Reinforcement learning for adaptive battery management of structural health monitoring IoT sensor network

Battery-powered wireless sensor networks (WSNs) provide an affordable and easily deployable option for Structural Health Monitoring (SHM). However, their long-term viability becomes challenging due to uneven battery wear across the sensor network, logistical planning difficulties for battery replacement, and maintaining the desired Quality of Service (QoS) for SHM. A system-level battery health management strategy is vital to extend the lifespan and reliability of WSNs, especially considering the expensive maintenance trips required for battery replacement. This study presents a reinforcement learning (RL) based framework to actively manage battery degradation at the system level while preserving SHM QoS. The framework focuses on group battery replacement, reducing logistical burdens, and enhancing WSN longevity without compromising desired QoS. To validate the RL framework, a detailed simulation environment was created for a real-world WSN setup on a cable-stayed bridge SHM. The simulation accounted for various environmental and operational factors such as weather-induced solar harvesting variability, communication uncertainties, lithium-ion battery degradation models, sensor power consumption, and duty cycle strategies etc. Additionally, a mode shape-based quality index was introduced for a SHM network. The RL agent was trained within this environment to learn optimal node selection for specific duty cycles. The results demonstrate the framework's effectiveness in optimizing battery replacement efforts by ensuring a similar end of lifetimes with more uniform battery degradation and allowing the longer and more reliable operation of WSNs under uncertainties.