Intelligent System for Selecting Technological Routes Based on a Graph Neural Network and Machine-Readable Product Model

The aim of the study is to develop an intelligent system for automating the selection of technological routes based on a graph neural network and an interface for encoding part features. The objective of the article is to present a method for formalizing the product structure in a machine-readable format and to develop a GCN model capable of classifying parts by structure and selecting a technological template. The research methods include manual input of features through an interface, constructing a directed graph, encoding parameters in one-hot format, training the neural net-work on a labelled dataset, and implementing a self-learning mode. The novelty lies in combining an interface for forming the part structure without using 3D models; a graph model with intersection nodes as separate entities; a GCN model for selecting routes; and a mechanism for incorporating user decisions into the learning process. The results show that the model is trained on 100 graphs and achieves an accuracy of 78.6%. The dynamics of loss and accuracy demonstrates stable convergence. Findings state that the system allows representing parts as a graph, auto-matically selecting processing templates, and accumulating technologist experience. The solution is oriented towards flexible production tasks and can be used as part of digital CAPP platforms.