Automated disassembly-oriented knowledge graph construction for retired battery packs using a candidate entity-based relational triple joint extraction method

Currently, the disassembly of retired electric vehicle battery packs relies on manpower and results in high cost, low efficiency, and poor stability. With the development of artificial intelligence, automated disassembly is an efficient method to largely reduce even completely replace human disassembly. However, the various kinds of battery packs and the uncertainty on their retired numbers and types lead to frequent changes of their disassembly processes. It is necessary to provide a method that can integrate valuable disassembly knowledge to enable automated disassembly. Thus, this study proposes an automated disassembly-oriented knowledge graph for retired battery packs which considers the properties of subassemblies (entities) and explicit physical connections/implicit associations among subassemblies (relations). A large amount of unstructured data exists regarding battery packs, such as product manuals and maintenance records, whereas the knowledge that can be available to guide the disassembly process is dispersed and sparse. To solve this, a candidate entity-based relational triple joint extraction method is developed to efficiently extract the disassembly knowledge, which consists of semantic feature learning, candidate entity recognition, and explicit/implicit relational triple identification. Finally, more than 10,000 sentences collected from multi-source unstructured texts are adopted to verify the proposed method. The experimental results demonstrate that our proposed method achieves an F1-score of 93.99% in candidate entity recognition and an F1-score of 95.6% in triple extraction. Also, the information of disassembly operations, disassembly tools, and subassembly properties can be recommended by the automated disassembly-oriented knowledge graph for retired battery packs.