A federated learning approach to mixed fault diagnosis in rotating machinery

Rotating machinery is ubiquitous in modern industrial systems. Ensuring optimal operating conditions for rotating machinery is essential to satisfy stringent requirements on safety, efficiency, and reliability. State-of-the-art performance for fault detection, identification, and isolation for rotating machinery has been achieved using deep learning-based methods which generally require large quantities of high-quality supervised learning data. Collecting, labeling, and maintaining such data is resource-intensive and sometimes cost-prohibitive. Distributed data across multiple factories or organizations can be pooled together to learn a powerful collective model; however, this is not always possible due to the sensitive and proprietary nature of the data. To simultaneously alleviate these constraints of data availability and privacy, this paper develops a federated learning (FL) framework for the diagnosis of mixed faults from multiple factories. A duplet classifier is constructed to separate the mixed fault classification task into parallel networks where each network is responsible for one component. This classifier is trained under the FL framework and its performance is thoroughly examined for different data distributions across 30 participating factories. Experimental results show that the proposed methodology yields excellent mixed fault classification accuracy for all participating factories even under highly unbalanced and heterogeneous distribution of fault labels. Further studies highlight the data efficiency of the proposed method and its robustness to previously unseen fault types.