Abstract: To address the strong reliance on annotated data and the limited ability to parse multi-hop relationships inherent in traditional extraction methods—issues arising from the multi-step nature and semantic ambiguity of remanufacturing knowledge—a multi-granularity knowledge extraction method for remanufacturing processes based on chained prompts in large language model (LLM) is proposed. By integrating prompt engineering and chain-of-thought reasoning with a semantic alignment mechanism, the method enables accurate extraction from coarse-grained to fine-grained knowledge. Initially, coarse-grained knowledge is extracted by guiding the LLM through prompt engineering to identify the core entities involved in remanufacturing processes. Subsequently, a chain-of-thought reasoning framework is designed to enable the LLM to interpret complex logical relationships among entities. Semantic alignment across heterogeneous expressions is achieved using cosine similarity, enhancing the consistency and accuracy of fine-grained knowledge matching. Experimental results demonstrate that the proposed chained prompt method achieves an F1-score of 88.0%, representing an improvement of over 30% compared to traditional approaches. Moreover, the coverage of multi-hop relationships reaches 89.2%, effectively mitigating the dependency on annotated data observed in conventional techniques.