关键词: 废旧机械零件, 广义生长, 方案生成, 异质图神经网络

Abstract: Aiming at the problems that the generation of generalized growth schemes of retired mechanical parts depends too much on the information of one single type of part due to the large difference of individual information,heterogeneous information and high-dimensional characteristics,which makes the reliability and efficiency of historical data reuse low and the accuracy of the scheme weak,a generalized growth scheme generation method of retired mechanical parts based on Heterogeneous Graph Neural Network (HetGNN) was proposed.The generalized growth scheme graph model of retired mechanical parts was established.The HetGNN heterogeneous graph neural network architecture consisting of One-hot/Word2Vec coding layer,Bi-directional Long Short Term Memory network (Bi-LSTM) aggregation layer and attention neural network combination layer was used to realize the vectorization representation of the graph model.Restart random walk sampling was used to extract the distribution of feature nodes of each type of parts,and One-hot/Word2Vec was used to encode node types and attribute contents,Bi-LSTM was used to aggregate the same type features of parts,and attention mechanism was introduced to combine different types of features of parts.The homology degree was designed to quantify the similarity between parts,and the generalized growth scheme of waste machinery parts was generated.The effectiveness of the proposed method was verified by taking the spur gear of a waste light motorcycle transmission as an example.

Key words: retired mechanical parts, generalized growth, scheme generation, heterogeneous graph neural network