IThing: Designing Next-Generation Things With Battery Health Self-Monitoring Capabilities for Sustainable IIoT

Abstract

An accurate and reliable technique to predict rechargeable battery health proves helpful in battery-operated, low-resourced industrial IoT devices. The existing data-driven battery health prediction techniques often require a comparatively large amount of computational power for predicting the State of Health (SOH) and the Remaining Useful Life (RUL) due to most methods being feature-heavy. Further, there are very limited works for battery RUL prediction in IoT nodes. To address this issue, this paper presents a unique IoT-based sensor node framework, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">iThing</i> , to predict the on-board battery SOH and RUL with the least computational and memory load. The <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">iThing</i> automatically extracts the voltage and time-based health indicators, which is then fed to the random learning algorithm-based methods with good learning performance for SOH and RUL prediction. The proposed Extreme Learning Machine (ELM) network provides SOH prediction with 0.0054 Root Mean Square Error (RMSE), and 0.0024 Mean Absolute Error (MAE). Random Vector Functional Link (RVFL) neural network predicted the RUL with 0.0282 RMSE and 0.021 MAE. The proposed method has been tested on three different battery datasets with varying charging policies with high accuracy. The models have been deployed successfully on an experimental hardware setup, proving its eligibility for real-time IIoT applications.