Work

Bearing Fault Diagnosis Method Based on Knowledge Graph and Multimodal Data Fusion

Year: 2026
Type: article
Abstract: Effective bearing fault diagnosis methods are crucial for the diagnosis and maintenance of modern industrial equipment. Relying solely on equipment failure mechanisms or data-driven approaches makes it difficult to address complex equipment failures and the associated knowledge and data. Furthermore, for complex industrial systems, single-modal signals are far from sufficient to describe the system’s complex health status; a single signal often omits critical failure information, resulting in insufficient diagnostic robustness. To address these issues, we propose a bearing fault diagnosis method based on a knowledge graph and multimodal data fusion. First, entity extraction is performed using knowledge rules derived from bearing data. Next, relationship extraction is achieved through the dual-branch multimodal data fusion network (DMDFN) proposed in this paper to construct a bearing fault diagnosis knowledge graph. Finally, after fault classification by the model, decision-making is aided by the knowledge graph to achieve a comprehensive bearing fault diagnosis. Ablation and comparison experiments were conducted on a public bearing dataset. The overall diagnostic accuracy reaches 98.39%, and the recognition rate for the compound inner-outer ring fault achieves 100%. Compared to single-modal signals, the multimodal data fusion method improves diagnostic accuracy and demonstrates superior performance compared to existing methods.
Source: Processes
Institution: Xinjiang University
Language: English
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7234486863
FWCI:
Cites: 36
Cited by:
Related to: 10