An improved memory networks based product model classification method

Abstract

ABSTRACT Existing CAD model classification methods are usually performed by extracting the geometric and topological information of the models. This is not suitable for the classification of Model-based Definition (MBD) models because of the lack of non-geometric semantic information. They cannot solve the problem, as the geometry and topology are similar but the product manufacturing information (PMI) is completely different. Firstly, this paper proposed a multi-granularity classification model based on MBD attribute adjacency graph (MAAG). And then an improved Long Short-Term Memory (LSTM) neural network with a new loss function and adaptive training times is put forward. This neural network model is trained to achieve fast and accurate classification of the MBD models. Finally, the experimental results show that the proposed method not only improves the accuracy of classification, but also effectively reduces the training time cost of the model.