Machine learning in feature recognition for manufacturing: taxonomy, analytical review, comparisons, trends, challenges, and outlook

Feature recognition is an important topic in the area of intelligent manufacturing and production research. This technique is utilised to extract and recognise functional components and geometric shapes, such as slots, holes, and pockets, from solid models. The term ‘learning-based feature recognition’ refers to the feature recognition methods implemented based on machine learning. This topic was examined three decades ago but has been growing quickly in the recent five years. In this paper, a taxonomy and review of learning-based feature recognition methods are presented. This paper categorises the literature, examines basic components of a learning-based feature recognition method, identifies the technical evolution of these components, examines their impact on the overall feature recognition process, provides a broader overview of existing methods, identifies the milestone learning paradigms, makes comparisons among different approaches, assesses their practical capabilities, and identifies the gaps between research and real-world applications. By reading this paper, researchers and industry professionals can understand different learning-based feature recognition methods and their underlying principles, gain insights into design decisions made in a learning-based system, and understand their applied situations.