부산대학교 도서관

The automation of the entire value chain of a product requires not only automated production and assembly, but also automated production and assembly planning. Today, assembly planning is mostly a manual task and is based on computer-aided design descriptions and technical drawings. For the automation of the assembly planning, an automated recognition of the assembly features of the goal product is essential. In this paper, a concept for an automatic recognition of form features as well as the assignment of the connections between them as joints to build the assembly feature is described. The feature recognition is based on point clouds and uses the PointNet architecture. A data set of point clouds for training and an approach for automatic recognition of shape and assembly features using a convolutional neural network is presented. This approach can be used to recognize 12 features and assign seven joints. The recognition rate for the form features is over 95%. Finally, the features and their connections are stored in a graph database.