A method for detecting process design intent in the process route based on heterogeneous graph convolutional networks

The process design intent is the concentration of the technologists’ design cognitive process which contains the experiential knowledge and skills. It can reproduce technologists’ design thinking process in process design and provides guidance and interpretability for the generation of process results. The machining process route, as a core component of a part's entire manufacturing process, contains substantial process design intent. If the process design intent embedded in the existing process route can be explicitly identified, subsequent technologists will be able to learn and understand the original designers’ thinking, methodologies, and intents. This understanding enables effective reuse of design thinking and logic in the process design of new parts, rather than merely reusing data. It can also promote the propagation of the expertise and skills inherent in the process design intent. However, existing research on process design intent lacks a detailed explanation of its formation and specific structure from the design cognition perspective, making it challenging to effectively predict the process design intent containing interpretable empirical knowledge in the process route. To address this issue, this paper provides a method for predicting process design intent in the process route using heterogeneous graph convolutional networks. First, the heterogeneous graph is used to represent the parts and their associated process routes in the dataset. The nodes in the graph are then labeled based on accumulated and summarized process design intent. The prediction of process design intent in the process route is then converted into a node classification issue with heterogeneous graphs. A node classification network model is constructed using a heterogeneous graph convolutional network where the input is the created heterogeneous graph, and the output is the design reason contained in the machining feature and the intent cognition embedded in the working step, both of which are part of the process design intent. After training, the proposed model accurately predicted design reasons for machining features and intent cognitions for working steps (95.13 % and 96.85 %, respectively). Finally, examples of actual process routes are analyzed to verify the method's feasibility and reliability. The method given in this article can help technologists gain a deeper understanding of process route generation, hence improving their process design capabilities.