LiRUL: A Lightweight LSTM Based Model for Remaining Useful Life Estimation at the Edge
Olumide Kayode, Ali Şaman Tosun · 2019
Unexpected downtime and sudden breakdown of IoT devices can be extremely destructive most especially for safety-critical systems. Condition monitoring and health state estimation are vital techniques for maintaining high reliability and availability. Many data-driven approaches to Remaining Useful Life (RUL) estimation using deep learning algorithms are computationally-intensive and often not feasible on resource-constrained devices. Time lag and network cost associated with massive data transfer to a centralized cloud for processing can be minimized by adopting edge computing. In this paper, we propose a lightweight Long Short-Term Memory (LSTM) based model called LiRUL, suitable for RUL estimation on an edge device. We also implement a tagging function and evaluate our approach on publicly available Turbofan Engine datasets.