GCM-FL: A Novel Granular Computing Model in Federated Learning for Fault Diagnosis

In the realm of industrial production, maintaining continuous monitoring and implementing precise diagnostics of mine ventilators (MVs) holds a critical role in minimizing faults and accidents. Hence, it becomes imperative to devise an efficient and precise fault diagnosis (FD) technique for MVs. This paper endeavors to address the FD challenge of MVs by integrating federated learning (FL) with granular computing models. FL offers a partial solution to the issues of security and privacy associated with data sharing, which ensures data security while establishing an FD system for MVs. This system harnesses the adjustable multi-granularity (MG) triangular fuzzy (TF) probabilistic rough set (PRS) model to enhance the model’s interpretability. In this study, the TF concept is introduced into the structure of three-way decisions (TWD) to tackle uncertainty and multi-performance attributes. We introduce the notion of an MG TF information system (IS) and propose an adjustable MG TF PRS model. The ELECTRE (Elimination Et Choice Translating Reality) method is employed to determine the optimal threshold. Furthermore, to validate the efficiency of the proposed model, we establish a TF multi-attribute group decision-making (MAGDM) approach using MV data within the MG and TWD frameworks. Finally, we verify the method’s applicability through comparative analysis experiments. The experimental outcomes demonstrate the method’s effectiveness and practicality in diagnosing faults for MVs.