are indispensable for characterizing a feature. This research presents a feature recognition method based on 2D convolutional neural networks (CNNs). First, a novel feature representation scheme based on heat kernel signature is developed. Heat Kernel Signature (HKS) is a concise and efficient pointwise shape descriptor. It can present both the topology and geometry characteristics of a 3D model. Besides informative and unambiguity, it also has advantages like robustness of topology and geometry variations, translation, rotation and scale invariance. To be inputted into CNNs, CAD models are discretized by tessellation. Then, its heat persistence map is transformed into 2D histograms by the percentage similarity clustering and node embedding techniques. A large dataset of CAD models is built by randomly sampling for training the CNN models and validating the idea. The dataset includes ten different types of isolated v features and fifteen pairs of interacting features. The results of recognizing isolated features have shown that our method has better performance than any existing ANN based approaches. Our feature recognition framework offers the advantages of learning and generalization. It is independent of feature selection and could be extended to various features without any need to redesign the algorithm. The results of recognizing interacting features indicate that the HKS feature representation scheme is effective in handling the boundary loss caused by feature interactions. The state-of-the-art performance of interacting features recognition has been improved.
Shi, Y.(2018). Manufacturing Feature Recognition With 2D Convolutional Neural Networks. (Master's thesis). Retrieved from https://scholarcommons.sc.edu/etd/5100