Context-Awareness on Federated Transformer-Based Prognostics for Equipment Life Assessment

Abstract

Federated Transformer enables privacy-preserving Remaining Useful Life (RUL) estimation across distributed data, helping optimize maintenance schedules for machinery and equipment. Accurate estimation of the RUL helps optimize maintenance schedules for machinery and equipment. However, centralized training methods for AI models raise concerns about privacy when dealing with sensitive data. Context-awareness enables the model to account for varying operational conditions, improving RUL prediction performance. This study proposes the application of a Transformer architecture within the federated learning framework, while also incorporating contextual information as input to enhance the model’s ability to capture degradation patterns in time series data. Experimental results show that the model trained in a distributed manner achieves performance equivalent. Moreover, our study demonstrates higher prediction accuracy compared to previous related works. The proposed approach not only preserves data privacy but also leverages the power of the modern Transformer architecture, highlighting its high potential for application in next-generation predictive maintenance systems.