A novel approach to part process route generation based on graph neural network encoding

Process planning is an important bridge between design and production. However, existing feature-based process planning methods usually treat process planning as a classification problem. This approach leads to inflexible process planning results and requires a lot of manual adjustments. To address the problem, this paper proposes a novel process planning method. The method combines a graph encoder and a sequence decoder based on an attention mechanism. First, this paper extracts the overall information of the part through the part feature attribute vectors and the baseline adjacency matrix. Then, the graph embedding is constructed by considering each manufacturing feature as a node and transforming the feature attribute vectors into node embeddings through the baseline adjacency matrix. Subsequently, the sequence decoder uses the graph embeddings and node embeddings as inputs to decode the generated sequence of process steps progressively. The attention mechanism is utilized to focus on different manufacturing features of the part while generating each step of the process. The whole model is optimized by joint training to maximize the conditional probability of the target process path. Finally, the superiority of this method in terms of accuracy and effectiveness is verified through comparison experiments with other state-of-the-art methods.