AAGNet: A graph neural network towards multi-task machining feature recognition - ScienceDirect

Fig. 2. The structure of Geometric Attributed Adjacency Graph (gAAG). The ‘Face 14’ denotes the index of this face is 14 in the B-Rep model and the ‘Edge 25’ denotes the index of this edge is 25. Table 1. Attributes of faces and edges. Table 2. Main features of different representations. Fig. 3. The semantic, instance, and bottom face label of MFInstSeg. Table 3. Some datasets related to machining feature recognition. Fig. 4. The architecture of AAGNet. The 𝑁 denotes the number of nodes, the 𝐸 denotes the number of edges, 𝐷 denotes the dimension of the input vector, the GAP denotes the Global Average Pooling layer, the ○ 𝑐 ○ denotes concatenate operator, the Norm denotes normalization, the Act denotes nonlinear activation function, and the ○ + ○ denotes add operator. Table 4. The details of surface encoder. 𝑘 = 3 denotes that convolution kernel size is set to 3. The data format of input and output is 𝑐 ℎ 𝑎 𝑛 𝑛 𝑒 𝑙 × ℎ 𝑒 𝑖 𝑔 ℎ 𝑡 × 𝑤 𝑖 𝑑 𝑡 ℎ . Table 5. The

Fig. 2. The structure of Geometric Attributed Adjacency Graph (gAAG). The ‘Face 14’ denotes the index of this face is 14 in the B-Rep model and the ‘Edge 25’ denotes the index of this edge is 25. Table 1. Attributes of faces and edges. Table 2. Main features of different representations. Fig. 3. The semantic, instance, and bottom face label of MFInstSeg. Table 3. Some datasets related to machining feature recognition. Fig. 4. The architecture of AAGNet. The 𝑁 denotes the number of nodes, the 𝐸 denotes the number of edges, 𝐷 denotes the dimension of the input vector, the GAP denotes th

saved by

• Jaffer R