Aiming at the challenges of multi-level and multi-granularity knowledge integration and conflict resolution in the conceptual design of complex customized products (CCPs), this paper proposes an innovative method based on a function-behavior-structure (FBS) multi-layer knowledge graph. By constructing a hierarchical knowledge graph (KG) model, combined with the BERT-BiLSTM-CRF model, knowledge extraction and entity alignment are realized. TuckER tensor decomposition is used to complete cross-domain knowledge mapping from function to behavior and behavior to structure, and TRIZ theory is used to solve design conflicts. Taking the high-speed train bogie as an example, the method's effectiveness in knowledge integration, scheme generation, and conflict optimization is verified. Experiments show that the F1 value of BERT-BiLSTM-CRF model entity recognition is 89.07%, significantly better than the traditional model. Combined with TRIZ invention principles, innovative solutions can be quickly generated. This method provides a systematic knowledge management and conflict resolution framework for complex product conceptual design and supports efficient innovation under dynamic requirements.